This book covers concepts and the latest developments on microscale flow and heat transfer phenomena involving a gas. Th

*338*
*61*
*9MB*

*English*
*Pages XIX, 365
[375]*
*Year 2020*

- Author / Uploaded
- Amit Agrawal
- Hari Mohan Kushwaha
- Ravi Sudam Jadhav

*Table of contents : Front Matter ....Pages i-xix Introduction to Microscale Flows and Mathematical Modelling (Amit Agrawal, Hari Mohan Kushwaha, Ravi Sudam Jadhav)....Pages 1-23 Microscale Flows (Amit Agrawal, Hari Mohan Kushwaha, Ravi Sudam Jadhav)....Pages 25-80 Microscale Heat Transfer (Amit Agrawal, Hari Mohan Kushwaha, Ravi Sudam Jadhav)....Pages 81-113 Need for Looking Beyond the Navier–Stokes Equations (Amit Agrawal, Hari Mohan Kushwaha, Ravi Sudam Jadhav)....Pages 115-123 Burnett Equations: Derivation and Analysis (Amit Agrawal, Hari Mohan Kushwaha, Ravi Sudam Jadhav)....Pages 125-188 Grad Equations: Derivation and Analysis (Amit Agrawal, Hari Mohan Kushwaha, Ravi Sudam Jadhav)....Pages 189-258 Alternate Forms of Burnett and Grad Equations (Amit Agrawal, Hari Mohan Kushwaha, Ravi Sudam Jadhav)....Pages 259-304 Overview to Numerical and Experimental Techniques (Amit Agrawal, Hari Mohan Kushwaha, Ravi Sudam Jadhav)....Pages 305-312 Summary and Future Research Directions (Amit Agrawal, Hari Mohan Kushwaha, Ravi Sudam Jadhav)....Pages 313-315Back Matter ....Pages 317-365*

Mechanical Engineering Series

Amit Agrawal Hari Mohan Kushwaha Ravi Sudam Jadhav

Microscale Flow and Heat Transfer Mathematical Modelling and Flow Physics

Mechanical Engineering Series Series Editor Francis A. Kulacki Department of Mechanical Engineering University of Minnesota Minneapolis, Minnesota, USA

The Mechanical Engineering Series presents advanced level treatment of topics on the cutting edge of mechanical engineering. Designed for use by students, researchers and practicing engineers, the series presents modern developments in mechanical engineering and its innovative applications in applied mechanics, bioengineering, dynamic systems and control, energy, energy conversion and energy systems, fluid mechanics and fluid machinery, heat and mass transfer, manufacturing science and technology, mechanical design, mechanics of materials, micro- and nano-science technology, thermal physics, tribology, and vibration and acoustics. The series features graduate-level texts, professional books, and research monographs in key engineering science concentrations.

More information about this series at http://www.springer.com/series/1161

Amit Agrawal • Hari Mohan Kushwaha Ravi Sudam Jadhav

Microscale Flow and Heat Transfer Mathematical Modelling and Flow Physics

123

Amit Agrawal Department of Mechanical Engineering Indian Institute of Technology, Bombay Mumbai, Maharashtra, India

Hari Mohan Kushwaha Department of Mechanical Engineering Indian Institute of Technology, Bombay Mumbai, Maharashtra, India

Ravi Sudam Jadhav Department of Mechanical Engineering Indian Institute of Technology, Bombay Mumbai, Maharashtra, India

ISSN 0941-5122 ISSN 2192-063X (electronic) Mechanical Engineering Series ISBN 978-3-030-10661-4 ISBN 978-3-030-10662-1 (eBook) https://doi.org/10.1007/978-3-030-10662-1 Library of Congress Control Number: 2019932162 © Springer Nature Switzerland AG 2020 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, express or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG. The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

This book is dedicated to past and current students of “Rarefied Gas Dynamics” and “Microfluidics” laboratories, IIT Bombay, and to our families.

Preface

This book has been written with the objective of making the readers familiar with the exciting developments on gaseous slip flow and heat transfer in microchannel. A large amount of work is being currently undertaken worldwide in these areas, with numerous potential applications. The subject is therefore topical and also particularly significant as it leads to the question about the validity of the NavierStokes equations, which is usually considered sacrosanct. It obviously leads to a subsequent question: if the Navier-Stokes equations are themselves not valid, then how do we model the flow and heat transfer? Although the answer to this last question is still not clear, it is important for the fluids and thermal community to first appreciate the limitations of the continuum approach (which leads to the Navier-Stokes equations), to better appreciate the ongoing search for the “beyond Navier-Stokes equations,” and to test some of the available equations for their accuracy, before the larger objective of finding beyond the Navier-Stokes equations can be practically met. The book is therefore organized in two parts: the first part is on the gaseous slip flow and heat transfer in microchannel (Chaps. 2 and 3). In the second part, we examine beyond the Navier-Stokes equations (Chaps. 5–7). Chapter 1 summarizes the characteristics of microscale flows and provides an introduction to the various modelling approaches available. Chapter 4 is a transition chapter between the two parts, where it is shown that simple extensions of the Navier-Stokes equations are not adequate. Recognizing that only analytical tool will not be adequate for studying flows in the slip and transition regimes, a brief overview to relevant numerical and experimental techniques is provided in Chap. 8. Finally, Chap. 9 summarizes our current understanding and provides suggestions for future research in this subject. The Knudsen number is the most important parameter, and several known solutions with Knudsen number as an additional parameter are compiled in the first part of the book. Interesting observations on Knudsen minima and flow separation are presented. However, it should become apparent that our understanding of heat transfer at the microscale is not that sound, as several additional effects such as axial conduction, pressure work, conduction in the substrate, viscous dissipation, etc. coexist, but it is virtually not possible to treat all these effects together and vii

viii

Preface

obtain an analytical solution for even the simplest problems. Only few experimental and direct simulation Monte Carlo (DSMC) data exist, and they do not agree well with that obtained from alternative approaches. A good portion of the second part of the book is devoted to deriving and understanding the Burnett and Grad equations. These two sets of equations form the most important “beyond the Navier-Stokes” equations and are generally referred to as “higher-order continuum transport equations.” The study of these equations is important for further development of the subject, as they involve several novel concepts and represent important breakthroughs in the subject. The derivation of these equations has not often been repeated, and it is expected that the stepwise derivation presented here will invoke wide readership. Special effort has been made to make the text readable through the insertion of a large number of figures. The hope is that with the help of this book, it should now be possible to derive the Burnett and Grad equations in a graduate class. A few problems are solved to illustrate the type of solution obtained from these equations. In the derivation of some solutions presented here, some minor error in the original source was noticed which has been corrected here. The first part of the book (Chaps. 1–3 along with Chaps. 8 and 9) should appeal to readers interested in understanding fundamental aspects of microscale flows and heat transfer, while the second part (Chaps. 4–7) is primarily for slightly advanced readers interested in understanding equations beyond the Navier-Stokes equations. Mumbai, India Mumbai, India Mumbai, India September 2018

Amit Agrawal Hari Mohan Kushwaha Ravi Sudam Jadhav

Acknowledgments

We are grateful to all the students who have worked in the “Rarefied Gas Dynamics” and “Microfluidics” laboratories at IIT Bombay, including those who were involved in purely theoretical or numerical aspects of the problem. It is they who slowly and patiently uncovered the various fine points of the subject, which finally culminated in the form of this book. We are also grateful to Prof. K. Muralidhar (IIT Kanpur) and Prof. F. Kulacki (University of Minnesota), the editors of this series, for first placing their trust in us for writing this book and then patiently waiting for it while we went past several deadlines. The book would not have resulted without the coaxing of Prof. Muralidhar to write on this subject. While preparing the book, we gained from the comments of Prof. Atul Sharma (IIT Bombay). Funding for this line of investigation has been provided by the Indian Institute of Technology Bombay (IITB), Indian Space Research Organisation (ISRO), Department of Science and Technology (DST), and recently by the Department of Atomic Energy (DAE) under its prestigious scheme—DAE-SRC Outstanding Investigator Award, awarded to the first author.

ix

Contents

1

2

Introduction to Microscale Flows and Mathematical Modelling . . . . . . . 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Applications of Microscale Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1 Cooling of Electronic Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.2 Micro-nozzles and Micro-thruster . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.3 Breath Analyser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.4 Microdevice for Conducting Blood Test . . . . . . . . . . . . . . . . . . . . . 1.3 Classification of Flow Regimes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4 Characteristics of Microscale Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.1 Rarefaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.2 Compressibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.3 Thermal Creep . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.4 Viscous Dissipation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.5 Property Variation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.6 Axial Conduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.7 Conjugate Heat Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5 Mathematical Modelling of Microscale Flows . . . . . . . . . . . . . . . . . . . . . . . 1.5.1 The Navier–Stokes Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5.2 Limitations of Conventional Equations and Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5.3 Approach to Modelling Microscale Flows. . . . . . . . . . . . . . . . . . . 1.6 Relevance and Scope of the Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1 2 3 4 4 6 6 8 9 9 10 11 12 13 14 14 15 16

Microscale Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Governing Equations for Fluid Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Tensorial Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.2 Constitutive Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.3 Compressible Navier–Stokes Equations . . . . . . . . . . . . . . . . . . . . . 2.2.4 Incompressible Navier–Stokes Equations . . . . . . . . . . . . . . . . . . .

25 25 26 26 28 30 31

16 18 21

xi

xii

Contents

2.3

Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 Maxwell’s Slip Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.2 Derivation of Higher-Order Slip Boundary Condition . . . . . . 2.3.3 Alternate Slip Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.4 Value of Slip Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Some Exact Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.1 Couette Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.2 Flow in a Microchannel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.3 Flow in a Microtube . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.4 Flow in an Arbitrary Cross Section Microchannel . . . . . . . . . . 2.4.5 Flow in the Annulus of Rotating Sphere and Cylinder . . . . . . Observations on Flow in Straight Passages. . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5.1 Appearance of Knudsen Minima . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5.2 Flow in Rough Microchannel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5.3 Transient Flow in a Capillary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Observations on Flow in Complex Passages . . . . . . . . . . . . . . . . . . . . . . . . . 2.6.1 Flow in Sudden Expansion/Contraction Microchannel . . . . . 2.6.2 Flow in Diverging/Converging Microchannel . . . . . . . . . . . . . . . 2.6.3 Flow in a Bend Microchannel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Useful Empirical Correlations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.7.1 Developing Length in Microtube and Microchannel . . . . . . . . 2.7.2 Friction Factor for Microchannel of Various Cross Sections. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

34 34 37 38 40 44 44 48 60 61 64 67 67 68 69 70 70 72 74 76 76

Microscale Heat Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Governing Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1 Derivation of First Order Temperature Jump Condition . . . . 3.4 Some Exact Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1 Heat Transfer in Microchannel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.2 Heat Transfer Analysis Through a Micropipe . . . . . . . . . . . . . . . 3.4.3 Heat Transfer Through a Micro-Annulus . . . . . . . . . . . . . . . . . . . . 3.5 Observations on Other Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.1 Variation in Thermophysical Properties . . . . . . . . . . . . . . . . . . . . . 3.5.2 Conduction in the Substrate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.3 Axial Conduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5.4 Flow Work and Shear Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6 Observations from Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.7 Application to Knudsen Pump . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.8 Useful Empirical Correlations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

81 81 82 83 84 85 86 92 97 106 106 107 108 109 110 110 112 113

2.4

2.5

2.6

2.7

2.8 3

78 79

Contents

4

xiii

Need for Looking Beyond the Navier–Stokes Equations . . . . . . . . . . . . . . . . 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Examples Needing a Look Beyond the Navier–Stokes Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Extended Navier–Stokes Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Modified Navier–Stokes Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5 Non-Fourier Heat Conduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.6 Shock Wave Flow Problem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

115 115

5

Burnett Equations: Derivation and Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Boltzmann Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.1 Macroscopic Quantities from Distribution Function . . . . . . . . 5.2.2 Physical Significance of Higher Order Moments. . . . . . . . . . . . 5.2.3 Evolution Equations for Moments . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.4 Conservation Laws. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.5 Constitutive Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 The Chapman–Enskog Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.1 Dimensionless Boltzmann Equation . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.2 Central Idea of the Chapman–Enskog Method . . . . . . . . . . . . . . 5.3.3 Derivation of Distribution Function . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4 Derivation of Hydrodynamic Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.1 Euler Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.2 Navier–Stokes Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.3 Burnett Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.4 Super-Burnett Equations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.5 Burnett Equations in Cylindrical Coordinates. . . . . . . . . . . . . . . . . . . . . . . . 5.5.1 Navier–Stokes Stress and Heat Flux . . . . . . . . . . . . . . . . . . . . . . . . . 5.5.2 Burnett Stress and Heat Flux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.6 Order of Magnitude Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.7 Some Exact Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.7.1 Iterative Approach of Agrawal and Singh . . . . . . . . . . . . . . . . . . . 5.7.2 Perturbation Based Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.8 Stability Analysis of Burnett Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

125 125 126 128 131 134 136 139 141 141 143 144 151 151 154 161 164 166 167 167 175 177 177 181 184 187

6

Grad Equations: Derivation and Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Definition of Moments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 20-Moment Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.1 Evolution Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.2 Expansion in Hermite Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.3 Derivation of Distribution Function . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.4 Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.5 20-Moment Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

189 189 190 192 194 200 203 204 205

116 117 119 120 122 123

xiv

Contents

6.4

6.5 6.6 6.7

6.8

6.9 6.10 6.11 6.12 7

The 13-Moment Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4.1 Evolution Equations for Stress Tensor and Heat Flux Vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4.2 Derivation of Distribution Function . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4.3 Closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4.4 13-Moment Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Boundary Conditions for 13-Moment Equations. . . . . . . . . . . . . . . . . . . . . Significance of 13-Moment Equations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Exact Solution of Stokes’ First Problem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.7.1 Navier–Stokes Based Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.7.2 Navier–Stokes Solution with Slip Boundary Condition. . . . . 6.7.3 13-Moment Equation Based Solution . . . . . . . . . . . . . . . . . . . . . . . . 6.7.4 Kinetic Theory Based Solution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.7.5 Comparison with Free-Molecular Regime Solution. . . . . . . . . 6.7.6 Significance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Exact Solution of Cylindrical Couette Flow. . . . . . . . . . . . . . . . . . . . . . . . . . 6.8.1 Governing Equations and Boundary Conditions . . . . . . . . . . . . 6.8.2 Linearization of 13-Moment Equations . . . . . . . . . . . . . . . . . . . . . . 6.8.3 Solution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.8.4 Discussion on the Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Derivation of Navier–Stokes Equations from 13-Moment Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Derivation of Cattaneo’s Equation from 13-Moment Equations . . . . Comparison of Chapman–Enskog and Grad Methods . . . . . . . . . . . . . . . Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Alternate Forms of Burnett and Grad Equations . . . . . . . . . . . . . . . . . . . . . . . . 7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Limitations of Burnett Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3 Variants of the Burnett Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.1 Augmented Burnett Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.2 BGK-Burnett Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.3 Other Forms of Burnett Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4 Limitations of Grad Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.5 Variants of Grad Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.5.1 Regularized 13-Moment Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 7.5.2 Regularized 26-Moment Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 7.6 Onsager Consistent Approach. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.6.1 Derivation of Distribution Function . . . . . . . . . . . . . . . . . . . . . . . . . . 7.6.2 Onsager–Burnett Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.6.3 Onsager-13 Moment Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.6.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

207 209 214 215 216 218 225 226 227 228 228 238 240 240 241 242 244 246 251 254 254 255 258 259 259 260 261 262 263 269 270 271 272 278 284 284 287 292 295

Contents

7.7

xv

Comparison of Various Forms of Equations. . . . . . . . . . . . . . . . . . . . . . . . . . 7.7.1 Couette Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.7.2 Force-Driven Poiseuille Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

295 295 296 303

8

Overview to Numerical and Experimental Techniques . . . . . . . . . . . . . . . . . . 8.1 Numerical Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.1.1 Direct Simulation Monte Carlo Method . . . . . . . . . . . . . . . . . . . . . 8.1.2 Alternative Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2 Experimental Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2.1 Measurement at the Microscale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2.2 Measurement in Rarefied Gas Flows. . . . . . . . . . . . . . . . . . . . . . . . . 8.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

305 305 305 307 307 307 310 312

9

Summary and Future Research Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313

7.8

A Appendix to Basic Tensor Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A.1 Introduction to Scalar, Vector, and Tensor. . . . . . . . . . . . . . . . . . . . . . . . . . . . A.2 Kronecker Delta . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A.3 Some Important Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A.4 Decomposition of Tensors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A.5 Some Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A.6 General Tensor Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

317 317 320 321 322 327 331

B Appendix to Burnett Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B.1 Derivation of f (0) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B.2 Derivation of f (1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B.2.1 Evaluation of Time Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B.3 Expression for D (1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

335 335 337 337 339

C Appendix to Grad Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343 C.1 Transformation of Boltzmann Equation: f (x, c, t) ⇒ f (x, C, t). . . 343 C.2 Note on Hermite Polynomials. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357

Nomenclature

Alphabetical m s

c

molecular velocity,

cs

speed of sound,

cx , cy , cz

components of molecular velocity c,

cp

specific heat at constant pressure,

cv

specific heat at constant volume,

C

peculiar velocity,

C1 , C2

Slip coefficients

Cx , Cy , Cz

components of peculiar velocity C,

d

molecular diameter, m

Dh

hydraulic diameter, m

e

specific energy,

F

external force,

f

particle distribution function,

fM

Maxwell-Boltzmann distribution,

h

convective heat transfer coefficient,

k

thermal conductivity,

kB

Boltzmann constant,

m s m s

J kg K J kg K

m s

m s

J kg

kg m s2 s3 m6 s3 m6 W m2 K

W mK J K

xvii

xviii

Nomenclature

L

characteristic length scale, m

m

mass of a molecule, kg

m ˙

mass flow rate,

n

number density,

p

hydrostatic pressure,

Pij

pressure tensor,

pij

divergence-free part of pressure tensor Pij ,

q

heat flux vector,

R

specific gas constant (= kB /m),

r, θ, z

cylindrical coordinate system

r, θ, φ

spherical coordinate system

t

time, s

T

absolute temperature, K

Tm

bulk mean temperature, K

Tw

wall temperature, K

u

bulk velocity vector,

u, v, w

components of bulk velocity u,

us

velocity of gas at the wall,

uw

wall velocity,

v¯

most probable speed,

x

position vector, m

x, y, z

Cartesian coordinates in physical space

kg s 1 m3 N m2

N m2

W m2 J kg.K

m s m s

m s

m s m s

Greek symbols α

thermal diffusivity,

γ

specific heat ratio,

λ

mean free path, m

δij

Kronecker delta

m2 s cp cv

N m2

Nomenclature

xix J kg

specific internal energy,

μ

dynamic viscosity,

μ2

second coefficient of viscosity,

μv

coefficient of bulk viscosity,

ν

kinematic viscosity,

m2 s

Φ

viscous dissipation,

kg m s3

ρ

mass density,

σT

thermal accommodation coefficient

σv

tangential momentum accommodation coefficient

σij

viscous stress tensor,

τij

stress tensor (= −Pij ),

N s m2 N s m2

N s m2

kg m3

N m2 N m2

Non-dimensional numbers Br

Brinkman number, Br =

Ec

Eckert number, Ec =

Kn

Knudsen number, Kn =

Ma

Mach number, Ma =

Nu

Nusselt number, Nu =

Pe

Peclet number, P e =

Pr

Prandtl number, P r =

Re

Reynolds number, Re =

μu2 k(Tw −Tm )

u2 cp (Tw −Tm ) λ L

√u γ RT hDh k

ρucp Dh k μcp k ρuDh μ

Abbreviations CFD

computational fluid dynamics

DSMC

direct simulation Monte Carlo

MEMS

micro-electro-mechanical systems

TMAC

tangential momentum accommodation coefficient

Chapter 1

Introduction to Microscale Flows and Mathematical Modelling

The purpose of this book is to understand fluid flow and heat transfer at the microscale. The focus is on gases, as the mean free path of the gas can become comparable to the passage dimensions, and an additional non-dimensional number (Knudsen number) starts affecting the dynamics of the flow and heat transfer behavior. The effect of Knudsen number turns out to be non-trivial, as it leads to not only several new physical phenomena, but also exposes the limitations of the celebrated Navier–Stokes equations in modelling such flows. The second major purpose of this book is therefore to examine equations which can model such high Knudsen number flows. Again, we first show that simple modifications to the Navier–Stokes equations and boundary conditions is not sufficient to fulfill this objective. There is perhaps no alternative but to shun the conventional way of deriving transport equations, and resort to the Boltzmann equation for derivation of “higher-order continuum transport equations.” The book presents a clear derivation of the Burnett and Grad equations starting from the Boltzmann equation, a discussion on the variants of the Burnett and Grad equations, and some known solutions of these higher-order continuum transport equations. Microscale flows therefore help put the higher-order continuum transport equations in better perspective, and provide reasons to study these “beyond the Navier– Stokes” equations. The higher-order continuum transport equations may not just help understand high Knudsen number flows better, rather, being superset of the Navier–Stokes equations, help understand fluid flow and heat transfer phenomena in ways that we cannot even imagine as of now; thereby making the study of these higher-order equations extremely relevant. In this chapter, we provide an introduction to microscale flows and their characteristics, application of microscale flows, and challenges in modelling such flows.

© Springer Nature Switzerland AG 2020 A. Agrawal et al., Microscale Flow and Heat Transfer, Mechanical Engineering Series, https://doi.org/10.1007/978-3-030-10662-1_1

1

2

1 Introduction to Microscale Flows and Mathematical Modelling

1.1 Introduction Microscale flows are in vogue and the last three decades have seen substantial advancement in their understanding. This interest in microscale flows is because of counter-intuitive flow physics that can be encountered at the microscale along with several potential applications of microdevices that can be built using such an understanding. A large number of application of microdevices are in the domain of medical diagnostics, and lie beyond the scope of this book. Here, we rather focus on the physics of flow in microdevices, which is interesting because of slipping of the gas at the passage walls and substantial compressibility effect (or variation in density) as the flow occurs. Such internal compressible flows involving slip at the wall constitute a new class of problems, which have not been well studied in the past. The reduction in length scale (small cross section of the passage across which the flow is occurring) means that a gas molecule experiences fewer inter-molecular collisions before hitting the opposite wall. This issue along with finite momentum at the wall (due to slip) is brought about due to rarefaction of the gas, with the amount of rarefaction being measured by a non-dimensional parameter—Knudsen number. The issue of relatively infrequent inter-molecular collisions leads to the gas not being in thermodynamic equilibrium close to the wall. The region of this thermodynamic non-equilibrium extends further into the bulk of the flow with an increase in the Knudsen number. The classical flow equations—the Navier– Stokes equations—are not applicable in the region of non-equilibrium, which makes modelling of the non-equilibrium region, and therefore flow at the microscale particularly challenging. Further complications arise as the typical no-slip boundary condition at the gas–wall interface is no longer applicable. The equation that needs to be solved in place of the Navier–Stokes equations to describe the flow is not entirely settled, although several equations have been proposed in the literature. These higher-order continuum transport equations arise from the Boltzmann equation and are derived by making suitable assumptions regarding the nature of the particle distribution function. Most noteworthy of these higher-order continuum transport equations are by Burnett and Grad, and bear the name of these scientists. The derivation of the Burnett and Grad equations is intimately linked to concepts in kinetic theory—a subject about which engineers are ill-aware of. The awareness of the Burnett and Grad equations, even among senior practitioners in the field, is therefore limited. This deficiency in understanding is recognized and addressed in this book. We have further invoked minimal concepts from the kinetic theory in our discussion of the Burnett and Grad equations, and have taken the viewpoint that the “output” of the kinetic theory should serve as the “input” based on which higher-order continuum transport equations can be derived. Currently, most rarefied gas flows are solved using the Direct Simulation Monte Carlo (DSMC) method of Bird. So can we not continue to work with the DSMC method, especially since the DSMC method is so well established? That is, why do we need higher-order continuum transport equations at all? DSMC is a numerical

1.2 Applications of Microscale Flows

3

method and all numerical methods yield solution for the specific case solved, as opposed to a general solution possible with the analytical approach. Therefore, unless one solves a large number of cases, one does not get an idea of the trend with respect to variation of a given parameter. Of further interest is to have a scaling of the solution with respect to the various governing parameters. Solving multiple cases requires effort and is computationally expensive in terms of time and resources. On the other hand, understanding the variation or scaling with respect to various governing parameters is straightforward after having obtained an analytical solution. Knowledge of the governing equation that needs to be solved is however a prerequisite for obtaining an analytical solution, pointing towards the need for obtaining accurate transport equations. Analytical solution from governing equations can further be applied to validate numerical results, and even experimental results especially when a new setup is being built. Further calibration of instruments can be undertaken in flows for which analytical solution is available and which are also simple and convenient to set up. In summary, having an alternative independent approach (knowledge of accurate higher-order continuum transport equations in the present case) is useful in many ways, and helps by enriching our understanding of the phenomena as it adds another dimension to the available tools (DSMC based simulations and experiments). The field of computational fluid dynamics (CFD), which solves the Navier–Stokes equations, is already well developed and its extension to solve the higher-order continuum transport equations should not be too difficult. Thus, the range of problems that can be solved using CFD can be substantially increased. In Chaps. 2 and 3 of this book, we have discussed the various fluid flow and heat transfer solutions of microscale flows. These examples bring out the difficulty with respect to employing the Navier–Stokes equations as discussed further in Chap. 4. The Burnett and Grad equations and few known solutions of these equations are discussed in Chaps. 5 and 6, respectively. The variants of the Burnett and Grad equations along with some other recent works on higher-order continuum transport equations are covered in Chap. 7. It is further recognized that the analytical approach is not sufficient to understand flows at the microscale. Therefore, a brief introduction to the relevant numerical and experimental techniques is provided in Chap. 8. Finally, scope for further work is identified in the last chapter.

1.2 Applications of Microscale Flows We begin by examining a few interesting applications of microscale flows. The ability to first design a microdevice towards its intended objective, followed by the fabrication of the micron size passages involved and their sealing, can lead to the development of numerous interesting microdevices. A good understanding of microscale flows is important for design of microdevices as most practical microdevices will involve movement of some fluid through them.

4

1 Introduction to Microscale Flows and Mathematical Modelling

Although numerous potential applications can be quoted, here we restrict ourselves to four applications from distinct domains. Some of the applications take advantage of physical effects which are specifically present at the microscale. While other applications exploit typical features of microdevices—large surface area to volume ratio, small amount of sample and reagents required, requirement of lower amount of power which leads to a solution at a reduced cost, and the possibility of having a microdevice which is portable and disposable. These advantages however come at a cost as the microdevices suffer from the requirement of relatively large pumping power, low rate of mixing of fluids, low throughput, among others. Several issues related to selection of material and cost of fabrication also show up in making of practical microdevices.

1.2.1 Cooling of Electronic Devices The electronics industry has seen remarkable advances in the last few decades because of which it is now possible to pack numerous advanced electronics circuits on a single chip. As the functionality of the electronics chip increases, the amount of heat being generated by the electronic circuit increases, and consequently the heat flux increases. The heat generated in the chip needs to be removed, so that the temperature stays below a threshold value and the functioning of the chip is not adversely affected. Although the amount of power that needs to be dissipated is only a few 100s of watt, the area over which this heat is generated is small (250 W/cm2 ). The conventional cooling techniques are not capable of handling such intense heat fluxes. In this context, flow in microchannel, with or without phase change, has been proposed as a potential cooling solution in the literature. Figure 1.1a,b shows a microchannel with a microheater on the other side of the wafer to mimic the heat generating chip. A simple calculation shows that the heat transfer coefficient with flow of water in a 100 µm passage can easily be greater than 10 kW/m2 K without phase change, and much greater than this with boiling (Fig. 1.1c). Such a high value of heat transfer coefficient is obtained only with phase change at the conventional scale, illustrating the advantage of cooling electronic devices (and other high heat flux generating components) using an array of microchannels, with water (or another appropriate fluid) flowing through the array of microchannels.

1.2.2 Micro-nozzles and Micro-thruster Micro/nano satellite technology has attracted increasing attention worldwide due to its desirable features of small size, light weight, low cost, short development cycle, excellent performance, and rapid launch. The existing propulsion systems

1.2 Applications of Microscale Flows

5

Fig. 1.1 (a) Microchannel made on silicon wafer and bonded with quartz. There are four parallel trapezoidal cross-section microchannels with dimensions of 78 µm (depth) by 273 µm (upper width) (159 µm lower width) by 20 mm (length), yielding a hydraulic diameter of 108.8 µm. Connectors for bringing in and taking out fluid from the microchannel are also seen in the figure [135]. (b) Microheater on the other side of the silicon wafer to supply a known amount of heat to the flowing fluid [135]. (c) Variation of heat transfer coefficient with position in the microchannel at two different flow rates of water [134]

and thrusters cannot meet the demands of micro-satellites due to the limitations in their size, mass, and thrust. Therefore, there is a need to develop micro-thrusters suitable for micro-satellites having features of low mass, small size, precise thrust, and low power consumption. Micro-thrusters are generally used in spacecraft that require small amounts of thrust, and can be classified as [10]: (1) mesoscopic dimensions (2) microscopic segments, and (3) nanoscopic structures. A microthruster generates 1–10 mN force and can be used for propulsion and attitude control in small space satellites. A smaller micro-thruster generating 10–1000 µN force is used for dynamic suppression/damping of vibrations in extended space structures. In the operation of the micro-thruster, micro-nozzle forms the basic component and therefore, fluid flow in micro-nozzles is of interest. Further, it is expected that

6

1 Introduction to Microscale Flows and Mathematical Modelling

the basic understanding of fluid flow characteristics in these relatively complex geometries will be beneficial in the design of other micro-devices for scientific applications.

1.2.3 Breath Analyser Generally, three types of biological media are used to assess the human health, i.e. breath, urine, and blood. Working with breath is obviously advantageous as it can be collected in a non-invasive manner obviating the need of a trained clinical personnel, while also being unlimited in timing and volume. Breath analyser can be used to determine the human metabolic rate, estimate the blood alcohol content, diagnose diseases (such as lung cancer), and provide information on physical fitness. Usually three types of breath matrices are used to assess the human health: exhaled breath, exhaled breath condensate, and exhaled breath aerosol. Exhaled breath analyzer has received much attention in the field of medical diagnostic technology due to the simple procedure in sampling analytes with sufficient information for further diagnosis [163]. In order to assess the human health, a wide variety of sample collection and detection methods are available. These sample collection methods include polymer bags, canisters, sorbent tubes, exhaled breath condensate (EBC) collection and exhaled breath aerosol (EBA) collection, while detection methods include sensors, mass spectrometry, and immunoassay instrumentation. Currently, different types of sensors and mass spectrometry instruments are available for breath analysis. Extensive research is currently being done on several breath-relevant topics including security and airport surveillance, cellular respiration, and canine olfaction. In this context, several breath-relevant microdevices have already been developed [106]; such as for detection of volatile organic compounds in breath; for detection of nitric oxide in exhaled breath (a marker of asthma), and halitosis (or bad breath) caused by ammonia, amines, and certain sulfur gases.

1.2.4 Microdevice for Conducting Blood Test Blood test is one of the most routinely undertaken procedures when a person is unwell. This is because the physiological condition of a person can be easily assessed by examining the concentration level of various analytes present in blood (particularly in plasma). The ability to perform blood test at the site of the patient itself—rather than the patient going to the hospital for the test—would clearly be of immense value. Several companies and research groups are therefore working to make such a microdevice—with sample preparation module and sensing module, as the two primary modules of the microdevice.

1.2 Applications of Microscale Flows

7

The sample preparation module is required because almost 45% of human blood comprises cellular elements (solid components in blood), which needs to be discarded before conducting the test. Red blood cells form the most important cellular component, while white blood cells and platelets are the next major cellular components. The liquid component of blood (plasma) contains a plethora of biomarkers and it is therefore vital to extract the plasma from blood for conducting the tests. The conventional method of plasma separation is through centrifugation, which is tedious, time consuming, requires human interface, and involves multiple stages of sample handling. Integrating a centrifuge with a microdevice is not easy and removes the advantages of microdevices (small sample size, portable, etc.). This requirement of separating plasma from blood in a simple and cost-efficient manner and without the use of a centrifuge has led research groups to develop technologies on microfluidics platform. An easy-to-use on-chip microdevice for extraction of high-quality plasma, developed at the Indian Institute of Technology Bombay, is shown in Fig. 1.2a, b. Note that several of the physical effects on which the separation principle is based kick-

Fig. 1.2 (a) Microchannel for separating plasma from blood developed at the Indian Institute of Technology Bombay. A two-rupee coin has been placed next to the fabricated device for comparison. (b) Close-up view of flow of blood through the microdevice, with plasma passing through one branch and the remaining blood passing through the other branch [118]. (c) Optical image of fabricated microfluidic devices embedded with a three electrode system consisting of Platinum (Pt) counter electrode, Silver/silver chloride (Ag/AgCl) reference electrode, and Gold (Au) working electrode on a PET substrate. Working electrode Au is functionalized with C-ZNO NFs. Finally, the AntiHRP II was covalently immobilized to the nanofiber integrated sensing surface for detection of Malaria parasite [111]

8

1 Introduction to Microscale Flows and Mathematical Modelling

in only at the microscale. Figure 1.2c shows a micro-sensor for detecting malaria, developed at the Indian Institute of Technology Hyderabad. Successful integration and field-testing of the components of the microdevices can revolutionalize the way blood tests are done in the future. Examples 2 and 3 given above involve flow of gases through microdevices, which is the subject of this book. We will briefly comment on flow in diverging and converging passages with relevance to design of micro-nozzle and microthruster, and flow in other complex passages with relevance to microdevice for breath analyser, in Sect. 2.6.

1.3 Classification of Flow Regimes The length scale of micro-scale systems is in micro-meter (or µm). Due to such small length scale, it is possible that the mean free path of the gas (λ) becomes comparable to the length scale (L). Formally, this possibility is quantified by defining a parameter Knudsen number as Kn =

λ . L

(1.1)

In situations (such as external flows) where definition of L is not obvious, one can employ spatial gradient of a relevant parameter (φ, such as density) for defining the length scale as L−1 =

dφ/dx . φ

(1.2)

As an example, consider flow of air in a centimeter size passage. The Knudsen number for this case is 7 × 10−6 since the mean free path of air under standard condition is about 70 nm. Therefore, Kn → 0 at the conventional scale because of relatively large value of L, whereas it can be easily verified that Kn becomes nonnegligible (> 10−3 ) for micro-scale flows. The presence of Knudsen number as an additional parameter leads to the interesting possibility of finding new flow physics in such microscale flow systems, as discussed further in this book. The Knudsen number can be used to demarcate the flow into four regimes: continuum (Kn < 10−3 ), slip (10−3 < Kn < 10−1 ), transition (10−1 < Kn < 101 ), and free-molecular (Kn > 10) (Table 1.1). The definition of the various regimes is therefore based on order of magnitude of the Knudsen number and should only be taken as indicative. Note that slip and transition regimes are sandwiched between two major regimes of continuum and free-molecular. Both these intermediate regimes span approximately two decades each. Some references refer to Kn between 0.01 and 0.1 as the slip regime. However since slip effect starts to show up for Kn > 0.001, Kn = 0.001 is taken here as the beginning of the slip regime, as also done in several other sources.

1.4 Characteristics of Microscale Flows

9

Table 1.1 Classification of flow regimes and models Knudsen number Kn < 10−3

Regime Continuum

10−3 < Kn < 10−1

Slip flow

10−1 < Kn < 10

Transition

Kn > 10

Free molecular flow

Fluid model Navier–Stokes equations with no slip boundary conditions Navier–Stokes equations with velocity slip and temperature jump boundary conditions Higher order continuum transport equations (e.g., Burnett, Grad, OBurnett, O13, etc.) Collisionless Boltzmann equation

In this book, we are primarily concerned with the slip and transition regimes.

1.4 Characteristics of Microscale Flows At the microscale, the interaction between the fluid and solid wall is different than that at the macroscale. Fundamental studies have shown that the continuum hypothesis may not be valid at microscale and some specific effects may be present that can alter the fluid flow and heat transfer characteristics significantly. At the microscale, surface area to volume ratio is much larger, which results in increased surface forces, which may produce large pressure drops, compressibility effects, and viscous dissipation; decreased inertial forces, which allows diffusion and conduction processes to become relatively more significant; and increased heat transfer, which may lead to variable fluid properties and creep flow. Therefore, these effects need to be considered to predict the fluid flow and heat transfer characteristics. A detailed discussion of these effects is presented in the following sections.

1.4.1 Rarefaction Rarefaction refers to the reduction of molecular density in the fluid flow medium and usually, it is encountered in micro and nano-scale flows. In such flows, the molecular mean free path may be of the same order as that of the characteristic dimension of the system. Microscale flows are characterized by Knudsen number (Eq. (1.1)). A larger Knudsen number arises either due to longer mean free path (rarefied gas flows) or small characteristic dimension of the system (microdevices). The Knudsen number is used to demarcate the fluid flow into four regimes: continuum regime

10

1 Introduction to Microscale Flows and Mathematical Modelling

Wall Flow direction

Continuum

Slip

Transition Free molecular Wall

Fig. 1.3 With mean free path increasing due to reduction in pressure, the flow transits through various regimes. (Here p denotes pressure and λ denotes mean free path and hollow circles represent the molecules in random motion)

Kn < 10−3 , slip regime 10−3 < Kn < 10−1 , transition regime 10−1 < Kn < 10, and free molecular regime Kn > 10. These regimes can occur in isolation, or as shown in Fig. 1.3 can even co-exist in a given passage. For very low Knudsen number, the number of collisions between the molecules is large compared to the number of collisions between the molecules and the wall. In such a case, the usual continuum concept is applicable and the Navier–Stokes equations and the Fourier heat conduction law are valid. The continuum flow is characterized by the Reynolds number and the Mach number only; the Knudsen number will not enter the problem explicitly since it is already considered to be very small. For sufficiently large Knudsen number, the continuum concept needs to be modified to predict the fluid flow and heat transfer characteristics. In such a case, first- (or second-) order slip boundary condition needs to be considered in the slip and early transition regimes. The Navier–Stokes equations with first-order slip boundary condition are relatively accurate up to Kn ≈ 0.1, beyond which second-order slip boundary conditions are more appropriate. At still higher Knudsen numbers, higher-order continuum transport equations need to be considered to analyze the fluid flow and heat transfer behavior. Rarefaction affects both fluid flow and heat transfer characteristics significantly. It is responsible for a decrease in friction factor and Nusselt number. With the increase in Kn, the Nusselt number reduces due to the increased temperature jump at the wall.

1.4.2 Compressibility Compressibility effect leads to density becoming a variable in the flow (Fig. 1.4), and therefore addition of another variable occurs in the problem. It leads to the linking of the momentum equation with the energy equation. Further, certain terms (such as pressure work) may have to be retained in the energy equation in this case. Therefore, the complexity of the problem increases substantially.

1.4 Characteristics of Microscale Flows

11

Fig. 1.4 Shading represents the gradient of density; when density difference is large, compressibility effects become important

Usually, the effect of compressibility is considered when the Mach number exceeds the value of 0.3, otherwise the flow is treated as incompressible. However, Ma > 0.3 criterion is only a sufficient condition for flow to be considered compressible. In case of flow through microdevices, the pressure drops rapidly along the flow direction because of viscous effect, even though the velocity may not be high enough for the Mach number to exceed the threshold of 0.3. The gas density would therefore change and these variations alter both the velocity and temperature profiles. Hence, the fluid flow and heat transfer behavior are affected significantly by compressibility effect. The Mach number is related to the Knudsen number and Reynolds number through the relation Kn =

π γ Ma 2 Re

(1.3)

obtained from the classical kinetic theory of gases where γ denotes the ratio of specific heats. That is, out of these three relevant non-dimensional numbers, only two of them can be independently specified in any problem.

1.4.3 Thermal Creep In case of microflows, the slip velocity is related to the velocity gradient at the wall. Additionally, if a temperature gradient exists in the flow, the slip velocity will be altered as depicted in Fig. 1.5. The slip velocity in terms of the above effects is given by the following expression 2 − σv ∂u 3 μ ∂T us − uw = λ + σv ∂y wall 4 ρT dx wall

(1.4)

where the last term represents the thermal creep or transpiration effect. In the above equation, us is velocity of gas at the wall, uw is wall velocity, σv is tangential momentum accommodation coefficient, λ is mean free path of the gas, x and y are directions parallel and normal to the wall, μ is viscosity, ρ is density, and T is absolute temperature.

12

1 Introduction to Microscale Flows and Mathematical Modelling

Fig. 1.5 A schematic diagram showing thermal creep phenomenon; since Tw1 < Tw2 , a macroscopic motion of gas molecules is observed from points 1 to 2 resulting in increased slip velocity at point 2

The thermal creep is defined as the macroscopic movement of rarefied gas molecules induced by a temperature gradient from lower to higher temperature zone. Note the difference in direction of gradient of the two terms. Usually, creep flow is negligible for large scale flows (since the term scales as Kn), fully developed and constant wall temperature flows, while it becomes significant for constant wall heat flux flows in the slip regime. The creep flow can alter the pressure losses and convective heat transfer rates significantly as compared to the cases where the effect of creep is negligible. The term also leads to a coupling of the momentum and energy equations.

1.4.4 Viscous Dissipation Due to the large velocity gradient in the microchannel (brought about by small length scale), viscous dissipation (proportional to square of velocity gradient) becomes substantial (Fig. 1.6). The ratio of heat generation because of viscous dissipation and heat exchange between fluid and the wall by molecular conduction is quantified by a dimensionless parameter, known as Brinkman number. Mathematically, Br =

μu2m μu2m or or EcP r k(Tw − Tm ) qw Dh

(1.5)

where μ is the dynamic viscosity, um is the flow velocity, k is the thermal conductivity, (Tw − Tm ) is the temperature difference at a particular axial location, Tm is the bulk fluid temperature; Tw is the wall temperature, qw is the wall heat flux, P r is the Prandtl number, and Ec is the Eckert number. Note that the definition of Brinkman number varies with the type of boundary condition at the wall. A larger value of Brinkman number corresponds to larger viscous dissipation compared to the rate of heat conduction, and hence larger is the local temperature rise. The positive value of Brinkman number indicates heat transfer from wall to the fluid (i.e. heating), while the opposite is true for negative value of Brinkman number (i.e. cooling). Although Br is usually neglected in low-speed and low-viscosity flows through conventionally sized channels of short lengths, it may become important for flows

1.4 Characteristics of Microscale Flows

13

Fig. 1.6 Schematic showing the significance of viscous dissipation

Fig. 1.7 With density (ρ), viscosity (μ), and thermal conductivity (k) being different in different regions of the flow, due to steep variation in temperature, the property variation in fluid can become substantial

through conventionally sized long passages. For flows in microchannel, the lengthto-diameter ratio is much larger as compared to flows through conventionally sized long passages; thus, Br may become important in microchannels. For macroscale flows, the effect of viscous dissipation is significant either for higher viscous flows or flows with higher velocity, while for microscale flows, the effect of viscous dissipation is significant even for low velocity flow because of smaller dimensions. Similar to thermal creep, viscous dissipation causes additional coupling between the momentum and energy equations.

1.4.5 Property Variation Since microscale flows are characterized by large temperature gradient, and thermophysical properties of fluid are a function of temperature, it can be expected that the thermophysical properties can vary in the flow as depicted in Fig. 1.7. Both the flow and heat transfer behavior can differ substantially in presence of property variation as compared to the constant property case. The property variation can affect both the developing and fully developed regions of the flow. The amount of difference is a function of Knudsen number, aspect ratio, and temperature difference between the channel inlet and walls. Property variation is therefore relevant for both macro- and microscale flows. This effect needs to be suitably accounted for through appropriate equations for temperature-varying property over the temperature range of interest. Of course, this introduces additional variables and equations in the problem, thereby increasing the complexity.

14

1 Introduction to Microscale Flows and Mathematical Modelling

Fig. 1.8 The second derivative of temperature with respect to axial direction (representing axial conduction) is usually non-negligible at the microscale

1.4.6 Axial Conduction In the analysis of flow at the conventional scale, it is customary to assume that 2 axial conduction can be neglected. This simplifies the analysis by removing the ∂∂xT2 term (where x is the streamwise coordinate) from the equation. The effect of axial conduction is quantified by Peclet number defined as Pe =

ρcp um Dh ρum Dh μcp = = ReP r k μ k

(1.6)

Thus, Peclet number signifies the strength of convection to the strength of conduction. In the case of macro-sized flow, because of relatively large value of Reynolds number, the Peclet number is usually large (P e > 150) and the effect of axial conduction may be neglected. On the contrary, in case of microscale flow with smaller channel dimensions and Reynolds number, the effect of axial conduction becomes much more significant (Fig. 1.8). Therefore, such an assumption may not be justifiable at the microscale.

1.4.7 Conjugate Heat Transfer Normally it is assumed that heat transfer to the fluid is not affected by heat transfer inside the solid walls surrounding the fluid. That is, the heat supplied to the walls is faithfully transferred to the fluid and any heat transfer in the walls is insignificant and therefore need not be accounted for. However, if the wall thickness becomes comparable to the flow passage length scale (a situation applicable to microscale flows) as shown in Fig. 1.9, this assumption ceases to apply. The implication of the relaxation of this assumption can be quite severe in that the boundary condition applied on the outer wall of the passage is not transmitted faithfully to the inner wall of the passage (the difference being due to heat conduction in the wall). In other words, the boundary condition seen by the fluid is not the boundary condition that is applied in the problem. Since the boundary condition has a strong impact on

1.5 Mathematical Modelling of Microscale Flows

15

Fig. 1.9 A schematic showing the conjugate effect at microscale where the wall thickness δw is comparable to the channel height H . The constant heat flux applied at the outer wall, qw and the heat flux at the inside the wall, q(x) are different because of the heat conduction taking place in the wall thickness region

the obtained solution, this conjugate effect needs to be suitably accounted for in the analysis. Conjugate effects are more pronounced at microscale as compared to macroscale because the hydraulic dimension is comparable to the channel thickness in the former case. Therefore, the standard correlations for idealized constant wall temperature or uniform wall heat flux boundary conditions have to be applied with more care. The conjugate effect can be quantified by a parameter defined as M=

ψcond ψconv

(1.7)

where ψcond is a heat flux characterizing axial heat transfer in the wall, ψconv is the total convective heat flux. The quantity M is therefore the ratio of conduction in the wall to the heat transferred as convection from the surface. For M > 0.01, the conjugate effect cannot be neglected. Problems involving substantial conjugate effect therefore requires simultaneous solution in both the solid and fluid domains. The coupling of the solid and fluid domains obviously increases the complexity. Other than the aforementioned generic characteristics of microscale flows, some specific characteristics of microscale flows are discussed in Chaps. 2 and 3.

1.5 Mathematical Modelling of Microscale Flows After examining the characteristics of microscale flows, we discuss ways to mathematically model microscale flows. We first provide a brief introduction to the Navier–Stokes equations, which are used for modelling conventional scale flows, followed by two examples illustrating the need for altering the boundary conditions and even the governing equations for modelling high Knudsen number flows.

16

1 Introduction to Microscale Flows and Mathematical Modelling

1.5.1 The Navier–Stokes Equations The Navier–Stokes equations are used to model the flow of fluids at the conventional scale. The Navier–Stokes equations are mathematical representation of the physical principle on conservation of momentum, as applied to a differential element in the flow. The left-hand side of the equation represents the rate of change of momentum of fluid in the control volume, while the right-hand side represents the external forces acting on the fluid. (The mathematical form of these equations is provided in the next chapter.) The external forces are due to gravity or electric/magnetic field (body forces), pressure or viscous effects (surface forces), surface tension (line force), etc. While the body and surface forces are part of the governing equation, the line force is generally accounted for through the boundary condition. Note that the body forces scale as L3 (where L is the characteristic dimension), while the surface and line forces scale as L2 and L1 , respectively. Since L3 L2 for small values of L, the body force (gravity, etc.) typically has only a minor role to play at the microscale. That is, the orientation of the microdevice is not expected to significantly alter the performance of the microdevice. Being differential equations, the Navier–Stokes equations require specification of initial and boundary conditions for their complete solution. The most important boundary condition of concern here is that at the fluid–wall interface. The no-slip boundary condition, whereby it is assumed that the tangential velocity of the fluid at the wall is equal to the velocity of the wall, is normally imposed in conventional scale flows. The fluid is therefore not allowed to slip on the wall (similarly, temperature of the fluid at the wall is assumed equal to the wall temperature, in case this is required during solution of the energy equation). Further, the wall-normal velocity component is zero for a non-porous wall. Although the Navier–Stokes equations refer to the momentum equations, in this book we have used the term “Navier–Stokes equations” to refer to the conventional mass, momentum, and energy equations taken together.

1.5.2 Limitations of Conventional Equations and Boundary Conditions In this section, we show that the solution for flow in the slip and transition regimes differs substantially from its continuum regime counterpart. Towards this, we examine two problems for which obtaining solution is relatively easy. We then give an overview of approaches available to model the flow in the slip and transition regimes. Upon imposing a uniform shear to the flow (for example, through the movement of one large plate with respect to another large stationary plate; the sides of the plates being large with respect to the normal distance between the plates), a linear velocity profile is expected to be set up in the flow (Fig. 1.10a). This expectation is

1.5 Mathematical Modelling of Microscale Flows

17

u/Uo

(a)

y/H

(b) Fig. 1.10 (a) Schematic of flow between two parallel plates, upper plate is moving at speed Uo in the positive x-direction while lower plate is stationary. (b) Velocity profile in different flow regimes obtained using DSMC (We are grateful to Mr. Abhimanyu Gavasane for these computations)

in line with the observation at the conventional scale (distance between the plates is much larger than the mean free path of the gas in the gap, or Kn ≈ 0). However, upon reducing the distance between the plates or increasing the mean free path of the gas (for example, by reducing the pressure), the velocity starts to deviate from the classical profile (Fig. 1.10b). In particular, a finite velocity slip is observed at the plates at Kn ≈ 10−3 . Further, the velocity profile becomes non-linear in the near plate region when the distance between the plates becomes comparable to the mean free path of the gas or Kn ≈ 10−1 .

18

1 Introduction to Microscale Flows and Mathematical Modelling

The above observation points to modelling two important aspects of the flow present at finite Knudsen numbers (Kn > 10−3 ): velocity slip at the wall and non-linearity in the velocity profile in the near plate region. Presence of wall-slip suggests that the no-slip boundary condition has to be relaxed at finite Knudsen numbers; we will examine several slip models in Sect. 2.3. It turns out that in most situations, we can model flow in a large part of the slip regime through the use of the Navier–Stokes equations but with modification of the boundary condition from no-slip to slip. The non-linear nature of the velocity profile at still higher Knudsen numbers, however, cannot be modelled within the framework of the Navier–Stokes equations. This is true for the present problem even after accounting for property variation, for example, due to variation in temperature if any. A similar observation is noted for flow between two large parallel plates, with the flow being driven by a pressure gradient between the inlet and outlet (Fig. 1.11a). In this situation, the velocity profile is expected to be parabolic with a constant pressure along the wall-normal direction; these are supported by observations. However, with an increase in the Knudsen number, the flow starts to slip at the wall in the slip regime (Fig. 1.11b). Besides the wall-slip, a non-linear pressure variation is observed along the wall-normal direction in the transition regime (Fig. 1.11c). As before, the first observation can be modelled through appropriate wall-slip models, while the latter observation is beyond the modelling capability of the Navier–Stokes equations.

1.5.3 Approach to Modelling Microscale Flows The two simple flow problems presented in the previous section bring us to the interesting question: How do we model the flow at high Knudsen number? These examples have exposed the limitation of the linear constitutive relation (relation between stress and strain rate) involved in the classical Navier–Stokes equations. A non-linear constitutive relation applies to non-Newtonian fluids; however, the behavior of non-Newtonian fluids is not dependent on the value of the Knudsen number, implying that non-Newtonian fluid models are not applicable in the present scenario. Other modifications to the Navier–Stokes equations have also not been successful as discussed in Chap. 4. We have to rather consider equations containing terms of higher order in Knudsen number, termed as “higher order continuum transport equations.” The considered equations should obviously reduce to the Navier–Stokes equations in the limit of low Knudsen number. It can be anticipated that the higher order continuum transport equations will contain additional terms as compared to the Navier–Stokes equations, with the additional terms being of the order Knudsen number squared (or higher). Also the non-linear nature of the constitutive equation will make these models much more involved to work with. It is however not obvious how to obtain higher-order models from the conventional method of deriving the transport equations. Although there are several distinct approaches in the conventional method to arrive at the transport equations,

(a)

u/Uo

(b)

p/pc

(c) Fig. 1.11 (a) Schematic of flow between two parallel plates, the flow being driven by a pressure gradient. (b) Velocity and (c) pressure profiles across the channel height in different flow regimes obtained using DSMC, profiles are non-dimensionalized using the centerline values. (We are grateful to Mr. Abhimanyu Gavasane for these computations)

20

1 Introduction to Microscale Flows and Mathematical Modelling

0

1

2

R26

Fig. 1.12 Different schemes for modelling of fluid flow

no extra term that could have been retained is seen in any of these derivation procedures. Therefore, these conventional approaches cannot be extended to obtain higher-order models. The alternative then is to change the starting point and resort to the Boltzmann equation. The conservation equations for mass, momentum, and energy can be obtained easily from the Boltzmann equation. Interestingly, as will be shown later in Sect. 5.2.4, the exact form of the distribution function (solution of the Boltzmann equation) will not even be needed to arrive at the conservation equations. However, the problem of closure shows up at this stage. This problem of closure is same as that seen with the conventional method. An empirically obtained relation between stress and strain rate (through Newton’s law of viscosity), and heat flux and temperature gradient (through Fourier’s law of heat conduction) are supplied in the conventional method to close the system of equations. This way of closure is however not followed in the alternative method which makes this Boltzmann equation based approach distinct from the conventional method. There are several approaches of deriving transport equations starting from the Boltzmann equation. An essential requirement of these approaches is that they should faithfully reproduce the known equations (the Navier–Stokes equations) at the lower-order, before the higher-order equations can be trusted. The two successful methods for solving the Boltzmann equation (and considered here) are the Chapman-Enskog expansion and the Grad moment method (Fig. 1.12). Both these methods provide a way to evaluate the distribution function, which can then be used to close the equations. In the Chapman-Enskog method, the distribution function is expanded in a perturbation series around equilibrium (i.e., Maxwell-Boltzmann distribution function

1.6 Relevance and Scope of the Book

21

corresponding to gas at rest). The method leads to both linear and non-linear constitutive relations depending on where the expansion is terminated. The linear constitutive relations are of Knudsen number order one, and substitution of the constitutive model in the conservation equation leads to the Navier–Stokes and energy equations. The non-linear terms are of Knudsen number order two (or higher); such equations lead to higher-order models, which are known as the Burnett (or super Burnett) equations. The inviscid Euler equations can also be obtained using this approach. A detailed derivation of the Burnett equations is given in Chap. 5. Note that the number of independent variables in all the above equations is five. These variables are density (ρ), velocity vector (vi ), and temperature (T ). (Note that tensor notation has been used extensively throughout the book and an introduction to tensors is included in Appendix A.) In contrast, the moment method does not seek a relation between stress and strain rate, rather treats stress (σij ) and heat flux (qi ) as independent variables. The number of independent variables therefore increases to 13 (or even 20 or 26) as separate evolution equations are written for each of these variables. The problem of closure is then transferred to the unknown terms appearing in the stress and heat flux evolution equations. Distribution function, written in terms of Hermite polynomials, is finally employed to obtain closure. The Grad 13-moment and 20-moment equations are discussed in detail in Chap. 6. Some variants of the Burnett and Grad equations and other alternative approaches are introduced in Chap. 7. Equations obtained from a recent approach, which employs a thermodynamically consistent distribution function, with the approach being very different from the two aforementioned classical approaches, are also presented in Chap. 7. This approach, although promising, still needs to be established and is hence shown by dotted line in Fig. 1.12.

1.6 Relevance and Scope of the Book The book considers several topics that have not been dealt with adequately in the past. For example, while some researchers know that the Navier–Stokes equations can be derived from the Boltzmann equation, how exactly can it be derived is not so well understood. Here, we provide the derivation in a relatively simple way. Such an understanding may also help better appreciate numerical methods such as the Lattice Boltzmann method, which is based on the Boltzmann equation and ChapmanEnskog expansion; the Lattice Boltzmann method has become very popular and is being widely used for solving flow problems. Similarly, while the need for looking beyond the Navier–Stokes equations (especially for cases where the continuum hypothesis is violated) has already been felt, the alternate set of equations to be solved is not known. Microscale flows put the need for having better higher-order transport equations in perspective. The subject of higher-order continuum transport equations is still under development, with research groups coming up with their own versions of the equations. It is generally difficult to follow all the physical and mathematical details behind these equations, due to

22

1 Introduction to Microscale Flows and Mathematical Modelling

the inherent complexity involved in the process of derivation. For people not so well versed with this line of investigation, it becomes further difficult to appreciate these new developments. The book attempts to fill this gap by exposing the required details behind the idea and derivation in relatively simple terms. The hope is that the reader will be better prepared to follow the newly proposed equations, and even contribute towards their derivation. Further, we realize that the various available equations need to be rigorously tested for their accuracy by numerous people by applying them to different problems. Obviously, this can happen only after a certain level of comfort has been achieved with the available equations. We also compile and discuss the few known exact solutions of the Navier–Stokes and higher-order continuum transport equations for problems in the slip regime. To the best of the authors’ knowledge, such a complete compilation (especially for higher-order continuum transport equations) is not previously available. The book is therefore organized in two parts, with the first part dealing with physical aspects of microscale flows and heat transfer. The second part deals with the mathematical modelling of microscale (and other high Knudsen number) flows, and application of obtained equations to a few simple problems. In Chap. 2, we discuss the various fluid flow solutions within the framework of the Navier–Stokes equations and slip boundary condition. The slip model is derived followed by a survey of the measured values of the slip coefficients. Analytical solution for a few useful cases is derived in reasonable detail, and a discussion of the obtained solution provides pertinent insights into the fluid flow behavior. A compilation of some interesting known observations in straight as well as complex passages is also provided. This is followed by a compilation of some useful correlations. In Chap. 3, we consider the energy equation along with temperature jump condition at the wall. We consider analytical solution for problems involving temperature jump and viscous dissipation, while the effect of other complicating parameters is only briefly commented upon. The objective therefore is not to provide all known solutions of microscale flows, rather discuss some typical effects at the microscale. Our approach is similar to what is followed in the literature, in that a single (or a few, but not all) complexities are dealt with in a given problem; however, since experiments and practical situations have many more complexities, we find that there is a further need for theoretical development of the subject. In Chap. 4, we compile problems for which the Navier–Stokes equations are found to be inadequate. Since such a compilation is being attempted for the first time, it is likely to be incomplete. We consider some known extensions of the Navier–Stokes equations; also a popular non-Fourier model is discussed. We argue that these extensions are still inadequate for studying problems in the transition regime, and we have to rather resort to an entirely different approach. Accordingly, the Burnett equations are introduced in Chap. 5. We understand that the awareness and appreciation of these equations is rather limited. The derivation is therefore explained in detail, and further summarized through a flow chart. An order of magnitude analysis and stability analysis of the Burnett equations are also

1.6 Relevance and Scope of the Book

23

presented. A novel iterative approach to solve these equations, which can further be applied to other partial differential equations, and other recent solutions of the Burnett equations are also provided, and results are compared against DSMC data. The Burnett equations in cylindrical coordinates are also provided for ready reference. The Grad equations are next introduced in Chap. 6. The derivation is again stepwise and aided with a flow chart, with most of the derivation being sufficiently original. The required boundary conditions are obtained next. The linearized Grad equations are then solved for an impulsively started plate problem, and a comparison of the solution with both the Navier–Stokes and kinetic equations provides useful insights. The problem of cylindrical Couette flow is also solved using the linearized Grad equations. There are only a handful of examples where analytical solution of the Grad equations are available, making this compilation original and useful. The strengths and limitations of the Burnett and Grad equations are discussed in Chap. 7. This helps appreciate the need for having variants of these equations. Some other promising recent works on higher-order continuum transport equations are also covered in this chapter. A comparison of the available variants for two problems is also provided. Several other examples where variants are compared are available in the literature, which can be further referred to. It is recognized that the analytical approach adopted in the earlier chapters is not always sufficient to understand flows at the microscale. Therefore, a brief introduction to the relevant numerical and experimental techniques is provided in Chap. 8. Some broad conclusions and scope for further work are documented in Chap. 9. The book therefore aims to expound on the influence of additional physical effects present at the microscale on the fluid flow and heat transfer aspects of gas flow in microchannel. The book also exposes the derivation and application of higher-order models, making them accessible to a wide readership.

Chapter 2

Microscale Flows

In this chapter, fundamentals of fluid flow and flow modelling aspects in the slip regime have been presented. This includes a brief introduction to the Navier–Stokes equations and the slip boundary condition. A few analytical solutions for flow in the slip regime are then obtained. Finally, we briefly examine the flow in some complex passages and present some useful empirical correlations. These solutions and insights help appreciate the flow physics in the slip regime better.

2.1 Introduction At the microscale, the interaction between the fluid and solid wall is different than that at the macro-scale, due to the relatively large value of Knudsen number. It is readily apparent from Eq. (1.1) that a larger Knudsen number can arise either due to long mean free path (rarefied gas flows) or for a small characteristic dimension of the system (micro-devices) or both. In this chapter, we present the concept of velocity slip along with the factors affecting slip and the value of the slip coefficient. Further, the solutions of the Navier–Stokes equations in the slip regime and implications of slip on the flow behavior are also presented. We also briefly discuss some experimental results for flows in relatively complex passages. While numerous studies are available on the fundamental properties of fluid flow at the macroscale, relatively few resources are available for the microscale. Such resources are important since microscale flows occur in a variety of practical applications operating in the slip regime. These applications include micro-motors, micro-actuators, micro-turbines, micro-sensors, micro-reactors, micro-valves, micro-pumps to name a few. Such resources are also relevant for rarefied gas flows and particularly low-pressure applications such as high-altitude aircraft flight and vacuum devices.

© Springer Nature Switzerland AG 2020 A. Agrawal et al., Microscale Flow and Heat Transfer, Mechanical Engineering Series, https://doi.org/10.1007/978-3-030-10662-1_2

25

26

2 Microscale Flows

2.2 Governing Equations for Fluid Flow Governing equations for fluid flow can be derived from first principle, based on the physical principles for conservation of mass, momentum, and energy [102]. The equations can be cast in either integral or differential form, corresponding to finite- and infinitesimal-size control volume, respectively. Only the differential form of mass (continuity), momentum (Navier–Stokes), and energy equations are given here.

2.2.1 Tensorial Form The continuity equation is ∂ ∂ρ + (ρui ) = 0 ∂t ∂xi

(2.1)

The equation of motion for a Newtonian fluid is obtained by substituting the following constitutive equation,

∂uj ∂ui τij = −pδij + μ + ∂xj ∂xi

+ μ2

∂uk δij ∂xk

(2.2)

into the Cauchy’s equation of motion, ρ

∂uj ∂uj + ui ∂t ∂xi

=

∂τij + ρgj ∂xi

to obtain ∂uj ∂uj ∂uj ∂ui ∂p ∂uk ∂ ρ =− μ +μ2 + ui + + δij +ρgj ∂t ∂xi ∂xj ∂xi ∂xi ∂xj ∂xk

(2.3)

(2.4)

Note that (∂p/∂xi )δij = ∂p/∂xj has been employed in the above derivation gj is external body force per unit mass. The second coefficient of viscosity, μ2 is related to dynamic viscosity μ and the coefficient of bulk viscosity μv through μ2 = (μv − 23 μ). As per Stokes’ hypothesis, the coefficient of bulk viscosity is zero giving μ2 = −(2/3)μ; this hypothesis can be shown to apply well to monatomic gases (see Sect. 5.4.2 and Eq. (5.129)). In general, the viscosities, μ and μv , can depend on the thermodynamic state. Indeed, μ for most fluids displays a rather strong dependence on temperature, decreasing with temperature for liquids and increasing more rapidly than T 1/2 (where T is the absolute temperature) for gases. As shown in the next section, the constitutive equations for normal and shear stresses (Eq. (2.2)) can be obtained as an analogy to generalized Hooke’s law in solid mechanics.

2.2 Governing Equations for Fluid Flow

27

Using μ2 = −2μ/3 as per Stokes’ hypothesis, Eq. (2.4) can be written as ρ

∂uj ∂uj + ui ∂t ∂xi

=−

∂uj ∂ui ∂p 2 ∂uk ∂ μ − μ + + δij +ρgj ∂xj ∂xi ∂xi ∂xj 3 ∂xk

(2.5)

The thermal energy equation in tensorial notation is expressed as ρ

∂uk D ∂qk −p +Φ =− Dt ∂xk ∂xk

(2.6)

where is the specific internal energy and Φ is the viscous dissipation given as

∂uk 2 Φ = 2μeij eij + μ2 ∂xk ∂uk 2 2 = 2μeij eij − μ 3 ∂xk 2 1 ∂uk = 2μ eij − δij 3 ∂xk

(2.7)

∂uj ∂ui is the strain rate tensor (symmetric part of velocity where eij = 12 ∂x + ∂xi j gradient tensor). Note that Stokes’ hypothesis, μ2 = −2μ/3 is employed in the second step. In the last step, with some algebra, complete square can be obtained which signifies that viscous dissipation, Φ is always positive. As evident in Eq. (2.7), viscous dissipation is proportional to dynamic viscosity and represents the irreversible conversion of kinetic energy into internal energy due to viscous effects. Equation (2.6) states that the internal energy can increase because of heat coming into the differential element (first term on RHS is negative of divergence of heat flux vector), volume compression and heating due to viscous dissipation. Expressing heat flux vector in terms of temperature gradient (qi = −k∂T /∂xi ) using Fourier’s law of heat conduction, Eq. (2.6) becomes ∂T ∂uk ∂ D k −p = +Φ (2.8) ρ Dt ∂xk ∂xk ∂xk The system of equations (Eqs. (2.1), (2.5), and (2.8)) form an unclosed set of five equations with seven variables namely, ui , p, ρ, T , and . Assuming the gas to be a perfect gas, we have equation of state as the sixth equation, p = ρRT

(2.9)

where R is the specific gas constant. The seventh equation can be obtained using a thermodynamic relation for a calorically perfect gas (constant specific heats) as = cv T

(2.10)

28

2 Microscale Flows

D Substituting for (Eq. (2.10)) and Φ (Eq. (2.7)) and using Dt = ∂t∂ + uk ∂x∂ k to replace the material derivative term, the final form of thermal energy equation can be written as ∂T ∂T ∂ ∂T ∂uk ρcv = k −p + uk ∂t ∂xk ∂xk ∂xk ∂xk 2 ∂uj 1 ∂uk 1 ∂ui − + 2μ + δij (2.11) 2 ∂xj ∂xi 3 ∂xk

Using Eq. (2.9), one can replace ρ with p, and obtain a closed set of five equations with ui , p, and T as the dependent field variables. To obtain solutions for these variables, second-order coupled non-linear partial differential equations (2.1), (2.5), and (2.11) need to be solved along with specification of appropriate number of boundary conditions. The specification of boundary conditions at a solid wall-gas interface is discussed in Sect. 2.3.

2.2.2 Constitutive Relations In this section, we derive the constitutive relation for Newtonian fluids. While the shear stress is more straightforward, the normal stresses are obtained in a manner analogous to that for solids [126]. As per Hooke’s law, a normal stress τxx will produce a strain εxx where εxx = τxx /E, with E being the modulus of elasticity. The normal stress also produces lateral strains given by εyy = εzz = −ντxx /E where ν is the Poisson’s ratio. Therefore, in general, when τxx , τyy , and τzz are all present, εxx will be given by (since both τyy and τzz will also produce a lateral strain) τ τzz τxx yy −ν + E E E τ τyy τzz τxx xx −ν + + = (1 + ν) E E E E τ¯ τxx − 3ν = (1 + ν) E E

εxx =

(2.12)

where τ¯ = (τxx + τyy + τzz )/3. Equation (2.12) can be rewritten as τxx =

E 3ν εxx + τ¯ . 1+ν 1+ν

(2.13)

Since shear modulus (G) is related to the modulus of elasticity and the Poisson’s E ratio, as G = 2(1+ν) we can write Eq. (2.13) (with the last term split into two terms) as

2.2 Governing Equations for Fluid Flow

29

τxx = 2Gεxx + τ¯ +

(2ν − 1) τ¯ . 1+ν

(2.14)

Similar to equation for εxx (Eq. (2.12)), we can write equations for εyy and εzz . Summing up the three equations for the normal strains, we obtain εxx + εyy + εzz =

1 − 2ν (τxx + τyy + τzz ) E

E (εxx + εyy + εzz ) = τ¯ . 3(1 − 2ν)

or,

(2.15)

Substituting Eq. (2.15) for the last term in Eq. (2.14): 2 τxx = τ¯ + 2Gεxx − G(εxx + εyy + εzz ) 3

(2.16)

which is the required equation for solids. For applying the above relation to fluids, we replace G with μ and the strain with the strain-rate (denoted by a dot over the relevant symbol). We further identify e˙xx , e˙yy , and e˙zz as ∂u/∂x, ∂v/∂y, and ∂w/∂z, respectively (u, v, w are the velocity components). Finally, we put τ¯ = −p, to obtain τxx

∂u ∂v ∂w ∂u 2 − μ + + = −p + 2μ ∂x 3 ∂x ∂y ∂z

(2.17)

which completes our derivation for normal stress in fluids. Note that the above equation is same as Eq. (2.2) (for i = 1 and j = 1) with μ2 = −(2/3)μ. The shear stress is directly proportional to the rate of shear strain for Newtonian fluids, with viscosity as the constant of proportionality. Therefore, τxy = μe˙xy ∂u ∂v + . =μ ∂y ∂x

(2.18)

The other normal and shear stress components can be obtained by a suitable change of variables as given in Table 2.1. Table 2.1 Base equation and change of variables to be followed while evaluating other components of stress tensor Variable τyy τzz τyz τzx

Base equation τxx τxx τxy τxy

Change of variables u → v, x → y, v → u, and y → x u → w, x → z, w → u, and z → x u → v, x → y, v → w, y → z, w → u, and z → x u → w, x → z, v → u, y → x, w → v, and z → y

30

2 Microscale Flows

2.2.3 Compressible Navier–Stokes Equations Strictly speaking, the Navier–Stokes equation is only a statement of the momentum balance given in Eq. (2.4). The expanded form of the equations in the compressible form is given in this section. Three separate equations can be written from Eq. (2.5) by successively taking j (the free index) as 1, 2, 3. On the other hand, i is the repeated index in Eq. (2.4) and, therefore, we need to sum over i (with i going from 1 to 3). For example, from Eq. (2.4) we can write the x-momentum (j = 1) equation as x-momentum: ∂u ∂u ∂u ∂u +u +v +w ∂t ∂x ∂y ∂z

∂ ∂w ∂u 2 ∂u ∂v ∂p + − + + μ 2 =− ∂x ∂x ∂x 3 ∂x ∂y ∂z ∂u ∂v ∂u ∂w ∂ ∂ + + μ + μ + ρgx + ∂y ∂y ∂x ∂z ∂z ∂x

ρ

(2.19)

Note that in the above equation, we replaced (u1 , u2 , u3 ) by (u, v, w), and (x1 , x2 , x3 ) by (x, y, z). Further, recall that δij = 1 for i = j , and zero otherwise. Similarly, the y- and z-momentum equations can be written by taking j = 2 and 3, respectively. y-momentum: ∂v ∂v ∂v ∂v +u +v +w ρ ∂t ∂x ∂y ∂z

∂ 2 ∂u ∂v ∂w ∂v ∂p + − + + μ 2 =− ∂y ∂y ∂y 3 ∂x ∂y ∂z ∂u ∂v ∂v ∂w ∂ ∂ + + μ + μ + ρgy + ∂x ∂y ∂x ∂z ∂z ∂y

z-momentum: ρ

∂w ∂w ∂w ∂w +u +v +w ∂t ∂x ∂y ∂z

(2.20)

2.2 Governing Equations for Fluid Flow

31

∂ ∂w ∂w 2 ∂u ∂v ∂p + − + + μ 2 =− ∂z ∂z ∂z 3 ∂x ∂y ∂z ∂w ∂u ∂v ∂w ∂ ∂ + + + μ + μ + ρgz ∂x ∂x ∂z ∂y ∂z ∂y

(2.21)

The continuity equation is a scalar equation (no free index) and summation over i yields ∂ρ ∂(ρu) ∂(ρv) ∂(ρw) + + + =0 ∂t ∂x ∂y ∂z

(2.22)

The thermal energy equation (Eq. (2.11)) in expanded form is ρcv

∂T ∂T ∂T ∂T +u +v +w ∂t ∂x ∂y ∂z

=

∂T ∂ ∂T ∂ ∂T k + k + k ∂x ∂y ∂y ∂z ∂z ∂w ∂u ∂v + + +Φ (2.23) −p ∂x ∂y ∂z ∂ ∂x

where the viscous dissipation term Φ in expanded form is Φ = 2μ

∂u ∂x

2

+

∂v ∂y

2

+

∂w ∂z

2

∂u ∂v 2 +μ + ∂y ∂x

∂u ∂v ∂w 2 ∂w 2 ∂v ∂w ∂u 2 2 + + + + +μ +μ − μ ∂z ∂y ∂x ∂z 3 ∂x ∂y ∂z

(2.24)

Compressibility is usually associated with high speed flows (Ma > 0.3, where Ma is Mach number). However, for low Mach number, the flow can still be compressible. For example, strong wall heating or cooling can cause a sufficient change in density and the incompressible flow criterion breaks down even at low speeds. Less known is the situation encountered in some micro-devices where the pressure may strongly change due to viscous effects even though the speeds may not be high enough for the Mach number to go above the traditionally applied threshold of 0.3 as also discussed in Sect. 1.4.2. Corresponding to the changes in pressure would be changes in density that must be taken into account when writing the equations of motion. The compressibility effects therefore become significant at high temperature or at high speed and in micro-devices involving gas flow.

2.2.4 Incompressible Navier–Stokes Equations The incompressible assumption applies when the fluid density can be assumed to be a constant over the entire flow domain. This assumption amounts to taking

32

2 Microscale Flows

Dρ Dt

∂ρ = ∂ρ ∂t + ui ∂xi ≈ 0, which, in turn, implies that the spatial and temporal gradients are necessarily small enough to be neglected. Thus, along with questioning the nature of the density field, one also needs to question the nature of the density gradients. For example, even in low speed microscale flows (with Ma < 0.3), the spatial gradient of density becomes large enough to invalidate the incompressibility assumption. Similarly, a cavity with a diaphragm vibrating at a high frequency causes sufficiently large density fluctuations in the vicinity of the diaphragm. This density fluctuation disallows the flow to be treated as incompressible even when the flow speed is enforced to be small. ∂ρ ∂ρ Upon using the incompressible flow assumption (i.e. Dρ Dt = ∂t + ui ∂xi ≈ 0), the continuity equation (2.1) simplifies to

∂w ∂u ∂v + + =0 ∂x ∂y ∂z

(2.25)

Using continuity equation ∂uk /∂xk , the momentum equation (2.5) in tensorial form simplifies as ρ

∂uj ∂uj + ui ∂t ∂xi

=−

∂uj ∂p ∂ ∂ui μ + ρgj + + ∂xj ∂xi ∂xi ∂xj

(2.26)

When dynamic viscosity is taken as a constant, the momentum equation in tensorial form simplifies as ∂uj ∂uj ∂p ∂ ∂uj ∂ui =− + ρgj + ui +μ + ρ ∂t ∂xi ∂xj ∂xi ∂xi ∂xj ∂ 2 uj ∂ui ∂p ∂ + ρgj =− +μ +μ ∂xj ∂xi ∂xi ∂xj ∂xi ∂ 2 uj ∂uj ∂uj ∂p =− + ui ∴ρ +μ + ρgj (2.27) ∂t ∂xi ∂xj ∂xi ∂xi where we have used continuity equation (∂ui /∂xi = 0) in the last step. In expanded form, these equations can be written as x-momentum: ρ

∂u ∂u ∂u ∂u +u +v +w ∂t ∂x ∂y ∂z

=−

2 ∂ u ∂ 2u ∂ 2u ∂p + ρgx +μ + + ∂x ∂x 2 ∂y 2 ∂z2 (2.28)

2.2 Governing Equations for Fluid Flow

33

y-momentum: ρ

∂v ∂v ∂v ∂v +u +v +w ∂t ∂x ∂y ∂z

2 ∂ v ∂p ∂ 2v ∂ 2v +μ =− + 2 + 2 + ρgy ∂y ∂x 2 ∂y ∂z (2.29)

z-momentum: ρ

∂w ∂w ∂w ∂w +u +v +w ∂t ∂x ∂y ∂z

=−

2 ∂p ∂ w ∂ 2w ∂ 2w + ρgz + + +μ ∂z ∂x 2 ∂y 2 ∂z2 (2.30)

Energy equation: ρcp

∂T ∂T ∂T ∂T +u +v +w ∂t ∂x ∂y ∂z

=

∂T ∂ ∂T k + k ∂x ∂y ∂y ∂T ∂ k +Φ (2.31) + ∂z ∂z ∂ ∂x

where the viscous dissipation term Φ is 2 ∂u 2 ∂v ∂w 2 ∂u ∂v 2 Φ = 2μ + + + +μ ∂x ∂y ∂z ∂y ∂x ∂w 2 ∂v ∂w ∂u 2 + + +μ +μ ∂z ∂y ∂x ∂z

(2.32)

The incompressible flows are not affected by the second coefficient of viscosity μ2 because the second-last term in Eq. (2.4) drops off due to ∂uk /∂xk = 0 (for incompressible flows, from continuity equation (2.25)). Note the appearance of cp instead of cv in the energy equation (Eq. (2.31)). The energy equation for incompressible flows is actually the convection of enthalpy and not internal energy that is balanced by heat conduction and viscous dissipation [57]. Although density is no longer a variable, the number of variables (ui , p, and T ) does not reduce. Note that the equation of state (2.9) is not applicable to incompressible flows. To illustrate this point, consider any isothermal incompressible flow. Here, T is constant (because of isothermal assumption), ρ is constant (because of incompressible flow assumption), yet, in general, pressure varies in the flow, clearly showing that p = ρRT is not applicable to this (and all incompressible flow) situation.

34

2 Microscale Flows

2.3 Boundary Conditions The no-slip boundary condition is used to analyze the fluid flow at macroscale. As per this condition, the velocity of the fluid at the wall is equal to the velocity of the wall. This boundary condition has been shown to apply well to a wide range of fluid flow problems at the conventional scale. However, the no-slip condition does not apply to micro-scale flows, where, rather, slip occurs at the boundary. Initially, the concept of slip boundary condition was presented by Navier. According to his concept, the magnitude of the slip velocity is proportional to the magnitude of shear rate at the wall. This idea was found tenable by Kundt and Warburg who performed experiments on rarefied gas flow. They found that slipping of gas is indeed proportional to the velocity gradient close to the wall, i.e., du us = ζ (2.33) dy wall where ζ is a constant, called the coefficient of slip and represents a length.

2.3.1 Maxwell’s Slip Theory Maxwell explored the concept of slip further and quantified the slip length near the wall. During gas–wall interaction, the gas molecules are classified into two streams relative to the wall—approaching and receding (Fig. 2.1a). Maxwell assumed that all

(a)

(b)

(c)

Fig. 2.1 (a) Gas–wall interactions, (b) specular interactions, and (c) diffuse interactions

2.3 Boundary Conditions

35

interactions can be classified as either specular or diffuse and defined a parameter, the tangential momentum accommodation coefficient (TMAC) to quantify the interaction as σv =

τi − τr . τi − τw

(2.34)

In the above equation, τi and τr are the tangential momentum of the incoming and reflected molecules, respectively, and τw is the tangential momentum of the wall. Notice that τw = 0 for a stationary wall; τw is also the tangential momentum of the re-emitted molecules that may have been adsorbed to the surface. The subscript v in σ is to emphasize that we are referring to accommodation of velocity (later in Chap. 3, we will introduce thermal accommodation coefficient). During specular interaction (which is equivalent to reflection of light from a surface; Fig. 2.1b), the tangential momentum of the incoming and reflected molecules is equal (while the normal component of their momentum is reversed). That is, τi = τr and from Eq. (2.34) σv = 0 for specular interaction. On the other hand, during diffuse interaction, since the molecule can move in any random direction after interaction with the wall, on an average its momentum is zero for a stationary wall (and equal to τw otherwise). That is, τr = τw for diffuse interaction and σv = 1. The specular and diffuse interactions therefore nicely define the limits of σv ; all practical values are expected to lie within these two limits. (Note that σv can become greater than unity, if a molecule is reflected back in the direction from where it came; a situation called back scattering. However, this situation does not arise frequently.) In any flow, if f is the fraction of diffuse interaction out of the total interactions, then the effective value of TMAC is σv = f × σv,diffuse + (1 − f ) × σv,specular = f × 1 + (1 − f ) × 0 = f.

(2.35)

Therefore, σv also denotes the fraction of diffuse interactions. During these interactions, the wall experiences a viscous drag due to the difference between the tangential momentum of the approaching stream and that carried away by the receding stream. Maxwell assumed that the approaching stream exhibits a velocity gradient extending uniformly up to some value u0 , while us is the slip velocity at the wall as shown in Fig. 2.2. The tangential momentum brought by the approaching stream to the wall (per unit time per unit area) can then be analyzed Fig. 2.2 Actual velocity profile extended linearly up to the wall

36

2 Microscale Flows

Fig. 2.3 Momentum of the incoming and receding molecular stream

as [82]: The momentum that is brought to the wall is equal in magnitude to that which is transmitted elsewhere in the gas by the corresponding molecular stream. 0 Therefore, the incoming momentum is μ2 du dy per unit area per unit time, where μ is the coefficient of viscosity, and y is normal to the wall. From kinetic theory of gases, the number of molecules incident per unit wall area is 14 nv¯ for a gas with n molecules per unit volume having a mean speed of v. ¯ These molecules will carry a momentum of mus where m is the mass of the molecule. Therefore, the momentum carried away by the gas per unit area per unit time due to the slip velocity us in the direction of the flow is 14 nmv¯ times us . The incoming momenta is therefore the sum of these two components. The above argument only applies to diffusively reflected molecules, while specularly reflected molecules do not contribute towards drag as apparent from Fig. 2.1(b). Since drag on the wall being difference of incoming and outgoing momenta, we obtain σv

1 du0 1 du0 μ + ρus v¯ = μ 2 dy 4 dy

(2.36)

where we have utilized ρ = mn the density. Hence, the slip velocity us can be written as 2 − σv μ du0 . (2.37) us = 2 σv ρ v¯ dy Upon inserting v¯ = 2(2RT /π )1/2 , a standard result from kinetic theory, in the above equation, we obtain us =

du 2 − σv λ σv dy w

(2.38)

where we have additionally utilized expressions for mean free path λ = (μ/p) √ π RT /2 and the equation of state in arriving at Eq. (2.38).

2.3 Boundary Conditions

37

v Note that only one slip coefficient σv (or C1 = 2−σ σv ) along with the first derivative of velocity is involved in the above equation. The above model is therefore known as “first-order” slip boundary condition, as opposed to higher-order slip models introduced in the next two sections. The accuracy of the above model has been verified by different researchers in various ways, including a direct comparison with DSMC data by Gavasane et al. [59].

2.3.2 Derivation of Higher-Order Slip Boundary Condition In this section, we derive a second-order velocity slip model. Although the derivation is somewhat heuristic, it gives important clues about how the model derived above can be extended. Considering momentum crossing a control surface located at a distance of λ/2 away from the wall, where λ is the mean free path of the gas (Fig. 2.4). One-half of the momentum crossing this control surface is brought by gas at a distance of λ from the surface (where the gas velocity is uλ ), while the other half is brought by gas after interaction from the wall. The magnitude of this latter momentum comprises components due to both specular and diffuse reflections at the wall. The diffuse component contributes σv uw while the specular contributes (1 − σv )uλ . Therefore, we obtain uλ/2 =

σv uw + (1 − σv )uλ uλ + . 2 2

If we expand uλ in Taylor series about λ/2 λ ∂u λ2 ∂ 2 u uλ = uλ/2 + + + ··· 2 ∂y λ/2 8 ∂y 2 λ/2 and substitute the above series into Eq. (2.39), we can easily obtain λ 2 − σv ∂u λ2 2 − σv ∂ 2 u (uλ/2 − uw ) = + + ··· 2 σv ∂y λ/2 8 σv ∂y 2 λ/2

(2.39)

(2.40)

(2.41)

If we now move the control surface from a distance of λ/2 to close to the wall, we will obtain Fig. 2.4 Second-order slip boundary condition

38

2 Microscale Flows

2 2 − σv ∂u 2 2 − σv ∂ u (uλ/2 − uw ) = λ +λ + ··· σv ∂y λ/2 σv ∂y 2 λ/2

(2.42)

Replacing uλ/2 by ug and defining ug −uw = us , we obtain (upon neglecting higher order terms) 2 − σv us = σv

2 ∂u 2 ∂ u λ +λ ∂y λ/2 ∂y 2 λ/2

(2.43)

which is the general second-order slip model. Note that the above derivation is not rigorous; more accurate derivation can be done in a manner discussed in Sect. 6.5.

2.3.3 Alternate Slip Models In the past, several researchers have extended the basic slip model of Maxwell. These extensions have been presented here for easy reference. Diessler [40] realized that the Maxwell’s model is applicable only when velocity and temperature profiles are linear over a mean free path. However, the model needs to be generalized in case this requirement is violated. He undertook detailed kinetic theory based calculations to arrive at the following generalized second order slip model: 2 ∂ u 2 − σv ∂u 9 ∂ 2u ∂ 2u (2.44) us = ± λ − λ2 2 2 + 2 + 2 σv ∂y w 16 ∂y w ∂x w ∂z w where x, y, and z refer to the stream wise co-ordinate, transverse co-ordinate along the height and transverse co-ordinate along the width, respectively and subscript w signifies that the partial derivatives are evaluated at the wall. Neglecting the tangential derivative of velocity, the above model reduces to us = ±

2 − σv ∂u 9 ∂ 2u λ − λ2 2 σv ∂y w 8 ∂y w

(2.45)

The above model was generalized by Sreekanth [137] as 2 ∂u 2 ∂ u us = −C1 λ − C2 λ ∂y w ∂y 2 w

(2.46)

This is a widely employed model and we refer to it as Sreekanth’s slip model. Since these early improvements by Deissler [40] and Sreekanth [137], new slip models have been proposed. For example, Beskok and Karniadakis [81] modified the second-order slip model as

2.3 Boundary Conditions

39

2 − σv Kn ∂u us = ± σv 1 − bKn ∂ y¯ w

(2.47)

to aid in the numerical implementation of the boundary condition. In the above equation, y¯ denotes dimensionless wall-normal coordinate, b is an empirical parameter which depends on the geometric configuration. The value of b can be determined as b = u0 /2u0 with u0 and u0 being first- and second-order derivative of velocity normal to the surface for no-slip level of approximation. Xue and Fan [55] expressed the second-order boundary condition in the following form: us = ±

2 − σv ∂u tanh(Kn) σv ∂ y¯ w

(2.48)

Jie et al. [80] reported a hybrid boundary condition, given as us = ±

∂u 2 − σv Kn ∂p Kn + Re σv ∂ y¯ w 2 ∂ x¯ w

(2.49)

where x¯ and y¯ denote dimensionless streamwise and normal coordinates, respectively, and Re denotes the local Reynolds number. Lockerby et al. [89] proposed the linearized Maxwell–Burnett boundary condition as given below. ∂u ∂u 2 − σv 3 μ ∂T us = ± λ + + σv ∂y ∂x w 4 ρT ∂x w μ ∂ 2 T 2 − σv μ ∂ 2ρ − + λ 2 2 σv ρT ∂x∂y w ρ ∂x∂y ∂ 2u 3 γ −1 ∂ 2v ∂ 2u + P rλ2 (45γ − 61) 2 + (45γ − 49) − 12 2 16π γ ∂x∂y ∂x ∂y w (2.50) where u and v are the velocity components tangential and normal to the wall. The second order slip models are derived from the kinetic theory by considering the momentum transfer rate of gas molecules impinging on the wall [168]. Chen and Tian [33] proposed us = (1 − α)ug + αuw ; where α =

1 1 + 4ωKn/p

(2.51)

ug is the velocity of the gas at the wall, and the role of ω in the above equation is similar to that of TMAC. This last model accounts for a finite residence time of the molecule at the wall before getting reflected from the wall. In this book, we adopt the model of Sreekanth (Eq. (2.46)), which is usually the case, given its general nature and ease of use. The model is similar to Eq. (2.43)

40

2 Microscale Flows

derived in the previous section, in that Sreekanth’s model is of second-order in terms of Knudsen number and involves the first- and second derivatives of velocity. There are, however, some subtle differences between the two models: note that the derivatives are calculated at the control surface (introduced in the derivation) in Eq. (2.43), whereas the derivatives are calculated at the wall in Sreekanth’s model, making the latter model more convenient to apply. This change of location where the gradients are calculated leads to the appearance of two independent slip coefficients as opposed to a single coefficient in Eq. (2.43). We examine the value of the slip coefficients proposed in the literature in the next section.

2.3.4 Value of Slip Coefficients In this section, we compile the values of slip coefficients and tangential momentum accommodation coefficient in Tables 2.2 and 2.3, respectively, as reported by different researchers. The information will be useful while using the first-order and second-order slip models (Eqs. (2.38) and (2.46)) introduced earlier. The value of slip coefficient C1 is usually related to TMAC through the following equation C1 =

2 − σv . σv

(2.52)

Most of the values provided in this section were compiled by Agrawal and Prabhu [5], who documented these values along with the parameter range applicable and the associated uncertainty in the measurements. The values have been suitably updated by the present authors. Compiling such information allows the trend in the values with respect to Knudsen number to be better gleaned from the data (Fig. 2.5). As evident from Fig. 2.5a, the value of TMAC for monatomic gases varies over a small range even with six-decade change in Knudsen number. Further, the variation in the value of TMAC for the five monatomic gases is small (Table 2.4). Therefore, Agrawal and Prabhu [4] suggested to assign a constant value of 0.926 for all monatomic gases. However, the data for non-monatomic gases in Fig. 2.5b seems to show a decreasing trend with Knudsen number. The following empirical correlation was proposed by Agrawal and Prabhu [4] based on data in the figure: σv = 1 − log(1 + Kn0.7 )

(2.53)

The value of TMAC being close to unity suggests that the interaction of the gas molecules with the wall is mostly diffuse in nature, with less than 8% interaction being specular for monatomic gases (see Eq. (2.35)) irrespective of the amount of rarefaction of the gas. Note that the value of 0.926 for monatomic gases applies to most surface materials and temperatures above the room temperature. The overly diffuse nature of the interaction suggests that the momentum of the reflected molecule is almost independent of the momentum of the incident molecule;

2.3 Boundary Conditions

41

Table 2.2 The values of TMAC in the slip and transition regimes as reported by different authors (updated from Agrawal and Prabhu [5]) Method Source Flow through Sreekanth [137] capillary or microchannel Colin et al. [35]

Spinning rotor gauge

Rotating cylinder Unsteady method

Gas Nitrogen

Kn range 0.007–0.03 0.03–0.13 0.13–0.237 Helium 0.029–0.22 Nitrogen 0.002–0.03 0.005–0.09 Ewart et al. [54] Helium 0.009–0.309 Argon 0.003–0.302 Nitrogen 0.003–0.291 Maurer et al. [93] Helium 0.06–0.8 Nitrogen 0.002–0.59 Ewart et al. [53] Helium 0.03–0.3 Arkilic et al. [12] Argon 0.1–0.41 Nitrogen 0.1–0.34 CO2 0.1–0.44 Jang and Wereley [79] Air 0.0017 Yamaguchi et al. [171] Argon 0.05–0.3 Nitrogen 0.05–0.3 Oxygen 0.05–0.3 Vadiraj et al. [72] Argon 0.07–1.2 Nitrogen 0.072–0.11 Oxygen 0.072–1.13 Tekasakul et al. [145] Helium 4.64 × 10−3 –5.83 × 10−1 and Bentz et al. [19] Argon 1.67 × 10−3 –2.10 × 10−1 Krypton 1.30 × 10−3 –1.63 × 10−1 Nitrogen 1.63 × 10−3 –2.58 × 10−2 Methane 1.30 × 10−3 –2.15 × 10−2 Agrawal and Prabhu [4] Dry air 0.1–8.3 (data of Kuhlthau 1949) 0.04–0.08 Suetin et al. [143] Helium 0.001–2 2.8–886 Neon 0.001–2 2.8–886 Argon 0.001–2 2.8–886 Porodnov et al. [117] Krypton 4.9 × 10−4 –9.6 × 10−3 Xenon 3.6 × 10−4 –7.0 × 10−3 Hydrogen 1.1 × 10−3 –2.2 × 10−2 Deuterium 1.1 × 10−3 –2.2 × 10−2 Nitrogen 6.0 × 10−4 –1.2 × 10−2 CO2 4.0 × 10−4 –7.8 × 10−3

σv 1 0.9317 0.9317 0.93 1 0.93 0.914 ± 0.009 0.871 ± 0.017 0.908 ± 0.041 0.91 ± 0.03 0.87 ± 0.03 0.910 ± 0.004 0.8 ± 0.1 0.83 ± 0.05 0.88 ± 0.06 0.85 0.869 ± 0.03 0.851 ± 0.02 0.851 ± 0.02 0.957 ± 0.04 0.955 ± 0.03 0.866 ± 0.04 0.9410 ± 0.0328 0.8968 ± 0.0354 0.8706 ± 0.0578 0.86 ± 0.027 1.04 ± 0.022 0.74 0.94 0.895 ± 0.004 0.935 ± 0.004 0.865 ± 0.004 0.929 ± 0.003 0.927 ± 0.028 0.975 ± 0.006 0.995 ± 0.026 1.010 ± 0.040 0.957 ± 0.015 0.934 ± 0.006 0.925 ± 0.014 0.993 ± 0.009

42

2 Microscale Flows

Table 2.3 The values of slip coefficients as reported by different authors (updated from Agrawal and Prabhu [5]) Source Maxwell [94] Schamberg [125] Albertoni et al. [9] Deissler [40] Cercignani [29] Chapman and Cowling [31] Hsia and Domoto [75] Mitsuya [97] Sreekanth [137]

Gas – – – – – – – – Nitrogen

Ewart et al. [54]

Helium Argon Nitrogen Helium Nitrogen Helium Argon Nitrogen Oxygen Argon Nitrogen Oxygen

Maurer et al. [93] Ewart et al. [53] Yamaguchi et al. [171]

Vadiraj et al. [72]

Kn range – – – – – – – – 0.007–0.03 0.03–0.13 0.13–0.237 0.009–0.309 0.003–0.302 0.003–0.291 0.06–0.8 0.002–0.59 0.03–0.7 0.05–0.3 0.05–0.3 0.05–0.3 0.07–1.2 0.072–0.11 0.072–1.13

C1 1 1 1.1466 1 1.1466 ≈1 1 1 1 1.1466 1.1466 1.052 ± 0.020 1.147 ± 0.042 1.066 ± 0.088 1.2 ± 0.05 1.3 ± 0.05 1.26 ± 0.022 1.30 ± 0.09 1.35 ± 0.06 1.35 ± 0.06 1.089 ± 0.08 1.093 ± 0.07 1.301 ± 0.09

C2 0 5π/12 0 1.6875 0.9756 ≈0.5 0.5 2/9 0 0 0.14 0.148 ± 0.014 0.294 ± 0.029 0.231 ± 0.057 0.23 ± 0.1 0.26 ± 0.1 0.17 ± 0.02 0.063 ± 0.018 0.031 ± 0.005 0.028 ± 0.004 0.140 ± 0.007 0.151 ± 0.003 0.155 ± 0.004

The first eight entries in the table are theoretical values, while the remaining are experimental

suggesting that gas molecules have sufficient time to interact with the wall. This can arise if the molecule which has been directed to the surface and the molecule coming out of the surface are not the same, and therefore momentum of the incoming molecule and that of the outgoing molecule are not correlated. Alternatively, the incident molecule may undergo multiple reflections at the surface, owing to the roughness of the wall; therefore, the momentum of the molecule effectively undergoes a random change in magnitude and direction. Based on available experimental data, Agrawal and Prabhu [5] suggested the following value of slip coefficients: C1 = 1.1465 and C2 = 0.164. It is suggested to use these values for all (monatomic and non-monatomic) gases in the range 0.002 ≤ Kn ≤ 0.8. A dependence of slip coefficient on the nature of the gas suggests that the flow behavior will additionally depend on the gas that we are working with. So does it mean that, for fully developed flow in a tube, the friction factor depends on additional parameters (specific to the gas) besides the Reynolds and Knudsen numbers? Of course, such a dependence will substantially complicate the analysis

2.3 Boundary Conditions

43 Tekasul (Helium) Tekasul (Argon) Tekasul (Krypton) Blanchard and Ligrani (Helium) Maurer (Helium) Colin (Helium) Suetin (Helium) Suetin (Neon) Suetin (Argon) Ewart (Channel; Helium) Ewart (Helium) Ewart (Argon) Porodnov (Krypton) Porodnov (Xenon)

1.05

TMAC

1

0.95

0.9

0.85 10

-3

10

-2

10

-1

10

0

10

1

10

2

10

3

Knudsen number

(a) Buckley and Loyalka (data of Millikan) Tekasul (Nitrogen) Tekasul (Methane) Agrawal and Prabhu (data of Kuhlthau) Blanchard and Ligrani (Air) Sreekanth (Nitrogen) Maurer (Nitrogen) Chiang Colin (Nitrogen) Arkilic (Nitrogen) Arkilic (CO2) Ewart (Nitrogen) Porodnov (Hydrogen) Porodnov (Deuterium) Porodnov (Nitrogen) Porodnov (CO2) Correlation (Eq. 13)

1.2

1.1

TMAC

1

0.9

0.8

0.7 10

-3

10

-2

10

-1

10

0

10

1

Knudsen number

(b) Fig. 2.5 TMAC versus Knudsen number as reported by Agrawal and Prabhu [5]. (a) Monatomic gases. (b) Non-monatomic gases

44

2 Microscale Flows

Table 2.4 Mean TMAC from all reported values for monatomic gases (nobs indicates the number of observations on which the mean value is based)

Gas σv nobs

Helium 0.923 14

Neon 0.912 6

Argon 0.896 16

Krypton 0.954 8

Xenon 0.950 5

These values appear valid for commonly employed surface materials and the entire range of Knudsen number (from Agrawal and Prabhu [5], with the value of Argon updated)

of slip flows. This question warrants a clear answer than what can be gleaned from the available literature. On the other hand, it also opens the exciting possibility of making some simple measurements and determining the nature of the gas involved; something which can be useful to the sensors community, interested, for example, in detecting minute amounts of explosives or harmful bio-gases.

2.4 Some Exact Solutions In this section, we will consider some problems from the slip flow regime for which analytical solution is possible. A discussion of the results can provide insights into the slip flow behavior. These cases also serve to be useful for initial benchmarking of numerical codes and experimental setups. The theory is also useful (and has been employed) for obtaining the value of TMAC reported in the previous section.

2.4.1 Couette Flow Couette flow is perhaps the simplest of the flow problems. Here, the upper plate is moving with velocity U while the lower plate is stationary; the distance between the plates being H (Fig. 2.6). We will assume that the flow between the plates is laminar, incompressible, steady, and fully developed, with no variation along the z-direction. The pressure is assumed to be constant along the streamwise direction (dp/dx = 0). The only difference with respect to the standard problem is that there is slip at the walls [110]. After applying the aforesaid assumptions, the x-momentum equation (2.13) reduces to d 2u =0 dy 2

(2.54)

u = A1 y + A2

(2.55)

On integrating this equation, we have

2.4 Some Exact Solutions

45

Fig. 2.6 Schematic diagram of Couette flow with upper plate moving with velocity U in xdirection and lower plate is stationary

where A1 and A2 are the two constants of integration. First-order slip (Eq. (2.38)) gives the following boundary conditions: 2 − σv ∂u us − uw = λ σv ∂y y=0 and y=H

(2.56)

With uw = 0 on the lower wall (y∂u= 0), uw = U on the upper wall (y = H ), and keeping in mind that ∂u ∂y y=0 = − ∂y y=H , we obtain 2 − σv λU U σv and A2 = . A1 = 2 − σv 2 − σv H + 2λ H + 2λ σv σv

Upon substituting the values of A1 and A2 in Eq. (2.55), we obtain 2 − σv λU Uy + σv u= 2 − σv H + 2λ σv

(2.57)

as the required velocity profile. We can express the above velocity profile in dimensionless form as 2 − σv y + Kn u H σv = (2.58) 2 − σv U 1 + 2Kn σv

46

2 Microscale Flows 1

1 s v = 0.2

sv = 1

0.6

0.6 Kn = 0.00 Kn = 0.02 Kn = 0.04 Kn = 0.06 Kn = 0.08 Kn = 0.10

0.4 0.2 0

y/H

0.8

y/H

0.8

0

0.2

0.4

0.6

0.8

Kn = 0.00 Kn = 0.02 Kn = 0.04 Kn = 0.06 Kn = 0.08 Kn = 0.10

0.4 0.2 0

1

0

0.2

0.4

0.6

u/U

u/U

(a)

(b)

0.8

1

Fig. 2.7 Velocity distribution across the channel height for different values of Knudsen number for (a) σv = 0.2 (b) σv = 1 1

1 Kn = 0.1

0.8

0.8

0.6

0.6 y/H

y/H

Kn = 0.02

0.4

sv sv sv sv sv

0.2 0

0

0.2

0.4

0.6

0.8

= 0.2 = 0.4 = 0.6 = 0.8 = 1.0

0.4

sv sv sv sv sv

0.2

1

0

0

0.2

0.4

0.6

u/U

u/U

(a)

(b)

0.8

= 0.2 = 0.4 = 0.6 = 0.8 = 1.0 1

Fig. 2.8 Velocity distribution across the channel height for different values of TMAC for (a) Kn = 0.02 (b) Kn = 0.1

where Kn = λ/H . The solution given by Eq. (2.58) indicates not only a velocity slip at the wall but also a correction to the slope of profile for the Couette flow in presence of slip at the wall. The velocity profile is a function of Knudsen number and TMAC as evident from Eq. (2.58). Notice that there is an increase in the slip velocity with an increase in the Knudsen number for a fixed value of tangential momentum accommodation coefficient (Fig. 2.7). Similarly, the slip velocity increases with a decrease in the value of tangential momentum accommodation coefficient (or the gas-wall interaction becoming more specular) as shown in Fig. 2.8.

2.4 Some Exact Solutions

47

The volume flow rate can be calculated using Eq. (2.57) as H ˙ udy Q=

(2.59)

0

which yields ˙ Q 1 = . UH 2

(2.60)

Notice that the volume flow rate depends only on the speed of the plate and the distance between the plates, and is independent of both the Knudsen number and TMAC. The slip flow solution therefore gives the same volume flow rate as that of the no-slip case. This occurs because the increase in velocity owing to slip in the lower-half is exactly compensated by a reduction in velocity in the upper-half; as evident from Fig. 2.7b. We next compute the skin friction coefficient from Cf =

τw (1/2)ρU 2

(2.61)

where τw (= μdu/dy) is the wall shear stress. For the case of slip at the wall we obtain ⎞ ⎛ Cf,slip =

2 ⎜ ⎜ Re ⎝

⎟ 1 ⎟ ⎠ 2 − σv Kn 1+2 σv

(2.62)

where Re = ρU H /μ. For the case of no-slip at the wall (Kn → 0), the above equation reduces to Cf,no-slip =

2 Re

(2.63)

Hence, Cf,slip = Cf,no-slip

1 2 − σv Kn 1+2 σv

(2.64)

Equation (2.62) can alternatively be written as ⎛ Cf,slip Ma =

⎞

⎟ ⎜ 1 2 ⎟ ⎜ √ Re ⎝ 2 − σv Ma ⎠ 1 + 2π γ Ma σv Re

(2.65)

48

2 Microscale Flows

1 0.6

Cf,slip Ma

0.4 0.2 0.1

sv = 1 s v = 0.9 s v = 0.7 s v = 0.5 s v = 0.3

0.06 0.04 0.02 0.01 0.1

sv = 0.1

0.2

0.4 0.6 1

2

4 6

Re/Ma

10

20

40 60 100

Fig. 2.9 Variation of Cf,slip Ma with respect to Re/Ma

where we have additionally used the relation: Kn =

π γ Ma 2 Re

(2.66)

In Fig. 2.9, Cf,slip Ma has been plotted versus Re/Ma with σv as the parameter. It is clear that the quantity Cf,slip Ma decreases with the increase in parameter Re/Ma. Equation (2.65) can therefore be used for determining the value of TMAC from measurement of skin friction coefficient and knowing the values of Reynolds and Mach numbers.

2.4.2 Flow in a Microchannel Several researchers have attempted to obtain analytical solution of gaseous slip flow in a microchannel. The problem is much more complicated than flow in a conventional channel because of both compressibility effect and slip at the wall. The two most important approaches to solve the problem are provided in this and the next subsection. The first approach, due to Arkilic et al. [11], involves solution of the Navier–Stokes equations, whereas the integral form of the momentum equation is solved in the other approach. A comparison of the two approaches is provided at the end of the next subsection.

2.4 Some Exact Solutions

49

Fig. 2.10 Schematic of flow between two parallel plates, driven by a pressure gradient

L 2.4.2.1

Perturbation of Navier–Stokes Equations

Following Arkilic et al. [11], we consider steady isothermal flow through a microchannel of large depth such that the two-dimensional flow assumption applies (Fig. 2.10). This assumption is valid for microchannels of high aspect ratio, and allows the w component of velocity and any variation along the z-direction to be neglected. The isothermal flow assumption simplifies the analysis, as it obviates the need for considering the energy equation. Isothermal condition applies if effects such as heat generated due to viscous dissipation and cooling due to the expansion of the gas are properly mitigated, for example, through the use of a good heat conducting material of the microchannel. Under the aforementioned assumptions, the continuity (2.22) and Navier–Stokes equations (2.27), (2.28) in the absence of body force terms reduce to ∂(ρu) ∂(ρv) + = 0, ∂x ∂y 2 ∂ u ∂ 2u 1 ∂ 2u ∂u ∂p ∂u ∂ 2v +v +μ =− , ρ u + + + ∂x ∂y ∂x 3 ∂x 2 ∂x∂y ∂x 2 ∂y 2

∂v ∂v ρ u +v ∂x ∂y

2 ∂ v ∂p ∂ 2v 1 ∂ 2v ∂ 2u +μ =− . + 2+ + ∂y 3 ∂y 2 ∂x∂y ∂x 2 ∂y

(2.67)

(2.68)

(2.69)

Here, u and v are the streamwise and wall-normal components of velocity, i.e. along the x- and y-directions, respectively. We non-dimensionalize the above equations using the following scales (the non-dimensional variables are represented by a ˜ symbol), x˜ =

x y u v p ρ ; y˜ = ; u˜ = ; v˜ = ; p˜ = ; ρ˜ = L H u¯ u¯ p¯ ρ¯

(2.70)

50

2 Microscale Flows

where L and H are the microchannel length and height, respectively, u¯ is the velocity at the centerline and at the exit of the microchannel, p and ρ are the exit pressure and density values, respectively. Invoking isothermal assumption and introducing equation of state for ρ˜ as p˜ = ρRT ˜ , we get the above equations in the non-dimensional form as H ∂(p˜ u) ˜ ∂(p˜ v) ˜ + = 0, L ∂ x˜ ∂ y˜

2 2 ρ¯ uH ¯ H ∂ u˜ H pH ¯ 2 ∂ p˜ ∂ u˜ ∂ u˜ ∂ 2 u˜ + v˜ + ρ˜ u˜ =− + μ L ∂ x˜ ∂ y˜ uLμ ¯ ∂ x˜ L ∂ x˜ 2 ∂ y˜ 2 2 H ∂ v˜ 1 H 2 ∂ 2 u˜ , + + 2 3 L ∂ x˜ L ∂ x∂ ˜ y˜

2 2 ρ¯ uH ¯ H ∂ v˜ H pH ¯ ∂ p˜ ∂ v˜ ∂ v˜ ∂ 2 v˜ + v˜ + ρ˜ u˜ =− + μ L ∂ x˜ ∂ y˜ μu¯ ∂ y˜ L ∂ x˜ 2 ∂ y˜ 2 2 2 H ∂ u˜ 1 ∂ v˜ . + + 2 3 ∂ y˜ L ∂ x∂ ˜ y˜

Introducing the ratio of the channel height to its length as ε = H /L (considered to be small),√the Reynolds number, Re = ρ¯ uH ¯ /μ and the exit Mach number, Ma = u/c ¯ (c = γ RT ), ε

∂(p˜ u) ˜ ∂(p˜ v) ˜ + = 0, ∂ x˜ ∂ y˜

(2.71)

2˜ ∂ u˜ εRe ∂ p˜ ∂ 2 u˜ 1 2 ∂ 2 u˜ ∂ 2 v˜ ∂ u˜ 2∂ u + v˜ + ε =− ε , + + + ε Re ρ˜ εu˜ ∂ x˜ ∂ y˜ 3 ∂ x∂ ˜ y˜ γ Ma 2 ∂ x˜ ∂ x˜ 2 ∂ y˜ 2 ∂ x˜ 2 (2.72) 2 ∂ v˜ ∂ v˜ Re ∂ p˜ ∂ 2 v˜ 1 ∂ 2 v˜ ∂ 2 u˜ 2 ∂ v˜ Re ρ˜ εu˜ + v˜ +ε =− . + 2+ +ε ∂ x˜ ∂ y˜ 3 ∂ y˜ 2 ∂ x∂ ˜ y˜ γ Ma 2 ∂ y˜ ∂ x˜ 2 ∂ y˜ (2.73) Note that the numerator of the Reynolds number is equal to the mass flow rate per unit depth through the microchannel, and the viscosity in the denominator is constant because of the isothermal assumption; therefore, the Reynolds number does not vary with position in the microchannel. The boundary conditions are such that the lateral velocity v˜ vanishes at the solid walls, and that the streamwise velocity u˜ is equal to the slip velocity. Therefore, Eq. (2.38) becomes: ∂ u˜ u| ˜ wall = C1 Kn . (2.74) ∂ y˜ wall

2.4 Some Exact Solutions

51

Additional boundary conditions required for the complete solution will be introduced later. For flow in conventional channels, it is assumed that ∂()/∂x = 0 (other than for pressure), which allows simplification of the equations. However, this cannot be done here, as the velocity can increase with x given the compressible nature of the flow. This behavior has led to a complex set of governing equations (2.71)–(2.73). In order to simplify the governing equations, we apply the perturbation technique and expand u, ˜ v, ˜ and p˜ in powers of ε (which is already assumed to be small): u˜ = u˜ o + εu˜ 1 + ε2 u˜ 2 + · · ·

(2.75)

v˜ = εv˜1 + ε2 v˜2 + · · ·

(2.76)

p˜ = p˜ o + εp˜ 1 + ε2 p˜ 2 + · · ·

(2.77)

Note that in order to satisfy the law of conservation of mass, Eq. (2.71) dictates that the highest order terms in u˜ and v˜ be of the same order in ε, i.e. v˜ must be of O(ε). Thus, the zeroth-order wall-normal velocity is not included in Eq. (2.76). Upon substituting the above series into the governing equations leads to a series of equations in power of ε, which are relatively easier to solve than the original governing equations. Notice that the coefficients of certain terms in Eqs. (2.71)–(2.73) depend on the values of Re and Ma, which can independently be O(ε), O(1), or O(1/ε), respectively, leading to nine different cases. Since Kn is related to Ma and Re through Eq. (2.66), its order gets fixed upon choosing the orders of Re and Ma. Here, we will consider Ma = O(ε) and Re = O(ε), which implies that Kn = O(1); this case can represent typical microscale flows well. Thus, Re/Ma 2 involved in the momentum equations is O(1/ε). It is apparent that the pressure term in Eq. (2.73) being of O(1/ε) is much larger than all the other terms in the equation. Therefore, Eq. (2.73) simplifies to ∂ p˜ =0 ∂ y˜

(2.78)

which leads to p˜ = p˜ o (x). ˜ Using this result, the x-momentum ˜ equation (2.72), at O(1), can be simplified as εRe d p˜ o ∂ 2 u˜ o = γ Ma 2 d x˜ ∂ y˜ 2

(2.79)

Notice that Eqs. (2.78), (2.79) are similar to what one would obtain for a conventional channel flow. In particular, the absence of inertial terms in Eq. (2.79) is underscored. However, in the present case, the streamwise velocity is dependent upon both the wall-normal (y) ˜ and streamwise (x) ˜ directions. This essential difference is due to the fact that the streamwise velocity in a microchannel keeps

52

2 Microscale Flows

increasing, leading to ∂ u/∂ ˜ x˜ > 0. The implication of this continuous flow acceleration is that the pressure variation is no longer linear (although it is still only a function of streamwise coordinate) and the lateral velocity component is no longer zero (although it is a few orders of magnitude smaller than the streamwise velocity component; see Eq. (2.76)). The x-momentum ˜ equation (Eq. (2.79)) can now be integrated twice with respect to y. ˜ The two constants of integration can be obtained by utilizing the symmetry condition (∂ u˜ o /∂ y˜ = 0 at the microchannel centerline) and the slip-flow boundary condition (Eq. (2.74)), to obtain u˜ o (x, ˜ y) ˜ =−

εRe d p˜ o 8γ Ma 2 d x˜

Kn . 1 − 4y˜ 2 + 4C1 p˜ o

(2.80)

The last term in the above solution goes to zero as Kn → 0; clearly, slip at the wall affects the velocity profile. The velocity profile in dimensional form is uo (x, y) = −

y2 λ(x) p 1 H 2 dpo (x) 1 − 4 2 + 4C1 8 μ dx H po (x) H

(2.81)

which shows the dependence of various terms involved in the above equation on either x or y. It is apparent that although the velocity profile is parabolic, the coefficients are a function of the streamwise coordinate. The solution for u˜ o can now be substituted into the continuity equation (2.71) to obtain v˜1 as 4 3 εRe 1 1 d 2 (p˜ o )2 d 2 p˜ o y˜ − y˜ + 4C1 Kny˜ . (2.82) ˜ y) ˜ = v˜1 (x, 3 8γ Ma 2 p˜ o 2 d x˜ 2 d x˜ 2 Since the wall-normal velocity must vanish at the wall (y˜ = ±1/2), the term in the bracket should go to zero, leading (finally!) to the following equation for pressure: d 2 (p˜ o )2 d 2 p˜ o + 12C Kn = 0. 1 d x˜ 2 d x˜ 2

(2.83)

The solution of Eq. (2.83) can be found by integrating the equation twice. The boundary conditions are p˜ o = 1 at the outlet (since pressure has been normalized by the outlet pressure) and the pressure ratio across the microchannel is specified as, say, P (i.e. p˜ = P at x = 0). The quadratic equation obtained upon integration can be easily solved to arrive at the following expression for the zeroth-order pressure distribution: p˜ o (x) ˜ = −6C1 Kn + (6C1 Kn)2 + (1 + 12C1 Kn)x˜ + (P2 + 12C1 KnP)(1 − x). ˜ (2.84) It is apparent from the above solution that the variation of pressure in a microchannel is non-linear, as already suspected.

2.4 Some Exact Solutions

53

Note that only the first-order term in the perturbation series (Eqs. (2.80)–(2.83)) has been explicitly obtained here. However, as will become apparent, this suffices for our purpose. Having obtained the complete solution, the derived quantities like mass flow rate and friction factor can be obtained. The dimensional mass flow rate through the microchannel is given by m ˙ slip

H 3 wPo2 2 P − 1 + 12C1 Kn(P − 1) = 24μLRT

(2.85)

where Po is the (dimensional) outlet pressure, and w is the micro channel width. Notice the particularly strong dependence of the mass flow rate on the height of the microchannel H , which is qualitatively similar to that observed in conventional channels. To see the effect of slip on mass flow rate, we take the ratio of the above obtained mass flow rate with mass flow rate (m ˙ no-slip ) obtained by putting Kn = 0 in the above expression. This yields m ˙ slip 12C1 Kn . =1+ m ˙ no-slip P+1

(2.86)

It is apparent from the above equation that the mass flow rate through the microchannel is enhanced due to wall slip. The amount of enhancement is larger at higher Knudsen number, larger value of slip coefficient (or more specular reflection at the wall), and smaller value of pressure ratio.

2.4.2.2

Integral Approach

In this section, we follow Dongari et al. [44] and obtain an analytical solution of gaseous slip flow in a long channel, by solving the integral form of the momentum equation. This methodology results in a more generalized solution due to the following reasons: a second-order slip boundary condition is applied thereby making the solution valid for a larger range of Knudsen number; inertial terms are retained to reveal a dependence on the Reynolds number. Just like the previous section, the assumptions of steady isothermal fully developed flow are applicable in the present case as well. Moreover, entry and exit effects along with viscous compressive stresses are assumed to be negligible. These assumptions allow us to safely assume a parabolic velocity profile which takes the form u(x, y) = A(x)y 2 + B(x)y + C(x)

(2.87)

Note that only the shape of the velocity profile is assumed and that the velocity field is still two-dimensional; also, the density (or pressure) field is a function of the streamwise coordinate only, as is demonstrated further.

54

2 Microscale Flows

For a microchannel of height H , it is worthwhile to note the following points at this stage: 1. The hydraulic diameter Dh = 2H is chosen as the characteristic length scale. λ Accordingly, the Knudsen and Reynolds numbers are Kn = 2H and Re = ρ u(2H ¯ ) where λ is the mean free path length and u¯ = u(x) ¯ is the cross-section μ averaged streamwise velocity. 2. The Knudsen number is directly proportional to the rarefaction or inversely proportional to the pressure (p), i.e. Kn ∝ (1/p) ⇒ p × Kn = constant. 3. The ideal gas equation is employed, i.e. p = ρRT , where R is the gas constant. For a fully developed flow, the maximum velocity for any streamwise location is evaluated to be the centerline velocity. Therefore, using the condition that ∂u ∂y |y= H2 = 0, we can evaluate B(x) = −AH . Further, (2.46) and noting at y = H , using Eq. λ that Kn = 2H , we get C(x) = (−AH 2 ) 2C1 Kn + 8C2 Kn2 . Lastly, keeping in H ˙ 1 mind that the flow fields are unit-depth quantities, we obtain u¯ = Q udy = A = H 0 ¯ we can (−AH 2 ) 16 + 2C1 Kn + 8C2 Kn2 . Thus, making use of B(x), C(x) and u, obtain the velocity profile as u(x, y) = u(x) ¯

y/H − (y/H )2 + 2C1 Kn + 8C2 Kn2 1/6 + 2C1 Kn + 8C2 Kn2

(2.88)

˙ f low Recalling that Re is constant in the streamwise direction (Re = m ˙ f low /μ, m is constant because of the steady state assumption and μ is constant due to the isothermal assumption) and using the equation of state, we can write u¯ as u(x) ¯ =

ReμRT p(x)Dh

(2.89)

Therefore Eq. (2.88) can be written as ReμRT y/H − (y/H )2 + 2C1 Kn(x) + 8C2 Kn2 (x) . (2.90) u(x, y) = p(x)Dh 1/6 + 2C1 Kn(x) + 8C2 Kn2 (x) Knowing the velocity profile, the shear stress at the wall can be obtained from Eq. (2.88) as τw = −μ

∂u 3Reμ2 . = ∂y y=H ρH 2 (1 + 12C1 Kn + 48C2 Kn2 )

(2.91)

In order to evaluate the pressure field, we make use of the integral formulation of the momentum balance in the streamwise direction. Consider a control volume of axial length δx located at a distance x from the leading edge of the microchannel. The forces acting on this elemental control volume are shown in Fig. 2.11. The momentum equation for this finite elemental volume spanning the cross section of the microchannel can be obtained using the integral momentum balance as

2.4 Some Exact Solutions

55

Fig. 2.11 Control volume for the integral momentum balance. M(x) = A ρu2 dA, M(x) is the momentum flux, A is the cross-sectional area into the plane of the page, p(x) is pressure and τw is wall shear stress

A p(x) − p(x + δx) + (−2τw × Alateral ) = M(x + δx) − M(x)

A dp + · · · − 2τw δx ∴ A p(x) − p(x) + δx dx H

dM = M(x) + δx + · · · − M(x) dx dp A dM(x) − 2 τw δx = δx + O(δx 2 ) dx H dx A 2 ρu dA + O(δx) ∴ −Adp − 2 τw dx = d H A ∴ −Aδx

(2.92)

In arriving at Eq. (2.92), we have used Taylor series expansion in the second step and finally cast the equation in a variable separable form. To proceed further, we make the approximation that since δx is small, we can neglect the O(δx) terms. Thus, the integral formulation of the momentum equation for a rectangular microchannel is 2Aτw 2 − Adp − dx = d ρu dA (2.93) H A To solve for p, we first make use of Eq. (2.88) to simplify the integral appearing in Eq. (2.93) as I=

H ρu dA = ρ

u2 dy = ρ u¯ 2 H χ

2

A

(2.94)

0

where 1 8 2 + C1 Kn + 4C12 Kn2 + C2 Kn2 + 64C22 Kn4 + 32C1 C2 Kn3 30 3 3 χ= (1/6 + 2C1 Kn + 8C2 Kn2 )2 (2.95)

56

2 Microscale Flows

Substituting τw from Eq. (2.91) and I from Eq. (2.94) into Eq. (2.93), we get 3Reμ2 2 dx = d ρ u ¯ H χ − H dp − 2 ρH 2 (1 + 12C1 Kn + 48C2 Kn2 )

(2.96)

where we have made use of the fact that A = H . Using Kn p = constant, we get Kn = (Kn0 × p0 )/p where Kn0 and p0 are known at some reference position x0 . Further, we have already established in Eq. (2.89) that u¯ = u(x). ¯ Thus, Eq. (2.96) can be expressed in a variable separable form to obtain the pressure field p(x) as

p p0

2

p p − 1 + 24C1 Kn0 − 1 + 96C2 Kn20 log p0 p0 p0 + 2Re2 βχ 12C1 Kn0 −1 p p0 2 p x − x0 2 − 1 − log + 24C2 Kn0 = −96Reβ p p0 Dh

(2.97)

where χ is given in Eq. (2.95) and β = (μ2 RT )/(p02 Dh2 ). Note the dependence of pressure on both Reynolds and Knudsen numbers from the above expression, with a more dominant effect of the latter non-dimensional number. The lateral component of velocity (v) can now be obtained by substituting the expressions for ρ(x) and u(x, y) into the continuity equation, which yields 1 ReμRT dp 12C1 Kn(x) + 96C2 Kn2 (x) 2 2 1 + 12C1 Kn(x) + 48C2 Kn (x) p (x) p(x)Dh dx 3y/H − 2(y/H )2 + 12C1 Kn(x) + 48C2 Kn2 (x) ×y 1 − 1 + 12C1 Kn(x) + 48C2 Kn2 (x)

v(x, y) =

(2.98) The mass flow rate per unit depth through the microchannel can be obtained from the Reynolds number as m ˙ =

Reμ . 2

(2.99)

However, if the pressure ratio across the microchannel (P) is specified for a given gas and microchannel, the mass flow rate per unit depth flowing through the microchannel can be determined from

2.4 Some Exact Solutions

57

m ˙ =

μ −a2 + 2

a22 − 4a1 a3

2a1

(2.100)

where a1 , a2 , and a3 are given as a1 = 2βχ 12C1 Kn0 (P − 1) + 24C2 Kn20 (P2 − 1) + log P , a2 =

48βL H

a3 = (1/P2 − 1) + 24C1 Kn0 (1/P − 1) − 96C2 Kn20 log P The Reynolds number can subsequently be calculated from Eq. (2.99).

Note on Use of the Equations A note on the use of the above solution has been added here for convenience. Given is a microchannel of known dimensions (verify that it is long and two-dimensional) and a known gas flows through it at a given temperature (assumed to be constant). The interest is in one of the two cases: (1) finding out the mass flow rate for given inlet and outlet pressures (pressure driven system), or (2) finding out the pressure drop for a given mass flow rate with either inlet or outlet pressure specified (mass flow driven system). Of further interest would be finding the complete velocity and pressure fields in the microchannel. First we pick values of slip coefficients: C1 = 1.1465 and C2 = 0.164 (Sect. 2.3.4). The solution proceeds by solving the equations in the following order: Case 1 (Pressure Driven System) Calculate outlet Kn (since outlet pressure is known) and χ from Eq. (2.98). Find the mass flow rate from Eq. (2.100) (since P is known). Now calculate Re from Eq. (2.99). Take different values of pressure (between inlet and outlet) and find the corresponding x from Eq. (2.97) to obtain the streamwise variation of pressure in the microchannel. Finally, find u(x, y) and v(x, y) from Eqs. (2.90) and (2.98), respectively, since all the required variables are known by this stage. Case 2 (Mass Flow Driven System with Specified Outlet Pressure) Calculate outlet Kn (since outlet pressure is known) and χ from Eq. (2.98). Calculate Re from Eq. (2.99) since m ˙ is known in this case. Guess different values of pressure (greater than the outlet pressure) and find the corresponding x from Eq. (2.97). Verify that the obtained value of x lies within the bounds of the microchannel. This gives the streamwise variation of pressure in the microchannel. Finally, find u(x, y) and v(x, y) from Eqs. (2.90) and (2.98), respectively, since all the required variables are known by this stage.

58

2 Microscale Flows −5

x 10 3

0.12

2

Wall normal velocity

Streamwise velocity (m/s)

0.14

0.1 0.08 0.06 0.04

1

0

−1

−2 0.02 0

−3

0 0

0.5

Wall normal position

20

10

60

40

100

80

0.2

20

40

0.4 60

Streamwise position

0.8

80

Wall normal position

1 100

Streamwise position

(a)

(b) 10

2

2.6

Experimental Data (by Pong) Cercignani et al. (2004) Cercignani & Daneri (1963) Present theory (C1=1, C2=0) Present theory (C1=1.4, C 2=0.7) Present theory (C1=1.875, C 2=0.05)

2.4 2.2 2

Q

Non dimensional pressure

0.6

1.8

101

1.6 1.4 1.2 1

Second order slip model C1 = 1.1466 C2 = 0.9756 Experimental data (Pong et al. 1994) 0

0

0.25

0.5

z/L

(c)

0.75

1

10 -3 10

10

-2

10

-1

10

0

10

1

10

2

Kn

(d)

Fig. 2.12 (a) Variation of streamwise velocity as predicted by the theory. (b) Variation of cross stream velocity as predicted by the theory. (c) Variation of pressure as predicted by the theory, and comparison against the experimental data of Pong et al. [115], for outlet Knudsen number of 0.059. (d) Variation of normalized volume flux versus Knudsen number, and comparison against the theoretical results of Cercignani et al. [30]. All figures taken from Dongari et al. [44]

2.4.2.3

Comments

The variations of streamwise velocity, cross stream velocity, and pressure with streamwise coordinate are plotted in Fig. 2.12. The streamwise velocity is maximum at the centerline and increases with streamwise coordinate owing to the acceleration of the flow (Fig. 2.12a). Notice the slip at the wall, with the magnitude of slip velocity increasing with streamwise coordinate. The cross stream velocity is zero at the centerline (due to symmetry, the flow cannot move towards or away from the centerline), and also zero at the walls (due to non-porous nature of the walls) (Fig. 2.12b). For the case considered, the magnitude of cross stream velocity is about four orders of magnitude smaller than the streamwise velocity. The small difference in the amount of flow acceleration with lateral position leads to a finite (but small) amount of cross stream velocity.

2.4 Some Exact Solutions

59

The pressure drops non-linearly with streamwise coordinate (Fig. 2.12c) because of additional contribution from the change in momentum of the gas (Eq. (2.131)). Pressure is easier to measure than velocities in a microchannel, although measurement of pressure at intermediate locations is much more involved than measuring pressures at the two ends of the microchannel. A comparison of the theoretical pressure variation with the experimental data of Pong et al. [115] is also shown in the figure. Figure 2.12d shows the variation of non-dimensional volume flow rate versus Knudsen number, and the theoretical results are seen to compare well against some available experimental and linearized Boltzmann equation based theoretical results. Note the appearance of a minima in the curve—known as the “Knudsen minima.” Since the expressions are given in terms of slip coefficients, one can change the value of the slip coefficients and probe the magnitude and location of the minima. Interestingly, it is seen that the minima does not appear upon choosing C2 = 0, i.e., for first-order slip model, while the minima appears for all cases tested with C2 = 0. The two sets of values for slip coefficients used in Fig. 2.12d, although hypothetical, help underscore that the Navier–Stokes equation based solution can be extended well into the transition regime with the help of a second-order slip model, as the proposed solution matches the linearized Boltzmann equation based theoretical results till Kn ≈ 1 and even higher depending on the value of C1 and C2 . However picking the right values of slip coefficients remains an issue in this approach. We now show that the present solution is more general than the one obtained in Sect. 2.4.2.1. On differentiating Eq. (2.97) twice, with respect to x and substituting C2 = 0, we obtain : d2 dx 2

2

24C1 Kno d 2 p po dx 2 24C1 Kn + 1 dp 2 12C1 Kn + 1 d 2 p 2 =0 + 2Re βχ − dx p p2 dx 2 p po

+

(2.101)

which can be compared to Eq. (2.83). The coefficient of d 2 p/dx 2 in Eq. (2.101) is different because of the use of hydraulic diameter as the length scale instead of H , which changed the definition of Kn by a factor of 2. The extra terms in Eq. (2.101) are because the inertial terms are retained in the integral analysis; the magnitude of the terms is proportional to the square of the Reynolds number. Since in Sect. 2.4.2.1 we assumed that Reynolds number is of order , the extra terms do not appear there, thereby confirming the more general nature of the solution in Sect. 2.4.2.2.

60

2 Microscale Flows

2.4.3 Flow in a Microtube The solution of gas flow in microtube essentially follows the same steps as for gas flow in a microchannel discussed in Sect. 2.4.2.2. The assumptions of the flow being two-dimensional, locally fully developed and isothermal with negligible viscous compressive stresses applies to the present case as well. For a second-order slip boundary condition, we can obtain the velocity distribution as 1 + 4C1 Kn + 8C2 Kn2 − r 2 /R 2 u = u¯ 1/2 + 4C1 Kn + 8C2 Kn2

(2.102)

Knowing the velocity profile, the shear stress at the wall can be obtained as τ¯w =

8Reμ2 2 1 Kn + 16C2 Kn )

(2.103)

ρD 2 (1 + 8C

where u¯ = mean velocity, Re = ρ uD/μ, ¯ Kn = λ/D, and D is the microtube diameter. As with Eq. (2.93), we can express the momentum balance between two cross sections of the microtube as D

π D2 dp − τ¯w π Ddz = d − 4

2

ρu2 (2π r)dr

(2.104)

0

where z is the coordinate along the axis of the tube. Integrating Eq. (2.104) with the help of Eqs. (2.102) and (2.103), we obtain the following expression for pressure:

p 2 p p − 1 + 16C1 Kno − 1 + 32C2 Kn2o log + 2Re2 βχ po po po

po 2 po z − zo p 2 = −64Reβ − 1 + 8C2 Kno 8C1 Kno − 1 − log p p po D (2.105)

where p0 is the pressure at some reference position z0 , β = χ=

μ2 RT po2 D 2

, and

1/3 + 4C1 Kn + 16C12 Kn2 + 8C2 Kn2 + 64C22 Kn4 + 64C1 C2 Kn3 . (1/2 + 4C1 Kn + 8C2 Kn2 )2

2.4 Some Exact Solutions

61

1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

0

5

10

15

Fig. 2.13 Pressure drop according to slip theory compared with the experimental results [137], solid line represents second order slip theory results (C1 = 1.1466, C2 = 0.14), dashed line represents first order theory results (C1 = 1.1466), circles correspond to experimental results for Kn = 0.265, Re = 0.147, squares correspond to experimental results for Kn = 0.254, Re = 0.093, [137]

The variation of pressure as a function of position in the microtube is plotted in Fig. 2.13, from where it is clear that the nature of the graph is qualitatively similar to the pressure variation in a microchannel (Fig. 2.12c). The theoretical results are also compared with the experimental data of Sreekanth [137] at reasonably high values of Kn and a good match is seen, especially when second-order slip model is employed. The solution for microtube obtained in this section can be used for a pressure driven or mass flow driven system in a similar manner as described for microchannels.

2.4.4 Flow in an Arbitrary Cross Section Microchannel In the above sections we have considered flow in passages which have a regular cross section. Since high aspect ratios could be realized for these cross sections, we consider the flow to be two-dimensional. This assumption however does apply, for example, where the side walls are not very far away from the centerline of the passage (i.e., the width of the rectangular cross section passage is comparable to its

62

2 Microscale Flows

height). Fabricating circular or rectangular cross section microchannels also turns out to be difficult, whereas trapezoidal (or triangular) cross section microchannels are more common in practice. This motivates us to examine flow in microchannel of alternate cross sections. Gas flow in an arbitrary cross section microchannel was solved by Duan [46] under the conditions that the flow is locally fully developed and isothermal. Any change in momentum of the flow was neglected; therefore, the pressure force only helps in overcoming the friction force but does not accelerate the flow. Duan assumed the slip model to be of the form: w|y=a,b

2 2 ∂ w 2 − σv ∂w 1 ∂ 2 w 2 2 − σv =− λ − C2 λ + σv ∂y y=a,b σv ∂y 2 y=a,b 2 ∂x 2 y=a,b (2.106)

where w is the streamwise velocity component, x and y are the coordinates along the lateral directions, and a and b are the base width and height of the microchannel. The velocity distribution and average velocity w¯ was obtained as b2 dp w(x, y) = − μ dz

∞

i,j =1

b2 dp w(x, ¯ y) = − μ dz

x y 4 ∗2 cos ψj + 2C2 Aij cos ϕi Kn (2.107) aε b (1 + ε)2

∞

Aij

ε sin

i,j =1

ϕi ε

sin ψj

ϕi ψj

+ 2C2

4 ∗2 (2.108) Kn (1 + ε)2

where ε is the effective aspect ratio of the microchannel (and related appropriately to a and b through the geometry of the cross section; Fig. 2.14), Kn∗ is modified Knudsen number defined as Kn∗ =

2 − σv Kn, σv

ϕi and ψj are eigenvalues obtained from the following equations 4 4 ϕi Kn∗ ϕi tan + 2C2 Kn∗2 ϕi2 = 1 (1 + ε) ε (1 + ε)2 4 4 Kn∗ ψj tan ψj + 2C2 Kn∗2 ψj2 = 1 (1 + ε) (1 + ε)2

2.4 Some Exact Solutions

63

and Aij =

1 ϕi2 +ψj2

1 ε

+

4 sin(ϕi /ε) sin ψj ϕi ψj

sin(2ϕi /ε) 2ϕi

+ cos2 ϕε 1 +

2 + 4C2 1+ε Kn∗

sin 2ψ 1 + 2ψj j −1 sin 2ψj 2ψj

sin(ϕi /ε) cos ψj ϕi

2 Kn∗ + 2C2 1+ε

+

cos(ϕi /ε) sin ψj ψj

cos2 ψj

1 ε

+

sin(2ϕi /ε) 2ϕi

The pressure as a function of position (z) can be related to the inlet and outlet pressures (pi and po , respectively) through

pi2 p2 p pi ∗ pi ∗2 + 2C − + 2α Kn − α Kn ln 1 2 2 o o po po p po2 po2 2 pi z pi ∗ pi ∗2 − 1 + 2α1 Kno − 1 + 2C2 α2 Kno ln = L po2 po p

(2.109)

where L is the length of the microchannel, and coefficients α1 and α2 are related to the aspect ratio of the microchannel through relations α1 = 12 − 10.598ε + 8.654ε2 − 2.231ε3 α2 = 48 − 112.7ε + 194.2ε2 − 155.7ε3 + 42.66ε4 The mass flow rate (m) ˙ through the microchannel can be found as m ˙ = ρ wA ¯ 2 5/2 pi p02 A ∗ pi − 1 + 2α1 Kno −1 = P μRT L(f Re√A )ns po2 po pi ∗2 (2.110) + 2C2 α2 Kno ln po where P is the√wetted perimeter of the microchannel, square root of the crosssectional area ( A) has been used as the length scale of the microchannel (instead of hydraulic diameter), and (f Re√A )ns is obtained by setting Kn∗ = 0 in Eq. (2.136). Duan [46] claims that the above result can be used for describing slip flow in circular and non-circular cross sections such as (Fig. 2.14): rectangular, trapezoidal, double-trapezoidal, triangular, rhombic, hexagonal, octagonal, elliptical, semielliptical, parabolic, circular sector, annular sector, rectangular duct with unilateral elliptical or circular end, and annular. Therefore, the analysis presented in this section can be applied to a large number of practical situations.

64

2 Microscale Flows

Fig. 2.14 Effective aspect ratio for different arbitrary cross sections [46]

2.4.5 Flow in the Annulus of Rotating Sphere and Cylinder In this section we will briefly examine the theory of the spinning rotor gauge in the slip regime. As seen in Sect. 2.3.4, spinning rotor gauge has been employed for measuring the value of TMAC, and therefore its study is relevant. The spinning rotor gauge comprises a small sphere magnetically levitated within a cylindrical tube

2.4 Some Exact Solutions

(a)

65

(b)

Fig. 2.15 Schematic of (a) spinning rotor gauge (b) two concentric cylinders

(Fig. 2.15). The sphere is levitated so as to avoid any physical contact between the sphere and the tube. The sphere is first rotated electromagnetically through the field generated by coils surrounding the tube. The driving force is then removed; owing to friction between the sphere and the surrounding gas, the sphere starts to decelerate. From measurement of deceleration of the sphere, obtained through variations in the field induced on the coils, the resulting torque on the sphere can be obtained, from which the value of TMAC can be deduced. The interest therefore is in obtaining an expression for torque in terms of TMAC. The solution of the spinning rotor gauge was obtained by Loyalka [90]. He neglected the inertial terms in the Navier–Stokes equations due to the low Reynolds number of the flow, and cast the resulting Stokes equations in spherical coordinates. The slip condition was applied at the surfaces of both the inner sphere and the outer cylinder. After applying Jeffery’s transformation and some involved algebra, the following expression was obtained: T 3C1 λ C1 λ = 8π R13 Co (R1 /R2 ) 1 + + μω R1 R2

(2.111)

where T is the torque acting on the sphere, ω is the angular speed of the sphere, R1 is the radius of the sphere, R2 is the radius of the tube in which the sphere is suspended, C1 is the slip coefficient (= (2 − σv )/σv ), and coefficient Co is given by

3 10 −1 R1 R1 Co (R1 /R2 ) = 1 − 0.79682417 − 0.060047040 . R2 R2 Interested readers can refer to the original paper for further details about the analysis.

66

2 Microscale Flows

Two concentric cylinders, with the inner cylinder rotating while the outer cylinder is stationary is another configuration used to obtain the value of TMAC, and therefore a brief description of this case is included here. Agrawal and Prabhu [4] obtained an expression for torque acting on the cylinder in slip regime. The cylindrical form of the Navier–Stokes equations for the problem simplifies to − d μ dr

u2θ 1 dp =− r ρ dr

(2.112)

1 d (ruθ ) = 0 r dr

(2.113)

where uθ is the tangential velocity and r is the radial coordinate. The above equations are momentum equations in radial and tangential directions, respectively. The slip condition at the walls gives uθ = ωR1 + C1 λ

uθ duθ − dr r

at r = R1

uθ duθ − uθ = −C1 λ dr r

(2.114)

at r = R2

(2.115)

where R1 and R2 are the radius of the inner and outer cylinder, respectively, and ω is the angular speed of the inner cylinder. Note that the slip velocity is related to the shear stress at the wall which leads to the extra term in cylindrical coordinate. The solution for velocity can be obtained from Eq. (2.113) in a manner similar to conventional flows (albeit with slip boundary condition) as

1 r

uθ =

ω A−B

A=

1 R22

2λ 1 − C1 , R2

B=

1 R12

2λ 1 + C1 R1

Ar −

(2.116)

where

from where the torque (T ) on the inner cylinder can be found as T = where L is the length of the cylinders.

4π μωL B −A

(2.117)

2.5 Observations on Flow in Straight Passages

67

2.5 Observations on Flow in Straight Passages Some interesting observations in microchannels such as Knudsen minima is reviewed in this section. We also comment on two other practical aspects: effect of transients and roughness on the flow.

2.5.1 Appearance of Knudsen Minima An interesting phenomenon exhibited by gas flow in microchannel is the appearance of a minimum in normalized flow rate plotted against rarefaction factor. This was first noted in the measurements of Knudsen in 1909 while investigating flow of carbon dioxide in small capillaries. Subsequent measurements have confirmed the presence of this “Knudsen minima” at Kn close to unity. For example, Fig. 2.16 (taken from Hemadri et al. [70]) shows measurements from two different research groups, with three different gases showing this minima. Steckelmacher [138] reviewed the numerous developments that immediately followed the seminal work of Knudsen. The paper presents formula for flow resistance in various cross section tubes, in the free molecular regime, as given by Knudsen and corrected by Smoluchowski. He also mentions that the appearance of Knudsen minima occurs only for relatively long tubes. Some researchers view the flow of gas under an imposed pressure gradient as a superimposition of flow due to convection and molecular diffusion [114, 161]. The contribution of convection dominates at higher pressure; however, its contribution Fig. 2.16 Comparison of non-dimensional mass flow rate G with respect to mean rarefaction parameter δm (δm is inversely proportional to Knudsen √ number and given as δm = π /(2Kn)), Hemadri et al. [70]

68

2 Microscale Flows

to the overall flow reduces with a reduction in pressure. On the other hand, the contribution of molecular diffusion is insignificant at high pressure, but the contribution increases with a reduction in pressure. Therefore, the overall flow rate first reduces before increasing, and leads to the appearance of a minima. Alternatively, Gu et al. [69] viewed the phenomenon in terms of centerline velocity and slip velocity. Recall, that for a given flow rate, in presence of a slip, the centerline velocity will be smaller as compared to its no-slip counterpart. Therefore, any change in flow rate in the slip regime will be accompanied by changes in both the centerline velocity and the slip velocity. It was observed that initially, the rate of decrease in centerline velocity is greater than the rate of increase in slip velocity, which leads to a reduction in the flow rate. Subsequently, the trends reverse and the flow rate starts increasing, leading to the appearance of the minima. It needs to be underscored that the dimensional flow rate decreases monotonically with increase in rarefaction; the minimum is observed only upon normalization. The Knudsen minima is seen to occur in passages of both uniform cross section and non-uniform cross section. The latter has been shown only recently through the experiments of Hemadri et al. [70].

2.5.2 Flow in Rough Microchannel Surface roughness (ε) is expected to play a significant role at the microscale and its role has been actively investigated. This is because the value of roughness in a relative sense (i.e., ε/Dh ) increases when the length scale (Dh ) decreases. A schematic diagram of cross section of rough microtube is shown in Fig. 2.17. Because of the random nature of the surface roughness, it is extremely difficult to accurately predict the local radius r. Instead, mean radius a which is the average value of the local radius, r, is used. As per our conventional understanding, surface roughness is not important in the laminar regime and relevant only in the turbulent regime (one can refer, for example, to Moodys chart in this context). Since, flows in the microscale are mostly laminar (because of the low value of Reynolds number), will the relatively large value of ε/Dh still affect the fluid flow and heat transfer, or can our expectation from conventional flows be simply extended to microscale, therefore becomes a relevant question.

Fig. 2.17 Schematic diagram of cross section of rough microtube

2.5 Observations on Flow in Straight Passages

69

Techniques used to enhance roughness are adherence of sand particles, microwire electrical discharge machining, laser drilling, electrochemical etching, and chemical etching. Acid based etching of the microchannels has also been suggested for producing rough microchannels, with the grade of roughness being varied by the etching time and acid concentration. Several instruments are available to characterize the features on a surface (such as white light interferometry, digital microscope, and stylus-based profilometer) that can help determine the value of surface roughness. Cleanliness of the surface plays a crucial role, since presence of impurities can change the surface roughness characteristics. Another way to characterize surface roughness qualitatively is by measuring the dynamic advancing and receding contact angles of the surface. The contact angle increases with increasing surface roughness due to enhanced wetting characteristics. Measurements on rough microchannel are problematic because of the difficulty in controlling and characterizing the amount of roughness. Further, roughening of the microchannel tends to change its hydraulic diameter; although this change is small, the effect gets amplified due to the high sensitivity of pressure drop on hydraulic diameter. However, the change in hydraulic diameter is much more difficult to characterize, primarily because of the difficulty in finding a reference from where the measurements of height and width should be undertaken. As the standard deviation of the hydraulic diameter measurements increases, the uncertainty in friction factor also increases. Turner et al. [151] did not find any significant effect of surface roughness in slip regime, because of high experimental uncertainty masking any differences observed due to surface roughness. Some authors found friction factor to enhance at microscale [1, 144, 162, 169], while some found friction factor to reduce [34, 113, 174]. It is ambiguous that this increment or decrement owes to compressibility, rarefaction, or roughness effect. On the other hand, simulations become difficult because of the requirement to resolve the region close to the roughness element well, which increases the requirement for the total number of grid points in the domain. Further, performing simulations with random roughness is difficult. Due to these reasons, many simulations have considered much larger roughness elements arranged periodically on a surface. This shows that there is no consensus at present in the literature about the effect of surface roughness on friction factor in gaseous flows at microscale. Even no claim can be made, on the effect of roughness shape geometries, height, width, spacing, density distribution, and angle of roughness inclination on friction factor in microchannel flows. The scales of microfluidic systems would continue to decrease in future, which would increase the relative significance of exploring the influence of surface roughness.

2.5.3 Transient Flow in a Capillary A long capillary connected to a reservoir at each of its two ends with the pressure of the reservoirs varying with time has also been considered. Interest is in finding the

70

2 Microscale Flows

evolution of the pressure in the two end-reservoirs as a function of the rarefaction parameter. Practically, the problem is relevant for finding out how rapidly a rarefied flow will respond to a sudden change to an externally imposed pressure gradient coming, for example, from a vacuum pump or a valve. Lihnaropoulos and Valougeorgis [87] solved the above problem by employing linearized unsteady BGK equation, along with the Maxwell’s boundary conditions (assuming purely diffuse interaction). The gas starts moving due to the suddenly applied pressure gradient across the ends of the tube, and the variation of velocity as a function of radial position and time can be calculated. With passage of time, the gas reaches the steady-state flow field in an asymptotic manner. The time required for the flow to reach its fully developed characteristics is minimum close to Knudsen minimum of unity. Sharipov and Graur [127] found that the time for attaining the equilibrium pressure is maximum in the transition regime (i.e., close of the value of Knudsen minimum) and decreases rapidly in the continuum regime.

2.6 Observations on Flow in Complex Passages A microchannel is unlikely to be straight in most practical applications; it would rather involve bends, sudden or gradual change in cross-sectional area, bifurcation, junction, and other complex features. The interest is therefore in understanding flows in “complex” microchannels comprising one of these features. A clear understanding of the flow behavior through such sections is unarguably of both practical and fundamental interest.

2.6.1 Flow in Sudden Expansion/Contraction Microchannel A sudden change in the cross-sectional area can have a strong effect on the local flow dynamics, and the disturbance created by the junction to the flow can travel several diameters upstream and downstream of the junction. The strength of the disturbance depends on the area ratio of expansion/contraction, the geometry of the cross section, the Reynolds number, and the Knudsen number of the flow. Appearance of a corner eddy at the sudden expansion/contraction section is the most noticeable feature of the flow in the continuum regime (Fig. 2.18a). Will the corner vortex still form if the Knudsen number is changed from zero to a finite value, and how far upstream and downstream will the disturbance propagate, are therefore natural questions to wonder about. The measurements of Varade et al. [156] in a scaled-up facility comprising two tubes of different cross-sectional areas connected to form a junction, surprisingly showed that corner eddies are not formed in the slip flow regime. The maximum Reynolds number covered in these measurements is 837 and area ratio is 64. Far from the junction, the pressure was found to drop non-linearly in each of the two tubes and the pressure variation compares well with the theoretical results

2.6 Observations on Flow in Complex Passages Fig. 2.18 (a) Schematic streamlines of laminar, incompressible, separated flow near the junction of a tube with a sudden change in cross-sectional area. (b) Schematic streamlines of rarefied gas flow near the junction. (c) Experimentally measured static pressure variation (normalized with outlet pressure) along the wall for flow of low pressure gas in a tube with a sudden change in cross-sectional area. All figures taken from Varade et al. [156]

71

(a)

P/P O

(b)

(c)

72

2 Microscale Flows

of Sect. 2.4.3. However, the pressure was found to drop more rapidly than the theoretical pressure close to the junction; the pressure profiles also exhibited a discontinuity in slope at the junction (Fig. 2.18c). The variation of pressure is in contrast to observation in the continuum regime, where no such discontinuity is seen. In fact, in the continuum regime, the pressure curve shows an adverse pressure gradient corresponding to the flow separation region, a feature which was not seen in the experiments of Varade et al. The authors explained the observations through an increased momentum owing to slip at the wall, and enhanced radial diffusion of flow downstream of the junction, due to enhanced diffusivity brought about by the reduction in density (or pressure) in their system (Fig. 2.18b). However, these measurements do not rule out the possibility of finding very small corner eddies, of size smaller than the resolution in their measurement. The following correlation was proposed for calculating the additional pressure drop (padd ) in sudden expansion case [156]: Kadd,SE =

16.36(AR)0.63 (Knj )0.45 (Res )1.2

(2.118)

where AR is the area ratio, Knj is the Knudsen number at the junction, Res is the Reynolds number in smaller tube. Knowing Kadd , the additional pressure drop can be calculated as padd = (1/2)Kadd ρj Uj2 , where ρj is the density of gas at the junction and Uj is the average velocity at the junction in the smaller tube just upstream of the junction. The above empirical correlation is valid for 3.74 ≤ AR ≤ 64; 0.0002 ≤ Knj ≤ 0.075; 0.32 ≤ Res ≤ 837. Similarly, the following correlation for sudden contraction case was proposed by Varade et al. [157]: Kadd,SC =

(AR)2.7 (Knj )0.50 . 3(Res )1.2

(2.119)

The above empirical correlation is valid for 12.43 ≤ AR ≤ 64; 0.0003 ≤ Knj ≤ 0.02; 0.56 ≤ Res ≤ 837. It is interesting to see from Eqs. (2.118) and (2.119) the same dependence of Reynolds number on causing an additional pressure drop. Further notice that the Knudsen number also has approximately the same value of exponent but with different roles in the two cases.

2.6.2 Flow in Diverging/Converging Microchannel At the conventional scale, flow in diverging or converging microchannel is referred to as Jeffery-Hamel flow, where the walls are replaced by a row of mass source or sink. The Navier–Stokes equations reduce to non-linear ordinary differential equation, which can subsequently be solved. The above approach however does not seem to have been extended to the slip regime. Graur et al. [65] obtained an approximate analytical solution by solving the Stokes equation, along with application of slip boundary condition.

2.6 Observations on Flow in Complex Passages

73

Diverging/converging microchannels were employed by Stemme and Stemme [139] to control the flow direction, instead of using micro-valves. Their construction therefore reduces the number of moving parts, thereby making it attractive. The flow rate in a diverging microchannel is different in a converging microchannel, for the same pressure difference acting across the length of the microchannel. This disparity in flow rate is utilized to cause a flow rectification, and can be quantified in terms of a parameter, diodicity D defined as [65]: D=

m ˙ conv /((p1conv )2 − (p2conv )2 ) m ˙ div /((p1div )2 − (p2div )2 )

(2.120)

where m ˙ refers to mass flow rate, p is pressure, superscripts “conv” and “div” refer to converging and diverging directions, respectively, and subscripts “1” and “2” refer to inlet and outlet of the microchannel, respectively. For a given flow rate, pressure drop in the converging microchannel is greater than its diverging counterpart [70, 71]. A large absolute value of diodicity leads to a better efficiency of valve-less micro-pump, with the effect vanishing at D = 1. The experiments of Duryodhan et al. [49] and numerical simulation of Graur et al. [65] suggest a maximum value of 1.2 for diodicity. A fundamental question to address while employing diverging/converging microchannel is definition of characteristic length, required, for example, in calculation of Reynolds number. The issue is not straightforward because of the continuous change in cross-sectional area and therefore the hydraulic diameter. Using the hydraulic diameter at the entrance or at the mid-length of the passage is not scientifically compelling. This issue was addressed by Duryodhan et al. [49] with liquid flow and the location where the hydraulic diameter should be calculated was expressed as a fraction of microchannel length. The location was determined by equating the pressure drop in the diverging (or converging) microchannel to an equivalent straight microchannel. Based on this equivalence, the hydraulic diameter at L/3 (where L is the length of the microchannel) from the narrower end was proposed as the “equivalent hydraulic diameter” (Fig. 2.19a). Interestingly, the proposed location is reasonably invariant of the divergence angle, length and crosssectional dimensions of the microchannel, and also the flow rate (or Reynolds number) but shows a dependence on the Knudsen number [70]. The location changes to L/3.6 from the narrower end in case of a converging microchannel, for the case of liquid flow (Fig. 2.19b). These length scales have also been employed in the calculation of Nusselt number. Note that the difference in location between the diverging and converging microchannels gives rise to diodicity. With an increase in Knudsen number, the location of the equivalent hydraulic diameter moves to the narrower end, for both diverging and converging microchannels [70]. This suggests that the diodic effect vanishes (i.e., D becomes unity) with increasing rarefaction of the flow, and the design of Stemme and Stemme [139] cannot be used to pump low-pressure gases. In Sect. 3.7, we will discuss a Knudsen pump which is used to pump low pressure gases based on temperature difference at the two ends of a microchannel.

74

2 Microscale Flows

Fig. 2.19 A schematic illustrating the concept of equivalent hydraulic diameter for (a) diverging and (b) converging microchannel. The pressure drop and heat transfer in diverging/converging microchannel with liquid flow is same as a straight microchannel, provided that the straight microchannel has the dimension at the location marked in the figure. Taken from Duryodhan et al. [51]

2.6.3 Flow in a Bend Microchannel Similar to flow in a sudden expansion/contraction passage, an eddy is expected to form at the corner of a bend microchannel. Does the formation of this corner eddy get delayed in presence of slip as in the case of a sudden expansion/contraction microchannel? Towards this, Varade et al. [158] conducted measurements in a scaled-up facility and obtained local wall pressure distribution along the inner and outer walls of a 90◦ bend tube. It was surprisingly found that an adverse pressure gradient occurs near the bend, indicating flow separation, at much lower value of Reynolds number as compared to conventional flow. The numerically obtained data of Varade et al. [158], Agrawal et al. [6], and White et al. [165] also confirm this observation. The numerically obtained velocity field near the bend plane clearly indicates secondary flows at Reynolds number as low as unity. The flow acceleration and the presence of secondary flows near the bend causes a larger pressure drop as compared with a straight tube. Based on the experimental data, the following correlation was proposed for f Re for a bend tube in presence of slip: f Re =

60 1 − 0.64 exp(−607Knm )

(2.121)

2.6 Observations on Flow in Complex Passages

75

Fig. 2.20 (a) Velocity vectors near the bend for Kn = 0.202, numbers denote the local Mach number (Agrawal et al. [6]). (b) Streamlines in the zoomed view of the bend region showing clearly the presence of the corner vortex (Agrawal et al. [6]). (c) Experimentally measured static pressure variation for bend microchannel compared with straight tube for Re = 0.67, Kn = 0.035 (Varade et al. [158])

Here, Knm is the mean Knudsen number defined as the average of the inlet (Kni ) and outlet (Kno ) Knudsen numbers (i.e., Knm = (Kni + Kno )/2). The above correlation applies to 0.27 ≤ Re ≤ 418.5; 0.0007 ≤ Kn ≤ 0.0359. The correlation suggests that f Re decreases with an increase in the Knudsen number; further, f Re becomes independent of rarefaction for Kno > 0.005. An empirical correlation for computing the additional pressure loss coefficient was also presented.

76

2 Microscale Flows

The contrasting behavior of flow in sudden expansion/contraction versus bend microtube in the slip regime is interesting and worth further study and confirmation.

2.7 Useful Empirical Correlations In this section, we provide some useful empirical correlations for calculating the developing length and friction factor in microchannel and microtubes.

2.7.1 Developing Length in Microtube and Microchannel When the flow enters a passage, it takes a certain length to develop, which is called the developing length. Once the flow becomes fully developed, the shape of the velocity profile does not change. The developing region is more difficult to analyze than the fully developed region as there is an additional change in inertia of the fluid in this region. Appropriate correlations have therefore been proposed to estimate the developing length in the literature. Information about the developing length is useful, for example, for deciding the location of the first pressure tap while making measurements in a long microchannel. For laminar flow in a pipe in the continuum regime, one of the following correlations can be employed to estimate the developing length (Ld ) normalized by the pipe diameter (D): Ld = [(0.619)1.6 + (0.0567Re)1.6 ](1/1.6) . D

(2.122)

The above correlation is by Durst et al. [48] while the following correlation is by Dombrowski [43]. Ld = 0.260 + 0.055Re + 0.379 exp(−0.148Re) D

(2.123)

Chen [32] proposed: Ld 0.6 = + 0.056Re. D 0.035Re + 1

(2.124)

All the above three correlations give reasonably close values of developing length for laminar continuum flow in a tube (i.e., within 6% of each other; except around Re = 10 where the difference increases to 12%) over the Reynolds number range of 0–3000. In presence of slip at the wall, Sreekanth [137] suggested to calculate the developing length as

2.7 Useful Empirical Correlations

77

Ld kRe = D 4

(2.125)

where k takes a value between 0.15 and 0.2 depending on the Reynolds number. Duan and Muzychka [47] modified the Chen’s formula given above to include the effect of slip at the wall as 2 0.60 2 − σv Ld 2 − σv = + 0.0396Re 1 + 3.7 Kn − 15 Kn D 0.035Re + 1 σv σv (2.126) where σ v is tangential momentum accommodation coefficient discussed in Sect. 2.3. Note that the form of the correlation, in both continuum and slip regimes, has to be chosen such that the developing length is not suggested to be zero when the Reynolds number is zero. For laminar flow in the continuum regime in a two-dimensional microchannel, one of the following correlations can be employed: Ld = [(0.631)1.6 + (0.0442Re)1.6 ](1/1.6) Dh

(2.127)

suggested by Durst et al. [48] or Ld 0.315 + 0.011Re = Dh 0.0175Re + 1

(2.128)

given by Chen [32]. In presence of slip, Duan and Muzychka [47] modified the Chen’s formula as 2 Ld 2 − σv 2 − σv 0.315 + 0.0112Re 1 + 6.7 = Kn − 37 Kn Dh 0.0175Re + 1 σv σv (2.129) Barber and Emerson [18] suggested ⎞ 2 − σv Kn 1 + 14.78 ⎟ ⎜ Ld 0.332 σv ⎟ + 0.011Re ⎜ = ⎠ ⎝ 2 − σv Dh 0.0271Re + 1 Kn 1 + 9.78 σv ⎛

(2.130)

Since the Reynolds number of the flow is small at the microscale, primarily because of the small length scale, the developing length is small in most cases; and the influence of the developing region on the overall pressure drop in a long microchannel can be neglected.

78

2 Microscale Flows

2.7.2 Friction Factor for Microchannel of Various Cross Sections Measurement of pressure drop as a function of flow rate is among the first exercise undertaken in practice to test a freshly fabricated microchannel. The pressure drop is normalized into friction factor and flow rate into Reynolds number, using the values of the geometric parameters of the passage and the fluid properties. The value of friction factor (or Poiseuille number, which is the product of friction factor and Reynolds number) is then compared against available data or empirically obtained correlations. A good comparison would point towards the experimental setup and measurement procedure being satisfactory. Similarly, numerically obtained value of friction factor is compared with its experimentally determined value. This shows the importance of having high-quality empirical correlations for friction factor. Each correlation has a validity over a certain parameter range, and the comparison should be limited to this range. The friction factor (f ) for flow in a microchannel can be calculated from the expression given by Wu and Little dp f ρ u¯ 2 d u¯ =− + ρ u¯ dx Dh 2 dx

(2.131)

which shows that the pressure drop depends on flow friction as well as acceleration effect caused by the change of density. Integrating the above equation across the length of the channel and using the equation of state, we obtain [44]: f =

P2 − 1 π 2Dh . + log P 2 2 2 L P 2Re Kn

(2.132)

where P is pressure ratio. For flow in microtube involving density variation and rarefaction, Verma et al. proposed the following correlation based on compilation of a large amount of data from the literature: f Re =

64 1 + 14.9Kn

(2.133)

Demsis et al. [42] proposed the following correlation based on measurements for flow in a smooth square tube: f Re =

57 1 + 34.5Kn

(2.134)

The above correlation is valid in the range 0.0022 < Kn < 0.024 and 0.54 < Re < 13.2 range. Both the above correlations give the correct value of f Re in the continuum regime (i.e., as Kn → 0).

2.8 Summary

79

Niazmand et al. [105] reported correlations to evaluate friction factor through rectangular and trapezoidal channels, given by

(f Re)f d = 13.9

90 φ

−0.07

+ 10.4 exp

90 − 3.25α φ

0.23 G1

(2.135)

where G1 = 1 − 2.48 Kn0.64 (1 − 0.2 tanh(3α)). The friction factor for an arbitrary cross section microchannel can be obtained from [46] f Re√A = =

√ 2(−A/P )(dp/dz) A μw¯ 4 !∞ √ sin(ϕi /) sin ψj 4Kn∗2 (1 + ) + 2C2 i,j =1 Aij ϕi ψj (1 + )2 (2.136)

where the symbols have been defined in Sect. 2.4.4. Clearly, the Poiseuille number (product of friction factor and Reynolds number) depends on the geometry of the cross section and rarefaction. Additionally, correlations for microchannel of various useful shapes have been documented in a tabular form in Morini et al. [101]. All the above correlations suggest that the friction factor reduces with an increase in rarefaction or Knudsen number.

2.8 Summary In this chapter, we have shown that slip at the wall (present in gaseous microscale flows) can bring about substantial difference in the obtained solution with respect to the no-slip case (conventional scale flows). Towards this, we reviewed the relevant slip models and the empirically obtained values of the slip coefficients. The available data for slip coefficients suggests some interesting trends—such as dependence of the slip coefficients on the atomicity of the gas, which needs further verification. The process of obtaining an analytical solution in presence of both slip at the wall and compressibility indeed becomes complicated, and therefore exact solutions are available for only a handful of cases, as reviewed here. A successful comparison of these theoretical solutions against experimental and DSMC data shows the applicability of the Navier–Stokes equations in the slip regime, but with a modification of the boundary condition to first-order slip, and with a further increase in Knudsen number to second-order slip. However, the validity of the Navier–Stokes equations in the transition regime even with second-order slip boundary condition becomes questionable.

80

2 Microscale Flows

Information of compressible slip flow in alternate cross section straight microchannels and complex passages of uniform cross section is rather limited. These are interesting cases to study from both practical and fundamental viewpoints. Available information suggests that flow in complex microchannels can be substantially different from its counterpart in the conventional range of parameters, and further study can help uncover other unexpected flow behavior in such complex scenarios.

Chapter 3

Microscale Heat Transfer

In this chapter, we present the fundamental aspects of microscale heat transfer for gas flow through different geometries. Such a study is motivated by interest in cooling of electronic components, energy conversion devices, and other MEMS and bio-medical applications. The heat transfer at microscale is different than that of macroscale primarily due to the presence of velocity slip and temperature jump at the wall. The physics pertinent to microscale heat transfer is reasonably complex and solutions of various simplified models are available in the literature. Here we confine our presentation to the slip flow regime for flow through three configurations: parallel plates, microtube, and micro-annulus. We also briefly discuss the effect of other complicating factors and comment on comparison with experiments. A discussion on Knudsen pump and useful empirical correlations are also provided.

3.1 Introduction This chapter deals with the fundamentals of heat transfer of gas flow at microscale with special attention to its basic concepts including slip velocity, temperature jump, rarefaction, and viscous dissipation. The presence of slip velocity at the wall can lead to an enhancement in Nusselt number, while the temperature jump tends to reduce the value of Nusselt number, as the flow now ‘feels’ a smaller temperature difference. A careful analysis of heat transfer in the slip regime therefore becomes particularly interesting. Further complications arise since the effect of axial conduction can also become significant due to the low Peclet number (P e = ReP r) brought about by the low value of Reynolds number. Variation in thermo-physical properties of the fluid, due to variation in temperature, is another concern especially when the temperature difference between fluid and wall is large. Several theoretical studies also assume a relatively large effect of viscous dissipation on heat transfer. The compressibility © Springer Nature Switzerland AG 2020 A. Agrawal et al., Microscale Flow and Heat Transfer, Mechanical Engineering Series, https://doi.org/10.1007/978-3-030-10662-1_3

81

82

3 Microscale Heat Transfer

Fig. 3.1 Non-continuum effects in gas micro-convection (adapted with modifications from Prabhu and Mahulikar [119])

effect already discussed in Chap. 2 is also relevant here. Conduction in the wall (conjugate effect) can also become important because the thickness of wall is comparable to the lateral dimensions of the flow passage. These various effects can be classified as rarefaction and non-rarefaction effects (Fig. 3.1), where rarefaction effects are due to the relatively large value of Knudsen number whereas nonrarefaction effects refer to other microscale effects. The study of heat transfer in the slip regime is therefore quite involved, as should be further evident from our discussion in this chapter.

3.2 Governing Equations The thermal energy equation (Eq. (2.11)), which is the primary governing equation for heat transfer problems is provided in this section in both tensorial and expanded forms. In tensorial notation, ∂T ∂ ∂T ∂uk ∂T = k −p + uk ρcv ∂t ∂xk ∂xk ∂xk ∂xk 2 ∂uj 1 ∂uk 1 ∂ui − + 2μ + δij (3.1) 2 ∂xj ∂xi 3 ∂xk

3.3 Boundary Conditions

83

and in expanded form, ρcv

∂T ∂T ∂T ∂T +u +v +w ∂t ∂x ∂y ∂z

=

∂T ∂ ∂T ∂ ∂T k + k + k ∂x ∂y ∂y ∂z ∂z ∂w ∂u ∂v + + +Φ (3.2) −p ∂x ∂y ∂z ∂ ∂x

where viscous dissipation term Φ in expanded form is Φ = 2μ

∂u ∂x

2

+

∂v ∂y

2

+

∂w ∂z

2

∂u ∂v 2 + +μ ∂y ∂x

∂u ∂v ∂w 2 ∂w 2 ∂v ∂w ∂u 2 2 + + + + +μ +μ − μ ∂z ∂y ∂x ∂z 3 ∂x ∂y ∂z

(3.3)

3.3 Boundary Conditions Usually, no-temperature jump boundary condition is considered in the study of heat transfer phenomena at the macroscale. However this boundary condition cannot be used at microscale, rather a temperature jump boundary condition is applicable here (Fig. 3.2). Similar to TMAC defined in Sect. 2.3, here we define thermal accommodation coefficient, σT =

Ei − Er Ei − Ew

(3.4)

where Ei and Er are the energy fluxes of incoming and reflected molecules per unit time, and Ew is the energy flux of re-emitted molecules corresponding to the Fig. 3.2 Schematic diagram showing temperature jump at the wall in a parallel plate channel; the walls are maintained at temperature Tw while the gas layer adjacent to the wall experiences a temperature Ts showing a jump in temperature (Ts − Tw ) at the wall

84

3 Microscale Heat Transfer

surface temperature Tw . Note that for the case of perfect energy exchange between the gas and wall, Er = Ew , and we obtain σT = 1. On the other hand, for no energy exchange between the gas and wall, the outgoing molecules carry the same energy that they came in with. Therefore, in this case, Er = Ei and σT = 0. So we see that σT is a parameter that nicely describes the amount of “accommodation” that the outgoing molecules have with the wall. The actual value of thermal accommodation coefficient depends on the wall temperature, gas pressure, surface finish, and surface roughness of the wall.

3.3.1 Derivation of First Order Temperature Jump Condition As done earlier with momentum, we consider the energy of incoming and outgoing streams separately [82] as shown in Fig. 3.3. The incoming stream, say, coming from a higher temperature region away from the wall, brings in additional energy with it. If (∂T /∂y)|w is the temperature gradient, then (k/2)(∂T /∂y)|w (where k is the thermal conductivity of the gas) is the excess energy brought in by the incoming stream. Here, we have assumed that half of the excess energy is brought to the wall, while the other half is transmitted elsewhere in the gas. The incoming gas also brings the internal and kinetic energies with it. Recall that from kinetic theory 14 nv¯ is the number of molecules incident per unit area in a gas, where n is the molecules per unit volume and v¯ is their mean speed. These molecules will carry an energy of m(cv + (1/2)R)Ts where m is the mass of the molecule, cv is the specific heat at constant volume, R is the gas constant, and Ts is the temperature of the gas slightly away from the wall (Fig. 3.2). Therefore, we obtain Ei = (k/2)(∂T /∂y)w + 14 nvm(c ¯ v+ (1/2)R)Ts . The outgoing gas will carry away their internal and kinetic energies with them. The energy term (Ew ) can be written as 14 nvm(c ¯ v + (1/2)R)Tw where Tw is the temperature of the wall.

Fig. 3.3 Energy of incoming and outgoing stream of molecules close to the wall

3.4 Some Exact Solutions

85

Therefore k ∂T 1 1 ¯ cv + R (Ts − Tw ) Ei − Ew = + nvm 2 ∂y w 4 2

(3.5)

Using ρ = mn, cv + R/2 = (γ + 1)cv /2, v¯ = 2(2RT /π )1/2 , we can write the above equation as, k ∂T γ + 1 cv (Ts − Tw )p Ei − Ew = + √ 2 ∂y w 2 2π RT

(3.6)

The Ei − Er term (energy delivered to the wall) is ∂T Ei − Er = k ∂y w

(3.7)

where Er denotes the energy carried away by the reflected stream. Upon substituting the above equations into Eq. (3.4), k

k ∂T ∂T γ + 1 cv p(Ts − Tw ) = σ + √ T ∂y w 2 ∂y w 2 2π RT

On re-arranging the above equation along with using λ = μcp k ,

μ p

π RT 2

(3.8) ,γ =

cp cv ,

Pr =

we obtain the desired form of the temperature jump condition, ∂T 2γ 2 − σT T s − Tw = λ σT P r(γ + 1) ∂y w

(3.9)

Equation (3.9) can also be written ∂T Ts − Tw = ζT ∂y w

(3.10)

where ζT denotes the temperature jump length, ζT =

2γ 2 − σT λ σT P r(γ + 1)

(3.11)

3.4 Some Exact Solutions As summarized in Fig. 3.1, there are several effects that can arise simultaneously in heat transfer problems; because of these effects many more terms (as compared to the corresponding problem at the conventional scale) have to be retained in the

86

3 Microscale Heat Transfer

governing equations. This makes obtaining analytical solution of the governing equations difficult. Here we consider only two such effects—rarefaction and viscous dissipation where obtaining analytical solution is relatively easy. A discussion of the other effects on the solution is provided briefly in a subsequent section. Specifically we present exact solutions for heat transfer of gas flow through microchannel, microtube, and microannulus. The flow is assumed to be steady, twodimensional, laminar, incompressible, and fully developed both hydrodynamically and thermally in all the cases. Thermophysical properties of the Newtonian fluid involved are assumed to be constant. The effect of axial conduction is neglected both in the fluid and through the wall.

3.4.1 Heat Transfer in Microchannel We present closed form expressions for temperature distribution and Nusselt number for microchannel gas flow, first derived by Aydin and Avci [15]. Consider flow between two stationary parallel plates at a distance W apart (Fig. 3.4). The flow is driven by a constant pressure gradient along the x-axis. Two cases can arise: Uniform heat flux is supplied to the walls of microchannel due to which there is a transfer of heat from wall to the flowing gas, or the walls are maintained at a temperature which is different from that of the flowing gas. For the present situation, the solution of the momentum equation is not coupled to the energy equation, and therefore the mass and momentum equations are solved before the energy equation. Further due to the fully developed flow assumption, the velocity u varies only in the y direction. The velocity profile, along with slip at the wall can be derived as in Chap. 2 to obtain u 3 1 + 4Kn − (y/w)2 (3.12) = um 2 1 + 6Kn where um denotes the mean velocity and Kn (=λ/2w) is the Knudsen number. Note that the first order slip model with complete accommodation (i.e., σv = 1) has been utilized in writing the above solution. Fig. 3.4 Schematic of microchannel gas flow with uniform wall heat flux boundary condition

3.4 Some Exact Solutions

87

The energy equation considering viscous dissipation is ∂ 2T ν ∂u 2 ∂T =α 2 + u ∂x cp ∂y ∂y

(3.13)

where the last term denotes the effect of viscous dissipation (obtained through simplification of Eq. (3.3)) and ν is the kinematic viscosity (= μ/ρ). The thermal boundary condition at the center of microchannel (due to symmetry) is ∂T = 0. (3.14) ∂y y=0 3.4.1.1

Uniform Heat Flux

We will first consider the case of uniform heat flux condition. Application of constant wall heat flux boundary condition at wall is ∂T (3.15) qw = k ∂y y=w where qw is positive when its direction is to the fluid (heated wall), otherwise it is negative (cooled wall). As in the conventional case, ∂T dTs dTm = = ∂x dx dx

(3.16)

where T is the absolute temperature, Ts is the temperature of gas at the wall, and Tm is the bulk mean temperature (defined below through Eq. (3.25)). Define the non-dimensional variables, Y =

y T − Ts μu2m Ts − T μu2m , θq = , θ= , Br = , Brq = w Ts − Tc qw w/k k(Ts − Tc ) wqw (3.17)

where Tc is the centerline temperature. In the fully developed region the axial temperature gradient is obtained by applying energy balance to an elemental area, qw P + μΦdA dTm = dx mc ˙ p

(3.18)

where P is heated perimeter. Using Eq. (3.18) the axial temperature gradient is evaluated, 3Brq dTm qw (3.19) = 1+ dx ρum cp w (1 + 6Kn)2

88

3 Microscale Heat Transfer

where P is heated perimeter. Using Eqs. (3.12), (3.15), (3.16), (3.17), and (3.19) the energy equation (3.13) in non-dimensional form is 3Brq d 2 θq 9Brq Y 2 1 3 2 (1 + 4Kn − Y + = ) − 2 1 + 6Kn (1 + 6Kn)3 dY 2 (1 + 6Kn)2

(3.20)

The non-dimensional boundary conditions are ∂θq = 1 at Y = 1 ∂Y θq = 0 at Y = 1

(3.21)

The non-dimensional energy equation (3.20) can be integrated twice with the thermal boundary condition equation (3.21), gives the temperature distribution in terms of modified Brinkman number, 3Brq 5 3 1 T − Ts 1 − + Y2 − Y4 = + θq = 3 qw w/k 1 + 6Kn (1 + 6Kn) 8 4 8 3Brq + 3Kn(Y 2 − 1) − (Y 4 − 1) (3.22) 4(1 + 6Kn)2 The above solution is however in terms of Ts ; since the temperature of the gas at the wall is not explicitly known, we express the above equation in terms of wall temperature Tw . For this, we utilize the temperature jump condition given in Eq. (3.9), which expressed in non-dimensional form becomes 4Knγ Ts − T w =− qw w/k P r(γ + 1)

(3.23)

where we have assumed σT = 1 in writing the above equation. The non-dimensional temperature distribution can therefore be written as T − Tw qw w/k 3Brq 1 5 3 2 1 4 2 + + Y Y − + 3Kn(Y − 1) = − 1 + 6Kn (1 + 6Kn)3 8 4 8

θq∗ =

−

3Brq 4Knγ (Y 4 − 1) − 2 P r(γ + 1) 4(1 + 6Kn)

(3.24)

where a “*” has been added to the symbol to emphasize that the solution incorporates the temperature jump condition, while subscript “q” emphasizes the use of modified Brinkman number.

3.4 Some Exact Solutions

89

The bulk mean temperature of the fluid is usually estimated on the basis of thermal energy transported by the fluid stream that passes through a given crosssectional area A. Tm can be calculated using the temperature distribution, ρcp uT dA Tm = A A ρcp udA

(3.25)

The non-dimensional bulk mean temperature finally becomes ∗ θq,m =

2Brq 11Brq 1 12Knγ T m − Tw =− 1+ − − qw w/k 3 P r(γ + 1) 35(1 + 6Kn)4 35(1 + 6Kn)3 −

2(1 + 21Brq ) 2 − 2 15(1 + 6Kn) 105(1 + 6Kn)

(3.26)

The Nusselt number is defined, Nuq =

qw (2w) 2 h(2w) = =− ∗ k k(Tw − Tm ) θq,m

(3.27)

where h is the convective heat transfer coefficient. The Nusselt number expression in terms of modified Brinkman number is 2Brq 12Knγ 1 N uq = 2 1+ + 3 P r(γ + 1) 35(1 + 6Kn)4 +

3.4.1.2

11Brq 2(1 + 21Brq ) 2 + + 15(1 + 6Kn) 35(1 + 6Kn)3 105(1 + 6Kn)2

−1 (3.28)

Constant Wall Temperature Case

The constant wall temperature case suggests that dTs /dx = 0, and as in the conventional case, T − Ts dTc ∂T = (3.29) ∂x Tc − Ts dx Using Eqs. (3.13), (3.17), and (3.29) we write d 2θ 9BrY 2 3a 1 + 4Kn − Y 2 θ + = 2 1 + 6Kn dY 2 (1 + 6Kn)2

(3.30)

where a = −[um w 2 /(α(Ts − Tc ))](dTc /dx). A closed form solution of the above equation is however not possible.

90

3.4.1.3

3 Microscale Heat Transfer

Parametric Variation

We examine the effect of viscous dissipation and rarefaction on the heat transfer characteristics for the uniform heat flux case. Prandtl number is taken as 0.71. The Knudsen number values are considered between 0 and 0.10, while the modified Brinkman number is varied from −0.65 to 0.65. The non-dimensional velocity profiles are plotted by using Eq. (3.12) for different values of Knudsen number as shown in Fig. 3.5. The velocity profile for no slip condition is shown by Kn = 0, while for slip condition the profile is denoted by non-zero Knudsen number. It is clear from Eq. (3.12) that the velocity varies in y direction only, which is due to the fully developed assumption. Also, it is interesting to note that the slip velocity increases with increasing Knudsen number, while the maximum velocity decreases with the increase in Knudsen number, therefore, the velocity profile becomes relatively flat. The variation of non-dimensional temperature distribution given by Eq. (3.24) at various modified Brinkman numbers is shown in Fig. 3.6. Notice that the temperature profile is significantly distorted for finite viscous dissipation and Knudsen number. However, the effect of modified Brinkman number on the temperature profile is more profound as commented below. Figure 3.7 depicts the variation of Nusselt number with the Knudsen number for different values of the modified Brinkman number. The Nusselt number decreases monotonically with an increase in Knudsen number. The Nusselt number also shows a strong dependence on modified Brinkman number, with an increase (decrease) in Nusselt number with negative (positive) value of modified Brinkman number. This can be understood as follows: the viscous dissipation term in the equation acts as a heat source in the flow. An increased dissipation increases the bulk temperature of the fluid due to internal heating. For the case of wall heating, this increase in the fluid temperature decreases the temperature difference between the wall and bulk fluid, which results in a decrease in the Nusselt number. While for the case of cooling, the 1

Fig. 3.5 Non-dimensional velocity profile at different Knudsen number

Y

0.5 Kn = Kn = Kn = Kn = Kn = Kn =

0 -0.5 -1

0

0.00 0.02 0.04 0.06 0.08 0.10 0.5

u/um

1

1.5

3.4 Some Exact Solutions

91

1

Fig. 3.6 Non-dimensional temperature profile for various modified Brinkman number

Kn = 0.10

0.5

Brq = -0.65 Brq = -0.35 Brq = 0.00

Y

0

Brq = 0.35 Brq = 0.65

-0.5 -1 -1.2

Fig. 3.7 Variation of Nusselt number with Knudsen number at different modified Brinkman number

5.5

-1

-0.8

q *q

-0.6

Brq = 0.01 Brq = 0.00

4.5 Nu

-0.2

Brq = -0.10

P r = 0.7

5

-0.4

Brq = 0.01

4

Brq = 0.10

3.5 3 2.5 2

0

0.02

0.04

Kn

0.06

0.08

0.1

reverse applies: the temperature difference is increased with an increase in modified Brinkman number, which leads to an increase in the Nusselt number. Figure 3.8 shows the variation of Nusselt number with modified Brinkman number for different values of the Knudsen number. The figure shows a singularity and even negative values of Nusselt number, with the location of singularity being dependent on the Knudsen number. These observations may appear to be counter-intuitive but can be explained as follows: as already seen, the temperature distribution is significantly distorted due to viscous dissipation, which alters the Nusselt number as well. The difference in mean fluid temperature and wall temperature changes its sign across the singular point, while the difference is zero at the singular point.

92

3 Microscale Heat Transfer

50

Fig. 3.8 Variation of Nusselt number with modified Brinkman number at different Knudsen number

P r = 0.7

Nu

25 0

Kn Kn Kn Kn Kn Kn

-25 -50

-4

-2

Brq

0

= = = = = =

0.00 0.02 0.04 0.06 0.08 0.10 2

Fig. 3.9 Schematic diagram of gas flow through a micropipe with uniform wall heat flux boundary condition

3.4.2 Heat Transfer Analysis Through a Micropipe We now consider a fully developed laminar gas flow through a micropipe; such an analysis was carried out earlier by Aydin and Avci [14]. The direction of fluid flow is along z axis, with r axis perpendicular to it. Note that the fluid flow is independent of z, the velocity has only component u in the r direction (Fig. 3.9). The fully developed velocity profile is obtained by solving the z-momentum equation by employing the first-order slip condition with complete accommodation at the wall. The velocity normalized by the mean velocity is u 1 + 4Kn − (r/ro )2 =2 um 1 + 8Kn

(3.31)

The energy equation including the effect of viscous dissipation is u

∂T α ∂ ∂T ν ∂u 2 = r + ∂z r ∂r ∂r cp ∂r

(3.32)

3.4 Some Exact Solutions

93

where the second term in the right-hand side quantifies the effect of viscous dissipation. The thermal boundary condition at the center (r = 0) is ∂T =0 ∂r r=0

(3.33)

The constant heat flux boundary condition at the wall is ∂T qw = k ∂r r=ro

(3.34)

where qw is taken to be positive when its direction is to the fluid (the hot wall), otherwise it is negative (the cold wall). The non-dimensional variables are defined, R=

μu2m r Ts − T T − Ts μu2m , Br = , Brq = , θ= , θq = ro Ts − Tc qw ro /k k(Ts − Tc ) Dqw (3.35)

Now using Eq. (3.18) the axial temperature gradient in the fully developed region is evaluated, 8Brq dTm 2qw (3.36) = 1+ dz ρum cp R (1 + 8Kn)2 Using Eqs. (3.16), (3.31), (3.34), (3.35), and (3.36), the energy equation (3.32) in non-dimensional form is dθq 32Brq R 3 32Brq d 4 3 (R + 4KnR − R R = ) − + dR dR 1 + 8Kn (1 + 8Kn)3 (1 + 8Kn)2 (3.37) The non-dimensional boundary conditions are ∂θq = 1 at R = 1 ∂R θq = 0 at R = 1

(3.38)

The solution of the non-dimensional energy equation (3.37), using the thermal boundary conditions given by Eq. (3.38) gives the temperature distribution in terms of modified Brinkman number,

94

3 Microscale Heat Transfer

32Brq 3 4 R2 R4 T − Ts 2 − = + + − + Kn(R − 1) θq = qw ro /k 1 + 8Kn (1 + 8Kn)3 16 4 16 −

2Brq (R 4 − 1) (1 + 8Kn)2

(3.39)

Utilizing Eq. (3.9), the temperature jump at the wall is 4Knγ Ts − Tw =− qw ro /k P r(γ + 1)

(3.40)

Using Eqs. (3.39) and (3.40) the non-dimensional temperature distribution with temperature jump in terms of modified Brinkman number is 32Brq 3 4 R2 R4 T − Tw ∗ 2 θq = − = + + − + Kn(R − 1) qw ro /k 1 + 8Kn (1 + 8Kn)3 16 4 16 −

2Brq 4Knγ (R 4 − 1) − 2 P r(γ + 1) (1 + 8Kn)

(3.41)

Using Eq. (3.25) the bulk mean temperature in terms of modified Brinkman number is Brq Brq 1 16Knγ T m − Tw ∗ =− 1+ − θq,m = − qw ro /k 4 P r(γ + 1) 3(1 + 8Kn)4 (1 + 8Kn)3 −

1 + 16Brq 1 − 6(1 + 8Kn) 24(1 + 8Kn)2

(3.42)

The Nusselt number is defined, Nuq =

2 hD =− ∗ k θq,m

(3.43)

Using Eqs. (3.42) and (3.43), the Nusselt number expression in terms of modified Brinkman number is Brq Brq 16Knγ 1 1+ + N uq = 2 + 4 4 P r(γ + 1) 3(1 + 8Kn) (1 + 8Kn)3 −1 1 + 16Brq 1 + + (3.44) 6(1 + 8Kn) 24(1 + 8Kn)2

3.4.2.1

Constant Wall Temperature Case

The constant wall temperature case suggests that dTs /dz = 0, and as in the conventional case,

3.4 Some Exact Solutions

95

∂T = ∂z

T − Ts T c − Ts

dTc dz

(3.45)

Using Eqs. (3.31), (3.32), (3.35), and (3.45) we write d dR

R + 4KnR − R 3 dθ 16BrR 3 R = 2a θ+ dR 1 + 8Kn (1 + 8Kn)2

(3.46)

where a = −[um ro2 /(α(Ts − Tc ))](dTc /dz). A closed form solution of the above equation is however not possible.

3.4.2.2

Parametric Variation

We presented the effects of modified Brinkman number and Knudsen number on Nusselt number for both hydrodynamically and thermally fully developed flow. Note that Kn = 0 represents the macroscale case, while Kn > 0 shows the microscale case, and Brq = 0 represents the case of no-viscous dissipation. We employed a constant heat flux boundary condition at the wall. The velocity profiles are scaled with um from Eq. (3.50) and plotted for different values of the Knudsen number, i.e. (Kn = 0.00, 0.02, 0.04, 0.06, 0.08, and 0.1) as shown in Fig. 3.10. Here, velocity profile drawn for Kn = 0 represents the no-slip condition at wall, while the velocity profiles drawn for non-zero Knudsen numbers represent the slip condition at wall. Note that the velocity varies in radial direction only, which is due to the fully developed assumption. As explained earlier, the slip velocity increases and maximum velocity decreases with the increase in Knudsen number, which makes the velocity profile flat. 1

Fig. 3.10 Velocity distribution across the non-dimensional radial coordinate in micropipe at different Knudsen numbers

0.8 0.6 R

Kn Kn Kn Kn Kn Kn

0.4 0.2 0

0

= = = = = =

0.00 0.02 0.04 0.06 0.08 0.10

0.5

1 u/um

1.5

2

96

3 Microscale Heat Transfer

1

Fig. 3.11 Temperature distribution across the non-dimensional radial coordinate at different modified Brinkman number

Kn = 0.10 0.5 Brq = -0.65

R

0

Brq = -0.35 Brq = 0.00

-0.5 -1

Brq = 0.35 Brq = 0.65 -2.5

-2

-1.5 q*q

-1

50

Fig. 3.12 Variation of Nusselt number with modified Brinkman number at different Knudsen number

-0.5

0

P r = 0.7

Nu

25 0

Kn Kn Kn Kn Kn Kn

-25 -50 -3

-2

-1

Brq

0

1

= = = = = =

0.00 0.02 0.04 0.06 0.08 0.10 2

Figure 3.11 depicts the non-dimensional temperature profiles obtained from Eq. (3.41) using different values of modified Brinkman number (Brq = −0.65, −0.35, 0.00, 0.35, and 0.65) and a fixed value of Knudsen number. The positive value of modified Brinkman number is used to represent the wall heating case, negative values to the wall cooling case, and Brq = 0 represents the case of no viscous dissipation. The viscous dissipation acts as an energy source and increases the fluid temperature, especially near the wall due to higher shear rate. Here, the effect of viscous dissipation on the non-dimensional temperature profile is clearly visible from Fig. 3.11. The variation of Nusselt number with the modified Brinkman number for different Knudsen number (Kn = 0.00 to 0.10) is shown in Fig. 3.12. As explained earlier, we observed the singularities (shown by vertical lines) in Nusselt number for a particular Knudsen number, which discloses the fact that the heat supplied by the wall to fluid is balanced with the heat generated due to the effect of viscous dissipation.

3.4 Some Exact Solutions

97

8

Fig. 3.13 Variation of Nusselt number with Knudsen number at different modified Brinkman number

7

Nu

Brq = - 0.10

P r = 0.7

Brq = - 0.01

6

Brq = 0.00

5

Brq = 0.10

Brq = 0.01

4 3 2

0

0.02

0.04

0.06

0.08

0.1

Kn

Figure 3.13 depicts the variation of Nusselt number with the Knudsen number at different modified Brinkman number, i.e. (Brq = −0.10, −0.01, 0.00, 0.01, and 0.10). The Nusselt number is found to decrease with the increase in Knudsen number. Also, for lower values of Knudsen number the effect of viscous dissipation is more pronounced as compared to the higher Knudsen number. For the case of no-viscous dissipation, i.e. Brq = 0, the Nusselt number decreases with the increase in Knudsen number. For positive values of modified Brinkman number the fluid is considered to be heated (Tw > Tm ) and opposite is true for the negative values of Brq . For the case of wall heating, viscous dissipation decreases the temperature difference between the wall and bulk fluid. On the contrary, for the cooling case it increases the temperature difference between both the wall and bulk fluid by increasing the fluid temperature more. It is interesting to note that the Nusselt number is higher for negative modified Brinkman number as compared to the positive values. The Nusselt number decreases with the increase in Knudsen number. However, for Brq = 0.10, the Nusselt number increases up to a certain value of Kn(≈ 0.01), beyond it the trend is found to be similar again.

3.4.3 Heat Transfer Through a Micro-Annulus We presented the heat transfer characteristics of a gas flow though a micro-annulus reported by Aydin and Avci [13]. Two types of thermal boundary conditions are considered, namely, outer wall at constant heat flux and inner wall insulated (case A, refer Fig. 3.14a); and outer wall insulated and inner wall at constant heat flux (case B, refer Fig. 3.14b).

98

3 Microscale Heat Transfer

(a)

(b)

Fig. 3.14 Schematic diagram of gas flow through a micro-annulus (a) Outer wall kept at constant heat flux and inner wall insulated (case A) (b) Outer wall insulated and inner wall kept at constant heat flux (case B)

The momentum equation in z-direction is first solved: 1 ∂ ∂u 1 dp = constant r = r ∂r ∂r μ dz The velocity slip boundary conditions at the wall are 2 − σv ∂u u = us = λ σv ∂r r=ri 2 − σv ∂u u = us = − λ σv ∂r r=ro

(3.47)

(3.48)

where σv denotes tangential momentum accommodation coefficient and taken as unity for most of the engineering problems. The non-dimensional variables are ri u r (3.49) R = , r∗ = , U = ro ro um The fully developed velocity profile is derived from the momentum equation by employing the first order velocity slip condition at the wall. By using Eqs. (3.48) and (3.49) the velocity distribution is determined as ∗2 ln R + A 1 − R 2 + 2rm U =2 (3.50) B where ∗2 A = 4Kn(1 − r ∗ )(1 − rm ), r ∗2 ∗2 1 ∗ + ln r B = 1 − r ∗2 − 4rm 2 1 − r ∗2 ∗2 ), +8Kn(1 − r ∗ )(1 − rm

3.4 Some Exact Solutions

99

∗ (= rm ) is the dimensionless mean Kn is the Knudsen number Kn = Dλh , and rm ro radius. This result will be used while solving the energy equation later on. The point of maximum velocity can be obtained by setting ∂u ∂r = 0. The nondimensional radius is ⎡ ⎤1/2 ⎢ ⎢ ∗ rm =⎢ ⎣

(1 − r ∗2 )(1 + 4Kn) ⎥ ⎥ ⎥ ∗2 r −1 ⎦ 1 2 ln ∗ − 4Kn r r ∗2

Further, for the limiting condition Kn = 0, the above equation reduces to 1 − r ∗2 1/2 ∗ rm = 2 ln(1/r ∗ )

(3.51)

(3.52)

which is a well-established result reported in the literature (Rohsenow et al. [121]). The energy equation including the effect of viscous dissipation is α ∂ ∂T ∂T ν ∂u 2 = u (3.53) r + ∂z r ∂r ∂r cp ∂r The constant heat flux boundary condition at the wall is ∂T qw = k ∂r r=ro

(3.54)

The non-dimensional variables are θ=

μu2m T − Ts T m − Tw , θm∗ = , Br = qw ro /k qw ro /k qw ro

(3.55)

Using Eqs. (3.16), (3.50), (3.54), and (3.55), the energy equation (3.53) in nondimensional form is dU 2 dθ 1 d R = aU − Br (3.56) R dR dR dR um kro dTs . α qw dz For case A, the non-dimensional boundary conditions are

where a =

∂θ θ = 0, =1 ∂R R=1 ∂θ =0 ∂R R=r ∗

(3.57)

100

3 Microscale Heat Transfer

The solution of non-dimensional energy equation (3.56) is obtained by integrating it twice with the thermal boundary conditions given by Eq. (3.57), a 3 R2 T − Ts ∗2 2 ∗2 = − − A + 2rm + R 1 + A − 2rm − − ln R θ (R) = qw ro /k 2B 4 4 Br ∗2 ∗2 1 + 2A − 2rm (1 + R 2 ) + 2 (1 − R 2 )(1 + R 2 − 8rm ) B ∗2 ∗4 2 (3.58) + 4 ln R(1 − 4rm ) − 8rm (ln R) + ln R where a=

∗2 ) + 32Br r ∗4 ln r ∗ −2B 2 + 8Br(r ∗2 − 1)(1 + r ∗2 − 4rm m ∗2 ∗2 ∗2 ∗2 r ∗2 ln r ∗ B(1 + 2A − 2rm − r )(r − 1) + 4B rm

For case B, the non-dimensional boundary conditions are θ = 0,

∂θ = −1 ∂R R=r ∗ ∂θ =0 ∂R R=1

(3.59)

Similarly, the solution of non-dimensional energy equation (3.56) is obtained by integrating it twice with the thermal boundary conditions given by Eq. (3.59), a T − Ts ∗2 = (R 2 − r ∗2 )(1 + A − 2rm ) − ((R 4 − r ∗4 )/4) − (ln R − ln r ∗ ) θ (R) = qw ro /k 2B Br ∗2 ∗2 2 ∗2 ∗ ∗2 (1 + 2A − 2rm ) + 2rm (R ln R − r ln r ) + 2 (R 2 − r ∗2 )(8rm B ∗2 ∗4 − (R 2 + r ∗2 )) + 4(ln R − ln r ∗ )(1 − 4rm ) − 8rm ((ln R)2 − (ln r ∗ )2 ) (3.60) where

∗2 ) + 32Br r ∗4 ln r ∗ −2B 2 r ∗ + 8Br(r ∗2 − 1)(1 + r ∗2 − 4rm m a= ∗2 − r ∗2 )(r ∗2 − 1) + 4B r ∗2 r ∗2 ln r ∗ B(1 + 2A − 2rm m

Using Eq. (3.9) the temperature jump at the wall can be written Ts − T w 4Knγ =− (1 − r ∗ ) qw ro /k P r(γ + 1)

(3.61)

3.4 Some Exact Solutions

101

Using Eqs. (3.58) and (3.61), the non-dimensional temperature distribution with temperature jump is a 3 T − Tw R2 ∗2 2 ∗2 = − − A + 2rm + R 1 + A − 2rm − θ (R) = qw ro /k 2B 4 4 Br ∗2 ∗2 − ln R 1 + 2A − 2rm (1 + R 2 ) + 2 (1 − R 2 )(1 + R 2 − 8rm ) B 4Knγ ∗2 ∗4 2 (1 − r ∗ ) + 4 ln R(1 − 4rm ) − 8rm (ln R) + ln R − P r(γ + 1) (3.62) ∗

Similarly, using Eqs. (3.60) and (3.61), the non-dimensional temperature distribution is a T − Tw ∗2 θ ∗ (R) = = (R 2 − r ∗2 )(1 + A − 2rm ) − ((R 4 − r ∗4 )/4) qw ro /k 2B ∗2 ∗2 ) + 2rm (R 2 ln R − r ∗2 ln r ∗ ) − (ln R − ln r ∗ )(1 + 2A − 2rm Br ∗2 ∗2 − (R 2 + r ∗2 )) + 4(ln R − ln r ∗ )(1 − 4rm ) + 2 (R 2 − r ∗2 )(8rm B 4Knγ ∗4 2 ∗ 2 (1 − r ∗ ) (3.63) − 8rm ((ln R) − (ln r ) ) − P r(γ + 1) Using Eq. (3.25) the non-dimensional bulk mean temperature for case A is θm∗ =

4(p1 + p2 + p3 ) 4Knγ (1 − r ∗ ) − ∗2 P r(γ + 1) 1 + 2A − 2rm

(3.64)

where ∗2 )(−3 − 4A + 8r ∗2 ) ∗2 )(2 + 6A − 3r ∗2 ) (3 + 6A − 6rm a (1 + 2A − 2rm m m + p1 = 2B 16 72 ∗2 ) ∗2 ) ∗2 ) (1+2A)(−3−4A+8rm r ∗2 (−32−45A+81rm (3+12A−4rm − + m − 72 16 72 ∗2 − 24r ∗2 (4 + 6A − 9r ∗2 ) Br 15 + 24A − 32rm m m p2 = 2 72 B ∗2 )(−3 − 4A + 8r ∗2 ) ∗4 (7 + 8A − 24r ∗2 ) (1 − 4rm rm m m + − 4 4

102

3 Microscale Heat Transfer

and p3 =

∗2 −5 − 9A + 18rm . 36

Similarly the non-dimensional bulk mean temperature for case B is θm∗ =

4(p4 + p5 ) 4Knγ (1 − r ∗ ) − ∗2 P r(γ + 1) 1 + 2A − 2rm

(3.65)

where p4 =

∗2 ∗2 2 + 6A − 3rm 1 + 2A − 2rm a ∗2 ) − r ∗2 (1 + A − 2rm 2B 24 4 ∗2 ∗2 1 3 + 12A − 4rm 1 + 2A − 2rm − − r ∗4 4 72 4 ∗2 ∗2 −3 − 4A + 8rm ∗2 ∗ 1 + 2A − 2rm − ln r − (1 + 2A − 2rm ) 16 4 ∗2 (−5 − 9A + 9r ∗2 ) rm m ∗2 ∗ ∗2 − r ln r (1 + 2A − 2rm ) + 2 36

and

∗2 −72r ∗4 −12A(1−12r ∗2 )−3 ∗2 ∗2 ∗2 52rm m m ∗2 (8rm −r )(1+2A−2rm ) −r 72 4 ∗2 ∗2 1 + 2A − 2rm −3 − 4A + 8rm ∗2 + 4(1 − 4rm − ln r ∗ ) 16 4 ∗2 ∗2 1 + 2A − 2rm ∗4 7 + 8A − 24rm − (ln r ∗ )2 (3.66) − 8rm 32 4

Br p5 = 2 B

Using expressions derived above, the Nusselt number can be obtained from, Nu =

3.4.3.1

qw Dh 2 = − ∗ (1 − r ∗ ) k(Tw − Tm ) θm

(3.67)

Parametric Variation

We presented the results of Avci and Aydin [13] for gas flow through a concentric cylindrical annulus by employing two different thermal boundary conditions, such as: uniform heat flux at the outer wall and adiabatic inner wall (case A) and uniform heat flux at the inner wall and adiabatic outer wall (case B). The effect of rarefaction and viscous dissipation on temperature distribution and Nusselt number is discussed.

3.4 Some Exact Solutions

103

1

Fig. 3.15 Variation of non-dimensional velocity with non-dimensional radial co-ordinate at different Knudsen number

R

0.8 Kn Kn Kn Kn Kn Kn

0.6

0.4

0.2

0

1

Fig. 3.16 Variation of non-dimensional temperature with non-dimensional radial co-ordinate at different Knudsen number (case A)

= = = = = =

0.9

0.00 0.02 0.04 0.06 0.08 0.10 0.5

U

1

Br = 0.6 P r = 0.7

R

0.8

Kn Kn Kn Kn Kn Kn

0.7 0.6 0.5 0.4 -1

1.5

-0.8

-0.6 q

*

-0.4

= = = = = =

-0.2

0.00 0.02 0.04 0.06 0.08 0.10

0

Figure 3.15 illustrates the non-dimensional velocity profiles obtained for different values of the Knudsen number at r ∗ = 0.2. The velocity values at R = 0.2 and R = 1 (denotes the inner and outer walls of the annulus) represent the non-dimensional values of the velocity slip. The velocity distribution and the slip velocities are not symmetrical at the walls. A point having maximum value of the velocity occurs near to the inner wall and can be obtained by setting ∂u ∂r = 0. Note that with the increase in Knudsen number, the slip velocity increases at the inner wall and outer wall of the micro annulus, while the maximum value of the velocity decreases with the increase in Knudsen number, which makes the velocity profile flat, as shown in Fig. 3.15. For both cases A and B, the variation of non-dimensional temperature with non-dimensional radial co-ordinate for different values of Knudsen numbers at Br = 0.6 is shown in Figs. 3.16 and 3.17, respectively. Here, Kn = 0 represents the temperature profile without rarefaction, while non-zero Kn represent the

104

3 Microscale Heat Transfer

1

Fig. 3.17 Variation of non-dimensional temperature with non-dimensional radial co-ordinate at different Knudsen number (case B)

0.9

Br = 0.6 P r = 0.7 Kn Kn Kn Kn Kn Kn

R

0.8 0.7 0.6

= = = = = =

0.00 0.02 0.04 0.06 0.08 0.10

0.5 0.4

-0.8

-0.6

-0.4 q*

-0.2

0

50

Fig. 3.18 Variation of Nusselt number with Knudsen number at different Brinkman number (case A)

r * = 0.2

Nu

25

0 Br Br Br Br Br

-25

-50

0

0.02

0.04

0.06

= = = = =

-0.10 -0.01 0.00 0.01 0.10

0.08

0.1

Kn temperature profile with the effect of rarefaction. The rarefaction is found to affect the temperature distribution in the presence of viscous dissipation. It is noticed that the increase in Knudsen number flattens the temperature distribution significantly. For case A (outer wall kept at constant heat flux and inner wall insulated), the variation of Nusselt number with Knudsen number for various values of Brinkman number at r ∗ = 0.2 is shown in Fig. 3.18. For Br = 0, the effect of viscous dissipation is absent and fully developed Nusselt number decreases with the increase in Knudsen number due to the increased rarefaction effect. As Knudsen increases the temperature jump at wall increases, which leads to a decrease in heat transfer. It is interesting to note that the effect of viscous dissipation is more pronounced at lower Knudsen number, while for higher values of Knudsen number this effect is insignificant. In addition, viscous dissipation causes singularity in the Nusselt number at a particular Knudsen number. For negative Brinkman number, the Nusselt

3.4 Some Exact Solutions

50.0 r* = 0.2 25

Nu

Fig. 3.19 Variation of Nusselt number with Brinkman number at different Knudsen number (case A)

105

0

Kn Kn Kn Kn Kn Kn

-25

-50 -0.1

-0.05

0

Br

= 0.00 = 0.02 = 0.04 = 0.06 = 0.08 = 0.10

0.05

0.1

number is found to be more as compared to the case of positive Brinkman number. This is because near the wall, fluid temperature increases due to viscous dissipation, therefore, the temperature difference between wall and fluid increases, which results in an enhanced heat transfer as compared to the case of no viscous dissipation. Note that the heat transfer behavior is affected by the combined effect of rarefaction and viscous dissipation as shown in Fig. 3.18. Similarly, for case A, the variation of Nusselt number with Brinkman number for various values of the Knudsen number at r ∗ = 0.2 is shown in Fig. 3.19. Here, the vertical lines show the singularity in Nusselt number at a particular Knudsen number, that is because the mean temperature reaches the wall temperature. While, the negative Nusselt number shows that the mean temperature is higher as compared to the wall temperature. In addition, it is observed that the effect of viscous dissipation, quantified by Brinkman number, is effective for lower values of Knudsen number, as shown in Fig. 3.19. For case B (outer wall insulated and inner wall kept at constant heat flux) Fig. 3.20 describes the variation of Nusselt number with Knudsen number for different values of the Brinkman number at r ∗ = 0.2. The effect of rarefaction, quantified by Knudsen number, is to reduce heat transfer due to the increased temperature jump, which lowers the fluid temperature due to large temperature gradients close to the wall. For case B (outer wall insulated and inner wall kept at constant heat flux) the variation of Nusselt number with the Brinkman number for different values of the Knudsen number is shown in Fig. 3.21. The vertical lines show the points of singularity in Nusselt number, however there is no physical significance of the singular points; these points indicate that Tm = Tw . It is observed that the effect of viscous dissipation is more pronounced at lower values of Knudsen number, while for high Kn value, it is insignificant as demonstrated in Fig. 3.21.

106

3 Microscale Heat Transfer

50

Fig. 3.20 Variation of Nusselt number with Knudsen number at different Brinkman number (case B)

r * = 0.2

Nu

25

0 Br Br Br Br Br

-25

-50

0

0.02

0.04

0.06

= = = = =

0.08

-0.1 -0.01 0.00 0.01 0.1 0.1

Kn 50

Fig. 3.21 Variation of Nusselt number with Brinkman number at different Knudsen number (case B)

r = 0.2

Nu

25

0

-25

-50 -0.1

Kn Kn Kn Kn Kn Kn

= = = = = =

0.00 0.02 0.04 0.06 0.08 0.10

-0.05

0

Br

0.05

0.1

3.5 Observations on Other Effects In the above section, we have obtained exact solution of the governing equation for a few relatively simple cases. However, in presence of additional effects—such as axial conduction, compressibility, wall conduction, property variation although governing equations can still be written—the analytical solution of the governing equations becomes difficult and one rather solves the equations numerically.

3.5.1 Variation in Thermophysical Properties Among the relevant thermophysical properties, viscosity, density, and thermal conductivity show a stronger dependence on temperature as compared to heat capacity. Therefore, the variation of these properties with temperature has to be

3.5 Observations on Other Effects

107

suitably accounted for in understanding the variation of velocity and temperature profiles and the Nusselt number. The influence of density and thermal conductivity variations due to temperature gradient in the flow is measured in terms of entropy generation. The rate of entropy generation per unit volume in the fluid S˙ is directly proportional to the amount of viscous dissipation (Φ, which depends on the square of velocity gradient; Eq. (3.3)) and the square of temperature gradient; mathematically, k Φ + 2 (∇T )2 S˙ = T T

(3.68)

The variation of properties leads to an increase in the velocity (and temperature) gradient in certain region and a decrease in some other region, with respect to the baseline case of constant property, which alters the rate of entropy generation per unit volume in the fluid. Therefore, the integral value of S˙ over the entire region is computed in order to access the effect of property variation better. In the context of flow in a microtube, Prabhu and Mahulikar [119] reported that the axial and radial variations in density cause a distortion in the velocity profile. The density gradients cause flattening of axial velocity profile leading to an effect that is similar to “flow un-development.” Similar to flow in the developing region, this induces a radial flow in order to satisfy the condition for continuity of mass. The presence of radial velocity also leads to a radial convection, which can significantly affect the overall convection in the microtube. The variation of thermal conductivity along the flow direction leads to axial conduction and an increase in the fluid temperature, while the radial variation of thermal conductivity causes a flattening of the temperature profile. The overall rate of entropy generation was found to increase in presence of property variation.

3.5.2 Conduction in the Substrate Frequently, the thickness of the substrate on which the microchannels are etched is relatively large compared to the flow passage. This leads to substantial amount of conduction of heat in the substrate itself. So if a uniform heat flux is applied on the outer wall of the microchannel, which is typically the case, the heat flux on the inner wall is non-uniform (Fig. 1.9). The amount of non-uniformity is a function of various parameters including wall thermal conductivity and wall thickness. Maranzana et al. [92] defined a parameter called the axial wall conduction number to quantify when this wall conduction will become large enough so that we need to account for it. This parameter, M, is a ratio of heat transferred as conduction in the wall (opposite to the direction of the flow) to the heat transferred as convection from the surface, and expressed mathematically, M=

Φcond kso Aso (Ts,o − Ts,i ) = Φconv Lmc ˙ p (Tf,o − Tf,i )

(3.69)

108

3 Microscale Heat Transfer

where Φcond is a heat flux characterizing axial heat transfer in the wall, Φconv is the total convective heat flux, kso , Aso , and L are, respectively, the thermal conductivity, cross-sectional area and length of the microchannel solid substrate. Further, Ts,o and Ts,i are the outlet and inlet temperature of substrate, while Tf,o and Tf,i are the outlet and inlet temperature of fluid. The threshold value of M beyond which the wall conduction effect should be considered is 0.01. A detailed analysis of this conjugate effect in the context of a microchannel has been made by Mohrana and Khandekar [100]. Further, the conjugate effect has been utilized to obtain a uniform wall temperature on the inner wall of a diverging microchannel, while supplying a uniform heat flux on the outer wall, in Duryodhan et al. [50]. The above studies are however with liquid flow. We note that although wall thickness at the microscale is typically smaller than that at the conventional scale in absolute sense, the relative thickness (ratio of thickness of wall to thickness of flow passage) is much larger at the microscale, making conjugate effect a very relevant issue in microchannel heat transfer problems. That is, at the conventional scale although wall conduction does occur, its relative importance with respect to convection in affecting the heat flux distribution is insignificant and therefore the conjugate effects have conventionally been neglected.

3.5.3 Axial Conduction The importance of axial conduction can be accessed from the value of Peclet numρc u D ber, P e. Since P e = p km h , upon multiplying the numerator and denominator of this equation by the axial fluid bulk temperature gradient (dTm /dx), we can write this expression in terms of mass flow rate (m) ˙ through the microchannel, Pe =

mc ˙ p Dh (dTm /dx) kAo (dTm /dx)

(3.70)

Thus, Peclet number signifies the ratio of rates of advection to diffusion or it represents the relative magnitude of the thermal energy transported to the fluid (as fluid enthalpy change) to the thermal energy axially conducted within the fluid. The Peclet number is also the product of the Reynolds and Prandtl numbers. Due to a reduction in Reynolds number, the Peclet number reduces, and the convection term no longer dominates over conduction of heat in the fluid. That is, the longitudinal (or axial) conduction of heat can no longer be neglected in comparison to lateral conduction or convection. Note that in the presence of axial heat conduction both Nusselt number and thermal entry length increases. Also, the effect of axial conduction on Nusselt number increases with a decrease in Knudsen number. The thermal entrance region increases with an increase in Peclet number. It is apparent from the theoretical analysis of Satapathy [124] that temperature starts to change much before the gas enters the heating zone for P e < 1. Further, a change in the value of Knudsen number does not seem to change the variation

3.5 Observations on Other Effects

109

of temperature in pre-heating zone much, although the effect is evident in the post heating zone. This analysis is however for uniform fluid velocity. A similar observation can be made from experimental data as discussed in Sect. 3.6.

3.5.4 Flow Work and Shear Work The finite velocity at the wall (due to slip) combined with the shear stress produces a shear work at the boundary (Fig. 3.22), while this effect is absent at the macroscale due to the no-slip condition. This shear work at the solid boundaries can make a significant contribution in microscale gas flows, especially when the slip effect is substantial. The steady energy equation considering the effects of viscous dissipation and flow work can roughly be written ρcp u

∂ 2T du ∂T dp =k 2 +u + τxy ∂x dx dy ∂y

(3.71)

dp where u dx and τxy du dy terms denote the flow work and heat generated due to viscous heating, respectively. Note that the two terms have opposite sign: The flow work is negative because pressure usually decreases with x, while viscous heating (μ(du/dy)2 ) is positive, and the shear work (τxy × us = μus du dy ) due to slipping of the gas at boundary balances the contribution of viscous dissipation and flow work. The contribution of shear work at macroscale is zero due to the assumption of novelocity slip boundary condition. At the microscale shear work plays a determining role in the convective heat transfer problems under the assumption of (locally) fully developed and constant heat flux boundary condition. Therefore, it should be considered in the calculation of total heat exchange with the walls, however, it has no direct effect on the temperature field as it happens at the boundary only [58]. dp The term u dx acts as a distributed heat sink, with the majority of thermal energy absorbed near the center of flow, due to the larger velocity magnitude in this region. On the contrary, shear work acts as a heat source due to the thermal energy generated by the flow. In general, the two opposite effects may not cancel each other, and therefore the implications of these works on the overall flow and heat transfer behavior have to be carefully analyzed.

Fig. 3.22 Contribution of shear work at wall due to slipping of the gas

110

3 Microscale Heat Transfer

Fig. 3.23 (a) The axial variation of centerline fluid temperature along the length of tube (b) Variation of Nusselt number with Knudsen number at different Reynolds number; Figures taken from [73]

3.6 Observations from Experiments The amount of experimental data for flow in the slip regime in any configuration is rather limited. Hemadri et al. [73] employed a counter-flow tube-in-tube heat exchanger for measurement of the overall heat transfer coefficient in the slip regime (of the type shown in Fig. 8.4c). Towards this, low-pressure nitrogen flowed through the conventional size inner tube of the heat exchanger and picked up heat from the hot water flowing in the outer jacket. Knowing the mass flow rate of nitrogen and water, and the respective inlet and outlet temperatures, allows the overall heat transfer coefficient to be determined. Further, local measurement of fluid temperature, with the help of thermocouples mounted at different axial locations at the centerline of the tube, allowed the local heat transfer coefficient to be determined. These measurements showed a large amount of heating of the incoming flow, owing to axial back conduction of heat in the gas (Fig. 3.23a). As evident from the figure, the temperature of the gas reached the wall temperature over a relatively short distance from the entry of the tube. The Nusselt number value was found to be much lower than expected from available theoretical analysis (Fig. 3.23b). Further, the value of Nusselt number showed a stronger dependence on Reynolds number rather than Knudsen number, as apparent from the figure.

3.7 Application to Knudsen Pump Thermal creep is an interesting flow phenomenon based on which a pump for pumping gases can be constructed. Consider two reservoirs containing same gas maintained at different temperatures, connected by a microchannel. The width of

3.7 Application to Knudsen Pump

111

microchannel is comparable to the mean free path of the gas. Then a flow of gas occurs from the cold reservoir to the hot reservoir, and this phenomenon is called thermal creep (or thermal transpiration). As per kinetic theory, the mass of gas molecules crossing unit area per unit time towards one side is given by p RT 1 Γm = ρ v¯ = ρ =√ (3.72) 4 2π 2π RT √ flow of where v¯ is the mean molecular speed (= 8RT /π ). Therefore, the mass √ molecules crossing from the two sides of the reservoir is proportional to p/ T . So if the reservoirs are maintained at the same pressure but different temperatures, the flux of molecules crossing from hot to cold side, and the flux of molecules crossing from cold to hot side will be different; with the latter flux being larger than the former flux. This leads to a net movement of gas from the cold to the hot reservoir. The cross-over of molecules to the other reservoir will continue until the pressures become equal on the two sides or ( p1 T1 = . (3.73) p2 T2 Based on this idea, McNamara and Gianchandani [96] constructed a pump (called the Knudsen pump). A schematic of their micro-pump is shown in Fig. 3.24. The pressure generated by the pump can be increased by having multiple-stages of the pump.

Fig. 3.24 Layout of a Knudsen pump

112

3 Microscale Heat Transfer

Note that the analysis presented here involves several simplifying assumptions. The flow in the microchannel is also required to be in the free-molecular or transition regime. However given this interesting application of gaseous flow in small passages, we have covered it in this section.

3.8 Useful Empirical Correlations Some useful empirical correlations are compiled in this section for ready reference. Tunc and Bayazitoglu [150] proposed the following expression for local Nusselt number (N ux ) for flow in a tube involving velocity slip, temperature jump, and viscous dissipation, Nux =

2 4Knγ − θm θs + P r(γ + 1)

(3.74)

where θs is the non-dimensional temperature of the fluid at wall and θm is the mean bulk temperature. The above equation is for uniform wall heat flux condition. The above correlation was obtained analytically by solving the governing equation using the integral transform method. Similarly, Nusselt number expression for constant wall temperature condition was proposed as, ∂θ −2 ∂η η=1 hx D = Nux = 4Knγ ∂θ k θm − P r(γ + 1) ∂η η=1

(3.75)

where θ is the non-dimensional temperature and η is the non-dimensional radial co-ordinate. Inman [77] proposed an expression to evaluate heat transfer characteristics of laminar gas flow in a parallel plate channel or circular tube by employing constant heat flux boundary condition in the slip flow regime,

ξ 17 + 84ζ + 105ζ 2 Nu = + 4 140(1 + 3ζ )2

−1 (3.76)

2 − σv 2 − σT 8Knγ . and ξ = σv σT P r(γ + 1) Miyamoto et al. [98] considered argon flow in a small parallel plate channel to study heat transfer characteristics by employing the constant heat flux boundary condition at the wall. The effect of velocity slip, temperature jump, and viscous

where ζ = 4Kn

3.9 Summary

113

dissipation is included. The expression for Nusselt number is 9Br 3 + 42Kn + 140Kn2 8Kn2 γ Nu = + 35(1 + 6Kn) P r(γ + 1) (1 + 6Kn)2 −1 17 + 168Kn + 420Kn2 Knγ + + P r(γ + 1) 140(1 + 6Kn)2

(3.77)

Hooman [74] proposed a correlation to estimate heat transfer characteristics of a gas flow between parallel plate microchannels by using superposition approach, −1 2 − σT 17σv +84Knσv +0.5Kn(14+70Kn)(24 − 12σv ) Knγ Nu = + σT P r(γ + 1) 140(1+6Kn)(σv +0.5Kn(24−12σv )) (3.78) Niazmand et al. [105] reported correlations to evaluate Nusselt number for slip flow through rectangular and trapezoidal channels, (N u)f d

−0.26 0.21 90 90 G2 G3 = 2.87 + 4.8 exp − 3.9α φ φ

(3.79)

where α denotes the aspect ratio, φ is the acute side angle, and G2 = 1 − 1.75 Kn0.64 (1 − 0.72 tanh(2α)), G3 = 1 + 0.075(1 + α) exp(−0.45ReP r).

3.9 Summary In this chapter we have discussed about the microscale heat transfer and factors affecting heat transfer characteristics, such as: velocity slip, temperature jump, and viscous dissipation. Some exact solution for these cases were obtained. The strong effect of viscous dissipation on heat transfer is particularly noteworthy; the Nusselt number can even become negative when the temperature difference between the wall and mean temperature changes its sign. However, these solutions neglected factors such as axial conduction, pressure work, and compressibility effects, which can be important as evident from the experimental data. Property variation and conjugate effects can also play a substantial role and affect the velocity and/or temperature profiles. It is therefore worthwhile to innovatively design and conduct experiments in the slip regime where the effect of individual parameters on Nusselt number can be isolated, and the influence of microscale parameters can be better quantified.

Chapter 4

Need for Looking Beyond the Navier–Stokes Equations

The need for looking beyond the Navier–Stokes equations is addressed in this chapter through specific examples where these equations fail. We also examine some extensions of the Navier–Stokes equations, which have been recently proposed in the literature. Similarly, attempts to modify the Fourier law to account for nonFourier effects are reviewed. An example of shock wave where these alternative forms of Navier–Stokes equations have been applied is also included.

4.1 Introduction In the two previous chapters, we examined the solutions for gaseous slip flow and heat transfer under various conditions. The obtained solutions are primarily applicable in the slip regime; the extension in the range of Knudsen number over which the solution is applicable is only modest. In order to model flows at still higher Knudsen number range (particularly in the transition regime), the Navier– Stokes equations are no longer applicable, and there is a need to look beyond these conventional equations. However, before examining extension of the flow model to higher Knudsen number (which is undertaken in the next three chapters), we examine some relatively simple extensions of the Navier–Stokes and Fourier-order energy equations in the present chapter. An important reason why such extension is being sought is that numerical methods for solving the Navier–Stokes equations (or the field of computational fluid dynamics) are already well developed [38, 128] and its extension to any additional terms in the conventional Navier–Stokes equations is deemed to be not that difficult.

© Springer Nature Switzerland AG 2020 A. Agrawal et al., Microscale Flow and Heat Transfer, Mechanical Engineering Series, https://doi.org/10.1007/978-3-030-10662-1_4

115

116

4 Need for Looking Beyond the Navier–Stokes Equations

4.2 Examples Needing a Look Beyond the Navier–Stokes Equations We start by citing few problems whose treatment within the Navier–Stokes formulation is known to be problematic. 1. Couette flow in the later slip and early transition regimes: As discussed in Sect. 1.5.2, DSMC simulations suggest a non-linear velocity profile in the later slip and early transition regimes (Fig. 1.10b). The Navier–Stokes equations predict a linear profile (Eq. (2.58)), suggesting that the non-linearity cannot be modelled within the framework of these equations. As an aside, we note that the DSMC data suggests that the velocity profiles become linear again at still higher Knudsen numbers [133]. 2. Poiseuille flow in the transition regime: As discussed in Sect. 1.5.2, for flow in a microchannel, the Navier–Stokes equations predict a uniform pressure in the lateral direction (Eq. (2.78)). However, DSMC data shows that pressure varies along the lateral direction (Fig. 1.11c). This non-uniform value of pressure along the lateral direction cannot be modelled within the framework of the Navier– Stokes equations. 3. Flow near the leading edge of a plate: For uniform flow encountering a stationary flat plate, the Navier–Stokes equations predict an infinite coefficient of friction and an infinite Nusselt number at the leading edge of the plate. While the Navier– Stokes equations based solution is well applicable to regions slightly downstream of the leading edge, the singularity at the leading edge needs to be modelled by the Grad equation (or another higher-order continuum transport equation). This example can be extended to include flows at the entrance of a passage where such singularity appears. 4. Impulsively started plate in a flow: The impulsively started flow problem is analogous to the problem of flow near the leading edge of a plate. The Navier– Stokes equations are not applicable to time which is just after the start of the plate motion, although the equations are applicable to slightly later times. This point is discussed at length in Sect. 6.7 where solutions from various approaches are compared. 5. Flow in sudden contraction: The Navier–Stokes equations do not capture the gross features of flow in a sudden contraction in the slip regime well. In contrast, numerical results (essentially solving the Navier–Stokes equations) for the case of sudden expansion (i.e., with flow direction reversed) at otherwise comparable values of governing parameters compare reasonably well with experimental data for both global and local flow features. (This point was noted by Vijay V. Varade but the numerical evidence for it is still unpublished.) Further, the absence of flow separation in sudden expansion while early onset of flow separation in bend passages are interesting observations, and it is not clear if these observations can be quantitatively described within the framework of the Navier–Stokes equations.

4.3 Extended Navier–Stokes Equations

117

6. Shock wave: The flow properties change rapidly across a shock with the gradients being confined in a region of the order of the mean free path of the gas on either side of the shock. The flow in the vicinity of the shock layer is far from local thermodynamic equilibrium, and the Navier–Stokes equations do not perform well in this region. Predicting the shock structure correctly therefore forms an important test case for any higher order transport equation. 7. Propagation of waves in atmosphere of Mars: The atmosphere of Mars is considerably more rarefied than that of other terrestrial environments. Therefore, as recognized in recent works (for example, see reference [8, 112]), the treatment of infrasonic and ultrasonic waves in the atmosphere of Mars cannot be based on the Navier–Stokes equations. These authors suggested employing higher orders of the Boltzmann equation for capturing the high Knudsen number regimes likely to be encountered in the transmission of these waves. 8. Heat transfer in microchannels and cavities: DSMC calculation of heat transfer in microchannel [16] and cavities [99] has shown several non-intuitive effects. For example, the direction of rotation of vortices in a cavity with Navier–Stokes equations was found to be opposite to that predicted by DSMC and some higherorder transport equations, even at moderate values of Knudsen number. Further, anti-Fourier heat transfer, i.e. transfer of heat from cold-to-hot regions of the flow field was noted from the DSMC simulations for heat transfer in a microchannel. The experimentally measured value of Nusselt number for flow of rarefied gas in a tube is much smaller than the predicted values, as seen in Sect. 3.6. Clearly, these effects are not within the ambit of the conventional equations. The two main extensions of the Navier–Stokes equations are mentioned in the next two sections. This is followed by a discussion on a popular non-Fourier model.

4.3 Extended Navier–Stokes Equations Brenner [25] proposed an extension to the Navier–Stokes equation, with the intention to explain the thermophoretic motion of particles. Brenner’s hypothesis is that the velocity involved in the constitutive equation (Eq. (2.2)) is not same as velocity which brings in mass in to a control volume (Eq. (2.1)). He argued that the former is volume velocity (uv ) whereas the latter is mass velocity (um ). Here, the boldface symbol denotes a vector quantity. Dye added to the flow will show volume velocity, whereas the mean velocity of the molecules comprising the fluid element is mass velocity. The two velocities are however related through the following relation: uv = um + jv

(4.1)

where jv = α∇(ln ρ) represents the diffusive flux of volume, and α is thermal diffusivity. Note that jv = 0 for incompressible flows, and we obtain uv = um for this special case. At the wall, the mass velocity is supposed to go to zero.

118

4 Need for Looking Beyond the Navier–Stokes Equations

This modification, if applicable, changes the constitutive relation to σ = σ N S + 2μ∇jv

(4.2)

where σ is the stress tensor and the overbar denotes the symmetric and traceless part of the tensor (equivalent to angular bracket in our tensor notation). Therefore, both the Navier–Stokes and energy equations change to: ρ ρCp

∂um + um · ∇um ∂t

∂T + um · ∇T ∂t

= −∇p + ∇ · (2μ∇uv ) + ∇(μ2 ∇ · uv )

= ∇ · (k∇T ) + ∇ · p(uv − um ) −

∂ ln ρ ∂ ln T

+ μ2 (∇ · uv )(∇ · um ) + 2μ∇uv : ∇um

(4.3)

Dm p p Dt (4.4)

where μ2 is the second coefficient of viscosity. The mass velocity satisfies the continuity equation; that is, Eq. (2.1) with u = um remains unchanged. The volume velocity satisfies the kinematic constraint that ∇ · uv = 0 irrespective of whether the flow is compressible or incompressible. The argument that the new constitutive relation is acceptable is advanced by comparing the additional term in the constitutive equation (4.2) or σ + = 2μ∇jv

(4.5)

with two Burnett order terms σ+ = −

μ2 (K1 ∇∇T + K2 T −1 ∇T ∇T ) ρT

(4.6)

(the Burnett equations are introduced formally in Chap. 5); these terms were argued to be same. Therefore, there is some justification for augmenting the Navier–Stokes equation in the manner suggested by Brenner. The above modification however leads to several difficulties. For example: 1. The question about the order of Kn for this additional term (Eq. (4.5)) arises. As shown explicitly in Sect. 5.6, the terms in Navier–Stokes and energy equations are of order Knudsen number, whereas Burnett order stress and heat flux terms are of order Knudsen squared. Therefore, an incompatibility in the order of Kn for the various terms in the equation arises. 2. Addition of term σ + (Eq. (4.6)) to the constitutive relation (Eq. (4.2)) implies that mechanical stress can be induced by a purely density/thermal gradient (even in the continuum framework—and not just at Burnett order). 3. Notice that Eq. (4.6) involves second derivative of temperature, which becomes of third-order term upon substituting it in the Cauchy’s equation of motion

4.4 Modified Navier–Stokes Equations

119

(Eq. (2.3)). This term is known to create problem of stability and the requirement of extra boundary condition in the Burnett equation. The issue of having to prescribe extra boundary condition and the problem of stability would thereby creep into the Navier–Stokes equation as well! 4. Further, notice that the diffusion velocity is in the direction of the density gradient, whereas customary it is in the opposite direction to the density and temperature gradients. 5. The last term in Eq. (4.4) (the viscous dissipation term) is no longer a perfect square. The implication of this will have to be carefully analyzed, in that, can it lead to viscous dissipation becoming negative. In essence, in this approach, selected terms from a higher order equation (Burnett equations) are added to a lower order equation (Navier–Stokes equations) with the aim of fixing certain deficiencies in the lower order equation. Such attempts have been made to fix the Burnett equation by adding selected terms from the super Burnett equations, as discussed in Sect. 7.3.1. In general, this approach does not lead to a better framework of equations because of the selective nature of terms being added to the base equation. Going all out to the higher order equation is rather a superior approach.

4.4 Modified Navier–Stokes Equations Another interesting attempt to extend the Navier–Stokes equation has been made in Sambasivam [123]. Observations on flow through microchannels and shock wave, which are not well predicted by the Navier–Stokes equations provided the motivation for this work. The hypothesis is that strong gradients of density and temperature affect the flow in these situations, and therefore modification to the Navier–Stokes equations is needed to cover these cases. A diffusive mass flux term (−∂ m ˙ D /∂xi ) was added to the right-hand side of the continuity equation. This term was obtained as

1 ∂p 1 ∂T − m ˙ = −μ p ∂xi 2T ∂xi D

(4.7)

suggesting that mass diffusion depends on both pressure and temperature gradients. Alternatively, the mass diffusion term can be written in terms of gradients of density and temperature, using the ideal gas law. The constitutive relations are also affected because of similar modification (as mass diffusion) to stress and heat flux terms. The modified equations are ∂m ˙D ∂ρ ∂(ρui ) + =− i ∂t ∂xi ∂xi

(4.8)

120

4 Need for Looking Beyond the Navier–Stokes Equations

∂σijT ∂(ρuj ) ∂(ρui uj ) ∂p + =− + + ρgj ∂t ∂xi ∂xj ∂xi

(4.9)

∂q T ∂uj ∂(ρ) ∂(ρui ) ∂ui + =− i −p − σijT ∂t ∂xi ∂xi ∂xi ∂xi

(4.10)

where the constitutive relations get modified to: σijT = μ

∂uj ∂ui + ∂xi ∂xj

2 ∂uk 2 ˙D ˙D − μδij − m˙D i uj − mj ui + δij mk uk 3 ∂xk 3 qiT = −k

∂T + m˙D i Cp T . ∂xi

(4.11)

(4.12)

Good agreement with the experimental data of Maurer for flow in a microchannel was demonstrated through these equations. However, the results from the extended Navier–Stokes equation for complex microchannel case are only in moderate agreement with the DSMC data. Note that the viscous dissipation term (last term in energy equation (4.10)) is no longer a perfect square—same problem as Brenner’s equation. The implication of this will have to be carefully analyzed, in that, can it lead to viscous dissipation becoming negative, and thereby violating thermodynamics. The continuity equation remains unaffected even in the case of higher order continuum transport equations—the Burnett and Grad equations and its variants, as discussed in Chaps. 5 and 6. Therefore, the added mass diffusion term is not of higher order, it is rather a correction to the transport equations, in general. Note that no need for correction to the Navier–Stokes equations has been felt otherwise in case of high pressure and temperature gradient situations. For example, this equation suggests onset of flow in the presence of a horizontal temperature gradient, whereas in such situations, buoyancy would set up a flow while no flow would occur in absence of gravity.

4.5 Non-Fourier Heat Conduction The Fourier law suggests that the heat flux is instantaneously setup (or vanishes) upon application (or removal) of a temperature gradient. Since no process is instantaneous, correction is clearly required to the Fourier’s law. Cattaneo and Vernotte suggested a simple modification, whereby a delay (= τ ) was introduced to Fourier’s relation, i.e., q(x, t + τ ) = −k∇T (x, t)

(4.13)

4.5 Non-Fourier Heat Conduction

121

Upon expanding the above equation in a Taylor series, and retaining only first-order terms, we obtain τ

∂q (x, t) + q(x, t) = −k∇T (x, t) ∂t

(4.14)

The above equation is known as Cattaneo’s equation, and is one of the simplest nonFourier models available in the literature. Insertion of the above heat flux relation yields the following hyperbolic heat conduction equation: 1 ∂ 2T α ∂g 1 ∂T 1 2 =∇ T + g+ 2 (4.15) + α ∂t k c2 ∂t 2 c ∂t where viscous dissipation and several other terms have been neglected for simplicity, g is volumetric heat generation, α is thermal diffusivity, and c is wave speed. Notice that the above equation suggests a wave-like nature of heat flow, instead of a diffusive nature embodied in Fourier’s law. As argued in Sect. 6.10, τ is expected to be of order 10−10 s for air under standard conditions. It is known that the Cattaneo’s equation becomes relevant in cases involving very short (femto second) laser pulses [152]. Another application where the relaxation time effect cannot be ignored [109] is exothermal catalytic reaction in crystals, where significant temperature rise can happen within a duration of the order of 10−13 s. The classical theory also breaks down at very low temperatures (i.e., near absolute zero) [109]. Again, in situations involving small length scales ([(1/T )(dT /dx)]−1 > 1, the collision integral term in Eq. (5.61) vanishes and we have free molecular regime where gas–surface interaction becomes dominant as compared to gas–gas collisions. In the other extreme condition, when Kn 0, an important physical constraint is placed on higher order approximations of distribution function. This constraint states that the primary variables: mass density, momentum density, and internal energy density (ρ, ρui and ρ = 3ρRT /2) must remain same at any level of approximation. For this constraint to hold, the zeroth approximation f = f (0) should reproduce these primary variables and contribution of further higher order approximations (f (1) , f (2) , etc.) to these primary variables should be exactly zero. These constraints are known as compatibility conditions and mathematically represented as m 3 m f (0) dc = ρ, m ci f (0) dc = ρui , C 2 f (0) dc = ρRT . 2 2 m (r) (r) For r > 0, m f dc = 0, m ci f dc = 0, C 2 f (r) dc = 0. 2 (5.67) In physical sense, these compatibility conditions make sure that the conservation laws are satisfied at each successive approximation. Hence, in the derivation of the higher order approximations to the distribution function, it is very critical to ensure that the higher approximations only correct the pressure tensor and heat flux vector without violating the basic conservation laws.

5.3.3 Derivation of Distribution Function In the present section, we derive the functional form of the distribution function at each successive approximation. Particularly, the Enskog’s method of successive approximation to solve the Boltzmann equation has been presented [31]. This section has been included for the sake of completeness but can be skipped in the first reading. What is of interest is to use the distribution function to derive the transport equations, which is elaborated upon in the next section. In the most general form, the Boltzmann equation (Eq. (5.1)) is expressed as ξ(f ) = J (f, f1 ) + Df = 0, where operator,

D=

∂ ∂ ∂ + ck + Fk ∂t ∂xk ∂ck

(5.68) (5.69)

and ξ(f ) denotes the result of certain operations performed on the single particle distribution function f . Note that the quadratic terms inside the collision integral (Eq. (5.57)) are reversed, which turns the negative sign into positive. Now substitute the series (5.62) in ξ(f ) and assume that when ξ operates on this series, the final result can be expressed as a series in which the rth term involves only the first r terms of the series in (5.62). That is, we write ξ(f ) = ξ(f (0) + f (1) + f (2) + · · · ) = ξ (0) (f (0) ) + ξ (1) (f (0) , f (1) ) + ξ (2) (f (0) , f (1) , f (2) ) + · · ·

(5.70)

5.3 The Chapman–Enskog Method

145

In addition to condition that the sum (5.62) is a solution of ξ(f ) = 0, functions f (r) are assumed to satisfy the following equations, ξ (0) (f (0) ) = 0,

(5.71)

) = 0,

(5.72)

ξ (2) (f (0) , f (1) , f (2) ) = 0,

(5.73)

ξ

(1)

(f

(0)

,f

(1)

························ ξ (n) (f (0) , f (1) , . . . f (n) ) = 0.

(5.74)

These conditions together ensure that ξ(f ) = 0. From Eq. (5.71), f (0) is determined and subsequently, f (1) , f (2) , . . . f (n) are determined from Eqs. (5.72), (5.73), . . . (5.74), respectively; since each equation contains only one unknown when all the previous equations have been solved. Substituting the series (5.62) in Eq. (5.68), we have 2 ∞ 3 ∞ ∞ ξ(f ) = J

(s)

f (r) 2

r=0 ∞

+D

f1 s=0

r=0

3

k

f (l) f1(k−l)

=J

f (r) ∞

+D

f (r)

k=0 l=0 ∞

r=0 ∞

k (k−l)

=

J (f (l) f1

)+

k=0 l=0 ∞

∞

=

r=0

r (r−l) J (f (l) f1 ) + Df (r)

= r=0

Df (r)

l=0

J (r) + Df (r)

r

J (r) =

where

r=0

(r−l) J f (l) f1

(5.75)

l=0

Therefore, for the r-th term of ξ , we can write ξ r = J (r) + Df (r)

(5.76)

Note that in the first step of the above equation, we applied Cauchy’s rule for the product of two series (given by Eq. (5.77) below), while in the third step, we have interchanged the order of integral (involved in the collision term see Eq. 5.57) with the summation. In the last step, the dummy variable in the first term has been changed from k to r. The Cauchy’s rule for the product of two series is ∞

∞ i=0

∞

k

bj =

ai j =0

al bk−l k=0 l=0

(5.77)

146

5 Burnett Equations: Derivation and Analysis

We now define the term J (r) in Eq. (5.75) as r (r−l)

J (r) =

J (f (l) f1

(r)

(r−1)

) = J (f (0) f1 ) + J (f (1) f1

(0)

) + · · · + J (f (r) f0 )

l=0

(5.78) For the term ΣDf (r) in Eq. (5.75), Enskog suggested to divide it into a series of parts D (r) instead of Df (r) . The expression for operator D (Eq. (5.69)) contains a time derivative term. Similar to f , this term is expanded in an infinite series as ∂ ∂0 ∂1 ∂2 = + + + ··· = ∂t ∂t ∂t ∂t

∞ l=0

∂l ∂t

(5.79)

Substituting the series for f (Eq. (5.62)) and time derivatives (Eq. (5.79)) in the expression for operator D (Eq. (5.69)), and again using Cauchy’s rule for product of two infinite series, we have ∂f ∂f ∂f + ck + Fk ∂t ∂xk ∂ck ∞ ∞ ∂l ∂ ∂ + ck = + Fk f (m) ∂t ∂xk ∂ck l=0 m=0 ∞ r ∞ ∂s f (r−s) ∂f (r) ∂f (r) + = + Fk ck ∂t ∂xk ∂ck r=0 s=0 r=0 r ∞ ∂s f (r−s) ∂f (r) ∂f (r) + ck = + Fk ∂t ∂xk ∂ck

Df =

r=0

s=0

∞

=

r

D (r) r=0

where D (r) = s=0

∂s f (r−s) ∂f (r) ∂f (r) + ck + Fk ∂t ∂xk ∂ck

(5.80)

According to Enskog, D (0) = 0, and for r > 0, D (r) is dependent only on f (0) , f (1) , . . . , f (r−1) . As such, (5.81) D (r) ≡ D (r) f (0) , f (1) , . . . , f (r−1) r−1

D (r) = s=0

=

∂s f ((r−1)−s) ∂f (r−1) ∂f (r−1) + ck + Fk ∂t ∂xk ∂ck

∂1 f (r−2) ∂(r−1) f (0) ∂f (r−1) ∂f (r−1) ∂0 f (r−1) + + ··· + + ck + Fk ∂t ∂t ∂t ∂xk ∂ck (5.82)

5.3 The Chapman–Enskog Method

147

Each term in the above expression can be determined since f (0) , . . . , f (r−1) are known. Hence, at every approximation for the distribution function, the equation that needs to be solved is ξ (r) = J (r) + D (r) = 0,

(5.83)

where J (r) is given by Eq. (5.78) and D (r) is given by Eq. (5.82). Substituting expressions for J (r) (Eq. (5.78)) and D (r) (Eq. (5.82)) in expression for ξ (r) (Eq. (5.83)), we obtain (r)

(0)

(r−1)

J (f (0) f1 ) + J (f (r) f1 ) = −D (r) − J (f (1) f1

(1)

) − . . . − J (f (r−1) f1 ) (5.84)

Note that we have r terms on the right-hand side and consists of f (0) , f (1) , . . . , f (r−1) which we already know are the solutions of the previous equations ξ (0) = 0, . . ., ξ (r−1) = 0. The function to be determined f (r) appears linearly only on the left-hand side of the equation. So, the task at hand is to solve Eq. (5.84) which still turns out to be a complicated integral equation. The detailed procedure to solve this integral equation is given in Chapman and Cowling [31]. All the important steps in the Chapman–Enskog method that we discussed previously are outlined in a detailed flowchart as shown in Fig. 5.4 for easy reference.

5.3.3.1

Zeroth Approximation: f (0)

For the zeroth approximation f (0) , we substitute r = 0 in Eq. (5.83) and the equation to be solved becomes ξ (0) = J (0) + D (0) = 0,

(5.85)

Since D (r) is a function of f (0) , f (1) , . . . , f (r−1) , we essentially have D (0) = 0 for r = 0, i.e., ∂f (0) ∂f (0) ∂f (0) + ck + Fk =0 ∂t ∂xk ∂ck

(5.86)

This state corresponds to a uniform, steady state with no external forces. To find (0) J (0) , we substitute r = 0 in Eq. (5.78) and we get J (0) = 2J (f (0) f1 ). Hence, the equation to be solved for f (0) reduces to (0)

J (f (0) f1 ) = 0

(5.87)

=0

(0)

= qi

=0

Enskog’s hypothesis:

(1)

f (1)

qi

(NS)

= qi

(Eu)

+ qi

(1)

∂ ui (1) ∂T , q = −k ∂ x j i ∂ xi

(r)

Pi j = m CiC j f (r) dc, m (r) CiC2 f (r) dc qi = 2

Navier-Stokes Equations: (NS) (Eu) (1) Pi j = Pi j + Pi j ,

Pi j = −2m

1st correction

r=1

x (1) f (0) , f (1) = 0

qi

(Bu)

= qi

(NS)

+ qi

(2)

= Eq 5.145

Burnett Equations: (Bu) (NS) (2) Pi j = Pi j + Pi j ,

qi

(2)

(2)

r=2

f (2)

2nd correction

Pi j = Eq 5.144,

x (2) f (0) , f (1) , f (2) = 0

Fig. 5.4 Flowchart showing the detailed procedure of Chapman–Enskog method

qi

(Eu)

Euler Equations: (Eu) (0) Pi j = Pi j = d i j p,

qi

(0)

Pi j = pd i j ,

(0)

r=0

f (0)

0th correction

x (0) f (0) = 0

x ( f ) = x f (0) + f (1) + f (2) + · · · = 0

x (0) f (0) + x (1) f (0) , f (1) + x (2) f (0) , f (1) , f (2) + · · · = 0

Boltzmann equation: f =0

x ( f ) = J( f f1 ) +

r=3

f (3)

3rd correction

(SB)

qi

(Bu)

= qi

(3)

+ qi

Super-Burnett Equations: (SB) (Bu) (3) Pi j = Pi j + Pi j ,

Linearized form available: (3) (3) Pi j = Eq 5.148, qi = Eq 5.149

x (3) f (0) , f (1) , f (2) , f (3) = 0

f = r=0 f (r) = f (0) + f (1) + f (2) + · · · f (0) : Maxwell-Boltzmann distribution

Chapman-Enskog method to determine Pi j and qi

148 5 Burnett Equations: Derivation and Analysis

5.3 The Chapman–Enskog Method

149

This condition implies that the collision integral is zero and collisions as a whole produce no net effect as the effect of every encounter is exactly balanced by that of inverse process. Expanding the collision integral (Eq. (5.57)) J (f (0) , f1(0) ) =

(0) (0) f1 f − f1(0) f (0) grdrdαdc1 .

(5.88)

For collision integral to be zero, the only possible solution is

(0)

(0)

= f (0) f1

(0)

= log f (0) + log f1

f (0) f1 log f (0) + log f1

(0)

(5.89)

Therefore, log f is a summational invariant for collisions, i.e., the quantity log f is conserved during collisions. During elastic collisions, mass, momentum, and energy are always conserved and they are the only three summational invariants. Therefore, the quantity log f4 should be a5 linear combination of the three summational invariants, ψ (i) = 1, mc, 12 mc2 . After doing some algebra (see Appendix B.1), i the final form of the distribution function for the zeroth approximation can be shown to be Maxwellian distribution as f (0) = fM = n

5.3.3.2

m 2π kB T

3 2

mC 2 . exp − 2kB T

(5.90)

First Approximation: f (1)

For the first approximation f (1) , we substitute r = 1 in Eq. (5.83) and the equation to be solved becomes ξ (1) = J (1) + D (1) = 0,

(5.91)

From Eq. (5.82), the differential part D (1) depends only on the zeroth approximation f (0) . An expression for differential part D (1) can be obtained as (see Appendix B.2)

D

(1)

=f

(0)

∂ui mC 2 5 ∂ log T m − Ci Cj Ck + 2kB T 2 ∂xk kB T ∂xj

(5.92)

where the angular brackets denote the symmetric and trace-free part of the tensor. The calculation of collision term J (1) is rather cumbersome, so we replace the non-linear collision integral by the Bhatnagar–Gross–Krook (BGK) kinetic model [20]. The mathematics involved in the evaluation of term J (1) with the original collision integral is complex and the essence of the Chapman–Enskog method

150

5 Burnett Equations: Derivation and Analysis

maybe lost in the mathematical rigor. To convey the usefulness of the Chapman– Enskog method in simpler terms, the BGK kinetic model is employed and the complexity associated with the collision integral is avoided. As we will see, the derivation of the distribution function is greatly simplified, but at the same time, the BGK model retains many of the qualitative features of the original collision integral. Mathematically, the BGK kinetic model is expressed as J (f, f1 ) = ν(fM − f )

(5.93)

where fM is the local Maxwellian and ν is the collision frequency that is assumed to be independent of the molecular velocity c. The structure of this particular kinetic model implies that any non-equilibrium distribution function relaxes to the local Maxwellian with a relaxation time equal to the inverse of the collision frequency. The BGK kinetic model has the following minimum requirements of the Boltzmann equation: (1) The model satisfies the conservation of mass, momentum, and energy. (2) It fulfills the H theorem (or second law of thermodynamics). (3) The distribution function reduces to Maxwellian in equilibrium. However, this model fails to predict the correct value of Prandtl number which we will show in the subsequent sections. Substituting the series of f up to first approximation (f = f (0) + f (1) ) and with the knowledge of zeroth approximation being Maxwellian (fM = f (0) ), the collision integral simplifies as J (f, f1 ) = ν(fM − f ) = ν(fM − f (0) − f (1) ) = −νf (1) .

(5.94)

Substituting Eqs. (5.92) and (5.94) in Eq. (5.91), we get an expression for the first approximation of the distribution function as ∂ui mC 2 5 ∂ log T m − Ci Cj Ck + 2kB T 2 ∂xk kB T ∂xj

mC 2 5 ∂ log T ∂ui m 1 2 ∂ui f (0) − Ck Ci Cj + − C δij =− ν 2kB T 2 ∂xk kB T ∂xj 3 ∂xj (5.95)

f (1) = −

f (0) ν

where in the second step, the angular bracket notation is expanded as Ci Cj

∂ui ∂ui ∂ui 1 = Ci C j − C 2 δij ∂xj ∂xj 3 ∂xj

Before proceeding to achieve closure, it is necessary to check whether the first approximation f (1) satisfies the compatibility conditions (Eq. (5.67)), i.e.,

m

f (1) dC = 0,

m

ci f (1) dC = 0,

m 2

C 2 f (1) dC = 0

(5.96)

5.4 Derivation of Hydrodynamic Equations

151

On evaluating these integrals, it can be shown that the first approximation f (1) indeed satisfies the compatibility conditions. With this functional form of the distribution function, we proceed to evaluate the first approximations to pressure (1) (1) tensor Pij and heat flux vector qi to achieve closure.

5.4 Derivation of Hydrodynamic Equations The functional form of the distribution function that we derived previously will now be utilized to evaluate the pressure tensor and heat flux vector at each approximation level. After substituting these constitutive relationships in the basic conservation laws, it will be shown that we get Euler equations at the zeroth level, Navier–Stokes– Fourier equations at the first level, Burnett equations at the second level, and superBurnett equations at the third level.

5.4.1 Euler Equations Utilizing the Maxwellian distribution function at the zeroth level (Eq. (5.90)), we now present a step by step procedure to evaluate the zeroth approximation (0) (0) of the pressure tensor Pij and heat flux vector qi . Readers interested in the mathematical details can follow this section, else this section can be skipped without loss of generality. Putting Maxwellian distribution (f (0) = fM ) in integral equation for pressure tensor (Eq. (5.65)) and proceeding as Pij(0) = m

Ci Cj fM dC

m = mn 2π kB T

3 2

mC 2 dC Ci Cj exp − 2kB T

(5.97)

Evaluation of the above triple integral is somewhat tricky, so we transform from Cartesian coordinate system {Cx , Cy , Cz } to spherical coordinate system {C, θ, φ}, where θ is the azimuthal angle and φ is the zenith angle [141, Appendix A.3] as shown in Fig. 5.5. Also, according to Jacobian transformation, we get dCx dCy dCz = C 2 sin φdCdθ dφ. Introducing the direction vector of the microscopic velocities as υi = Ci /C = {cos θ sin φ, sin θ sin φ, cos φ}i , we have Pij(0) = mn

m 2π kB T

3 2

∞

π

C=0 φ=0

mC 2 C 2 sin φdCdθ dφ C 2 υi υj exp − 2kB T θ=0 (5.98)

2π

152

5 Burnett Equations: Derivation and Analysis

Fig. 5.5 Transformation from Cartesian {Cx , Cy , Cz } to spherical {C, θ, φ} coordinate system; Cx = C cos θ sin φ, Cy = C sin θ sin φ, Cz =C cos φ, where

C=

Cx2 + Cy2 + Cz2

Iij =

Let,

1 4π

π

2π

υi υj sin φdθ dφ

(5.99)

φ=0 θ=0

Therefore, integral equation for the pressure tensor in the compact notation becomes

(0) Pij

m = mn 2π kB T

3

2

∞

4π 0

mC 2 dC Iij C 4 exp − 2kB T

(5.100)

One can evaluate the integral Iij for its six distinct components. This simple symbolic integration can easily be performed with the aid of an appropriate software (such as Mathematica or Maple). This simple exercise gives all the off-diagonal terms of Iij as zero and diagonal elements as 1/3. Alternatively, one can realize that Iij is symmetric and isotropic, and since the Kronecker delta δij is the only such second-order tensor, we can express Iij in terms of δij as Iij = βδij

(5.101)

To find β, we multiply both sides of Eq. (5.101) by δij as Iij δij = βδij δij ∴

Iii = βδii

⇒

Iii = 3β

[∵ δii = 3]

Further, we evaluate the integral Iii as 1 Iii = 4π where ∴

Iii = 1

π

2π

υi υi sin φdθ dφ φ=0 θ=0

υi υi = (cos θ sin φ)2 + (sin θ sin φ)2 + (cos φ)2 ⇒

β=

1 3

(5.102)

5.4 Derivation of Hydrodynamic Equations

∴

153

1 δij . 3

Iij =

(5.103)

With this result, Eq. (5.100) simplifies as (0) Pij

3 ∞ 2 m 1 mC 2 4 dC = δij 4π mn C exp − 3 2π kB T 2kB T 0

(5.104)

The integral to be evaluated is a standard Gaussian integral defined as

∞

Gn =

C n exp[−aC 2 ]dC =

Γ

0

n+1 2

2a

n+1 2

(5.105)

where Γ is a standard gamma function often referred to as the generalized factorial function defined as ∞ Γ (x) = t x−1 e−t dt, (x > 0) 0

√ In the present case, we have a = m/2kB T , n = 4, and Γ (5/2) = 3 π /4. Substituting these expressions in Eq. (5.104), we get Pij(0) = nkB T δij = pδij .

(5.106)

For evaluation of heat flux vector at the zeroth approximation level, we substitute for r = 0, f (0) = fM in Eq. (5.66) and proceed as (0)

qi

=

m 2

Ci C 2 fM dC

(5.107)

Following the same procedure and expressing the integral in the compact notation as 3 ∞ 2 m mC 2 dC Ii 4π C 3 exp − 2π kB T 2kB T 0 π 2π 1 Ii = υi sin φdθ dφ (5.108) 4π φ=0 θ=0

(0)

qi where

= mn

Performing this symbolic integration for each of the three components of vector υi , we get Ii = 0,

⇒

(0)

qi

= 0.

(5.109)

154

5 Burnett Equations: Derivation and Analysis

Thus, at the zeroth level, we obtain constitutive relationships for the pressure tensor and heat flux vector as Pij(Eu) = Pij(0) = δij p, (Eu)

qi

(0)

= qi

=0

(5.110)

where the subscript “Eu” stands for Euler, corresponding to zeroth level of approximation. Substituting these expressions in the basic conservation equations (5.53)– (5.55) we obtain the well-known inviscid Euler equations: ∂ ∂ρ + (ρui ) = 0, ∂t ∂xi ∂ui ∂ui 1 ∂p + uj + = Fi , ∂t ∂xj ρ ∂xi ∂ui ∂T ∂T 3 +p ρR + uj = 0. 2 ∂t ∂xj ∂xi

(5.111)

5.4.2 Navier–Stokes Equations An expression for the first approximation of the distribution function is obtained as (Eq. (5.95)) f

(1)

f (0) =− ν

mC 2 5 ∂ log T ∂ui m 1 2 ∂ui − Ck Ci Cj + − C δij 2kB T 2 ∂xk kB T ∂xj 3 ∂xj (1)

(1)

dC,

(1) qi

We now proceed to evaluate Pij and qi (1) Pij

=m

Ci Cj f

(1)

by utilizing f (1) as m = 2

Ci C 2 f (1) dC

(5.112)

(1)

For first approximation of pressure tensor Pij , Pij(1)

=m

Ci Cj f (1) dC

=m

Ci C j

mC 2 5 ∂ log T − Ck 2kB T 2 ∂xk ∂uk ∂uk m 1 Ck Cl dC + − C 2 δkl kB T ∂xl 3 ∂xl −f (0) ν

(5.113)

5.4 Derivation of Hydrodynamic Equations

155

Note that we changed the dummy indices, i and j in the second term of the Eq. (5.95) since we already have indices i and j as the free indices. (1) Pij

mC 2 5 m ∂ log T (0) − dC =− Ci Cj Ck f ν ∂xk 2k T 2 ) *+ B , +

m ∂uk kB T ∂xl

1 − δkl 3

I1

Ci Cj Ck Cl f (0) dC ) *+ , I2

Ci Cj C 2 f (0) dC ) *+ ,

(5.114)

I3

We encounter some common integrals in kinetic theory of gases of the type as

m Ci1 Ci2 . . . Cin fM dC = n 2π kB T

3 2

mC 2 dC Ci1 Ci2 . . . Cin exp − 2kB T (5.115)

When we have odd number of peculiar velocities C, i.e., for n = 1, 3, 5, . . . , the integral always vanishes. With this general rule, integral I1 vanishes and integral (1) expression for Pij reduces to (1) Pij

m m ∂uk 1 (0) 2 (0) =− Ci Cj Ck Cl f dC − δkl Ci Cj C f dC ν kB T ∂xl 3 ) ) *+ , *+ , I2

I3

(5.116) As before, integrals I2 and I3 are evaluated by transforming from Cartesian to spherical coordinate system. Introducing the direction vector of the microscopic velocities as υi = Ci /C = {cos θ sin φ, sin θ sin φ, cos φ}i , we evaluate integrals I2 and I3 separately as I2 =

Ci Cj Ck Cl f (0) dC

3 2 m mC 2 dC =n Ci Cj Ck Cl exp − 2π kB T 2kB T 3 ∞ π 2π 2 m mC 2 =n C 2 sin φdCdθ dφ C 4 υi υj υk υl exp − 2π kB T 2k T B C=0 φ=0 θ=0

156

5 Burnett Equations: Derivation and Analysis

m =n 2π kB T

3

mC 2 dC Iij kl 4π C exp − 2kB T C=0 π 2π 1 where Iij kl = υi υj υk υl sin φdθ dφ 4π φ=0 θ=0

2

∞

6

(5.117) The integral Iij kl is a fourth order symmetric and isotropic tensor and any fourth order isotropic tensor can be represented in terms of Kronecker delta tensor as Iij kl = α δij δkl + δik δj l + δil δj k

(5.118)

To determine the value of α, we perform the contraction twice by multiplying δij δkl , Iij kl δij δkl = α [δii δkk + δik δik + δik δik ] ∴

⇒

Iiikk = α[3 × 3 + δii + δii ]

Iiikk = 15α

(5.119)

and the integral Iiikk is evaluated as Iiikk

1 = 4π

and

υi υi υk υk sin φdθ dφ

Iiikk = 1 Iij kl =

2π

φ=0 θ=0

where ∴

π

2 υi υi υk υk = (cos θ sin φ)2 + (sin θ sin φ)2 + (cos φ)2 ⇒

α=

1 15

1 δij δkl + δik δj l + δil δj k 15

(5.120)

Substituting the final result of Iij kl in integral equation for I2 ,

m I2 = n 2π kB T

3 2

1 mC 2 dC δij δkl + δik δj l + δil δj k 4π C exp − 2k T 15 B C=0 (5.121)

∞

6

Evaluating the underlined Gaussian integral for a = m/2kB T , n = 6, and Γ (7/2) = 15 √ π , 8

√ − 7 2 m 15 π mC 2 dC = C exp − 2kB T 16 2kB T C=0 ∞

6

Summarizing all the results, we get expression for the integral I2 as

(5.122)

5.4 Derivation of Hydrodynamic Equations

I2 = n

kB T m

2

157

δij δkl + δik δj l + δil δj k

(5.123)

The evaluation of integral I3 is relatively straightforward, and we proceed as I3 =

Ci Cj C 2 f (0) dC

mC 2 dC Ci Cj C 2 exp − 2kB T 3 ∞ π 2π 2 m mC 2 2 2 =n C 2 sin φdCdθ dφ C υi υj C exp − 2π kB T 2kB T C=0 φ=0 θ=0

=n

m 2π kB T

m =n 2π kB T

3 2

3

mC 2 dC Iij 4π C exp − 2kB T C=0 π 2π 1 1 where Iij = υi υj sin φdθ dφ = δij 4π φ=0 θ=0 3 2

∞

6

(5.124) The Gaussian integral and the integral Iij are already evaluated previously in Sect. 5.4.1, hence, the expression for integral equation I3 is obtained as

kB T I3 = 5n m

2 δij .

(5.125)

Substituting the results for integral equation I2 (Eq. (5.123)) and I3 (Eq. (5.125)) (1) in expression for Pij (Eq. (5.116)), we obtain (1) Pij

kB T 2 m m ∂uk n δij δkl + δik δj l + δil δj k =− ν kB T ∂xl m 2 5 kB T − δkl n δij 3 m

5 m m kB T 2 ∂uk δij δkl + δik δj l + δil δj k − δkl δij =− n ν kB T m ∂xl 3 m nkB T ∂uk ∂uk ∂uk 5 ∂uk =− δij δkl + δik δj l + δil δj k − δkl δij ν m ∂xl ∂xl ∂xl 3 ∂xl ∂uj nkB T ∂uk ∂ui 5 ∂uk =− δij + + − δij ν ∂xk ∂xj ∂xi 3 ∂xk

158

5 Burnett Equations: Derivation and Analysis

∂uj nkB T ∂ui 2 ∂uk =− + − δij ν ∂xj ∂xi 3 ∂xk

∂uj 1 ∂uk nkB T 1 ∂ui − + δij = −2 ν 2 ∂xj ∂xi 3 ∂xk

(5.126)

In terms of symmetric tensorial notation, and identifying the coefficient term nkνB T as absolute viscosity (recall that, here, ν is the collision frequency and not kinematic (1) viscosity), we obtain expression for Pij as Pij(1) = −2μ

where μ =

∂ui ∂xj

(5.127)

∂uj ∂ui nkB T 1 ∂uk 1 ∂ui − and = + δij ν ∂xj 2 ∂xj ∂xi 3 ∂xk

(5.128)

From the definition of decomposition of the pressure tensor (Eq. (5.22)), we can say that the first approximation to the pressure tensor is the divergence-less viscous stress tensor, i.e., pij = Pij(1) . This relation between the viscous stress tensor and velocity gradient is well-known as Navier–Stokes law. Therefore, at the first approximation, the expression for pressure tensor is Pij(N S) = Pij(Eu) + Pij(1) = pδij − 2μ

∂ui . ∂xj

(5.129)

Note that the second coefficient of viscosity and the coefficient of bulk viscosity (introduced in Sect. 2.2.1) has not appeared in Eq. (5.129). This is because the above derivation is for monatomic gases. However, polyatomic gases involve extra internal degrees of freedom and the coefficient of bulk viscosity does alter the above constitutive relation, as discussed in Kogan [83, p. 216]. To evaluate expression for q (1) , we substitute the first approximation distribution function f (1) (Eq. (5.95)) in first approximation of heat flux vector (Eq. (5.66)) and proceed as (1) qi

m = 2 m = 2

Ci C 2 f (1) dC

mC 2 5 ∂ log T − Ck Ci C 2kB T 2 ∂xk ∂uk ∂uk m 1 dC Ck Cl + − C 2 δkl kB T ∂xl 3 ∂xl 2

−f (0) ν

5.4 Derivation of Hydrodynamic Equations

=−

159

⎧ ⎪ ⎪ ⎪ ⎨

⎫ ⎪ ⎪ ⎪ ⎬

m m ∂ log T 5 Ci Ck C 2 f (0) C 2 dC − Ci Ck C 2 f (0) dC ⎪ 2ν ∂xk ⎪ 2k T 2 B ⎪ ⎪ ⎪ ) ) *+ , *+ ,⎪ ⎩ ⎭ I1

I2

⎧ ⎫ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎬ m m ∂uk ⎨ 1 2 (0) 2 2 (0) − Ci Ck Cl C f dC − δkl Ci C C f dC ⎪ 2ν kB T ∂xl ⎪ 3 ⎪ ⎪ ⎪ ) *+ , *+ ,⎪ ⎩) ⎭ I3

I4

For integrals I3 and I4 , we have odd number tensors of peculiar velocities and hence, they are exactly zero. Therefore, the equation for q (1) reduces to

∴

(1)

qi

=−

⎧ ⎪ ⎪ ⎪ ⎨

⎫ ⎪ ⎪ ⎪ ⎬

m m ∂ log T 5 Ci Ck C 2 f (0) C 2 dC − Ci Ck C 2 f (0) dC ⎪ 2ν ∂xk ⎪ 2kB T 2 ⎪ ⎪ ⎪ ) ) *+ , *+ ,⎪ ⎩ ⎭ I1

I2

(5.130) Following the same procedure as outlined previously, the integrals I1 and I2 are evaluated as I1 = Ci Ck C 2 f (0) C 2 dC

3 2 m mC 2 4 =n Ci Ck C exp − dC 2π kB T 2kB T 3 ∞ π 2π 2 m mC 2 2 4 C 2 sin φdCdθ dφ =n C υi υk C exp − 2π kB T 2k T B C=0 φ=0 θ=0 3 ∞ 2 m mC 2 dC Iik =n 4π C 8 exp − (5.131) 2π kB T 2kB T C=0 We already know Iik = δik√ /3 and evaluating the Gaussian integral for a = m/2kB T , n = 8, and Γ (9/2) = 105 π, 16

∞

mC 2 C exp − 2kB T C=0 8

√ − 9 2 m 105 π dC = 32 2kB T

Substituting this result, we obtain the expression for I1 as

kB T I1 = 35n m

3 δik

(5.132)

160

5 Burnett Equations: Derivation and Analysis

The integral I2 has already been obtained previously (Eq. (5.125)) as

I2 =

2 (0)

Ci Ck C f

kB T dC = 5n m

2 δik

(5.133)

Finally, substituting the integrals I1 (Eq. (5.132)) and I2 (Eq. (5.133)) in equation for heat flux vector (Eq. (5.130)), we have qi(1)

=− ∴

(1)

qi

2

m ∂ log T =− 2ν ∂xk

3 m kB T 3 kB T 2 5 35n δik − 5n δik 2kB T m 2 m

5 kB nkB T ∂T 2 m ν ∂xi

= −k

∂T , ∂xi

where k =

5 kB nkB T . 2m ν

(5.134)

Therefore, at the first approximation, we obtain expression for heat flux vector as (N S)

qi

(Eu)

= qi

(1)

+ qi

=0−k

∂T ∂xi

(5.135)

This linear relationship between the heat flux vector and temperature gradient is famously known as Fourier’s law. From the expressions for transport coefficients, μ and k, it is clear that they depend on atomic structure and only on temperature, but not on pressure, a finding well supported by the experiments. The Prandtl number is then calculated as Pr =

μCp = 1, k

[where Cp = 5R/2 = 5kB /2m]

(5.136)

Therefore, using the BGK kinetic model, we get unity Prandtl number while the measured value of Prandtl number for a monatomic gas is equal to 2/3. This inability of the BGK model to predict the correct Prandtl number results in discrepancies in the flow description of the gas. Nonetheless, BGK kinetic model greatly simplifies the method and help us to appreciate the significance of the Chapman–Enskog method. Summarizing the results, the constitutive relationships for the pressure tensor and heat flux vector at the first approximation are obtained as (N S)

= pδij − 2μ

(N S)

= −k

Pij qi

∂T ∂xi

∂ui , ∂xj

(5.137) (5.138)

5.4 Derivation of Hydrodynamic Equations

161

The closure is achieved by substituting these constitutive relationships in the basic conservation laws (Eqs. (5.53)–(5.55)), and the resulting equations are the famous Navier–Stokes equations for constant thermo-physical properties. ∂ρ ∂ (ρui ) = 0, + ∂t ∂xi

∂ui ∂ui 1 ∂p μ + uk =− + Fi + ∂t ∂xk ρ ∂xi ρ

∂ 2 ui 1 ∂ + ∂xj ∂xj 3 ∂xi

(5.139)

∂uk ∂xk

,

(5.140)

2 3 3 ∂uk 2 ∂ui ∂uj ∂uk ∂T ∂T 2 ∂uk 2 = −p ρR + uk +μ + − 2 ∂t ∂xk ∂xk ∂xk ∂xj ∂xi 3 ∂xk +k

∂ 2T ∂xk ∂xk

(5.141)

5.4.3 Burnett Equations The complete second order approximation to the distribution function was determined by Burnett [26] in 1935. However, extreme complexity is involved in deriving f (2) and therefore we do not wish to include those details in this text. As it turns out, there is no need to solve the equation for f (2) . Based on an order of magnitude of the terms, Chapman and Cowling [31, Chapter 15] obtained the additional terms introduced by f (2) into the equations for pressure tensor and heat flux vector. Here, we present only the final expressions at the third approximation which read (Bu)

(2)

(5.142)

qi(Bu) = qi(N S) + qi(2) ,

(5.143)

Pij

(N S)

= Pij

+ Pij ,

where (2)

Pij = ω1

∂uk μ2 ∂uk μ2 D0 Sij μ2 ∂ 2 T −2 Sij + ω2 Sj k + ω3 p ∂xk p Dt ∂xi ρT ∂xi ∂xj

μ2 ∂T ∂p μ2 ∂T ∂T μ2 + ω5 + ω6 Ski Sj k 2 pρT ∂xi ∂xj p ρT ∂xi ∂xj

∂uk ∂T μ2 ∂uk ∂T μ2 D0 ∂T − = θ1 + θ2 ρT ∂xk ∂xi ρT Dt ∂xi ∂xi ∂xk + ω4

(2)

qi

(5.144)

162

5 Burnett Equations: Derivation and Analysis

+ θ3

μ2 ∂p μ2 ∂Sik μ2 ∂T Sik Sik + θ4 + 3θ5 ρp ∂xk ρ ∂xk ρT ∂xk

(5.145)

where Sij is symmetric and trace-free part of the velocity gradient tensor defined as Sij =

∂ui 1 ∂ui 1 ∂uj 1 ∂uk = + − δij ∂xj 2 ∂xj 2 ∂xi 3 ∂xk

The Burnett equations are obtained by substituting Eqs. (5.144)–(5.145) into Eqs. (5.142)–(5.143), respectively, and the ensuing result in the conservation laws (Eqs. (5.53)–(5.55)). This complicated set of equations was first derived by Burnett using the Chapman–Enskog method. These equations contain the material ∂T derivatives terms of Sij and ∂x and this particular form of equations is known as i original Burnett equations. Chapman and Cowling revisited the derivation of Burnett equations using order of magnitude analysis and replaced the material derivatives terms with the spatial gradients using the inviscid Euler equations. This substitution maintained the second order accuracy of the equations and hence, was deemed reasonable. Owing to its relatively simple nature of the equations (no material derivatives terms in the equations), this set of equations as proposed by Chapman and Cowling became more prevalent in the literature and is known as conventional Burnett equations. With this substitution, the conventional Burnett equations are given as Pij(2)

∂uk ∂uj μ2 ∂uk μ2 ∂ 1 ∂p ∂uk Fj − − = ω1 Sij + ω2 −2 Sj k p ∂xk p ∂xi ρ ∂xj ∂xi ∂xk ∂xi μ2 ∂ 2 T μ2 ∂T ∂p μ2 ∂T ∂T μ2 Ski Sj k + ω4 + ω5 + ω 6 ρT ∂xi ∂xj pρT ∂xi ∂xj p ρT 2 ∂xi ∂xj (5.146)

∂uk ∂uk ∂T 2 ∂ μ2 ∂uk ∂T μ2 T −2 − = θ1 + θ2 ρT ∂xk ∂xi ρT 3 ∂xi ∂xk ∂xi ∂xk

+ ω3

qi(2)

+ θ3

μ2 ∂p μ2 ∂Sik μ2 ∂T Sik Sik + θ4 + 3θ5 ρp ∂xk ρ ∂xk ρT ∂xk

(5.147)

Note that the coefficients ω’s and θ ’s involved in original as well as conventional Burnett equations are pure numbers and their values are given in Table 5.1 for Maxwell molecules (exact values) and for rigid hard spheres. The complicated structure of the equations is the reason why these equations have not been studied extensively in the past. The numerous terms which are there in Eqs. (5.144) and (5.145) are not so apparent because of the tensor notation adopted here. The terms in expanded form are given in Sect. 5.5 albeit in cylindrical coordinates to further emphasize the complexity of these terms. The stress and heat flux terms at the Burnett order

5.4 Derivation of Hydrodynamic Equations

163

Table 5.1 Coefficients for Burnett stress and heat flux terms [31] Coefficients ω1 ω2 ω3 ω4 ω5 ω6 θ1 θ2 θ3 θ4 θ5

Maxwell molecules 4/3 [7/2 − (T /μ)(dμ/dT )] 2 3 0 3(T /μ)(dμ/dT ) 8 (15/4) [7/2 − (T /μ)(dμ/dT )] 45/8 −3 3 35/4 + (T /μ)(dμ/dT )

Hard-sphere molecules 1.014 × 4/3 [7/2 − (T /μ)(dμ/dT )] 2.028 2.418 0.681 0.806(3T /μ)(dμ/dT ) − 0.990 7.424 11.644 5.822 −3.090 2.418 1.627 × 3 [35/4 + (T /μ)(dμ/dT )]

depend on various combinations of the product of gradient of velocity, temperature, and pressure—making these terms non-linear. (In contrast, the stress and heat flux terms at the Navier–Stokes order are proportional to velocity gradient, while the stress and heat flux terms at super Burnett order involve triple gradients of velocity, temperature, and pressure.) Note that the number of terms will increase further upon differentiation of the terms, as gradient of stress and heat flux terms enter the momentum and energy equation, respectively. In the Navier–Stokes equations, non-linearity is due to the inertial terms on the left-hand side of the equations. In the Burnett equations, along with the non-linear inertial terms, the gradient of the stress and heat flux terms are also non-linear, owing to the non-linear constitutive relations as already noted above. Further, notice the presence of third-order derivative terms in the Burnett equations due to the first term of ω2 and ω3 terms in Eq. (5.146) and first term of θ2 and θ4 terms in Eq. (5.147); again referring to terms given in Sect. 5.5.2 will be helpful to see this point more clearly. The presence of third-order derivatives leads to the requirement of having to prescribe additional boundary conditions for velocity components, temperature, and pressure as compared to the Navier–Stokes equations. The Burnett equations are therefore third-order non-linear coupled partial differential equations. Notice that the stress is non-zero even if all the velocity and pressure gradients in the flow is zero; this is due to the dependence of stress on temperature gradient through the ω3 and ω5 terms in Eq. (5.146). This seems strange and unphysical, as it suggests the possibility of setting up a velocity field through purely a temperature gradient. Similarly, the heat flux term depends on velocity gradient, through first of θ2 and θ4 terms in Eq. (5.147). At the Navier–Stokes order, the coupling between velocity and temperature fields is through the variation of fluid properties (most notably, viscosity), compressibility, or buoyancy. In contrast, the coupling between velocity and temperature fields is much more involved at the Burnett order as evident from the above discussion. (This coupling can further be through the boundary condition, for example, if the velocity slip is dependent on both normal velocity gradient and tangential temperature gradient at the wall (Eq. (1.4)).

164

5 Burnett Equations: Derivation and Analysis

5.4.4 Super-Burnett Equations The third order approximation of the distribution function gives a superset of Burnett equations, known as Super-Burnett equations which are third order accurate in Knudsen number. However, because of the extreme complexity involved, full threedimensional non-linear Super-Burnett equations are not available. Here we present the linearized form of Super-Burnett equations as obtained by Shavaliyev [129] for weakly perturbed gas flows: ∂ur μ3 5 ∂ 2 4 ∂ 2 ∂ui (3) pij = (5.148) − pρ 3 ∂xi ∂xj ∂xr 3 ∂xr ∂xr ∂xj

(3)

qi

=−

μ3 ρ2

157 1 ∂ 2 ∂T 5 1 ∂ 2 ∂ρ + 16 T ∂xr ∂xr ∂xi 8 ρ ∂xr ∂xr ∂xi

(5.149)

In the same work [129], Shavaliyev derived the non-linear super-Burnett equations for purely one-dimensional flows. Accordingly, the third order approximation to the stress tensor and heat flux vector is given as (I )pxx

(3) pxx

(I )qx

qx(3)

(I I )p

xx + ,) * + ,) * 2 3 ∂u 32 RT ∂ 2 ρ ∂u μ 47 R ∂T ∂ρ ∂u 40 RT ∂ρ + = 2 − 3 ρ ∂x ∂x ∂x 3 ρ 2 ∂x ∂x 3 ρ ∂x 2 ∂x p 2 RT ∂ρ ∂ 2 u R ∂T 2 ∂u 47 ∂T ∂ 2 u − R − − 7 3 ρ ∂x ∂x 2 T ∂x ∂x 9 ∂x ∂x 2 ∂ 3 u 16 ∂u 3 31 ∂ 2 T ∂u 2 (5.150) + RT 3 + − R 2 9 ∂x ∂x 9 27 ∂x ∂x

(I I )qx

(I I I )q

x + ,) * + ,) * ,) * 2 2 + 3 271 1 ∂ρ ∂u 421 ∂u ∂ 2 u μ 9005 1 ∂T ∂u = + + − pρ 168 T ∂x ∂x 21 ρ ∂x ∂x 42 ∂x ∂x 2 917 R ∂ρ ∂T 2 1137 R ∂T ∂ρ 2 397 R ∂ρ ∂ 2 T + − + 8 ρT ∂x ∂x 16 ρ 2 ∂x ∂x 16 ρ ∂x ∂x 2 701 R ∂T ∂ 2 ρ 813 R ∂T 3 1451 R ∂T ∂ 2 T + − − 16 ρ ∂x ∂x 2 16 T 2 ∂x 16 T ∂x ∂x 2 157 ∂ 3 T 41 RT ∂ρ ∂ 2 ρ 5 RT ∂ 3 ρ 23 RT ∂ρ 3 R − − − + 16 ∂x 3 8 ρ 2 ∂x ∂x 2 8 ρ ∂x 3 4 ρ 3 ∂x (5.151)

5.4 Derivation of Hydrodynamic Equations

165

Table 5.2 Values of the coefficients as reported in the literature Equation (3)

pxx

(3)

qx

Term (I )pxx (I I )pxx (I )qx (I I )qx (I I I )qx

Shavaliyev [129] −40/3 32/3 −9005/168 271/21 421/42

Fiscko and Chapman [56] −64/9 40/9 −8035/336 166/21 949/168

Torrilhon and Struchtrup [148] −64/9 40/9 −2913/112 188/21 199/56

Shavaliyev [129] believed that the solution of the problem of the structure of the shock wave can be improved further by considering these super-Burnett order terms. However, Fiscko and Chapman [56] reported that the solutions of the super-Burnett equations are actually inferior to those of the Burnett equations. A possible reason for this is to an algebraic error in the coefficient of one of the terms of these equations. It is worth noting that the coefficients of the marked terms used in Fiscko and Chapman work are different than those reported by Shavaliyev. All the remaining coefficients agree with those given by Fiscko and Chapman [56] except those for the five overlined terms. The three-dimensional linearized super-Burnett equations [148] derived from R13 equations (which we will discuss in the next chapter) were found to be same as derived by Shavaliyev (Eqs. (5.148)–(5.149)). Once again, there was discrepancy in the values of the coefficients of the first three underlined terms of the heat flux equation (Eq. (5.151)) as reported by Torrilhon and Struchtrup [148]. Table 5.2 gives the overall summary of the values of the coefficients as reported by these researchers. Note that the coefficients differ only in magnitude and not in sign.

5.4.4.1

Summary of Chapman–Enskog Method

The entire essence of the Chapman–Enskog method is captured in a detailed flowchart as shown in Fig. 5.4. Briefly summarizing, the particle distribution function is expanded in an infinite series around Maxwellian distribution with the assumption that this series converges. This series is then substituted into the Boltzmann equation. By performing some subtle mathematical operations, Enskog hypothesized that the problem can be divided into smaller bits. The solution of the zeroth level approximation is Maxwellian distribution and with the aid of this solution, higher level approximations to the distribution function can be determined. With known functional form of the distribution function at each level, the constitutive relationships for the pressure tensor and heat flux vector are evaluated. These constitutive relationships are then substituted into Cauchy’s equation of motion and the necessary closure is achieved. At zeroth, first, second and third level of approximation, we obtain Euler [O(Kn0 )], Navier–Stokes [O(Kn1 )], Burnett [O(Kn2 )],and super-Burnett [O(Kn3 )] equations, respectively.

166

5 Burnett Equations: Derivation and Analysis

5.5 Burnett Equations in Cylindrical Coordinates In this section, the Burnett equations presented in Sect. 5.4.3 have been cast in cylindrical coordinates. Since the terms are in expanded form, equations in cylindrical coordinates help to appreciate the significance of the different terms and the associated complexity. The conservation equations mass, momentum, and energy equations cast in cylindrical coordinates are 1 ∂ρur 1 ∂ρv ∂ρw ∂ρ + + + =0 ∂t r ∂r r ∂θ ∂z ∂ρu 1 ∂(ρuu + p)r 1 ∂prr r 1 ∂ρuv + + + ∂t r ∂r r ∂r r ∂θ ∂ρwu ∂prz ρvv + pθθ 1 ∂prθ + + + =0 + r ∂θ ∂z ∂z r ∂ρv 1 ∂ρuvr 1 ∂pθr r 1 ∂(ρvv + p) + + + ∂t r ∂r r ∂r r ∂θ ∂ρwv ∂pθz ρuv − pθr 1 ∂pθθ + + + =0 + r ∂θ ∂z ∂z r

(5.152)

(5.153)

(5.154)

∂ρw 1 ∂ρuwr 1 ∂pzr r 1 ∂ρwv 1 ∂pzθ ∂(ρww + p) ∂pzz + + + + + + =0 ∂t r ∂r r ∂r r ∂θ r ∂θ ∂z ∂z (5.155) ∂e 1 ∂(e + p)ur 1 ∂(uprz + vpθr + wprr − qr ) + + ∂t r ∂r r ∂r ∂(ρ(e + p)w) 1 ∂(e + p)v + + r ∂θ ∂z 1 ∂(uprθ + vpθθ + wpzθ − qθ ) ∂(uprz + vpθz + wpzz − qz ) + = 0. + r ∂θ ∂z (5.156) where ρe = ρ( + u2 /2) = ρ(cv T + u2 /2). As before, the stress tensor and heat flux vector are expanded in a series as Eu NS Bu SB + pij + pij + pij + O(Knn+1 ) pij = pij

(5.157)

qi = qiEu + qiN S + qiBu + qiSB + O(Knn+1 )

(5.158)

where n represents the accuracy in terms of O(Kn).

5.5 Burnett Equations in Cylindrical Coordinates

167

Note that the Euler order stress and heat flux terms are zero as seen in Sect. 5.4.1.

5.5.1 Navier–Stokes Stress and Heat Flux The Navier–Stokes order stress and heat flux terms are included here for completeness: ∂u 1 ∂v ∂w v NS prr = −μ δ1 + δ2 + δ2 + δ2 (5.159) ∂r r ∂θ ∂z r ∂u 1 ∂v ∂w u NS + δ1 + δ2 + δ1 (5.160) = −μ δ2 pθθ ∂r r ∂θ ∂z r ∂u 1 ∂v ∂w u NS + δ2 + δ1 + δ2 (5.161) pzz = −μ δ2 ∂r r ∂θ ∂z r ∂v v 1 ∂u NS NS (5.162) prθ = pθr = −μ − + ∂r r r ∂θ 1 ∂w ∂v NS NS (5.163) pθz = pzθ = −μ + r ∂θ ∂z ∂u ∂w NS NS (5.164) pzr = pzr = −μ + ∂z ∂r ∂T ∂r 1 ∂T = −k r ∂θ ∂T = −k ∂z

qrN S = −k

(5.165)

qθN S

(5.166)

qzN S

(5.167)

where the coefficients are given as: δ1 = 1/3 and δ2 = 2/3.

5.5.2 Burnett Stress and Heat Flux Singh and Agrawal [130] used the coordinate transformation technique on stress and heat flux terms in Cartesian coordinates (Eqs. (5.146) and (5.147)) to derive the following Burnett order stress and heat flux terms in cylindrical coordinates (some of the typographical errors appeared in [126] have been corrected in the following equations).

168

5 Burnett Equations: Derivation and Analysis

5.5.2.1

Burnett Stress Terms

In this section, we document all the six components of the stress tensor (out of which five are independent given the traceless nature of the stress tensor).

Bu prr

μ2 = p

2ω1 14ω2 2ω6 − + 3 9 9

∂u ∂r

2

+

2ω2 2ω6 ω1 + − 3 9 9

u ∂u r ∂r

2 ∂u v ∂u ω6 1 ω6 2ω2 + − 3 12 ∂θ 3 6 r2 r 2 ∂θ ω ω6 ∂u 2 7ω2 ω6 u 2 ω1 2 + + + − + − 3 12 ∂z 3 9 9 r 2 ∂v 1 ∂v 2 2ω2 ω1 ω6 7ω2 ω6 − − − − + 3 12 ∂r 3 9 9 r ∂θ 2 ω ∂v v ∂v ω6 ω6 2ω2 2 − − + + 3 6 ∂z 3 6 r ∂r u ∂v 14ω2 2ω6 ω6 v 2 ω2 2ω1 − + + + − 3 9 9 3 12 r r 2 ∂θ 2 ω 1 ∂w ω6 2 − + 3 6 r ∂θ ∂w 2 ∂w u 7ω2 ω6 4ω2 4ω6 ω1 2ω1 − + + − − − 3 9 9 ∂z 3 9 9 ∂z r ∂u ∂w 2ω2 1 ∂u ∂v 2ω2 2ω6 ω6 ω1 + − − − + 3 9 9 ∂r ∂z 3 6 r ∂θ ∂r 1 ∂v ∂w ω1 1 ∂u ∂v 4ω2 4ω6 2ω2 2ω6 2ω1 + − + − + − 3 9 9 r ∂θ ∂z 3 9 9 r ∂r ∂θ 2 2ω2 ∂w ∂u 2ω2 ∂w ω6 ω6 − − − − 3 6 ∂r ∂z 3 12 ∂r 1 ∂w ∂v ω6 2ω3 R ∂ 2 T ω3 R ∂T 4ω2 − + − + 3 3 r ∂θ ∂z 3 r ∂r 3 ∂r 2 +

ω

2

+

−

ω3 R ∂ 2 T ω3 R ∂ 2 T − 2 3 ∂z 3 r 2 ∂θ 2

+

2ω4 R ∂T ∂p ω4 R ∂T ∂p ω4 1 R ∂T ∂p 2ω5 R − − + 3 p ∂r ∂r 3 p ∂z ∂z 3 r 2 p ∂θ ∂θ 3 T

∂T ∂r

2

5.5 Burnett Equations in Cylindrical Coordinates

−

ω5 R 3 T

∂T ∂z

2 −

ω5 1 R 3 r2 T

∂T ∂θ

169

2 −

2ω2 1 ∂ 2 p ω2 1 ∂ 2 p + 3 ρ ∂r 2 3 ρ ∂z2

ω2 1 1 ∂ 2 p ω2 1 1 ∂p 2ω2 1 ∂p ∂ρ ω2 1 ∂p ∂ρ + + − 3 r 2 ρ ∂θ 2 3 r ρ ∂r 3 ρ 2 ∂r ∂r 3 ρ 2 ∂z ∂z ω2 1 1 ∂p ∂ρ − 3 r 2 ρ 2 ∂θ ∂θ

+

Bu pθθ

(5.168)

2 ∂u u ∂u ω1 μ2 ω1 7ω2 1ω6 2ω2 2ω6 = + − − + + − p 3 9 9 ∂r 3 9 9 r ∂r 2 ∂u 1 14ω2 2ω6 u 2 ω6 2ω2 + − − 2 9 9 r 3 12 ∂θ r 2 ∂u v ∂u ω6 ω6 ω6 ∂v 2 ω2 ω2 − + + + 6 3 6 ∂z 3 12 ∂r r 2 ∂θ ω6 v ∂v 2ω2 ω6 v 2 − − 6 r ∂r 3 12 r 2 1 ∂v u ∂v 14ω2 2ω6 28ω2 4ω6 4ω1 + − + + 9 9 r ∂θ 3 9 9 r 2 ∂θ ω ω6 ∂v 2 2 + + 3 12 ∂z ω 1 ∂w 2 ω6 ∂w 2 ω6 2ω2 2 − − + − 3 6 ∂r 3 12 r ∂θ ∂u ∂w 4ω2 4ω2 4ω6 ω6 ∂u ∂w 2ω1 + − − + − 3 9 9 ∂r ∂z 3 3 ∂z ∂r 1 ∂u ∂v ω1 1 ∂v ∂w ω6 2ω2 2ω6 2ω2 − + − + − 3 6 r ∂θ ∂r 3 9 9 r ∂θ ∂z 1 ∂u ∂v 2ω2 1 ∂w ∂v 2ω2 2ω6 ω6 ω1 + − + − + 3 9 9 r ∂r ∂θ 3 6 r ∂θ ∂z

2ω1 − + 3 4ω2 − + 3 2ω2 − + 3 2ω1 − + 3

ω3 R ∂ 2 T 2ω3 R ∂T ω3 R ∂ 2 T 2ω3 R ∂ 2 T ω4 R ∂T ∂p − − + − 2 2 2 2 3 r ∂r 3 ∂r 3 ∂z 3 r ∂θ 3 p ∂r ∂r 2 ω4 R ∂T ∂p 2ω4 1 R ∂T ∂p ω5 R ∂T ω5 R ∂T 2 + − − − 3 p ∂z ∂z 3 p r 2 ∂θ ∂θ 3 T ∂r 3 T ∂z 2 ∂w u 2ω2 2ω6 ω1 2ω5 R 1 ∂T + − + + 3 T r 2 ∂θ 3 9 9 ∂z r +

170

5 Burnett Equations: Derivation and Analysis

− −

7ω2 ω6 ω1 − + 3 9 9

∂w ∂z

2 +

ω2 1 ∂ 2 p ω2 1 ∂ 2 p + 3 ρ ∂r 2 3 ρ ∂z2

2ω2 1 1 ∂ 2 p 2ω2 1 1 ∂p − 3 ρ r 2 ∂θ 2 3 ρ r ∂r

ω2 1 ∂p ∂ρ ω2 1 ∂p ∂ρ 2ω2 1 1 ∂p ∂ρ − − + 2 2 3 ρ ∂r ∂r 3 ρ ∂z ∂z 3 r 2 ρ 2 ∂θ ∂θ

Bu pzz

(5.169)

2 ∂u u ∂u 7ω2 1ω6 4ω2 4ω6 2ω1 μ2 ω1 − + + − = − − p 3 9 9 ∂r 3 9 9 r ∂r 2 ∂u 7ω2 ω6 u 2 ω2 ω6 1 ω1 − + − + − 2 3 9 9 r 3 6 ∂θ r 2 v ∂u 2ω2 ∂u ω6 ω6 2ω2 − − − − 2 3 3 3 12 ∂z r ∂θ 2 ω ω ω6 ∂v ω6 v 2 2 2 − − + + 3 6 ∂r 3 6 r 2 1 ∂v u ∂v 7ω2 ω6 14ω2 2ω6 ω1 2ω1 − + − + − − 3 9 9 r ∂θ 3 9 9 r 2 ∂θ 2 2ω2 ∂v ω6 ω6 ∂w 2 ω2 − − + + 3 12 ∂z 3 12 ∂r 2 ω 1 ∂w ω6 14ω2 2ω6 ∂w 2 2ω1 2 + − + + + 3 12 r ∂θ 3 9 9 ∂z 1 ∂u ∂v ω1 ω6 2ω2 2ω6 ∂w ∂u 4ω2 − + − + + 3 3 r ∂θ ∂r 3 9 9 ∂z ∂r 2ω2 2ω6 1 ∂v ∂w −2ω2 ω6 ∂u ∂w ω1 + − + + + 3 9 9 r ∂θ ∂z 3 6 ∂z ∂r 1 ∂u ∂v 4ω2 4ω6 ω6 v ∂v ω2 2ω1 + − + + − 3 9 9 r ∂r ∂θ 3 3 r ∂r 2ω2 1 ∂w ∂v ω6 ω4 1 R ∂T ∂p ω5 R ∂T 2 + + − + 3 6 r ∂θ ∂z 3 p r 2 ∂θ ∂θ 3 T ∂r 2 2 2ω2 2ω6 ∂w u ω1 2ω5 R ∂T ω5 R 1 ∂T + − + − + 3 T ∂z 3 T r 2 ∂θ 3 9 9 ∂z r

+

ω2 1 ∂ 2 p 2ω2 1 ∂ 2 p ω2 1 1 ∂ 2 p ω2 1 1 ∂p ω2 1 ∂p ∂ρ − − + + 3 ρ ∂r 2 3 ρ ∂z2 3 r 2 ρ ∂θ 2 3 r ρ ∂r 3 ρ 2 ∂r ∂r

5.5 Burnett Equations in Cylindrical Coordinates

+

ω2 1 1 ∂p ∂ρ ω3 R ∂T ω3 ∂ 2 T 2ω2 1 ∂p ∂ρ − − − R 3 ρ 2 ∂z ∂z 3 r 2 ρ 2 ∂θ ∂θ 3 r ∂r 3 ∂r 2

2ω3 R ∂ 2 T ω3 R ∂ 2 T ω4 R ∂T ∂p 2ω4 R ∂T ∂p + + − − 2 2 2 3 ∂z 3 r ∂θ 3 p ∂r ∂r 3 p ∂z ∂z

Bu Bu prθ = pθr =

171

(5.170)

ω1 v ∂u 5ω2 ω6 μ2 − + − p 2 3 6 r ∂r u ∂u ω1 2ω2 ω6 2ω2 ω6 uv ω1 − + − − − + 2 3 6 2 3 6 r 2 ∂θ r2 u ∂v ω1 v ∂v 5ω2 ω6 2ω2 ω6 ω1 − + − + − + 2 3 6 r ∂r 2 3 6 r 2 ∂θ ω ω ω2 ω6 v ∂w ω2 ω6 1 ∂u ∂w 1 1 + − + − + − 2 3 3 r ∂z 2 3 3 r ∂θ ∂z 1 ∂u ∂u 5ω2 ω6 ω2 ω6 ∂v ∂w ω1 ω1 − + + − + + 2 3 6 r ∂r ∂θ 2 3 3 ∂r ∂z ∂u ∂v ω1 2ω2 ω6 2ω2 ω6 ω1 − + − + + + 2 3 6 ∂r ∂r 2 3 6 ω1 1 ∂v ∂v 5ω2 ω6 1 ∂u ∂v − + + × 2 2 3 6 r ∂r ∂θ r ∂θ ∂θ ∂v ∂w 1 ∂u ∂w ω6 ω6 ω6 − ω2 − − ω2 − − ω2 − 4 ∂z ∂r 4 r ∂z ∂θ 4 ω4 R 1 ∂p ∂T R ∂ 2T R ∂T 1 ∂w ∂w − ω3 2 + + ω3 × r ∂r ∂θ r ∂r∂θ 2 r p ∂θ ∂r r ∂θ ω4 1 R ∂p ∂T ω6 ∂u ∂v 1 R ∂T ∂T 1 1 ∂p + ω5 + + ω2 2 2 r p ∂r ∂θ T r ∂r ∂θ 4 ∂z ∂z r ρ ∂θ 2 ω2 1 1 ∂p ∂ρ ω2 1 1 ∂p ∂ρ 11 ∂ p + + (5.171) −ω2 2 r ρ ∂r∂θ 2 r ρ ∂θ ∂r 2 r ρ 2 ∂r ∂θ +

Bu pθz

=

Bu pzθ

5ω2 ω6 ω1 u ∂v μ2 ω6 v ∂u ω2 − − + + = p 4 r ∂z 2 3 6 r ∂z v ∂w ω1 u ∂w 2ω2 ω6 ω6 − + + + ω2 − 4 r ∂r 2 3 6 r 2 ∂θ ∂v ∂w ω1 ω1 2ω2 ω6 ω2 ω6 ∂u ∂v + + − + + − 2 3 6 ∂z ∂z 2 3 3 ∂r ∂z

172

5 Burnett Equations: Derivation and Analysis

ω6 1 ∂u ∂u ω6 ∂u ∂v − ω2 − − ω2 − 4 r ∂θ ∂z 4 ∂z ∂r 1 ∂v ∂v 5ω2 ω6 ω1 ω6 1 ∂u ∂w − + − ω2 − + 2 3 6 r ∂z ∂θ 4 r ∂θ ∂r ω1 1 ∂w ∂w 5ω2 ω6 ω6 ∂v ∂w + − + + 4 ∂r ∂r 2 3 6 r ∂z ∂θ ω 1 ∂u ∂w ω1 ω2 ω6 2ω2 ω6 1 + − − + + + 2 3 3 r ∂r ∂θ 2 3 6 2 ω4 1 R ∂T ∂p ω4 1 R ∂T ∂p R ∂ T 1 ∂v ∂w + + + ω3 × 2 r ∂θ ∂z 2 r p ∂z ∂θ 2 r p ∂θ ∂z r ∂θ ∂θ

ω2 1 1 ∂p ∂ρ 1 R ∂T ∂T 1 1 ∂ 2p − ω2 + T r ∂z ∂θ r ρ ∂θ ∂z 2 r ρ 2 ∂θ ∂z ω2 1 1 ∂p ∂ρ + 2 r ρ 2 ∂z ∂θ

+ ω5

Bu Bu prz = pzr =

(5.172)

ω2 ω6 u ∂u μ2 ω1 ω6 v ∂v + − + ω2 − p 2 3 3 r ∂z 4 r ∂z ω u ∂w ω1 ∂u ∂w ω2 ω6 2ω2 ω6 1 + − − + + + 2 3 3 r ∂r 2 3 6 ∂z ∂z 1 ∂u ∂v ∂u ∂u 5ω2 ω6 ω1 ω6 − + − ω2 − + 2 3 6 ∂r ∂z 4 r ∂θ ∂z ∂v ∂v 1 ∂u ∂v ω2 ω6 ω1 ω6 + − + − ω2 − 4 ∂z ∂r 2 3 3 r ∂z ∂θ ∂w ∂w ω1 5ω2 ω6 2ω2 ω6 ω1 − + − + + + 2 3 6 ∂z ∂r 2 3 6 1 ∂v ∂w ω2 ω6 ω6 v ∂w ω1 ∂u ∂w + − + − × ∂r ∂r 2 3 3 r ∂θ ∂r 4 r 2 ∂θ ω6 1 ∂u ∂w ω6 1 ∂v ∂w ∂ 2T − ω + ω + − R 2 3 4 r 2 ∂θ ∂θ 4 r ∂r ∂θ ∂r∂z R ∂T ∂T 1 ∂ 2p ω4 R ∂T ∂p ω4 R ∂p ∂T + + ω5 − ω2 2 p ∂r ∂z 2 p ∂r ∂z T ∂r ∂z ρ ∂r∂z ω2 1 ∂p ∂ρ ω2 1 ∂p ∂ρ + (5.173) + 2 ρ 2 ∂z ∂r 2 ρ 2 ∂r ∂z +

5.5 Burnett Equations in Cylindrical Coordinates

5.5.2.2

173

Burnett Heat Flux Terms

In this section, we document the three components of the heat flux vector. As shown in Sect. 5.6, these terms are of order Knudsen square as also the stress tensor terms in the previous section. qrBu

2 θ4 ∂ 2 u θ4 1 ∂ 2 u 2θ2 ∂ u 2θ4 μ2 + − − = + ρ 3 3 2 ∂z2 2 r 2 ∂θ 2 ∂r 2 1 ∂u 2θ2 u 2θ2 2θ2 2θ4 2θ4 θ4 + − − − − − 3 3 r ∂r 3 3 3 6 r2 2θ2 1 ∂v 2θ2 ∂ 2w 7θ4 θ4 1 ∂ 2v − − + − × r ∂r∂θ 3 6 3 6 ∂r∂z r 2 ∂θ u 1 ∂T θ3 u 1 ∂p θ3 v 1 ∂p 3θ5 v 1 ∂T 2θ2 − θ5 − − − + θ1 − 2 3 r T ∂r 2 r T ∂θ 3 r p ∂r 2 r 2 p ∂θ 1 ∂u ∂T 3θ5 1 ∂u ∂T 3θ5 1 1 ∂u ∂T 8θ2 + 2θ5 + + + θ1 − 3 T ∂r ∂r 2 r 2 T ∂θ ∂θ 2 T ∂z ∂z θ3 1 ∂u ∂p θ3 1 1 ∂u ∂p θ3 1 ∂w ∂p + − 2 p ∂z ∂z 2 r 2 p ∂θ ∂θ 3 p ∂z ∂r 1 1 ∂v ∂T θ3 1 1 ∂v ∂p 3θ5 − + θ1 − 2θ2 + 2 r T ∂r ∂θ 3 r RT ∂θ ∂r 1 1 ∂T ∂v 1 ∂w ∂T 2θ2 3θ5 − θ5 − 2θ2 − + θ1 − 3 r T ∂r ∂θ 2 T ∂r ∂z 1 ∂w ∂T 1 ∂p ∂v θ3 θ3 1 ∂w ∂p 2θ2 + + θ1 − − θ5 + 2 p ∂r ∂z pr ∂θ ∂r 2 3 T ∂z ∂r 1 ∂p ∂u 2θ3 + p ∂r ∂r 3 +

qθBu

(5.174)

7θ4 2θ2 1 ∂u θ4 ∂ 2 v μ2 θ4 ∂ 2 v + + = + ρ 3 6 2 ∂z2 2 ∂r 2 r 2 ∂θ θ4 1 ∂v 1 ∂ 2v 2θ4 θ3 1 ∂v ∂p θ4 v 2θ2 + − + + − 3 3 2 r ∂r 2 p ∂r ∂r 2 r2 r 2 ∂θ 2 u 1 ∂T θ3 v 1 ∂p 2θ2 − θ5 − + θ1 − 3 2 r p ∂r r 2 T ∂θ 2 2θ2 ∂ u θ4 1 1 ∂u ∂T 2θ3 u 1 ∂p 2θ2 − − − θ + θ + − 1 5 3 r 2 p ∂θ 3 6 ∂r∂θ 3 r T ∂r ∂θ

174

5 Burnett Equations: Derivation and Analysis

θ3 1 1 ∂u ∂p θ3 1 1 ∂u ∂p 3θ5 1 ∂v ∂T 3θ5 1 ∂v ∂T + + + 3 r p ∂r ∂θ 2 r p ∂θ ∂r 2 T ∂r ∂r 2 T ∂z ∂z 1 1 ∂v ∂T 8θ2 θ3 1 ∂v ∂p 2θ3 1 1 ∂v ∂p + + θ + 2θ − + 1 5 2 p ∂z ∂z 3 r 2 RT ∂θ ∂θ 3 r 2 T ∂θ ∂θ 2θ2 1 1 ∂w ∂T 1 ∂ 2w 2θ2 θ4 − − − θ5 + θ1 − 3 6 r ∂z∂θ 3 r T ∂z ∂θ 1 1 ∂w ∂T θ3 1 1 ∂w ∂p v 1 ∂T −3θ5 3θ5 + + 2θ2 − − 2θ2 − 2 r T ∂θ ∂z 3 p r ∂z ∂θ r T ∂r 2 3θ5 1 1 ∂T ∂u (5.175) + −2θ2 + 2 r T ∂r ∂θ

−

qzBu =

2 2θ2 1 ∂u ∂ w θ4 2θ4 θ4 ∂ 2 w 2θ2 μ2 − − − + − 2 ρ 3 6 r ∂z 2 ∂r 3 3 ∂z2 u 1 ∂T θ3 u 1 ∂p θ4 1 ∂ 2 w 2θ2 − θ − + + θ − 1 5 2 2 2 r ∂θ 3 r T ∂z 3 r p ∂z 1 ∂u ∂T 2θ2 3θ5 1 1 ∂u ∂T − θ5 + −2θ2 + + θ1 − 3 T ∂r ∂z 2 r T ∂z ∂r 2 2θ2 ∂ u θ4 θ3 1 ∂u ∂p θ3 1 ∂u ∂p − − + − 3 p ∂r ∂z 3 6 ∂r∂z 2 p ∂z ∂r 2 1 ∂ v θ4 θ3 1 1 ∂v ∂p 2θ2 − + − 3 6 r ∂z∂θ 2 r p ∂z ∂θ 1 1 ∂v ∂T θ3 1 1 ∂v ∂p 3θ5 1 ∂w ∂T 2θ2 − θ5 + − + θ1 − 3 r T ∂θ ∂z 3 p r ∂θ ∂z 2 T ∂r ∂r 1 ∂w ∂T 2θ3 1 ∂w ∂p θ3 1 ∂w ∂p 8θ2 + θ1 − + 2θ5 + + 2 p ∂r ∂r 3 T ∂z ∂z 3 p ∂z ∂z θ3 1 1 ∂w ∂p 3θ5 1 1 ∂w ∂T + 2 r 2 p ∂θ ∂θ 2 r 2 T ∂θ ∂θ θ4 1 ∂w 1 1 ∂v ∂v 3θ5 + + − 2θ2 2 r ∂r r T ∂z ∂θ 2

+

(5.176)

The above equations involve 11 coefficients—six ω’s and five θ ’s. The values of these coefficients for both Maxwell and hard sphere molecules are tabulated in Table 5.1. The above Navier–Stokes and Burnett-order stress and heat flux terms when substituted into the conservation equations (Eqs. (5.152)–(5.156)) give the complete Burnett equation in cylindrical coordinates.

5.6 Order of Magnitude Analysis

175

5.6 Order of Magnitude Analysis An order of magnitude analysis can provide important insights about the various terms in an equation. Such an analysis can be utilized to remove the less significant terms from the equations, thereby simplifying the process of analysis. Specifically, we will examine the various terms in the Burnett-order stress tensor and heat flux vector, and compare them with the corresponding terms of the Navier–Stokes order. We select U , ρ, and L as the velocity, density, and length scale, respectively. Pressure difference in a flow scales with ρU 2 , while from ideal gas law, pressure scales as ρc2 (where c is the speed of sound). In an analogous manner to pressure, the temperature difference scales as ρU 2 /R while the absolute temperature scales as ρc2 /R where R is the gas constant. Normalizing stress terms by ρU 2 and heat flux terms by ρU 3 , we obtain NS pij

ρU 2

=−

2μ ∂ui μ U 1 Kn ∼ = ∼ Re Ma ρU 2 ∂xj ρU 2 L

qiN S 1 Kn k ∂T k ρU 2 ∼ ∼ =− ∼ 3 3 3 ReP r Ma ρU ρU ∂xi ρU RL where P r is Prandtl number, which is order unity for gases. It is therefore seen that N S and q N S are of order Kn. both pij i Repeating the above exercise for the various terms in Eq. (5.146) other than the external force term, we obtain Bu pij,1

ω1 μ2 ∂uk μ2 U 2 Ma 2 Sij ∼ = ∼ Kn2 2 2 2 2 ρU p ∂xk ρU ρc L Re2 Bu pij,2 μ2 ρU 2 ω2 μ2 ∂ 1 ∂p Ma 2 ∼ = − = ∼ Kn2 ρU 2 ρU 2 p ∂xi ρ ∂xj ρU 2 ρc2 ρL2 Re2 ρU 2

Bu pij,3

ρU 2 Bu pij,4

ρU 2 Bu pij,5

ρU 2 Bu pij,6

ρU 2

=

=−

ω2 μ2 ∂uk ∂uj μ2 U 2 Ma 2 ∼ = ∼ Kn2 ρU 2 p ∂xi ∂xk ρU 2 ρc2 L2 Re2

=−

ω2 2μ2 ∂uk μ2 U 2 Ma 2 Sj k ∼ = ∼ Kn2 2 2 2 2 ρU p ∂xi ρU ρc L Re2

=

ω3 μ2 ∂ 2 T μ2 U 2 Ma 2 ∼ = ∼ Kn2 ρU 2 ρT ∂xi ∂xj ρU 2 ρc2 L2 Re2

=

ρU 4 ω4 μ2 ∂T ∂p μ2 Ma 4 ∼ = ∼ Ma 2 Kn2 ρU 2 pρT ∂xi ∂xj ρU 2 ρc2 ρc2 L2 Re2

176

5 Burnett Equations: Derivation and Analysis Bu pij,7

ρU 2 Bu pij,8

ρU 2

=

U4 ω5 μ2 ∂T ∂T μ2 Ma 4 ∼ = ∼ Ma 2 Kn2 ρU 2 ρT 2 ∂xi ∂xj ρU 2 ρc2 c2 L2 Re2

=

ω6 μ2 μ2 U 2 Ma 2 = ∼ Kn2 Ski Sj k ∼ 2 2 2 2 ρU p ρU ρc L Re2

Similarly, repeating the above exercise for the various terms in Eq. (5.147), we obtain Bu qi,1

ρU 3

θ1 μ2 ∂uk ∂T μ2 U 3 Ma 2 ∼ = ∼ Kn2 ρU 3 ρT ∂xk ∂xi ρ 2 U 3 c 2 L2 Re2

=

We decompose the first term in the curly braces of Eq. (5.147) into two terms ∂uk 2 ∂ 2 uk 2 ∂T ∂uk 2 ∂ T =− T − − 3 ∂xi ∂xk 3 ∂xi xk 3 ∂xi ∂xk and proceed as Bu qi,2

ρU 3 Bu qi,3

ρU 3 Bu qi,4

ρU 3 Bu qi,5

ρU 3 Bu qi,6

ρU 3 Bu qi,7

ρU 3

=−

θ2 2 μ2 ∂ 2 uk μ2 U 1 Kn2 ∼ 2 3 2 = ∼ 3 2 ρU 3 ρ ∂xi ∂xk ρ U L Re Ma 2

=−

θ2 2 μ2 ∂uk ∂T μ2 U 3 Ma 2 ∼ = ∼ Kn2 ρU 3 3 ρT ∂xk ∂xi ρ 2 U 3 c 2 L2 Re2

= −2

θ2 μ2 ∂uk ∂T μ2 U 3 Ma 2 ∼ = ∼ Kn2 ρU 3 ρT ∂xi ∂xk ρ 2 U 3 c 2 L2 Re2

=

∂p θ3 μ2 μ2 ρU 3 Ma 2 Sik ∼ 2 3 2 2 = ∼ Kn2 3 ∂xk ρU ρp ρ U ρc L Re2

=

θ4 μ2 ∂Sik μ2 U 1 Kn2 ∼ = ∼ ρU 3 ρ ∂xk ρ 2 U 3 L2 Re2 Ma 2

=3

∂T θ5 μ2 μ2 U 3 Ma 2 S ∼ = ∼ Kn2 ik ∂xk ρU 3 ρT ρ 2 U 3 c 2 L2 Re2

The above exercise clearly shows that all the Burnett terms are of order Kn2 . The stress terms in Eq. (5.146) contain eights terms, of which six terms are of O(Kn2 ) while two terms are of O(Ma 2 Kn2 ). Therefore, one can neglect the two terms of O(Ma 2 Kn2 ) at lower Mach numbers. The heat flux terms in Eq. (5.147) contain seven terms, of which five terms are of O(Kn2 ) while two terms are of O(Kn2 /Ma 2 ). Therefore, one can neglect the two terms of O(Kn2 /Ma 2 ) at higher values of Mach numbers.

5.7 Some Exact Solutions

177

5.7 Some Exact Solutions Given the complex nature of the equations, it is not easy to find an analytical solution of the Burnett equations. Additionally, since the Burnett equations involve thirdorder derivatives, its solution requires additional boundary conditions. Prescribing the boundary conditions accurately is crucial, as the solution strongly depends on the boundary conditions. However, finding these additional boundary conditions is not easy and is therefore an additional impediment in obtaining analytical solution of the Burnett equations. Only recently, the Burnett equations have been solved analytically using the two approaches as given below.

5.7.1 Iterative Approach of Agrawal and Singh It is apparent from Eqs. (5.157)–(5.158) that the Burnett equations is a superset of the Navier–Stokes equations. That is, all the terms of the Navier–Stokes equations are contained in the Burnett equations, along with additional terms being present in the latter equation. This observation was advantageously used by Agrawal and Singh to “build-up” solution of the Burnett equations starting from the solution of the Navier–Stokes equations for two different problems. The primary steps of their approach are as follows (Singh et al. [132]—see also Fig. 5.6) (i) Start with the solution of the Navier–Stokes equations and assume that this is a solution of the Burnett equations. (The approach is applicable only for cases for which the Navier–Stokes solution is available.) (ii) Substitute this preliminary solution in the Burnett equations and evaluate the order of magnitude of various terms in the Burnett equations. (The fact that explicit expressions for density, velocity, pressure, etc. are available allows determination of the Burnett-order stress terms, which are essentially product of the gradients of these quantities.) (iii) Compare the magnitude of each term with the highest-order term in the governing equations from which the preliminary solution has been obtained. Therefore, the significance of each of the additional term from the Burnett equations can be gauged, and allows identification of term(s) of significance. (For a physical problem with specified dimensions of the flow passage and gas properties, the magnitude of each term can be obtained and compared against the term in the equation with the highest magnitude.) (iv) If the additional term(s) are significant, augment the governing equations with this additional term(s), and re-solve the augmented governing equations. (The approach works only if the intermediate equation obtained after augmenting the Navier–Stokes with the additional terms identified in this step can still be solved analytically. The order of the equation can also go up at this stage if

178

5 Burnett Equations: Derivation and Analysis Lower Order Model, LOM (Eqs. 5.159 to 5.167) for i = 1, IM is LOM Solution of Intermediate Model, IM

Intermediate Model, IM

substitution augmentation Higher Order Model, HOM (Eqs. 5.168 to 5.176)

Normalized Residue, ε

if ε ≥ 1%, update i

Highest Order Term

if ε < 1% Solution of Higher Order Model, HOM Fig. 5.6 Procedure for obtaining the solution of Burnett equations starting from the solution of Navier–Stokes equations, Singh and Agrawal [130]

a third-order derivative gets added to the governing equations, necessitating specification of additional boundary conditions.) (v) Repeat steps (ii)–(iv) above till convergence is achieved. Convergence can be achieved either when no additional term remains in the Burnett equations or the remaining terms are insignificant. In the latter case, the remaining terms can be neglected and convergence is deemed to have been achieved. Although the above procedure has been given in the context of Burnett and Navier–Stokes equations, it is general enough to be applicable to other nonlinear partial differential equations, although the truth of this statement has to be established by applying it other equations. Note that the procedure given above has to be applied at the point where the solution of the Burnett equations is expected to be substantially different than the solution for the Navier–Stokes equations, i.e., close to the wall. In practice, it is recommended to apply the above procedure at several points in the flow field.

5.7 Some Exact Solutions

179

Notice that per the above approach, additional boundary condition required for the solution of the Burnett equations is deferred till a third-order term gets added to the governing equation. That is, the procedure allows for a preliminary solution to be formulated, and defers prescription of the additional boundary conditions to the point that a third-order term gets added to the governing equations. The above procedure was successfully applied by Singh et al. [132] to isothermal flow in a microchannel, and the following solution was obtained: ReμRT u(x, y) = p(x)Dh

y/H − (y/H )2 + 2C1 Kn(z) + 8C2 Kn2 (z) . 1/6 + 2C1 Kn(z) + 8C2 Kn2 (z)

2

p p − 1 + 96C2 Kn20 log p0 p0 p0 2 + 2Re βχ 12C1 Kn0 −1 p p0 2 x − x0 p 2 + 24C2 Kn0 = −96Reβ − 1 − log p p0 Dh

p p0

(5.177)

− 1 + 24C1 Kn0

(5.178)

where 2Kn2 1 β= χ= π A =

2 u dA ¯ A u

1/30 + 23 C1 Kn + 4C12 Kn2 + 83 C2 Kn2 + 64C22 Kn4 + 32C1 C2 Kn3 . (1/6 + 2C1 Kn + 8C2 Kn2 )2

The above solution is same as that obtained in Sect. 2.4.2.2. This is because the procedure given above converged after the first step itself, and the magnitude of the all terms in the Burnett stress equation was negligible ( 3. Notice that the functional form of the distribution function (Eq. (6.9)) is such that (n) its dependence on space and time is only through the Hermite coefficients, ai and (n) the molecular velocity dependence is through the Hermite polynomials, Hi1 i2 ···in . Once the distribution function is determined, it can be used to compute the unknown higher order moments Ii1 i2 ...in k appearing in the flux term and the production term J¯i1 i2 ···in , and the necessary closure is achieved.

6.3 20-Moment Approximation In the present chapter, we first present a detailed derivation of the 20-moment equations, followed by the more popular 13-moment equations. A brief outline of the sections to follow is shown in Fig. 6.1. For the 20-moment equations, we adhere to the classic paper [62], so as to bring forth the essence of the moment method. However, it is more appropriate to consider a set of first 13 moments

6.3 20-Moment Approximation

193

Fig. 6.1 20- and 13-moment equations from Boltzmann equation using Grad’s distribution function

1 n

u (x,t) = i

ci f (x, c,t)dc

r (x,t) =

m

T (x,t) = 1 m2 C2 f (x, c,t)dc 3 rkB

f (x, c,t)dc Grad 20 moment approximation

Pi j (x,t) =

Si jk (x,t) =

m CiC j f (x, c,t)dc

m

m

CiC jCk f (x, c,t)dc

Qi jkl (x,t) = CiC jCkCl fG|20 (x, c,t)dc

Fig. 6.2 Expressions for the variables considered in the Grad 20-moment approximation; variables in solid circles are the primary variables while those in dashed circles are the closure variables

(ρ, ui , T , pij , and qi ) having physical intuitive meaning rather than the 20-moment approximation. The transition from 20-moment to 13-moment approximation can be done by contracting the third order moment Sij k as Sikk which actually is twice the heat flux vector (see Eq. (6.6)). In 20-moment approximation, the primary variables considered are {ρ, ui , (= 3RT /2), Pij , Sij k } as shown in Fig. 6.2. The higher order moments Pij and Sij k are treated on par with the thermodynamic variables and separate evolution equations are generated for these moments. However, in the evolution equation for Sij k , a fourth order moment Qij kl is present in the flux term which needs to be evaluated in terms of primary variables to achieve the necessary closure.

194

6 Grad Equations: Derivation and Analysis

6.3.1 Evolution Equations The higher order moments Pij , Sij k , and Qij kl are defined with respect to peculiar velocity (Eqs. (6.3)–(6.5)). Hence, in order to generate the evolution equations for these moments, it is convenient to select peculiar velocity C as an independent variable. Accordingly, the Boltzmann equation (Eq. (5.1)) needs to be converted to some appropriate form in which peculiar velocity replaces the molecular velocity as an independent variable. The derivation of transformed Boltzmann equation [f (x, c, t) ⇒ f (x, C, t)] is given in Appendix C.1 and is mathematically represented as Df ∂f ∂f ∂uj Duk ∂f + Fk − − Ck = J (f, f1 ) (6.10) + Ck Dt ∂xk Dt ∂Ck ∂Cj ∂xk To generate the evolution equation for higher order moment, say, Ria1 i2 ...in

=m

C 2a Ci1 Ci2 . . . Cin f (x, C, t)dC

we multiply the transformed Boltzmann equation (Eq. (6.10)) by mC 2a Ci1 Ci2 . . . Cin and integrate over the velocity space. We have already generated the evolution equations for the first five variables {ρ, ui , (= 3RT /2)} in Sect. 5.2.4 which read as ∂(ρuk ) ∂ρ + = 0, ∂t ∂xk

(6.11)

∂ui ∂ui 1 ∂p 1 ∂pik + uk + + = Fi , ∂t ∂xk ρ ∂xi ρ ∂xk

(6.12)

3 ∂T 3 ∂T ∂ui ∂ui ∂qk Rρ + Rρuk +p + pij + = 0. 2 ∂t 2 ∂xk ∂xi ∂xj ∂xk

(6.13)

The evolution equations for the higher order moments are more involved and the detailed procedure is explained in the next section.

6.3.1.1

Evolution Equation for Pij

In order to obtain the evolution equation for the pressure tensor Pij , we multiply the transformed Boltzmann equation (Eq. (6.10)) by mCi Cj and then integrate over the velocity space as

6.3 20-Moment Approximation

195

Df ∂f ∂f Duk Fk − dC + m Ci Cj Ck Ci Cj m C i Cj dC + m dC Dt ∂xk Dt ∂Ck ∂f ∂ul dC = m Ci Cj J (f, f1 )dC (6.14) − m Ci C j C k ∂Cl ∂xk

Since we have j as free index in Eq. (6.14), we changed the dummy index j in the fourth term of (Eq. (6.10)) to l. Applying rules of integration by parts and Leibnitz’s rule to take out the differentiation outside the integral, we evaluate term by term as (a) = m

Ci C j

Df dC Dt

Df Ci Cj DCi Cj =m dC − m f dC Dt Dt DCj D DCi = m f Ci Cj dC − m f Ci + Cj dC Dt Dt Dt DPij = − m f (0 + 0) dC [∵ C, t are independent variables] Dt DPij Df (a) = m Ci Cj dC = (6.15) Dt Dt

∴

(b) = m

Ci C j C k

=m

∂f dC ∂xk

∂(f Ci Cj Ck ) dC − m ∂xk

f

∂(Ci Cj Ck ) dC ∂xk

∂Sij k −0 [∵ C, x are independent variables] ∂xk ∂Sij k ∂f (b) = m Ci Cj Ck dC = (6.16) ∂xk ∂xk =

∴

Assuming the external force to be independent of the peculiar velocity, the term in the rounded bracket can be taken outside the integral sign as ∂f Duk Fk − (c) = m Ci Cj dC Dt ∂Ck ∂f Duk dC = Fk − m C i Cj Dt ∂Ck ∂Ci Cj ∂Ci Cj f Duk Duk m m f dC − Fk − dC = Fk − Dt ∂Ck Dt ∂Ck

196

6 Grad Equations: Derivation and Analysis

For the first term, we apply Gauss divergence theorem to convert the volume integral into a surface integral as

+∞ −∞

∂ ∂Ck

= Duk Duk Fk − Fk − mCi Cj f dC = mCi Cj f nk dA Dt Dt c→∞ (6.17)

This surface integral needs to be performed over the control surface for C → ∞. As there are no particles having infinite molecular velocities (or peculiar velocity), i.e., f (C → ∞) = 0, this surface integral vanishes. ∂Ci Cj Duk ∴ (c) = 0 − Fk − m f dC Dt ∂Ck ∂Cj ∂Ci Duk dC m f Ci + Cj = − Fk − Dt ∂Ck ∂Ck Duk = − Fk − m f Ci δj k + Cj δik dC Dt Duk δj k m f Ci dC + δik m f Cj dC = − Fk − Dt =0 [∵ m f Ci dC = 0] ∴

(c) = m

∂f Duk Fk − Ci Cj dC = 0 Dt ∂Ck

(d) = −m

Ci C j C k

∂ul =− m ∂xk =−

∂ul m ∂xk

∂f ∂ul dC ∂Cl ∂xk

C i Cj Ck

(6.18)

∂f dC ∂Cl

∂Ci Cj Ck f ∂ul dC + m ∂Cl ∂xk

f

∂Ci Cj Ck dC ∂Cl

Converting the volume integral in the first term from to the surface integral and using the same arguments as used for first term in (c), the term vanishes. ∂Cj ∂ul ∂Ck ∂Ci dC ∴ (d) = 0 + m f Ci Cj + Ci C k + Cj C k ∂xk ∂Cl ∂Cl ∂Cl ∂ul = m f Ci Cj δkl + Ci Ck δj l + Cj Ck δil dC ∂xk

6.3 20-Moment Approximation

197

∂ul δkl m f Ci Cj dC + δj l m Ci Ck dC + δil m Cj Ck dC = ∂xk ∂uj ∂uk ∂ui + Pik + Pj k ∂xk ∂xk ∂xk ∂uj ∂f ∂ul ∂uk ∂ui ∴ (d) = −m Ci Cj Ck dC = Pij + Pik + Pj k ∂Cl ∂xk ∂xk ∂xk ∂xk = Pij

(6.19)

For now, let’s denote the production term as m

(2) Ci Cj J (f, f1 )dC = J¯ij

(6.20)

Substituting for the Lagrangian derivative in the first term (a) and combining results of all the terms, we obtain the evolution equation for the pressure tensor as ∂Sij k ∂Pij ∂uj ∂Pij ∂uk ∂ui (2) + uk + + Pij + Pik + Pj k = J¯ij ∂t ∂xk ∂xk ∂xk ∂xk ∂xk or,

∂Pij ∂Sij k ∂uj ∂ui ∂ (2) (uk Pij ) + + Pik + Pj k = J¯ij + ∂t ∂xk ∂xk ∂xk ∂xk

(6.21)

Decomposing the pressure tensor Pij into isotropic pressure p and stress tensor pij using the relation pij = Pij − pδij , the above equation can be written as ∂Sij k ∂pij ∂p ∂ ∂ + δij + (uk pij ) + δij (uk p) + ∂t ∂t ∂xk ∂xk ∂xk + pik

∂uj ∂uj ∂ui ∂ui (2) + δik p + pj k + δj k p = J¯ij ∂xk ∂xk ∂xk ∂xk

(6.22)

For the underlined terms, we use Eq. (5.51)) (energy equation in terms of pressure) where dummy indices are changed accordingly to avoid conflict with the free indices, ∂p ∂ ∂ui 2 1 ∂Sk 2 ∂ui + − pij − (puk ) = − p ∂t ∂xk 3 ∂xi 3 ∂xj 3 ∂xk

(6.23)

Equation (6.22) for stress tensor then becomes ∂Sij k ∂pij ∂ ∂uk 2 2 ∂uk 1 ∂Sk + (uk pij ) + − δij pkl − δij p − δij ∂t ∂xk ∂xk 3 ∂xl 3 ∂xk 3 ∂xk + pik

∂uj ∂uj ∂ui ∂ui (2) +p + pj k +p = J¯ij ∂xk ∂xi ∂xk ∂xj

(6.24)

198

6 Grad Equations: Derivation and Analysis

Rearranging terms, we finally obtain the evolution equation for the stress tensor as ∂Sij k ∂uj ∂pij ∂ui 1 ∂Sk ∂ (uk pij ) + − δij + pik + pj k + ∂t ∂xk ∂xk 3 ∂xk ∂xk ∂xk ∂uj ∂uk ∂ui 2 2 ∂uk (2) = J¯ij . − δij pkl +p + − δij 3 ∂xl ∂xj ∂xi 3 ∂xk

(6.25)

Evolution Equation for Sij k

6.3.1.2

Following a similar procedure as given above, we obtain the evolution equation for the third order moment Sij k . For this, we multiply the transformed Boltzmann equation (Eq. (6.10)) by mCi Cj Ck and integrate over the velocity space to obtain m

Df ∂f dC dC + m Ci Cj Ck Cr Dt ∂xr ∂f Dur Fr − +m Ci Cj Ck dC Dt ∂Cr ∂f ∂ul dC = m Ci Cj Ck J (f, f1 )dC − m Ci C j C k C r ∂Cl ∂xr Ci Cj Ck

(6.26)

Using the same arguments, integration by parts and Leibnitz’s rule, we evaluate term by term as (a) = m

Ci C j C k

=m

Df dC Dt

Df Ci Cj Ck dC − m Dt

f

DCi Cj Ck dC Dt

DSij k −0 Dt ∂f (b) = m Ci Cj Ck Cr dC ∂xr ∂(Ci Cj Ck Cr ) ∂(f Ci Cj Ck Cr ) dC − m f dC =m ∂xr ∂xr =

∂Qij kr −0 ∂xr ∂f Dur Fr − (c) = m Ci Cj Ck dC Dt ∂Cr ∂f Ci Cj Ck ∂Ci Cj Ck Dur m dC − m f dC = Fr − Dt ∂Cr ∂Cr =

6.3 20-Moment Approximation

199

∂Ci Cj Ck Dur m f = 0 − Fr − dC Dt ∂Cr Dur m f Ci Cj δkr + Ci Ck δj r + Cj Ck δir dC = − Fr − Dt Using the momentum conservation equation (Eq. (5.40)), Fr −

∴

Dur Dt

=

1 ∂Prl ρ ∂xl

1 ∂Prl Pij δkr + Pik δj r + Pj k δir ρ ∂xl ∂Pj l 1 ∂Pkl ∂Pil =− Pij + Pik + Pj k ρ ∂xl ∂xl ∂xl ∂f ∂ul (d) = −m Ci Cj Ck Cr dC ∂Cl ∂xr ∂Ci Cj Ck Cr ∂ul ∂ul ∂ f C i Cj Ck Cr dC + = −m m f dC ∂Cl ∂xr ∂xr ∂Cl ∂ul m f Ci Cj Ck δrl + Ci Cr Ck δj l + Ci Cj Cr δkl + Cj Ck Cr δil dC = 0+ ∂xr (c) = −

= Sij k

∂uj ∂ur ∂uk ∂ui + Sirk + Sij r + Sj kr ∂xr ∂xr ∂xr ∂xr

(6.27)

Denoting the production term as (e) = m

(3) Ci Cj Ck J (f, f1 )dC = J¯ij k .

(6.28)

Combining and rearranging the terms, the evolution equation for third order moment Sij k is obtained as ∂uj ∂Sij k ∂ ∂uk ∂ui + (ur Sij k + Qij kr ) + Sij r + Sirk + Sj kr ∂t ∂xr ∂xr ∂xr ∂xr ∂Pj r ∂Pkr ∂Pir 1 = J¯ij(3)k , Pij − + Pik + Pj k ρ ∂xr ∂xr ∂xr

(6.29)

The evolution equations for pressure tensor Pij (6.21), stress tensor pij (6.25), and third order moment Sij k (6.29) are exact. However, the system of equations is not closed since fourth order moment Qij kr appears in the divergence term of the evolution equation for Sij k (6.29).

200

6 Grad Equations: Derivation and Analysis

6.3.2 Expansion in Hermite Polynomials A brief introduction to the Hermite polynomials is given in Appendix C.2. Readers are advised to go through the Appendix before proceeding to this section. Grad proposed to expand the function ω−1/2 g in orthogonal tensorial Hermite functions as (the superscript n represents an nth order tensor) ∞

g(x, v, t) = ω(v) n=0

1 (n) (n) a (x, t)Hi (v), n! i

(6.30)

where we have introduced the dimensionless microscopic velocity v and the dimensionless distribution function g(x, v, t) as C , v= √ RT

f (x, C, t) g(x, v, t) = n (RT )3/2

(6.31)

and the weight function in three-dimensional space (N = 3 in definition of weight function, Eq. (C.9)) is ω(v) =

2 −v 1 exp 3/2 2 (2π )

(6.32)

The purpose of introducing the term ω−1/2 in the function ω−1/2 g is to get the Maxwellian distribution function around which the expansion is sought. Also, notice carefully the functional dependence of each of the variable, especially the Hermite (n) coefficients ai , which depend on the space coordinates and time and not on the microscopic velocity. It will become clear in the ensuing sections that the first few significant Hermite coefficients are actually the macroscopic variables. In expanded notation, the dimensionless distribution function is represented as 1 (2) (2) 1 (3) (3) (1) (1) (0) (0) g=ω a H + ai Hi + aij Hij + aij k Hij k + · · · 2! 3!

(6.33)

The Maxwellian distribution f (0) (using R = kB /m in Eq. (5.90) ) can be written in terms of the weight function ω as 1 −C 2 n exp 2RT (RT )3/2 (2π )3/2 2 1 n −v = exp 3/2 3/2 2 (RT ) (2π )

f (0) =

[Using Eq. (6.31)]

6.3 20-Moment Approximation

∴ f (0) =

201

n ω (RT )3/2

[Using Eq. (6.32)]

(6.34)

Consequently, the expansion for distribution function f can be written in the form f = =

n g (RT )3/2 n ω(v) (RT )3/2 ∞

= f (0) n=0

∞ n=0

1 (n) (n) a (x, t)Hi (v) n! i

1 (n) (n) a Hi n! i

1 (2) (2) 1 (3) (3) (1) (1) = f (0) a (0) H (0) + ai Hi + aij Hij + aij k Hij k + · · · 2! 3!

(6.35)

where the Hermite coefficients ai(n) are given by (n) ai

=

gHi

=

1 n

=

1 n

(n)

dv

f Hi

(n)

dc

[In dimensional form]

f Hi

(n)

dC

[Using dc = dC]

(6.36)

The first few Hermite polynomials [61] are obtained in Appendix C.2 as H (0) = 1, Hi

(1)

= vi Ci , =√ RT

(2)

Hij

(6.37)

(6.38)

= vi vj − δij =

Ci Cj − δij , RT

(6.39)

(3)

Hij k = vi vj vk − (vi δj k + vj δik + vk δij ) =

Ci C j C k 1 (Ci δj k + Cj δik + Ck δij ), −√ 3/2 (RT ) RT

(6.40)

202

6 Grad Equations: Derivation and Analysis (4)

Hij kl = vi vj vk vl − (vi vj δkl + vi vk δj l + vi vk δj l + vj vk δil + vj vl δik + vk vl δij ) + (δij δkl + δik δj l + δil δj k ) =

Ci C j C k C l (RT )2 −

1 (Ci Cj δkl + Ci Ck δj l + Ci Ck δj l + Cj Ck δil + Cj Cl δik + Ck Cl δij ) RT

+ (δij δkl + δik δj l + δil δj k )

(6.41)

Using the dimensional form of expression for Hermite coefficients (Eq. (6.36)), and substituting the expressions for Hermite polynomials (Eqs. (6.37)–(6.41)), the first few Hermite coefficients in terms of moments are obtained as 1 (0) a (0) = f Hi dC n 1 = f dC ρ = 1, (1)

ai

(6.42)

=

1 n

=

1 √ ρ RT

f Hi

(1)

dC

f Ci dC

=0

(2)

aij = =

1 n

m ρRT

f Hi

(2)

dC

f Ci Cj dC − δij

Pij − δij p pij , = p

(6.43)

1 n

f dC

[∵ ρ = nm]

=

(3)

aij k = =

1 n(RT )3/2 Sij k , √ p RT

(6.44) Ci Cj Ck f dC − 0 (6.45)

6.3 20-Moment Approximation (4)

aij kl =

203

Qij kl 1 − (Pij δkl + Pik δj l + Pil δj k + Pj k δil + Pj l δik + Pkl δij ) pRT p + (δij δkl + δik δj l + δil δj k )

=

Qij kl 1 − (pij δkl + pik δj l + pil δj k + pj k δil + pj l δik + pkl δij ) pRT p − (δij δkl + δik δj l + δil δj k )

(6.46)

where the usual definitions of mass density (Eq. (5.10)), pressure tensor (Eq. (5.5)), third order moment Sij k (Eq. (6.4)), fourth order moment Qij kl (Eq. (6.5)), ideal gas equation (p = ρRT = nkB T ) are used along with Eqs. (5.15) and (5.22). The second order Hermite coefficient aij(2) (Eq. (6.44)) actually turns out to be a trace-free part of the pressure tensor, i.e., stress tensor pij . Similarly, when we (3) contract third order Hermite coefficient aij k (Eq. (6.45)), we obtain an expression in terms of heat flux vector as 1 m 2qi (3) 2 (6.47) aikk = C f dC = C Ci C 2 f dC = √ i n(RT )3/2 ρ(RT )3/2 p RT

6.3.3 Derivation of Distribution Function In this section, we derive the full third order approximation of the distribution function fG|20 . We will utilize this form to evaluate the unknown higher order moment Qij kl appearing in the evolution equation of third order moment Sij k . The basic idea of the third order approximation is to truncate the series (Eq. (6.35)) after (n) the first four terms by assuming the higher order Hermite coefficients ai to be zero (n) for n > 3. Substituting the expressions for Hermite coefficients ai (Eqs. (6.42)– (n) (6.45)) and Hermite polynomials Hi (Eqs. (6.37)–(6.40)), we obtain

1 (2) (2) 1 (3) (3) (1) (1) fG|20 = f (0) a (0) H (0) + ai Hi + aij Hij + aij k Hij k 2! 3!

1 (2) (2) 1 (3) (3) (1) [∵ a (0) = H (0) = 1, ai = 0] = f (0) 1 + aij Hij + aij k Hij k 2 6

Ci Cj Ck 1 pij Ci Cj 1 Sij k (0) 1+ − δij + =f √ 2 p RT 6 p RT (RT )3/2 1 (Ci δj k + Cj δik + Ck δij ) −√ RT

204

6 Grad Equations: Derivation and Analysis

=f

(0)

Sij k 1 pij 1 pii Ci Cj − + Ci Cj Ck 2 pRT 2 p 6p(RT )2 (Ci Sijj + Cj Sij i + Ck Siik )

1+

1 6pRT

= f (0) 1 + −

pij Sij k Si Ci C j + Ci Ci Cj Ck − 2 2pRT 2pRT 6p(RT )

[∵ pii = 0] (6.48)

Hence, the final form of the Grad 20-moment distribution function fG|20 in terms of first 20 moments is obtained as

pij Sij k Si (0) 1+ (6.49) Ci C j + Ci fG|20 = f Ci Cj Ck − 2pRT 2pRT 6pR 2 T 2

6.3.4 Closure To obtain closure and reduce them to third order, Qij kr must be replaced by its third order equivalent, i.e., the unknown Qij kr must be expressed in terms of primary (n) variables. For full third order approximation, the Hermite coefficients ai must be (4) zero for n > 3. Recalling the equation for coefficient aij kr (Eq. (6.46)) which we derived previously, (4)

aij kr =

Qij kr 1 − (pij δkr + pik δj r + pir δj k + pj k δir + pj r δik + pkr δij ) pRT p − (δij δkr + δik δj r + δir δj k )

(6.50)

(4)

and substituting aij kr = 0, we obtain the constitutive relationship for moment Qij kr and achieve the necessary closure as Qij kr = RT (pij δkr + pik δj r + pir δj k + pj k δir + pj r δik + pkr δij ) + pRT (δij δkr + δik δj r + δir δj k )

(6.51)

An important remark needs to be made here. We could have utilized the distribution function fG|20 to evaluate the fourth order moment Qij kr as Qij kr = m

Ci Cj Ck Cr fG|20 dC

(6.52)

which is however not done here. Rather, the knowledge of the higher order Hermite coefficients helped us to derive the necessary constitutive relationship for Qij kr .

6.3 20-Moment Approximation

205

6.3.5 20-Moment Equations To obtain a closed set of equations, we substitute the constitutive relationship for Qij kr (Eq. (6.51)) in Eq. (6.29). Recalling the evolution equation for Sij k , ∂ur Sij k ∂Qij kr ∂uk ∂uj ∂Sij k ∂ui + + + Sij r + Sirk + Srj k ∂t ∂xr ∂xr ∂xr ∂xr ∂xr ) *+ , (i)

∂Pj r ∂Pkr ∂Pir 1 (3) Pij − + Pik + Pj k = J¯ij k , ρ ∂xr ∂xr ∂xr ) *+ ,

(6.53)

(ii)

Evaluating terms (i) and (ii) separately as (i) =

∂Qij kr ∂(RT pij ) ∂(RT pik ) ∂(RT pir ) ∂(RT pj k ) = + + δj k + ∂xr ∂xk ∂xj ∂xr ∂xi + δik

=

∂(RT pij ) ∂(RT pik ) ∂(RT pj k ) ∂(RT pkr ) ∂(RT pir ) + + + δij + δj k ∂xk ∂xj ∂xi ∂xr ∂xr + δik

=

∂(RT pj r ) ∂(RT pkr ) ∂ pRT δij δkr + δik δj r + δir δj k + δij + ∂xr ∂xr ∂xr

∂(RT pj r ) ∂(pRT ) ∂(pRT ) ∂(pRT ) + δij + δik + δj k ∂xr ∂xk ∂xj ∂xi

∂(RT pij ) ∂(RT pik ) ∂(RT pj k ) ∂RT + + + [pkr δij + pir δj k + pj r δik ] ∂xk ∂xj ∂xi ∂xr ∂pj r ∂pkr ∂pir +RT δij + δj k + δik ∂xr ∂xr ∂xr ) *+ , +p

(i−a)

∂p ∂p ∂p ∂RT ∂RT ∂RT δij + δik + δj k +RT δij + δik + δj k ∂xk ∂xj ∂xi ∂xk ∂xj ∂xi ) *+ , (i−b)

(6.54) ∂Pj r ∂Pkr ∂Pir 1 Pij (ii) = − + Pik + Pj k ρ ∂xr ∂xr ∂xr ∂Pj r 1 ∂Pkr ∂Pir = − (pij + pδij ) + (pik + pδik ) + (pj k + pδj k ) ρ ∂xr ∂xr ∂xr

206

6 Grad Equations: Derivation and Analysis

∂Pj r ∂Pkr ∂Pir 1 pij =− + pik + pj k ρ ∂xr ∂xr ∂xr ∂Pj r 1 ∂Pkr ∂Pir − pδij + pδik + pδj k ρ ∂xr ∂xr ∂xr ∂(pj r + pδj r ) 1 ∂(pkr + pδkr ) ∂(pir + pδir ) =− pij + pik + pj k ρ ∂xr ∂xr ∂xr ∂(pj r + pδj r ) 1 ∂(pkr + pδkr ) ∂(pir + pδir ) − pδij + pδik + pδj k ρ ∂xr ∂xr ∂xr ∂pj r 1 1 ∂pkr ∂pir ∂p ∂p ∂p − =− pij pij + pik + pj k + pik + pj k ρ ∂xr ∂xr ∂xr ρ ∂xk ∂xj ∂xi ∂pj r ∂pkr ∂pir ∂p ∂p ∂p + δik + δj k −RT δij + δik + δj k −RT δij ∂xr ∂xr ∂xr ∂xk ∂xj ∂xi ) *+ ,) *+ , (ii−a)

(ii−b)

(6.55) Substituting the expressions for (i) and (ii) in Eq. (6.53) and noting that the terms (i − a) and (i − b) of Eq. (6.54) cancel with the terms (ii − a) and (ii − b) of Eq. (6.55), respectively, and rearranging the terms, we obtain an equation for Sij k as ∂uj ∂Sij k ∂ ∂uk ∂ui + (ur Sij k ) + Sij r + Sirk + Srj k ∂t ∂xr ∂xr ∂xr ∂xr ∂RT pj k ∂RT pij ∂RT ∂RT pik + + + + [pkr δij + pir δj k + pj r δik ] ∂xk ∂xj ∂xi ∂xr ∂pj r 1 ∂pkr ∂pir ∂RT ∂RT ∂RT pij δij + δik + δj k − + pik + pj k +p ∂xk ∂xj ∂xi ρ ∂xr ∂xr ∂xr ∂p ∂p ∂p 1 (3) = J¯ij k pij + pik + pj k (6.56) − ρ ∂xk ∂xj ∂xi (2) (3) Notice that without explicit expressions for the production terms J¯ij , J¯ij k , the equation is still incomplete. The evaluation of the production terms is complicated. However, in the case of Maxwell molecules, even without knowing the full functional form of the distribution form, the exact values can be computed [149] and they read as p J¯ij(2) = − pij μ 3p 1 (3) ¯ Sij k − (Si δj k + Sj δik + Sk δij ) Jij k = − 2μ 5

(6.57) (6.58)

6.4 The 13-Moment Approximation

207

We are now ready with the full 20-moment equations in closed form ∂ ∂ρ + (ρui ) = 0, ∂t ∂xi

(6.59)

∂ui ∂ui 1 ∂p 1 ∂pik + uk + + = Fi , ∂t ∂xk ρ ∂xi ρ ∂xk

(6.60)

3 ∂T 3 ∂T ∂ui ∂ui ∂qk Rρ + Rρuk +p + pij + = 0, 2 ∂t 2 ∂xk ∂xi ∂xj ∂xk ∂pij ∂Sij r ∂uj ∂ui 2 ∂qk ∂ (ur pij ) + − δij + pir + pj r + ∂t ∂xr ∂xr 3 ∂xk ∂xr ∂xr ∂uj p ∂uk ∂ui 2 2 ∂uk = − pij , − δij pkl +p + − δij 3 ∂xl ∂xj ∂xi 3 ∂xk μ

(6.61)

(6.62)

∂Sij k ∂uj ∂ ∂uk ∂ui + (ur Sij k ) + Sij r + Sirk + Srj k ∂t ∂xr ∂xr ∂xr ∂xr ∂RT pj k ∂RT pij ∂RT ∂RT pik + + + + [pkr δij + pir δj k + pj r δik ] ∂xk ∂xj ∂xi ∂xr ∂pj r 1 ∂pkr ∂pir ∂RT ∂RT ∂RT pij δij + δik + δj k − + pik + pj k +p ∂xk ∂xj ∂xi ρ ∂xr ∂xr ∂xr 3p ∂p ∂p ∂p 1 1 =− pij Sij k − (Si δj k + Sj δik + Sk δij ) + pik + pj k − ρ ∂xk ∂xj ∂xi 2μ 5 (6.63) This set of 20 scalar equations determines the evolution of 20 third-order variables, viz., ρ, u, T , pij , and Sij k (1 + 3 + 1 + 5 + 10 = 20) for Maxwellian molecules. The limitation of Maxwellian molecules arises because the production terms are known only for these molecules. This completes the derivation of Grad 20-moment equations.

6.4 The 13-Moment Approximation In this section, we present the derivation of Grad 13-moment equations. The variables defining the state of the gas are ρ, u, T , pij , and q which are thirteen (1 + 3 + 1 + 5 + 3 = 13) in number. This set of Grad 13-moment equations is

208

6 Grad Equations: Derivation and Analysis

a more preferred choice over 20-moment equations since physical significance can be attributed to each of these variables. Remember, in 20-moment approximation, second order moment turned out to be the stress tensor, but, third order moment Sij k does not seem to have any physical significance. However, the contracted third order moment Si actually turns out to be twice the heat flux vector (see Eq. (6.6)) as Sikk = m Ci C 2 f dC = 2qi (6.64) Hence, it is more appropriate to consider moments having physical intuitive meaning rather than the full third order approximation. Accordingly, we include stress tensor pij and heat flux vector qi in the primary variables set in addition to density ρ, bulk velocity ui , and temperature T . The expressions for the primary variables and unknowns appearing in the 13-moment approximation are shown in Fig. 6.3. The present section is outlined as follows: The evolution equations for stress tensor (pij ) and heat flux vector (qi ) will be generated. The higher order moments appearing in these equations will be made divergence-free, a standard currently followed in the literature. The distribution function fG|13 is then derived and utilized to evaluate the unknowns appearing in the evolution equations for pij and qi to obtain a closed set of equations.

1 n

u (x,t) =

i

ci f (x, c,t)dc

r (x,t) =

m

T (x,t) = 1 m2 C2 f (x, c,t)dc 3 r kB

f (x, c,t)dc

pi j (x,t) =

Grad 13 moment approximation

m CiC j f (x, c,t)dc

1 rik (x,t) =

m

m 2

2

qi (x,t) =

CiC2 f (x, c,t)dc

w2 (x,t) =

m C ( fG|13 − fM )dC 4

CiCkC fG|13 dC si jk (x,t) =

m

CiC jCk fG|13 dC

Fig. 6.3 Expression for the variables considered in the Grad 13-moment approximation; variables in solid circles are the primary variables while those in dashed circles are the closure variables evaluated using Grad 13-moment distribution function fG|13

6.4 The 13-Moment Approximation

209

6.4.1 Evolution Equations for Stress Tensor and Heat Flux Vector The evolution equation for stress tensor pij was derived in Sect. 6.3.1.1 as ∂Sij k ∂uj ∂pij ∂ui 1 ∂Sk ∂ (ur pij ) + − δij + pik + pj k + ∂t ∂xr ∂xk 3 ∂xk ∂xk ∂xk ∂uj ∂uk ∂ui 2 2 ∂uk = J¯ij(2) − δij pkl +p + − δij 3 ∂xl ∂xj ∂xi 3 ∂xk

(6.65)

However, the third order moment Sij k although symmetric is not divergence-free. To make Sij k divergence-free, we use the following mathematical identity for the decomposition of any third order tensor Wij k , wij k = W(ij k) −

1 W(ill) δj k + W(j ll) δik + W(kll) δij 5

(6.66)

where the angular brackets denote the symmetric and trace-free part of the tensor and round brackets denote the symmetric part of the tensor which is obtained as W(ij k) =

1 Wij k + Wkij + Wj ki + Wikj + Wj ik + Wkj i 6

(6.67)

Since Sij k is already symmetric, we get rid of the round brackets and its trace-free part is then given as sij k = Sij k −

1 Sill δj k + Sj ll δik + Skll δij 5

(6.68)

With this decomposition, the underlined terms in Eq. (6.65) are evaluated as 1 ∂Sk ∂sij k ∂Sij k 1 ∂Sk 1 ∂ Sill δj k + Sj ll δik + Skll δij − δij − δij = + ∂xk 3 ∂xk ∂xk 5 ∂xk 3 ∂xk =

∂sij k 1 ∂Si 1 ∂Sj 1 ∂Sk 1 ∂Sk + + + δij − δij ∂xk 5 ∂xj 5 ∂xi 5 ∂xk 3 ∂xk

∂sij k 2 ∂qi 2 ∂qj 2 ∂qk 2 ∂qk + + + δij − δij [∵ Si = 2qi ] ∂xk 5 ∂xj 5 ∂xi 5 ∂xk 3 ∂xk ∂sij k ∂qj 1 ∂qk 4 1 ∂qi − δij = + + ∂xk 5 2 ∂xj ∂xi 3 ∂xk

=

=

∂sij k 4 ∂qi + ∂xk 5 ∂xj

(6.69)

210

6 Grad Equations: Derivation and Analysis

(2) p Introducing the Lagrangian derivative and using J¯ij = − μ pij for Maxwellian molecule (from Eq. (6.58)), the final evolution equation for the stress tensor is obtained as

∂sij k Dpij ∂uj ∂ui ∂uk 4 ∂qi p + pij + + + 2pri + 2p = − pij Dt ∂xk ∂xk 5 ∂xj ∂xr ∂xj μ

(6.70)

To obtain the evolution equation for the heat flux vector qi , we multiply the transformed Boltzmann equation (Eq. (6.10)) by m2 Ci C 2 and integrate over the velocity space as (The argument for certain integrals being zero is same as given in Sect. 6.3.1.1 and therefore these arguments are not repeated here.) m m Df ∂f ∂f m Duk dC + Ci C 2 Fk − dC + dC Ci C 2 Ci C 2 Ck 2 Dt 2 ∂xk 2 Dt ∂Ck m ∂f ∂uj m − dC = (6.71) Ci C 2 Ck Ci C 2 J (f, f1 )dC 2 ∂Cj ∂xk 2 Applying integration by parts and Leibnitz’s rule to take out the differentiation outside the integral, we evaluate term by term as m Df dC (a) = Ci C 2 2 Dt m m Df Ci C 2 DCi C 2 = dC − dC f 2 Dt 2 Dt Dqi −0 (6.72) Dt ∂f m (b) = dC Ci C 2 Ck 2 ∂xk m m ∂f Ci Ck C 2 ∂Ci C 2 Ck dC − dC = f 2 ∂xk 2 ∂xk 1 1 ∂Rik 1 −0 [Say Rik = m f Ci Ck C 2 dC] (6.73) = 2 ∂xk ∂f Duk m Ci C 2 Fk − dC (c) = 2 Dt ∂Ck m Duk m ∂f Ci C 2 ∂Ci C 2 dC − dC = Fk − f Dt 2 ∂Ck 2 ∂Ck Duk m ∂Ci Cl Cl dC [Since C 2 = Cl Cl ] = 0 − Fk − f Dt 2 ∂Ck Duk m = − Fk − f [Ci Cl δlk + Ci Cl δlk + Cl Cl δik ] dC Dt 2 =

6.4 The 13-Moment Approximation

211

Using the following form of momentum conservation equation (Eq. (5.40)), 1 ∂Pkl Duk 1 ∂p 1 ∂pkl Fk − = = + Dt ρ ∂xl ρ ∂xk ρ ∂xl

[∵ pkl = Pkl − pδkl ]

1 ∂p 1 ∂pkl m (c) = − + f 2Ci Ck + C 2 δik dC ρ ∂xk ρ ∂xl 2 m 1 ∂p 1 ∂pkl m 2 =− + f 2Ci Ck dC + C δik dC ρ ∂xk ρ ∂xl 2 2 m 1 ∂pkl 3 1 ∂p Pik + pδik ∵p= + =− C 2 f dc ρ ∂xk ρ ∂xl 2 3 1 ∂p 1 ∂pkl 3 + =− pik + pδik + pδik ρ ∂xk ρ ∂xl 2 1 ∂p 1 ∂pkl 5 pik + pδik =− + ρ ∂xk ρ ∂xl 2

∴

5 p ∂p pik ∂pkl 5 p ∂pil pik ∂p − − − ρ ∂xk 2 ρ ∂xi ρ ∂xl 2 ρ ∂xl ∂u ∂f ∂uj m m ∂ j 2 f C i C 2 Ck (d) = − dC = − dC Ci C Ck 2 ∂Cj ∂xk 2 ∂Cj ∂xk ∂uj m ∂uj m ∂Ci Ck Cr Cr dC = 0 + + f ∂xk 2 ∂Cj ∂xk 2 × f Ci Ck Cr δj r + Ci Ck Cr δj r + Ci Cr Cr δj k + Ck Cr Cr δij dC =−

=

∂uj m ∂xk 2

f 2Ci Ck Cj + Ci C 2 δj k + Ck C 2 δij dC

∂uj m δj k 2f Ci Ck Cj dC + f Ci C 2 dC ∂xk 2 ∂uj m ∂uj ∂uk ∂ui + δij Sij k + qi + qk f Ck C 2 dC = ∂xk 2 ∂xk ∂xk ∂xk

∂uj m = ∂xk 2

(6.74)

To make Sij k divergence-free, we use decomposition (Eq. (6.68)) for the first term ∂uj ∂xk Sij k and proceed as

212

6 Grad Equations: Derivation and Analysis

∂uj ∂uj 1 sij k + (Si δj k + Sj δik + Sk δij ) Sij k = ∂xk ∂xk 5 ∂uj 2 sij k + (qi δj k + qj δik + qk δij ) = ∂xk 5 =

[∵ Si = 2qi ]

∂uj 2 ∂uk 2 ∂uj 2 ∂ui sij k + qi + qj + qk ∂xk 5 ∂xk 5 ∂xi 5 ∂xk

Substituting the above expression in the (d) term, we have (d) =

∂uj 7 ∂uk 7 ∂ui 2 ∂uj sij k + qi + qk + qj ∂xk 5 ∂xk 5 ∂xk 5 ∂xi

(6.75)

The production term for the Maxwellian molecules can be computed without knowing the actual form of the distribution function as m (e) = 2

Ci C 2 J (f, f1 )dC = −

2p qi 3μ

(6.76)

Combining all the terms, the equation for heat flux vector qi is obtained as 1 Dqi 1 ∂Rik pik ∂p 5 p ∂p pik ∂pkl 5 p ∂pil + − − − − Dt 2 ∂xk ρ ∂xk 2 ρ ∂xi ρ ∂xl 2 ρ ∂xl ) *+ , ) *+ , (i)

+

(ii)

∂uj 7 ∂uk 7 ∂ui 2 ∂uj 2p qi , sij k + qi + qk + qj =− ∂xk 5 ∂xk 5 ∂xk 5 ∂xi 3μ

(6.77)

1 appearing in term (i) is, however, still not divergenceThe higher order moment Rik 1 free. To make Rik divergence-free, we decompose it as

(i) =

1 1 ∂Rik 2 ∂xk

1 ∂ = m 2 ∂xk 1 ∂ = m 2 ∂xk

f Ci Ck C 2 dC f Ci Ck C 2 dC

1 4 C2 δik f C δik dC ∵ Ci Ck = Ci Ck + 3 3 1 1 ∂rik 1 ∂ 4 2 1 = + m f C dC Say, m f Ci Ck C dC = rik 2 ∂xk 6 ∂xi (6.78) 1 ∂ m + 2 ∂xk

6.4 The 13-Moment Approximation

213

For term (ii) of Eq. (6.77), we use ideal gas equation p = ρRT to replace the pressure term as (ii) = − =−

pik ∂p 5 p ∂p − ρ ∂xk 2 ρ ∂xi pik ∂ρRT 5 ∂ρRT − RT ρ ∂xk 2 ∂xi

= −pik R

∂T pik 5 ∂ρ(RT )2 5 ∂ρ ∂RT − − + ρRT RT ∂xk ρ ∂xk 2 ∂xi 2 ∂xi

(6.79)

The third term on the right-hand side of the above equation turns out to be 5 ∂ρ(RT )2 1 ∂ = m 2 ∂xi 6 ∂xi

fM C 4 dC

where fM is a Maxwellian distribution function. ∴ (ii) = −

5 p ∂p pik ∂p − ρ ∂xk 2 ρ ∂xi

∂T pik 1 ∂ ∂ρ = −pik R − − m RT ∂xk ρ ∂xk 6 ∂xi

5 ∂RT fM C 4 dC + ρRT 2 ∂xi (6.80)

Combining the results for terms (i) (Eq. (6.78)) and (ii) (Eq. (6.80)), we proceed (i) + (ii) = =

1 1 ∂Rik pik ∂p 5 p ∂p − − 2 ∂xk ρ ∂xk 2 ρ ∂xi 1 ∂ρ ∂RT 1 ∂rik 1 ∂w 2 ∂T pik 5 RT + − pik R − + ρRT 2 ∂xk 6 ∂xi ∂xk ρ ∂xk 2 ∂xi

(6.81) where we have introduced w2 = m C 4 (f − fM )dC. Substituting results for terms (i)+(ii) (Eq. (6.81)) into Eq. (6.77), we finally have the evolution equation for the heat flux vector along with stress tensor equation (for the sake of completeness) as ∂sij k ∂uj Dpij ∂ui ∂uk 4 ∂qi p + pij + + + 2pri + 2p = − pij Dt ∂xk ∂xk 5 ∂xj ∂xr ∂xj μ

(6.82)

214

6 Grad Equations: Derivation and Analysis

1 Dqi 1 ∂w 2 ∂T pik 5 pik ∂pkl 1 ∂rik ∂ρ ∂RT + − pik R − + ρRT − + RT Dt 2 ∂xk 6 ∂xi ∂xk ρ ∂xk 2 ∂xi ρ ∂xl

∂uj ∂pik 5 7 ∂uk 7 ∂ui 2 ∂uj 2p qi , − RT + sij k + qi + qk + qj =− 2 ∂xk ∂xk 5 ∂xk 5 ∂xk 5 ∂xi 3μ (6.83) This set of coupled equations does not form a closed set due to the unknowns 1 =m rik

Ci Ck C 2 f dC

(6.84)

C 4 (f − fM )dC

(6.85)

Ci Cj Ck C 2 f dC

(6.86)

w2 = m sij k = m

To evaluate these unknowns and achieve closure, we should know the distribution function which we derive in the next section.

6.4.2 Derivation of Distribution Function To derive the distribution function fG|13 for 13-moment approximation, we start with the distribution function for the Grad’s 20-moment approximation fG|20 (Eq. (6.49)) which reads as 2 fG|20 = f

(0)

Sij k pij Si Ci C j + Ci Ci Cj Ck − 1+ 2 2 2pRT 2pRT 6pR T

3 (6.87)

The third order symmetric tensor Sij k is decomposed into divergence-less part as 1 sij k = Sij k − (Sill δj k + Sj ll δik + Skll δij ) 5

(6.88)

Now, in order to obtain the closure, the number of coefficients considered in the series (Eq. (6.35)) must be equal to the primary basic variables considered. Hence, in order to reduce the 20-moment distribution function fG|20 to 13-moment distribution function fG|13 , it becomes mandatory to take the divergence-less part of Sij k (i.e., sij k ) as zero. Hence, with sij k = 0, we proceed as Sij k =

1 (Sill δj k + Sj ll δik + Skll δij ) 5

(6.89)

6.4 The 13-Moment Approximation

215

Substituting Eq. (6.89) into the underlined term of full third order distribution function (Eq. (6.87)), Sij k 1 (Sill δj k + Sj ll δik + Skll δij ) Ci Cj Ck = Ci Cj Ck 2 2 30 6pR T pR 2 T 2 =

1 Sill Ci C 2 + Sj ll Cj C 2 + Skll Ck C 2 30 pR 2 T 2

=

1 Si Ci C 2 10 pR 2 T 2

=

1 qi Ci C 2 5 pR 2 T 2

[∵ Si = 2qi ]

(6.90)

The final form of the distribution function fG|13 in terms of first 13 moments is then obtained as

pij C2 qi Ci Cj − Ci 1 − fG|13 = f (0) 1 + 2pRT pRT 5RT [∵ pij Ci Cj = pij Ci Cj ]

(6.91)

According to Grad [62], the stresses do not alter the shape of distribution function fG|13 radically; the spherical symmetry of f (0) is distorted into an ellipsoid and the heat flux vector introduces skewness in the distribution function fG|13 . Grad also believed that the expansion (Eq. (6.91)) describes the majority of flow problems in which the macroscopic variables do not vary by a large amount over a mean free path with sufficient accuracy. He believed that the expansion (6.91) is sufficient to study shock waves of medium strength, which unfortunately proved to be false in his later studies [63].

6.4.3 Closure With the knowledge of the distribution function fG|13 (Eq. (6.91)), we can now proceed to evaluate the unknowns as 1 rik

=m

fG|13 Ci Ck C 2 dC

(6.92)

C 4 (fG|13 − fM )dC

(6.93)

fG|13 Ci Cj Ck dC

(6.94)

w2 = m sij k = m

216

6 Grad Equations: Derivation and Analysis

Notice that replacement of f by fG|13 makes the above equations approximate. As mentioned previously, it becomes necessary to take sij k = 0, hence, there is no further need to evaluate the third unknown. The procedure to evaluate the remaining two integrals is same as outlined in the previous chapter; hence, we do not show a step-by-step evaluation of these integrals. The final result can be obtained as 1 rik = 7RT pik ,

w 2 = 0,

sij k = 0.

(6.95)

6.4.4 13-Moment Equations Substituting the expressions for the unknowns (Eq. 6.95) in the evolution equations for the stress tensor and heat flux vector (Eqs. (6.82) and (6.83)), closure is finally achieved. Below we present Grad’s 13-moment equations along with the conservation laws: ∂(ρuk ) ∂ρ + =0 ∂t ∂xk

(6.96)

∂ui ∂ui 1 ∂p 1 ∂pik + uk + + = Fi , ∂t ∂xk ρ ∂xi ρ ∂xk

(6.97)

3 ∂T 3 ∂T ∂ui ∂ui ∂qk Rρ + Rρuk +p + pij + = 0, 2 ∂t 2 ∂xk ∂xi ∂xj ∂xk

(6.98)

Dpij ∂uj ∂ui ∂uk 4 ∂qi p + pij + + 2pri + 2p = − pij Dt ∂xk 5 ∂xj ∂xr ∂xj μ

(6.99)

Dqi 5 ∂ρ ∂RT ∂T pik 5 pik ∂pkl ∂pik + pik R RT − + ρRT − + RT Dt 2 ∂xk ρ ∂xk 2 ∂xi ρ ∂xl ∂xk 7 ∂uk 7 ∂ui 2 ∂uj 2p qi . + qi + qk + qj =− 5 ∂xk 5 ∂xk 5 ∂xi 3μ

(6.100)

The detailed flowchart of the Grad’s moment method is shown in Fig. 6.4 for convenience. The distribution function is expanded in terms of orthogonal tensorial Hermite polynomials. Depending on the desired accuracy, the series is truncated at an appropriate order of approximation. The number of Hermite coefficients considered must be equal to the number of primary variables considered. The evolution equations for the primary variables are generated from the transformed Boltzmann equation. The unknowns appearing in these equations are then evaluated using the distribution function and the necessary closure is achieved.

Grad 20 moment equations: r , ui , T, pi j , Si jk [1 + 3 + 1 + 5 + 10 = 20]

pi j

Divergence-free

Pi j

y= mCiC j

Si jk

Evaluate unknowns: Qi jkr

(n) ai

i

si jk = 0

fG|13

(n)

Closure

Evaluate unknowns: 1 , w2 , s rik i jk

= 0; (for n > 3)

1 (n) n! ai

Grad moment method: f in terms of orthogonal Hermite polynomials

f|G = f (0) Σ n=0

fG|20

Closure

y= mCiC jCk

Fig. 6.4 Flowchart depicting the Grad moment method

Cauchy’s equations

y= m m, mci , Ci2 2

Evolution equations for moments

Mulitply by y and integrate over velocity space

Transformed Boltzmann Equation: f (x, C,t)

Boltzmann Equation: f (x, c,t)

qi

y= m CiC2 2

Grad 13 moment equations: r , ui , T, pi j , qi [1 + 3 + 1 + 5 + 3 = 13]

pi j

Divergence-free

Pi j

y= mCiC j

Cauchy’s equations

y= m m, mci , Ci2 2

Evolution equations for moments

Mulitply by y and integrate over velocity space

Transformed Boltzmann Equation: f (x, C,t)

Boltzmann Equation: f (x, c,t)

6.4 The 13-Moment Approximation 217

218

6 Grad Equations: Derivation and Analysis

6.5 Boundary Conditions for 13-Moment Equations The equations derived above need appropriate number of boundary conditions for their solution. This is further required to bring the higher order continuum transport equations within the framework of computational fluid dynamics and for gaining mainstream adoption. In this section, we present the boundary conditions as given by Grad in his seminal work [62, 64]. Grad derived these boundary conditions for a two-dimensional flow problem purely on a mathematical basis. However, here we follow the method outlined by Yang [172] which has a more physical basis. The method is based on conservation laws and give the same results as that obtained by Grad. When a gas molecule strikes a solid boundary, there are a number of ways that the interaction can occur: the gas molecule may be specularly or diffusively reflected; it may be adsorbed at the solid boundary; it may react with the wall molecules to form a chemical bond; it may simply dissociate or ionize; or it may even dislodge the solid molecule. This interaction may further get affected by the nature of the solid boundary and its surface finish, temperature, etc. Because of these numerous possibilities, an accurate prediction of the effective interaction of the gas molecules with the solid boundary is particularly difficult. This is the primary reason because of which an accurate prediction of the wall boundary conditions has remained largely elusive. As such, the development of this research topic is still in progress and the theory presented in this section should not be considered as the definite answer. As higher order continuum transport equations are derived from the Boltzmann equation, it is natural that the boundary conditions should also originate from the Boltzmann equation. Consider a stream of gas molecules that is about to collide against a solid wall perpendicular to the x1 -direction as shown in Fig. 6.5. To simplify the problem, we make the following assumptions [28, 84]:

Fig. 6.5 Schematic showing the gas–wall interaction

6.5 Boundary Conditions for 13-Moment Equations

219

1. The solid wall is stationary. Otherwise, the boundary conditions are formulated in a frame where the solid wall is at rest. 2. No chemical reaction takes place between the gas and wall molecules. 3. The gas molecule remains in contact with the wall for a much lesser time, i.e., the adsorption time is much smaller than the characteristic time scale. 4. After collision, the gas molecule is re-emitted into the flow from the same point. With these assumptions, consider a molecule at time t striking the wall with a velocity between cI and cI + dcI at a certain point x and re-emitted at exactly the same point into the flow with velocity between cR and cR + dcR . Let f + (x, c, t) and f − (x, c, t) be the incident and reflected distribution functions, respectively. Then, we can write the distribution function at the wall as a sum of these two contributions as f (x, c, t) = f + (x, c, t) + f − (x, c, t).

(6.101)

Now, for the molecular stream, the molecules with x1 component of molecular velocity less than zero will not collide with the solid boundary (see Fig. 6.5 for definition of the axis introduced). That is, these molecules do not belong to the incident stream. In a similar way, the molecules with x1 component of molecular velocity greater than zero does not belong to the reflected stream. Mathematically, these conditions are represented as (functional dependence of distribution function on space and time is suppressed for convenience from now on) 2 f (c), for c1 > 0, + f (c) = (6.102) 0, for c1 < 0, 2 −

f (c) =

0,

for c1 > 0,

f (c),

for c1 < 0,

(6.103)

The appropriate form of the boundary condition should be such that it specifies the unknown distribution f − of the outgoing molecules from the known distribution of the incoming molecules f + . Now, upon collision, a certain fraction α of the incoming molecules is specularly reflected and the remaining molecules are perfectly absorbed by the wall and reemitted with a Maxwellian distribution. Mathematically, this boundary condition can be represented as f + (c1 , c2 , c3 ) = αf − (−c1 , c2 , c3 ) 3/2 1 ρw c2 , + exp − m 2π RTw 2RTw = f (c1 , c2 , c3 ),

c1 > 0 c1 < 0

(6.104)

220

6 Grad Equations: Derivation and Analysis

Substituting the coefficient of the second term on the left-hand side as k = 3/2 ρw 1 , we write as m 2π RTw c2 , f (c1 , c2 , c3 ) = αf (−c1 , c2 , c3 ) + k exp − 2RTw +

−

c1 > 0

(6.105)

This coefficient k is determined in such a way that the wall does not collect molecules, i.e., u1 = 0. Note that α = 1 − σv , with σv being the tangential momentum accommodation coefficient (TMAC). For a two-dimensional flow problem, the third component of velocity vector vanishes, i.e., u3 = 0 and the stress tensor and heat flux vector reduce to ⎡ p11 p12 pij = ⎣p12 p22 0 0

⎤ 0 0⎦ ,

with

p11 + p22 = 0

(6.106)

0 qi = q1 , q2 , 0 .

(6.107)

Specifying the distribution of the incoming molecules as the Grad’s distribution function (Eq. (6.91)), i.e., f + = fG|13 and expanding by considering u3 = 0, we obtain 2 + (c − u )2 + c2 (c − u ) ρ 1 1 2 2 3 f + (c1 , c2 , c3 ) = exp − 2RT m (2π RT )3/2 p11 (c1 − u1 )2 p11 (c2 − u2 )2 p12 (c1 − u1 )(c2 − u2 ) × 1+ − + 2pRT 2pRT pRT −

q1 (c1 − u1 ) q2 (c2 − u2 ) q1 (c1 − u1 )3 q1 (c1 − u1 )(c2 − u2 )2 − + + pRT pRT 5p(RT )2 5p(RT )2

q1 (c1 − u1 )c32 q2 (c2 − u2 )c32 q2 (c1 − u1 )2 (c2 − u2 ) q2 (c2 − u2 )3 + + + + 5p(RT )2 5p(RT )2 5p(RT )2 5p(RT )2 (6.108) For specular reflection, only the normal component of the molecular velocity changes sign while the tangential component remains unchanged. Hence, the molecular distribution function for the specularly reflected molecules can simply be obtained by replacing c1 by −c1 in Eq. (6.108). For molecules re-emitted with Maxwellian distribution with wall temperature Tw , we can write

c2 + c22 + c32 fw = k exp − 1 2RTw

,

where

ρw k= m

1 2π RTw

3/2 (6.109)

6.5 Boundary Conditions for 13-Moment Equations

221

Only those molecules having c1 > 0 collide with the wall, hence, the mass of the molecules striking the unit area of the wall can be written as m+ = m

+∞

dc3

−∞

+∞

+∞

dc2

−∞

c1 f + (c1 , c2 , c3 )dc1

(6.110)

0

Substituting the distribution function for the incident molecules (Eq. (6.108)) in the above equation, and performing the integration with u1 = 0 at the wall, we obtain the final expression for m+ as 2p + p11 m+ = m √ 2 2π RT

(6.111)

Similarly, using the distribution function for the reflected molecules, the mass of molecules leaving the unit area of wall in specular reflection is

−

αm = αm

+∞ −∞

dc3

+∞ −∞

∞

dc2

c1 f − (−c1 , c2 , c3 )dc1

0

2p + p11 = αm √ 2 2π RT

(6.112)

The mass of molecules leaving with Maxwellian distribution (Eq. (6.109)) can be written as +∞ +∞ ∞ dc3 dc2 c1 fw (c1 , c2 , c3 )dc1 kmw = km −∞

= km

+∞

−∞

exp −

0

c32

+∞

dc3 2RTw −∞ ∞ c2 × dc2 c1 exp − 1 dc1 2RTw 0 −∞

c2 exp − 2 2RTw

= km2π(RTw )2

(6.113)

Since the solid wall does not accumulate any molecules (u1 = 0), according to the law of conservation of mass, the mass of molecules leaving the wall (Eqs. (6.112) + (6.113)) should be exactly equal to the mass of molecules striking the wall (Eq. (6.110)). Accordingly, we get 2p + p11 2p + p11 =α √ + k2π(RTw )2 √ 2 2π RT 2 2π RT 1 − α 2p + p11 ∴ k= √ 2π(RTw )2 2 2π RT

(6.114)

222

6 Grad Equations: Derivation and Analysis

This is an equation for k derived using the basic law of conservation of mass that ensures that the wall does not accumulate any gas molecule. We now apply laws of conservation of momentum and energy at the wall to derive additional boundary conditions. As the incident molecules approach the wall, they bring tangential momentum to the unit area of the wall as M+ = m

+∞ −∞

dc3

+∞

−∞

+∞

(c2 − u2 )dc2

q2 p12 + √ =m 2 5 2π RT

(c1 − u1 )f + (c1 , c2 , c3 )dc1

0

(6.115)

For specularly reflected molecules, the tangential momentum carried away from the unit area of the wall is +∞ +∞ ∞ αM − = αm dc3 (c2 − u2 )dc2 (c1 − u1 )f − (−c1 , c2 , c3 )dc1 −∞

−∞

q2 p12 + √ = αm − 2 5 2π RT

0

(6.116)

and for those molecules that are emitted with Maxwellian distribution (Eq. (6.109)) is +∞ +∞ ∞ kMw = km dc3 (c2 − u2 )dc2 (c1 − u1 )fw (c1 , c2 , c3 )dc1 −∞

= km 0

∞

+∞

−∞

−∞

exp −

c32 2RTw

0

dc3

c2 (c1 − u1 ) exp − 1 2RTw

+∞ −∞

c2 (c2 − u2 ) exp − 2 2RTw

dc2

dc1

= −km2π(RTw )2 u2

(6.117)

According to law of conservation of momentum, the momentum brought to the unit area of the wall (Eq. (6.115)) is equal to sum of the momentum carried away by the specularly reflected molecules (Eq. (6.116)) and that by diffusively reflected molecules (Eq. (6.117)) as

∴

M + = αM − + kMw q2 q2 p12 p12 − k2π(RTw )2 u2 + √ + √ =α − 2 2 5 2π RT 5 2π RT

(6.118)

6.5 Boundary Conditions for 13-Moment Equations

223

Substituting the expression for k (Eq. (6.114)), we obtain after simplification as 2(1 − α)u2 p12 + √ p (1 + α) 2π RT

1+

p11 2p

+

2(1 − α)q2 =0 √ 5(1 + α)p 2π RT

(6.119)

Finally, we apply law of conservation of energy at the wall to obtain the final boundary condition. The stream of incident molecules brings kinetic energy to the unit area of the wall which is given as (with respect to the mean motion u) +∞ +∞ +∞ 1 m dc3 dc2 (c1 − u1 ) 2 −∞ −∞ 0 × (c1 − u1 )2 + (c2 − u2 )2 + c32 f + (c1 , c2 , c3 )dc1 q1 RT 1/2 3 + 2p + p11 =m 2 2 2π

E+ =

(6.120)

For specularly reflected molecules, the kinetic energy (with respect to the mean motion u) carried away from the unit area of the wall is +∞ +∞ ∞ 1 αE = αm dc3 dc2 (c1 − u1 ) 2 −∞ −∞ 0 × (c1 − u1 )2 + (c2 − u2 )2 + c32 f − (−c1 , c2 , c3 )dc1 RT 1/2 3 q1 = mα − + 2p + p11 2 2 2π −

(6.121)

and for those molecules that are emitted with Maxwellian distribution (Eq. (6.109)) is +∞ ∞ km +∞ kEw = dc3 dc2 (c1 − u1 ) 2 −∞ −∞ 0 × (c1 − u1 )2 + (c2 − u2 )2 + c32 fw (c1 , c2 , c3 )dc1 (6.122) = kmπ(RTw )2 4RTw + u22 According to law of conservation of energy, the kinetic energy brought to the unit area of the wall (Eq. (6.120)) is equal to sum of the kinetic energy carried away by the specularly reflected molecules (Eq. (6.121)) and that by diffusively reflected molecules (Eq. (6.122)) as

224

6 Grad Equations: Derivation and Analysis

E + = αE − + kEw RT 1/2 RT 1/2 3 3 q1 q1 + 2p + p11 ∴ = α − + 2p + p11 2 2 2π 2 2 2π + kπ(RTw )2 4RTw + u22 (6.123) Substituting the expression for k (Eq. (6.114)), we obtain the final boundary condition after simplification as

2π RT +

1/2

q1 p

p11 3 Tw Tw p11 (u2 /2)2 4(1 − α) + − 1− − 1+ =0 1+α T 2p 2 T 2p RT (6.124)

The appropriate boundary conditions for the Grad’s 13-moment equations can therefore be summarized as follows: u1 = 0 p12 2(1 − α)u2 + √ p (1 + α) 2π RT

2π RT

1/2

p11 1+ 2p

+

2(1 − α)q2 =0 √ 5(1 + α)p 2π RT

(6.125)

(6.126)

q1 p

p11 3 Tw Tw p11 (u2 /2)2 4(1 − α) + − 1− − 1+ =0 + 1+α T 2p 2 T 2p RT (6.127)

The first boundary condition (Eq. (6.125)) simply states that the solid wall does not accumulate or emit gas molecules. The stress boundary condition (Eq. (6.126)) states that the tangential stress (p12 ) or tangential heat flux (q2 ) at the wall boundary is related to tangential slip velocity of the stream. The third boundary condition (Eq. (6.127)) states that the normal component of the heat flux vector (q1 ) is roughly associated with a temperature jump between the solid wall and the adjacent gas molecules.

6.6 Significance of 13-Moment Equations

225

For flows that are not too far from equilibrium, we can safely say that p11 √ ; 21 ao = 0,

5 y 0 √ + 0.211 2 2 ⎥ ⎦ μo ao ⎪ 21 ao ⎪ ⎭ ⎛

5 y 0 √ × t−√ ⎥ ⎪ a μ a 21 o 21 ao o o⎪ ⎪ ⎦ ⎪ ⎪ ⎭ = 0,

5 y 0 O(100). A diffusive √ behaviour for the three variables is seen, with the viscous layer growing as t, in agreement with the solution suggested by the Navier–Stokes equations. For fixed small values of time, the variation of tangential stress (pxy ), tangential flow energy (qx ), and tangential gas velocity (u) in the direction normal to the plate surface is shown in Figs. 6.7 and 6.8, where the numbers 1, 2, and 3 represent the number of terms considered in evaluating the functions. From Eqs. (6.175)–(6.177), a parameter in the small time solution is defined as p0 t 3 Re = μ0 5 M02

(6.181)

where Re is instantaneous Reynolds number and M0 is Mach number given as Re =

ρ0 U 2 t , μ0

M0 =

U U = 5p0 a0 3ρ0

(6.182)

6.7 Exact Solution of Stokes’ First Problem

237

10-8

2

Fixed small time t = 1.8

10-10 s

y

1.5

1 2

0.5

0

0

200

400

1

3

600

800

1000

pxy Fig. 6.8 Variation of pxy with y for small values of time; numbers 1, 2, and 3 denote the number of terms retained in evaluating the corresponding expression

For small values of time, a2t t p0 t ∼ 0 ∼ O(100). It is noted that the solutions given by Eqs. (6.178)– (6.180) are convergent. The solution obtained here is not the most accurate solution known for the problem. Nonetheless, the above example illustrates the use of moment equations to a reasonably complex problem.

238

6 Grad Equations: Derivation and Analysis

10-8

2

Fixed small time t = 1.8

10-10 s

y

1.5

1

0

3

2

0.5

0

0.4

0.8

1

1.2

u Fig. 6.9 Variation of u with y for small values of time; numbers 1, 2, and 3 denote the number of terms retained in evaluating the corresponding expression

6.7.4 Kinetic Theory Based Solution The above solution has been improved by Gross and Jackson [67] using a kinetic model. They argued that the gas molecules which have interacted with the plate are affected by the movement of the plate. However, molecules which are yet to impinge on the plate remain unaffected by the movement of the plate (see Fig. 6.10). The distribution function for these two groups of molecules is different, and needs to be suitably accounted for, in order to obtain the correct dynamics immediately at the onset of motion of the plate. The following solution was obtained by Gross and Jackson [67] using kinetic theory: For short time: −b1 y −b1 y u = U 0.139H (t − a1 y)e + 0.361H (t − a2 y)e t − a1 y t − a2 y H (t − a1 y)e−b1 y + 7.36 H (t − a2 y)e−b2 y +U − 6.10 λ λ (6.183)

6.7 Exact Solution of Stokes’ First Problem

239

Fig. 6.10 Showing that the distribution function close to the plate is different from that away from the plate

nU nU −b1 y −b2 y 0.616H (t − a + y)e + 0.384H (t − a y)e 1 2 2(πβ)1/2 2(πβ)1/2 t − a1 y t − a2 y − 23.33 H (t − a1 y)e−b1 y + 20.25 H (t − a2 y)e−b2 y λ λ (6.184)

pxy =

where a1 = 0.799, b1 = 3.57/λ, a2 = 3.33, b2 = 7.42/λ (λ being mean free path of the gas). For large time: 1/2 2 2 1/2 −y d λ yd λ 4t − 1.24U u(y, t) = U erfc e + 0.292U e−cy πt πt 2t 1/2 (6.185)

pxy (y, t) =

1/2 −y 2 d 2 2 λ nU −cy 4t e − 0.170 e πt 2(πβ)1/2 d(t)1/2

(6.186)

where c = 6.86/λ and d = 1.54/(λ)1/2 . Note the presence of two step functions in the solution for small time (as opposed to a single step in the solution of Grad equation). The solution is further multiplied by exp(−by) (and therefore the strength of the step decreases on moving away from the plate); this multiplication is, however, missing in the Grad solution. This underscores the complexity of the problem in hand.

240

6 Grad Equations: Derivation and Analysis

6.7.5 Comparison with Free-Molecular Regime Solution The following solution applies to the problem in free-molecular regime [67]: √ y β U u(y, t) = erfc 2 t −βy 2 ρU pxy (y, t) = exp 2(πβ)1/2 t2

(6.187) (6.188)

where β = 1/(2RT ). The free molecular solution suggests that as t → 0, we get u→

U ρU ; pxy → 2 2(πβ)1/2

(6.189)

where we have utilized the following limiting properties of the complementary error function: 2x erfc(x) → 1 − √ , π erfc(x) →

exp(−x 2 ) , √ πx

as x → 0, as x → ∞.

(6.190)

In contrast, the Navier–Stokes solution with no-slip boundary condition predicts u → U and pxy → ∞. The Grad equations predicts u → 0.373U ; pxy → 1.1

ρU . 2(πβ)1/2

(6.191)

That is, the predicted value of initial flow velocity from Grad equations is about 26% smaller than the true (free-molecular) velocity, while the shear stress is about 10% larger. The kinetic theory predicts the correct values of both the velocity and shear stress, although the variation of shear stress in space is not correctly captured even with the kinetic theory.

6.7.6 Significance The solution of this problem is particularly interesting for understanding the more general case of arbitrary impulse on the surrounding fluid. The problem is also interesting as an analogy exists between the present problem and laminar flow over a flat plate. The boundary layer developing over the plate cannot be described by

6.8 Exact Solution of Cylindrical Couette Flow

241

the Navier–Stokes equations at the leading-edge. This is because the gas molecules hitting the plate and the gas molecules slightly away from the plate, will equilibrate only after a certain distance away from the leading edge. A similar situation arises for two plates placed close to each other (distance of the order of the mean free path). As discussed above (see Fig. 6.10), the distribution function of molecules impinging on a plate will primarily be governed by the velocity and temperature of the other plate, while the distribution function of molecules leaving the plate will primarily be governed by the velocity and temperature of the plate. The inability of the Navier–Stokes and Grad equations to treat the two distribution functions differently leads to error in the initial dynamics. However, as the density of the gas increases, the collision between the molecules increases, making the two distribution functions virtually identical; and the Navier– Stokes equations apply.

6.8 Exact Solution of Cylindrical Couette Flow We will now consider the problem of steady cylindrical Couette flow, with outer cylinder (radius = b) stationary while the inner cylinder (radius = a) is rotating as shown in Fig. 6.11. We will assume that the cylinders are long such that there is no variation along the length of the cylinders. The steady state assumption allows ∂ setting ∂t∂ = 0 while assumption of axisymmetry allows setting ∂θ = 0. All the quantities are therefore a function of radial position r only, and the partial differential equations reduce to ordinary differential equations. The mean quantities involve r and θ components of velocity vector, ur , uθ ; stress components, prr , prθ , pθθ ; and r and θ components of energy flux, qr , qθ ; along with other flow variables, mass density ρ, pressure p, and temperature T . Under these assumptions, as shown by Ai [7], we can obtain a solution of the problem from moment equations with some additional simplifications introduced later. The Navier–Stokes based solution is already provided in Sect. 2.4.5. Fig. 6.11 Schematic of cylindrical Couette flow

242

6 Grad Equations: Derivation and Analysis

6.8.1 Governing Equations and Boundary Conditions 13-moment equations in two-dimensional polar coordinate form are: Continuity: d (ρur r) = 0 dr

(6.192)

Momentum: u2 dp dprr prr − pθθ dur + + + ρur −ρ θ =0 dr dr r dr r

(6.193)

dprθ prθ uθ ur duθ +2 + ρur +ρ =0 dr r dr r

(6.194)

Energy: dur ur 2 dur duθ ur uθ 5 p + + prθ + pθθ − prθ + prr 3 dr r 3 dr dr r r qr 2 dqr dp + =0 (6.195) + ur + 3 dr r dr Equation of State: p = ρRT

(6.196)

Stresses: 2 dur ur qr uθ 4 dqr dprr dur 7 p 2 − − − 2 prθ + prr + 2 + ur 3 dr r 15 dr r dr r 3 dr 4 prθ prr ur p duθ 2 pθθ ur 2 − uθ − + = − prr − prθ 3 dr 3 r 3 r r μ p

uθ duθ − dr r

+ prr

+

(6.197)

qθ 2 dqθ dprθ prr − pθθ − + uθ + ur 5 dr r dr r

pθθ uθ p duθ ur dur + 2prθ + 2prθ − = − prθ dr r dr r μ

(6.198)

dur ur qr 4 4 dqr 2 2 dpθθ duθ −2 −2 + prθ − + uθ prθ + ur − p 3 dr r 15 dr r r dr 3 dr

6.8 Exact Solution of Cylindrical Couette Flow

p 7 dur uθ ur dur 2 − prθ + pθθ = − pθθ prr + pθθ − 3 dr r 3 r dr μ

243

(6.199)

Heat Flux: dur 5 dRT dprr dqr qθ prr − pθθ ur p + RT + + − uθ + qr + ur 2 dr dr r dr r dr r 2 prr dp 7 7 uθ 11 dur ur duθ dRT + + qθ − + prr qr − qθ + qr 5 dr 5 r dr 5 r ρ dr 2 dr prr − pθθ prθ prθ dprθ 2p prr dprr + +2 qr − =− − ρ dr r ρ dr r 3μ (6.200)

prθ dprθ +2 RT dr r

dur ur 11 ur dqθ uθ + + qr + qθ + qθ + ur dr r dr r 5 r 7 duθ dur uθ dRT 2 prθ dp 7 + qr + − qr + prθ qθ − 5 dr 5 dr r ρ dr 2 dr prr − pθθ prθ pθθ dprθ 2p prθ dprr + +2 qθ − =− − ρ dr r ρ dr r 3μ (6.201)

The above 10 equations involve the following 10-variables: ρ, p, T , ur , uθ , qr , qθ , prr , prθ , pθθ . The relevant boundary conditions are given below. At the outer stationary cylinder: ur (b) = 0

(6.202)

prr (b) 1−α uθ (b) 2 1−α prθ (b) 1 + +2 + p(b) 1 + α (2π RT (b))1/2 2p(b) 5 1+α

−

×

qθ (b) =0 (2π RT (b))1/2 p(b)

(6.203)

1/2 2π 1−α prr (b) 3 T2 T2 qr (b) − +4 − − 1− RT (b) p(b) 1+α T (b) 2p(b) 2 T (b) prr (b) (uθ (b))2 =0 (6.204) + 1+ 2p(b) 2RT (b)

244

6 Grad Equations: Derivation and Analysis

In the above equations, α = 1 − σv , with σv being the tangential momentum accommodation coefficient (TMAC). At the inner rotating cylinder (r = a) ur (a) = 0

(6.205)

prr (a) 1−α uθ (a) − U 2 1−α prθ (a) 1+ +2 + p(a) 1 + α (2π RT (a))1/2 2p(a) 5 1+α ×

qθ (a) =0 (2π RT (a))1/2 p(a)

(6.206)

1/2 2π 1−α prr (a) 3 T1 T1 qr (a) +4 + − 1− RT (a) p(a) 1+α T (a) 2p(a) 2 T (a) prr (a) (uθ (a) − U )2 =0 (6.207) + 1+ 2p(a) 2RT (a)

Equation (6.192) can be solved with the boundary conditions Eq. (6.202) or (6.205) as ur = 0, which helps get rid of several terms from the governing equations. Further, Eq. (6.194) can be readily integrated to obtain prθ =

B r2

(6.208)

where B is a constant of integration. The shear stress can be converted into a torque per unit length by multiplying it by 2π r 2 . Therefore, Eq. (6.208) shows that the torque per unit length is constant across the annulus. The above results can be used to simplify and integrate Eq. (6.195), and we obtain r(qr + prθ uθ ) = c

(6.209)

with c being another constant of integration. The equation therefore shows that the summation of heat flux and work done by the shear stress is a constant.

6.8.2 Linearization of 13-Moment Equations The remaining equations are difficult to solve analytically due to the non-linear and coupled nature of the equations. However, as seen in the previous section, it is possible to obtain a solution upon linearization of the equations and the boundary conditions. Linearization is possible when the inner cylinder is rotating slowly and

6.8 Exact Solution of Cylindrical Couette Flow

245

the temperature difference between the cylinder is kept small, i.e. Ma = T1 −T2 T1

√U γ RT1

0.4, thereby overestimating the mass flow rate in force-driven Poiseuille flow. However, by virtue of expanded variables set, R26 equations can overcome these limitations and increase the Knudsen number envelope over which accurate solutions can be obtained. However, note that the increased accuracy comes at the cost of substantial complexity involved both in the theoretical formulation and in the numerical computation. Moreover, obtaining numerical solution of R26 equations for twodimensional and three-dimensional flow problems seems extremely difficult at the moment.

284

7 Alternate Forms of Burnett and Grad Equations

7.6 Onsager Consistent Approach The Chapman–Enskog approach and Grad approach to arrive at Burnett equations and moment equations, respectively, were presented in the previous chapters. These equations have a mathematical basis, but suffer from limitations as discussed in Sects. 7.2 and 7.4. In view of these limitations, Singh and Agrawal [131] presented a novel approach to obtain Burnett-like and Grad-like equations utilizing an alternate form of the distribution function wherein Onsager’s reciprocity principle is additionally involved in the derivation. Instead of representing the distribution function in an infinite series in terms of Knudsen number as formal smallness parameter () as done in Chapman–Enskog approach or in terms of orthogonal Hermite polynomials as done in Grad approach, the idea is to cast the distribution function in terms of thermodynamic forces and fluxes.

7.6.1 Derivation of Distribution Function For any fluid flow transitioning from one conserved non-equilibrium state to another, the process is accompanied by entropy generation σ (Ji , Xi ) due to its thermodynamic force Xi and associated conjugate thermodynamic flux Ji . According to Onsager, the thermodynamic force is defined as derivative of entropy density σ with respect to thermodynamic variable αi as Xi =

∂σ , ∂αi

i = 1, 2, 3, . . .

(7.74)

The subscript i represents the type of thermodynamic force and is not a tensorial index. For example, thermodynamic force associated with stress tensor and heat flux vector are represented by Xτ and Xq , respectively. The entropy generated in the process can be expressed in terms of thermodynamic forces and fluxes as n

σ =

Ji Xi

(7.75)

i=1

Each flux is a linear homogeneous function of all the forces of the same tensorial order, n

Ji =

Lik Xk

(7.76)

k=1

where Lik are phenomenological coefficients and obey Onsager’s reciprocal relationship [107, 108], Lik = Lki with i, k = 1, 2, . . . , n.

7.6 Onsager Consistent Approach

7.6.1.1

285

First Order Correction to Distribution Function

The first order distribution function can be represented in terms of thermodynamic forces and microscopic fluxes around local Maxwellian distribution function f0 as [39, 131] f (1) = f0 − (Υτ : Xτ + Υq · Xq ) where

Υj Xj = tr(j )

∂f0 + ∇x .(c f0 ) ∂t

(7.77)

(7.78) Xj =0 ∀j =i

The macroscopic thermodynamic flux Ji can be obtained from its microscopic counterpart Υi as Ji =

Υ¯j f dc

(7.79)

where Υj = −f0 tr(j ) Υ¯j

1 2 ¯ Υτ = − C ⊗ C − |C| (γ − 1) I 2 5 2 ¯ Υq = − − |C| C 2β Xτ = β ∇ ⊗ u + (∇ ⊗ u)T Xq = ∇β

(7.80) (7.81) (7.82) (7.83) (7.84)

In the above equations, represents full contraction of the tensors of same order, ⊗ denotes the outer product, tr(τ ) and tr(q) are relaxation times for momentum [= μ/p] and energy [= k(γ − 1)/(Rγp)] transport, respectively γ is ratio of specific heat, and β = 1/(2RT ). The standard BGK collision model is employed in the above formulation. However, note that the introduction of different time scales for momentum and thermal diffusion ensures the correct value of Prandtl number. This first-order correction to the distribution function yields Navier–Stokes constitutive relationships but at the same time satisfies the Onsager’s symmetry principle. However, the second order correction to the distribution function as obtained in Chapman–Enskog approach which yields Burnett equations does not satisfy Onsager’s symmetry principle [95, 122]. This inconsistency with respect to symmetry principle may be the source of problems that plaque Burnett equations. With this form of distribution function in terms of forces and fluxes, we derive the second order correction to the distribution function in the next section.

286

7 Alternate Forms of Burnett and Grad Equations

7.6.1.2

Second Order Correction to Distribution Function

Keeping the same functional form of the distribution function in terms of thermodynamic forces and fluxes, the second-order correction to the distribution function as obtained by Singh and Agarwal [131] is Xq ) · Xq ) f|O = f0 − (Υτ : Xτ + Υq · Xq + (Υττ Xτ ) : Xτ + (Υqq (7.85) X are where expressions for Υττ Xτ and Υqq q

Υ¯ττ Xτ = Υ¯τ τ Xτ +

Δ

(Υq . Xq )Υ¯τ +

5 − |C|2 2β

tr(τ )

ϕ ¯ × Λ Υτ [ϕ(γ − 1) − γ ]∇ . u + C . ∇β + C . ∇g β ∂u ∂u T (7.86) + C⊗ + Ci C ⊗ ∂xi ∂xi

Υ¯qq Xq = Υ¯qq Xq +

Δ tr(q)

(Υτ : Xτ )Υ¯q +

5 2 − |C| (ΩC . ∇βC) 2β (7.87)

where Ω, Δ, Λ are defined in Eq. (7.91) and φ is viscosity exponent as in μ = μ0 (T /T0 )φ . Note that there are two parts in each of the above equations. The terms Υ¯τ τ Xτ and Υ¯qq Xq as proposed by Mahendra and Singh [91] are secondorder correction terms to the distribution function. However, explicit expressions for these terms and other terms on the right-hand side which are correction terms were obtained by Singh and Agarwal [131]. The terms Υ¯τ τ Xτ and Υ¯qq Xq are ∂u 1 ∂u T T ¯ + C ⊗ ∇g + (C ⊗ ∇g) Υτ τ Xτ = −Ci C ⊗ + C⊗ ∂xi ∂xi 2β 1 1 (γ − 1)C . ∇g − (γ − 1)(C ⊗ C) : Xτ I − 2β 2β 1 − Υ¯τ Υj Xj t(r)τ j

ϕ ¯ + Υτ [ϕ(γ − 1) − γ ]∇ . u + C . ∇β + C . ∇g β

(7.88)

7.6 Onsager Consistent Approach

Υ¯qq Xq = −Υ¯q

287

1

Υj Xj

t(r)q

−

j

1 1 [C . ∇g] − (C ⊗ C) : Xτ β β

1 5 5 5 (γ − 1)∇ . u + 2 (C . ∇β) C − − |C|2 ∇g 2β 2β 2β 2β 5 − |C|2 C ⊗ ∇u − 2β

ϕ ¯ (7.89) + Υq [ϕ(γ − 1) − γ ] ∇ . u + C . ∇β + C . ∇g β +

where β and g are given as β=

1 , 2RT

g = log

ρ β

(7.90)

However, since the distribution function in terms of Υ¯τ τ Xτ and Υ¯qq Xq does not satisfy the compatibility conditions (Eq. (5.67)), it becomes necessary to modify the distribution function. This task is not trivial and Singh and Agrawal [131] adopted a correction procedure that appended additional terms to the expressions Υ¯τ τ Xτ and Υ¯qq Xq . Note that the added terms are not appended on an ad hoc basis, these terms are necessary in satisfying the compatibility conditions without breaking the symmetry principle. The coefficients Ω, Λ, and Δ appearing in Eqs. (7.86) and (7.87) are given as tr(q) 1 tr(τ ) 2 −ϕ Ω = ϕ + 2; Λ = − ; Δ = − +2 5 2 tr(q) tr(τ )

(7.91)

The final form of the distribution function (Eq. (7.85)) satisfies the linearized Boltzmann equation and compatibility conditions without breaking the Onsager’s symmetry principle. This particular form of distribution function f|O (where the subscript O stands for Onsager) is termed as “Onsager consistent distribution function.” This form is then utilized to derive Onsager–Burnett (OBurnett) equations and Onsager-13 (O13) equations as shown in Fig. 7.3.

7.6.2 Onsager–Burnett Equations In this section, we utilize the Onsager consistent distribution function (Eq. (7.85)) to derive expressions for pressure tensor and heat flux vector as Pij = pδij + pij = m

Ci Cj f|O dC,

qi =

m 2

C 2 Ci f|O dC

(7.92)

288

7 Alternate Forms of Burnett and Grad Equations

Fig. 7.3 Flowchart showing the detailed procedure of Onsager consistent approach to obtain OBurnett and O13 equations

Onsager consistent approach

Distribution function in terms of thermodynamic forces Xi s and conjugate fluxes Ji s

Onsager consistent distribution function:

Burnett-like equations

Grad-like equations

Pi j = m CiC j f|O dC, m 2 qi = C Ci f|O dC 2

1 = m f C C C2 dC, rik |O i k 2 w = m C4 ( f|O − fM )dC, si jk = m f|OCiC jCk dC

Cauchy’s equations of motion

Evolution equations for pi j and qi

Onsager-Burnett equations

Onsager-13 moment equations

The expression for the Onsager consistent distribution function (Eq. (7.85)) contains a large number of terms. After evaluating the integrals, the constitutive relationships for normal stress pxx , shear stress pxy , and heat flux component qx are obtained as [136] NS B + pxx pxx = pxx 2

2 ∂u ∂u μ2 β ∂u ∂v ∂w + δ2 + δ2 +4 α1 = μ δ1 + α2 ∂x ∂y ∂z ρ ∂x ∂y 2 2 2 ∂u ∂w ∂v ∂u ∂v ∂u ∂w + α5 + α6 + α3 + α4 + α7 ∂z ∂y ∂x ∂z ∂x ∂x ∂x 2 2 ∂v ∂w ∂u ∂v ∂v ∂w ∂u ∂w + α9 + α12 + α8 + α10 + α11 ∂x ∂y ∂y ∂z ∂y ∂z ∂x ∂z 2 2 ∂w ∂v ∂v ∂w (7.93) + α14 + α15 + α13 ∂z ∂y ∂y ∂z

7.6 Onsager Consistent Approach

289

NS B pxy = pxy + pxy

= μδ3

μ2 β ∂u ∂v ∂u ∂u ∂v ∂v ∂u ∂v + μδ3 +4 + β2 + β3 β1 ∂y ∂x ρ ∂x ∂y ∂x ∂y ∂z ∂z

∂u ∂v ∂u ∂v ∂w ∂w ∂v ∂w ∂u ∂w + β5 + β6 + β7 + β8 ∂x ∂x ∂y ∂y ∂x ∂y ∂z ∂x ∂z ∂y ∂u ∂w ∂v ∂w + β10 (7.94) + β9 ∂y ∂z ∂x ∂z

+ β4

qx = qxN S + qxB = δ4 k

μ2 β 1 ∂g ∂u 1 ∂β ∂v 1 ∂β ∂w 1 ∂β + 4 + γ2 2 + γ3 2 γ1 ρ β ∂x ∂x 2Rβ 2 ∂x β ∂x ∂y β ∂x ∂z

+ γ4

1 ∂g ∂u 1 ∂g ∂v 1 ∂g ∂w 1 ∂β ∂u 1 ∂β ∂u + γ5 + γ6 + γ7 2 + γ8 2 β ∂y ∂y β ∂y ∂x β ∂z ∂x β ∂x ∂x β ∂y ∂y

1 ∂β ∂u 1 ∂β ∂v 1 ∂β ∂w 1 ∂g ∂v + γ10 2 + γ11 2 + γ12 β ∂x ∂y β 2 ∂z ∂z β ∂y ∂x β ∂z ∂x 2k(γ − 1) 2 1 1 ∂g ∂w ∂β ∂v ∂β ∂w + γ15 + γ14 + γ13 β ∂x ∂z Rγ ρβ ∂y ∂x ∂z ∂x + γ9

+ γ16

∂β ∂u ∂β ∂u ∂β ∂u ∂β ∂v ∂β ∂w + γ17 + γ18 + γ19 + γ20 ∂x ∂x ∂y ∂y ∂z ∂z ∂x ∂y ∂x ∂z

(7.95)

where g and β are defined through Eq. (7.90). The coefficients, α s, β s, γ s, and δ s, with numerical subscripts are functions of the type of gas and the interaction potential between the molecules. The constants appearing in front of the derivatives of flow field variables are 125γ 2 − 576γ + ϕ(110 − 160γ + 50γ 2 ) + 643 4 2 , δ1 = − , δ2 = , α1 = 3 3 40 4 4 1 1 3 3 , α3 = , α4 = , α5 = , α6 = − , α7 = − , 5 5 5 5 5 5 2 2 100γ − 484γ + ϕ(90 − 140γ + 50γ ) + 459 α8 = , 20 125γ 2 − 392γ + ϕ(70 − 120γ + 50γ 2 ) + 291 , α9 = 40

α2 =

α10 = α9 , α11 = α9 +

18 2 , α12 = α11 , α13 = − , 40 5

290

7 Alternate Forms of Burnett and Grad Equations

1 α14 = − , α15 = α14 , 5 23γ + 5γ ϕ − 5ϕ − 37 14 , β2 = β1 , β3 = −1, β4 = β1 − , 10 10 3 β5 = β4 , β6 = 4β4 , β7 = − , β8 = β7 , β9 = β1 10

δ3 = −1, β1 =

δ4 = 1, γ1 =

−47 + 25γ 8

, γ2 = ϕ

49 − 35γ 8

, γ 3 = γ2 ,

77 − 35γ 1 7ϕ , , γ8 = γ4 = − , γ5 = γ4 , γ6 = γ4 , γ7 = +ϕ 2 8 4 −39 + 25γ , γ9 = γ8 , γ10 = γ9 , γ11 = γ8 , γ12 = 8 −77 + 35γ + 10ϕ(−1 + γ ) , γ13 = γ12 , γ14 = −1, γ15 = γ14 , γ16 = 8 7 −59 + 35γ + 10ϕ(−1 + γ ) γ17 = − , γ18 = γ17 , γ19 = , γ20 = γ19 . 4 8 The Burnett contribution for other components of stress tensor and heat flux vector can be obtained by applying suitable change of variables in an appropriate base equation as given in Table 7.3. Substitution of these stress and heat flux terms in Cauchy’s equations of motion and energy conservation equation completes the derivation of the OBurnett equations.

Table 7.3 Base equation and change of variables to be followed while evaluating Burnett order contribution for other components of stress tensor and heat flux vector Variable

Base equation

Change of variables

B pyy

B pxx

u → v, x → y, v → u, and y → x

B pzz

B pxx

u → w, x → z, w → u, and z → x

B pyz

B pxy

u → v, x → y, v → w, y → z, w → u, and z → x

B pzx

B pxy

u → w, x → z, v → u, y → x, w → v, and z → y

qyB

qxB

u → v, x → y, v → u, and y → x

qzB

qxB

u → w, x → z, w → u, and z → x

7.6 Onsager Consistent Approach

7.6.2.1

291

Linear Stability Analysis

A linear stability analysis of the OBurnett equations is performed for a onedimensional wave about the following equilibrium state: u = 0; ρ = ρ0 and T = T0 . Assuming small perturbations around equilibrium, we can write u = u (x, t), ρ = ρ0 + ρ (x, t),

(7.96)

T = T0 + T (x, t) where denotes small quantities away from equilibrium. Substituting Eq. (7.96) in the Cauchy’s equations of motion and energy conservation equation, the following set of equations is obtained upon linearization: ∂ρ ∂u + ρ0 = 0, ∂t ∂x ρ0

4 ∂ 2 u ∂u ∂ρ ∂T + ρ0 R + RT0 + μ 2 = 0, ∂t ∂x ∂x 3 ∂x ∂ 2T ∂u ∂T 3 ρ0 R − k 2 + p0 = 0. 2 ∂t ∂x ∂x

(7.97)

(7.98)

(7.99)

Notice that there are no Burnett-order terms in the above equations and the form of the equations is similar to that of perturbed Navier–Stokes equations. Since perturbed Navier–Stokes equations are unconditionally stable, OBurnett equations are also unconditionally stable which in itself is a remarkable feature considering the fact that the Burnett equations are plagued with stability issues. This stability feature is due to the fact that Eqs. (7.93)–(7.95) involve only the cross-product of derivative of velocity, temperature, and pressure. Therefore, Burnett order contribution to the B , p B , and q B ) becomes identically zero for stress and heat flux components (pxx xy x small perturbation. In contrast, the presence of second-derivatives of velocity and temperature in stress tensor and heat flux vector in Burnett equations lead to the problem of stability.

7.6.2.2

Features of OBurnett Equations

The OBurnett equations have several attractive features: 1. Fewer boundary conditions: The constitutive relationships for OBurnett equations do not contain second and higher order derivatives of velocity as compared to conventional Burnett equations (which contain third order derivatives of velocity). Therefore, the OBurnett equations need the same number of boundary

292

2. 3.

4.

5.

7 Alternate Forms of Burnett and Grad Equations

conditions as the Navier–Stokes equations. Hence, the need of prescribing additional boundary conditions is mitigated. Unconditionally stable: As per the linear stability analysis, the equations are unconditionally stable. This point is already demonstrated in Sect. 7.6.2.1. Reduced coupling: The absence of temperature gradient terms in the constitutive relationships for stress tensor reduces the amount of coupling between the momentum and energy equations. Hence, in certain cases like incompressible forced convection, rarefied gas flow problems, one can obtain velocity field before solving the energy equation. Correct value of Prandtl number: The formulation of distribution function involves two different relaxation times for momentum and energy transport which ensures that the correct value of Prandtl number can be recovered for any gas. Larger Knudsen number envelope: The derivation of the OBurnett equations does not involve Chapman–Enskog like expansion, hence the OBurnett equations are expected to give accurate solutions in a larger Knudsen number envelope.

7.6.3 Onsager-13 Moment Equations In this section, we present the derivation of Onsager-13 (O13) moment equations utilizing the Onsager consistent distribution function (Eq. (7.85)). The derivation of O13 moment equations can be outlined as: The evolution equations for the stress tensor and heat flux vector are generated as derived in Sect. 6.4.1. However, instead of using Grad distribution function (fG|13 ), we employ the Onsager-consistent 1 , w 2 and s , appearing distribution function (f|O ) to evaluate the unknowns, rik ij k in the evolution equations to achieve the closure as 1 =m rik

f|O Ci Ck C 2 dC

1 ρ μ 2 p ∂ui [342 + 35ϕ + = 7μ ρ ∂xk 20 β 2 p

∂ui ∂ul ∂ui ∂ul ∂ui ∂uk ∂ul ∂ul − 7γ (31 + 5ϕ)] + 27 + 45 − 18 ∂xk ∂xl ∂xl ∂xk ∂xl ∂xl ∂xi ∂xk 2 ∂β ∂β ∂β ∂g 7 ρ k(γ − 1) 2 (7.100) +β + 5 8β Rγp ∂xi ∂xk ∂xi ∂xk

w2 = m

C 4 (f|O − fM )dC

7.6 Onsager Consistent Approach

293

p ∂uj 5 ρ k(γ − 1) 2 7 ∂β ∂β 2 ∂g ∂g + + β ρ ∂xj 2 β5 Rγp 2 ∂xj ∂xj ∂xj ∂xj 2 ∂uj 2 7 ρ μ + 2 (−5 + 3γ )(−63 − 10ϕ + 5γ (7 + 2ϕ)) ∂xj β p 16 (7.101)

= (50 − 30γ )μ

sij k = m =

f|O Ci Cj Ck dC

2 2 μ ∂g ∂uj k(γ − 1) 2 ∂β ∂uj μ 3 ρ β + ϕ − 2 β3 p ∂xi ∂xk Rγp p ∂xi ∂xk (7.102)

1 , w 2 , and s Substituting the expressions for higher order moments rik ij k (Eq. (7.102)) in the evolution equations for stress tensor and heat flux vector, we obtain O13 moment equations as

2 ∂pij μ ∂g ∂uj 3 ∂ β + ∂t 2 ∂xk p ∂xi ∂xk 2 2 k(γ − 1) ∂β ∂uj ρ μ + 3 ϕ − Rγp p ∂xi ∂xk β + uk

∂pij ∂uj ∂ui ∂uk 4 ∂qi p + + 2pki + 2p + pij = − pij , ∂xk 5 ∂xj ∂xk ∂xj ∂xk μ

(7.103)

5 p ∂pik p ∂ui ∂qi ∂qi 5 p ∂p p2 ∂ρ pik ∂p ∂ − μ + uk + − 2 − +7 ∂t ∂xk 2 ρ ∂xi 2 ρ ∂xk ρ ∂xk ∂xk ρ ∂xk ρ ∂xi

2 ∂ui ∂ul ∂ui ∂ul 1 ρ μ ∂ 7 ∂ui [342 + 35ϕ − 7γ (31 + 5ϕ)] + + 27 + qk 5 ∂xk ∂xk 20 β 2 p ∂xk ∂xl ∂xl ∂xk ∂ui ∂uk 7 ρ k(γ − 1) 2 ∂β ∂β ∂ul ∂ul ∂β ∂g + 2 + 45 − 18 + β ∂xl ∂xl ∂xi ∂xk 8 β5 Rγp ∂xi ∂xk ∂xi ∂xk

2 p ∂uj 7 7 ∂uk 2 ∂uk 1 ∂ ρ μ (−5 + 3γ ) (50 − 30γ )μ + qi + qk + + 2 5 ∂xk 5 ∂xi 6 ∂xi ρ ∂xj β p 16 ∂uj 2 5 ρ k(γ − 1) 2 7 ∂β ∂β × (−63 − 10ϕ + 5γ (7 + 2ϕ)) + ∂xj 2 β5 Rγp 2 ∂xj ∂xj

294

7 Alternate Forms of Burnett and Grad Equations

2 pij ∂pj k μ ∂g ∂uj ∂g ∂g 3 ρ ∂uj − β + ∂xj ∂xj ρ ∂xk 2 β 3 ∂xk p ∂xi ∂xk 2 2 ∂β ∂uj 2p μ k(γ − 1) =− qi − + ϕ Rγp p ∂xi ∂xk 3μ

+ β2

(7.104)

with β, g, γ , φ as defined in Sect. 7.6.1.2. The basic conservation laws (Eqs. (7.1)– (7.3)), the transport equations for stress tensor (Eq. (7.103)) and heat flux vector (Eq. (7.104)) form a complete closed set of O13 moment equations.

7.6.3.1

Comparison with Grad Closure

In this section, Onsager consistent closure is compared with the Grad closure. Most 1 , w 2 , and s of the terms in the expressions for the unknowns rik ij k (Eqs. (7.100)– (7.102)) have (μ/p)2 as their coefficient, indicating that these are higher order terms. The coefficient [k(γ − 1)/Rγp]2 can be written as 1/P r 2 (μ/p)2 , signifying that the terms with this coefficient are also higher order terms. With this result, Onsager consistent closure to the first order can be written as 1 rik = 7μ

p ∂ui ρ ∂xk

w 2 = (50 − 30γ )μ

(7.105) p ∂uj ρ ∂xj

(7.106)

sij k = 0

(7.107)

The comparison of Onsager consistent closure to the first order with the Grad closure is shown in Table 7.4. Note that the term μ∂ui /∂xk simplifies to pik to the first order and unknown w2 reduces to zero using γ = 5/3 for a monatomic gas. Therefore, Onsager consistent closure to the first order agrees with the Grad closure which enhances confidence in the O13 moment equations. Including second order terms [(μ/p)2 terms] in the constitutive relationships 1 , w 2 , and s , we get additional terms in the O13 moment equations for rik ij k when compared with Grad 13-moment equations. These terms include products of

Table 7.4 Comparison of Onsager consistent closure to the first order with the Grad closure. Note that the Grad equations are for monatomic gases Unknowns

Grad closure

Onsager consistent closure

1 rik w2

7RTpik

7μ pρ

0

(50 − 30γ )μ pρ

sij k

0

0

∂ui ∂xk

= 7RTpik ∂uj ∂xj

= 0 for γ =

5 3

7.7 Comparison of Various Forms of Equations

295

derivatives of velocity, temperature, and density and with inclusion of these higher order terms, the applicability of O13 equations is expected to be higher than Grad 13-moment equations. Since we have first order derivatives of stress tensor and heat flux vector (as opposed to second order derivatives in R13 equations), we need fewer boundary conditions as compared to R13 equations. This point becomes clear when we compare the constitutive relationships obtained for both the cases. With respect to R13 moment equations, we note the presence of stress and heat flux terms in the expressions for , Qij , and mij k (Eqs. (7.49)–(7.51)). However, in case of O13 moment equations, we have only first order derivatives of velocity, pressure, and 1 , w 2 , and s , thereby, reducing the need of temperature in the expressions for rik ij k prescribing additional boundary conditions.

7.6.4 Summary We presented a derivation of Onsager consistent distribution function which satisfies the compatibility conditions, H -theorem, and linearized Boltzmann equation. The form of the distribution function is such that the Onsager’s symmetry principle is preserved, an important and rigid constraint that was somehow ignored in the previous derivations. This form of distribution function is then utilized to derive OBurnett and O13 equations. With all the significant features the Onsager consistent distribution function offer, it is safe to say that the derivation of OBurnett and O13 moment equations has strong roots in the fundamental principles of the nonequilibrium thermodynamics and is believed to capture the strong non-equilibrium effects at higher Knudsen numbers.

7.7 Comparison of Various Forms of Equations In this section, we compare some of the available variants of Burnett and Grad equations for two relatively simple problems for which analytical (or semi-analytical) solution is possible.

7.7.1 Couette Flow The various forms of the Burnett equations have been compared for some specific problems. Here we will follow the analysis of Singh et al. [133] and compare the results for the Couette flow problem. The solution is further compared with the Grad’s equations and R13 equations.

296 Table 7.5 Value of A in Eq. (7.108)

7 Alternate Forms of Burnett and Grad Equations Higher order continuum model Burnett equations Augmented Burnett equations

A 0 1 2 6 Kn

Super Burnett equations (Agarwal et al. [2])

Kn2

Super Burnett equations (Shavaliyev [129])

− 23 Kn2

Grad’s equations R13 equations

0 9 2 5 Kn

The non-dimensional governing equations involving the shear stress (pxy ) can be written in the following general form: pxy = −

d 3u du +A 3 dy dy

(7.108)

where the value of A varies with the type of the equation; its value is given in Table 7.5. The above equation was solved by setting the value of velocity u and its curvature to be zero at the centerline (due to symmetry), and with a slope of α at the centerline. The value of α was determined from DSMC calculation. The solution of the above equation can be readily obtained as: √ y for A ≥ 0 u = pxy y + A(pxy + α) sinh √ A √ y u = pxy y + A(pxy + α) sin √ for A < 0. (7.109) A In the above equation, pxy is non-dimensional shear stress whose value is evaluated as the harmonic mean of values in the continuum and free-molecular regimes, as −1 2 − σv 2 pxy = 1 + 2 Kn (where σv = 1). (7.110) σv π The value of pxy and α depends only on Kn. Therefore, the solution from different equations can be readily compared for different values of Knudsen number. The DSMC data serves as the benchmark against which the obtained analytical solutions are compared. Figures 7.4 and 7.5 show such comparison for different sets of equations. Note that the equations are not able to capture accurately the non-linearity in the velocity profile (Knudsen layer) observed in the near wall region.

7.7.2 Force-Driven Poiseuille Flow We now consider flow between two parallel plates, with the flow driven by a constant external force, a [78]. Normally such flows are pressure driven; however,

7.7 Comparison of Various Forms of Equations

297

Fig. 7.4 Velocity distribution across the channel for Couette flow at Kn = 1 for (a) augmented Burnett equations [177], (b) super-Burnett equations [2], (c) super-Burnett equations [129] and (d) R13 equations; dashed line represents the DSMC data (all figures taken from Singh et al. [133])

the solution of pressure driven flow is difficult to solve analytically (pressure being the function of two space-coordinates, x and y), and therefore we resort to an external force driving the flow. As in Poiseuille flows, we consider the plates to be infinitely long and the flow to be steady. See Fig. 7.6 for additional details of the problem setup. The analysis proceeds by making the assumptions suggested by Uribe and Garcia [154]: the velocity normal to the stationary walls, v = 0, and further that the flow variables are functions of y-direction only. The latter assumption can be justified due to the flow being fully developed. With these assumptions, the basic conservation laws reduce to

298

7 Alternate Forms of Burnett and Grad Equations

0.22

0.22

0.21

0.21

R13 Solution

Super Burnett (Agarwal 2001)

u

0.2

u

0.2

0.19

0.19

0.18

0.18

0.17

0.17

0.4

0.42

0.44 y

0.46

0.48

0.4

0.22

0.22

0.21

0.21

Augmented Burnett

0.44 y

0.46

0.48

Super Burnett (Shavaliyev 1993)

0.2 u

u

0.2

0.42

0.19

0.19

0.18

0.18

0.17

0.17

0.4

0.42

0.44 y

0.46

0.48

0.4

0.42

0.44 y

0.46

0.48

Fig. 7.5 Zoomed in view of the velocity profile in the near wall region according to R13 equations, super-Burnett equations [2], augmented Burnett equations [177] and super-Burnett equations [129]; dashed line represents the DSMC data (all figures taken from Singh et al. [133])

Fig. 7.6 Schematic of plane Poiseuille flow driven by an external force, “a”

7.7 Comparison of Various Forms of Equations

dpxy = ρa, dy

299

[x-momentum equation]

dp dpyy + = 0, dy dy dq2 du + pxy = 0. dy dy

(7.111)

[y-momentum equation]

(7.112)

[energy equation]

(7.113)

Since the constitutive relationships are different for the Navier–Stokes, conventional Burnett, and OBurnett equations, the analysis differs from this point onwards. 1. Navier–Stokes framework: According to the linear constitutive relationships of the Navier–Stokes equations, we obtain pxy = −μ

du , dy

pyy = 0,

qy = −k

dT dy

Hence, the closed form of the equations is obtained as du p d μ = − a, dy dy RT dp = 0, dy 2 d du dT k +μ = 0, dy dy dy where we have utilized the ideal gas equation p = ρRT to replace density in the x-momentum equation. 2. Conventional Burnett equations: In conventional Burnett equations, the xmomentum and the energy equation remain unchanged. However, because of non-zero Burnett contribution to normal stress pyy in Eq. (7.112), we get closed form of equations as du p d a, μ = − dy dy RT CBu dp dpyy + = 0, dy dy 2 du dT d k +μ = 0. dy dy dy

(7.114) (7.115) (7.116)

300

7 Alternate Forms of Burnett and Grad Equations

CBu (with superscript depicting conventional Burnett equations) is given where pyy as CBu pyy

2 ω2 μ2 d 2 p 2 ω2 μ2 du 2 2 ω3 μ2 d 2 T 2 ω2 μ2 dp dρ − = − + 3 pρ dy dy 3 pρ dy 2 3 p dy 3 ρT dy 2 2 ω5 μ2 dT 2 2 ω4 μ2 dp dT 1 ω6 μ2 du 2 + + + (7.117) 3 pT ρ dy dy 3 ρT 2 dy 12 p dy

Note that the Burnett contribution to shear stress pxy and normal heat flux qy is zero. 3. OBurnett equations: Again, in case of OBurnett equations, the x-momentum and the energy equation remain unchanged. However, the Burnett contribution to the normal stress pyy is particularly different as compared to that of conventional Burnett equations and given as OBu pyy

du 2 μ2 β =4 α7 ρ dy

(with superscript depicting OBurnett equations). The closed set of equations is then obtained as [78] du d μ = − dy dy dp d μ2 du 2 = + 2α7 dy dy p dy 2 d du dT k +μ = dy dy dy

p a, RT

(7.118)

0,

(7.119)

0.

(7.120)

The analysis is performed by considering the temperature dependence of the transport coefficients. Accordingly, the dynamic viscosity (μ) and thermal conductivity (k), for a dilute gas of rigid spheres are given as [31], μ=

5cμ mkT (y) 1/2 75cλ k 3 T (y) 1/2 , k = , π πm 16σ 2 64σ 2

(7.121)

where σ is the particle diameter and accurate values of cμ = 1.016034 and cλ = 1.02513 are known. The variables are non-dimensionalized as p∗ (s) =

p(y) T (y) u(y) y , T ∗ (s) = , u∗ (s) = , where s = . p(0) TR L (2kTR /m)1/2 (7.122)

7.7 Comparison of Various Forms of Equations

301

Taking into account the temperature dependence of the transport coefficients and performing non-dimensionalization, we obtain coupled ordinary differential equations as 1. Navier–Stokes framework: 1 dT ∗ du∗ d 2 u∗ + = − 2T ∗ ds ds ds 2 dp∗ =0 ds 8cμ du∗ 2 d 2T ∗ = − − 15cλ ds ds 2

b0 p∗ , T ∗3/2

(7.123) (7.124)

∗ 2 dT 1 . 2T ∗ ds

(7.125)

where the coefficient b0 is given as b0

√ 8 L2 a 2π p(0)mσ 2 =− , 5 cμ k 2 TR2

(7.126)

2. Conventional Burnett equations: b p ∗ d 2 u∗ 1 dT ∗ du∗ + 0∗3/2 , =− ∗ 2 2T ds ds ds T

(7.127)

b1 p∗2 b2 p∗2 b1 p∗3 1 dp∗ 2 1 dp∗ dT ∗ d 2 p∗ = + − + − p∗ ds T ∗ ds ds ds 2 T ∗2 T ∗2 T ∗2 ∗ 2 8ω3 p∗ cμ du∗ 2 2p∗ du∗ 2 dT ω3 p∗ − ∗ − − T ds 15ω2 T ∗ cλ ds 2ω2 T ∗2 ds ω5 p∗ dT ∗ 2 ω4 dp∗ dT ∗ ω6 p∗ du∗ 2 + + + (7.128) ω2 T ∗ ds ds 4ω2 T ∗ ds ω2 T ∗2 ds ∗ 2 8cμ du∗ 2 dT d 2T ∗ 1 =− − . (7.129) ∗ 2 15cλ ds 2T ds ds where the coefficients b0 , b1 , and b2 are given as b0

√ 8 L2 a 2π p(0)mσ 2 384 L2 p(0)2 π σ 4 =− , b = − , 1 2 k2T 2 5 25 ω2 cμ cμ k 2 TR2 R d 2p L2 2 T (0)2 dy s=0 b2 = p(0)TR2

(7.130)

302

7 Alternate Forms of Burnett and Grad Equations

3. OBurnett equations: b0 p∗ d 2 u∗ 1 dT ∗ du∗ = − , (7.131) + 2T ∗ ds ds ds 2 T ∗3/2

∗ 2 2T ∗ du∗ du bo p∗ 1 dT ∗ du∗ 1 dT ∗ − + + − dp∗ p∗ ds 2T ∗ ds ds ds p∗ ds T ∗3/2 , =

ds T ∗ du∗ 2 b2 − ∗2 ds p (7.132) ∗ 2 8cμ du∗ 2 d 2T ∗ dT 1 = − − . (7.133) ∗ 2 15cλ ds 2T ds ds The coefficients are b0

√ 8 L2 a 2π p(0)mσ 2 64 p(0)L σ 2 2 π =− , b2 = , 5 25 kTR cμ α7 cμ k 2 TR2

(7.134)

Solving this problem of coupled differential equations as an initial value problem, the results of Navier–Stokes equations, conventional Burnett equations, OBurnett equations, and NCCR theory [52, 103] are validated against the direct simulation Monte Carlo (DSMC) and molecular dynamics (MD) simulations results as shown in Fig. 7.7. An important point to note is that the Burnett equations (which are second order accurate in Knudsen number) are not able to capture the temperature dip at the center which is a super-Burnett effect. Another important observation is the bi-modal pressure profile predicted by the conventional Burnett equations and confirmed by the DSMC results. However, we get monotonic pressure profile according to OBurnett equations, NCCR theory, and confirmed by more accurate MD results. Hence, we note that, even for a simpler looking one-dimensional flow problem like force-driven Poiseuille flow, there are many subtleties in the flow physics and the results of different equations set sometimes give qualitatively different results. Once we obtain the solution for velocity, temperature, and pressure, the results for stresses and heat fluxes can be readily obtained and they are shown in Fig. 7.8. For more mathematical details of these variables, interested readers can refer to Uribe and Garcia [154] in case of conventional Burnett equations, Jadhav et al. [78] in case of OBurnett equations, and Myong [104] in case of NCCR equations. For stresses and heat fluxes, the results of the conventional Burnett equations, OBurnett equations, and NCCR theory are qualitatively similar and in reasonable agreement with the DSMC and MD results but discrepancies are observed in the near-wall region. It is important to emphasize the failure of the Navier–Stokes equations which ∗ , p ∗ , p ∗ ) and tangential heat flux (q ∗ ) essentially predict the normal stresses (pxx yy zz 1 to be zero, as clearly evident from Figs. 7.8b, c, d and f.

7.8 Summary

303

1.1 1.12

1

1.1

0.9

1.08

0.7

r*

u*

0.8

0.4 0.3

1.06 1.04

0.6 0.5

Onsager−Burnett NSF Burnett NCCR DSMC MD

Onsager−Burnett NSF Burnett NCCR DSMC MD

−0.5 −0.4 −0.3 −0.2 −0.1

0

s

0.1

1.02 1 0.98

0.2

0.3

0.4

−0.5 −0.4 −0.3 −0.2 −0.1

0.5

1.06 1.05

0.2

0.3

0.4

0.5

0.2

0.3

0.4

0.5

1.01 Onsager−Burnett NSF Burnett NCCR DSMC MD

1 0.99

T*

1.04

p*

0.1

(b)

(a) 1.08 1.07

0

s

1.03

0.98 0.97

1.02 1.01

0.96

1 0.95

0.99 −0.5 −0.4 −0.3 −0.2 −0.1

0

s

(c)

0.1

0.2

0.3

0.4

0.5

Onsager−Burnett NSF Burnett NCCR DSMC MD

−0.5 −0.4 −0.3 −0.2 −0.1

0

s

0.1

(d)

Fig. 7.7 Variation of (a) velocity, (b) density, (c) pressure, and (d) temperature in the cross stream direction in force-driven compressible plane Poiseuille flow at Kn = 0.1. Comparison is made for OBurnett equations, Navier–Stokes equations, conventional Burnett equations, and NCCR equations and validation is performed against the DSMC and MD results. Figures taken from Jadhav et al. [78]

7.8 Summary In this chapter, the need for looking at variants of the Burnett and Grad equations is first explained and available variants of these equations are included. While several of these variants are obtained in a somewhat ad hoc manner, thermodynamically consistent Burnett-like and Grad-like equations have also been recently proposed. These new equations being consistent with Onsager’s principle are called OBurnett and O13 equations, respectively. The equations have been applied to two specific cases and satisfactory agreement of the available variants against benchmark data is noted. However, the results from these cases cannot be generalized in that if a particular variant works better in a given case than the others, it does not imply that variant is actually better. In fact, we urge the readers to apply the various available variants for their specific problem, and also try to arrive at better versions of these equations.

304

7 Alternate Forms of Burnett and Grad Equations 0

Onsager−Burnett NSF Burnett NCCR DSMC MD

0.3 0.2 0.1

−0.01 −0.02

* s yy

* s xy

−0.03 0

−0.04

−0.1

−0.05

−0.2

Onsager−Burnett NSF Burnett NCCR DSMC MD

−0.06

−0.3

−0.07

−0.5 −0.4 −0.3 −0.2 −0.1

0

s

0.1

0.2

0.3

0.4

0.5

−0.5 −0.4 −0.3 −0.2 −0.1

(a)

0

s

0.1

0.2

0.3

0.4

0.5

0.2

0.3

0.4

0.5

(b)

0.12 Onsager−Burnett NSF Burnett NCCR DSMC

0.1

0 −0.01 −0.02

0.08

* s zz

* s xx

−0.03

0.06

−0.04

0.04 −0.05 0.02

Onsager−Burnett NSF Burnett NCCR DSMC

−0.06 −0.07

0 −0.5 −0.4 −0.3 −0.2 −0.1

0

s

0.1

0.2

0.3

0.4

−0.5 −0.4 −0.3 −0.2 −0.1

0.5

0

s

0.1

(d)

(c) 0.12

q *y /[p(0) (RT (0)]

0.1 0.05

Onsager−Burnett NSF Burnett NCCR DSMC MD

0.1 [q *x ]0 /[p(0) (RT (0)]

0.15

0 −0.05 −0.1

0.06

0.04

0.02

−0.15 −0.5 −0.4 −0.3 −0.2 −0.1

0.08

Onsager−Burnett NSF Burnett NCCR DSMC MD

0 0

s

(e)

0.1

0.2

0.3

0.4

0.5

−0.5 −0.4 −0.3 −0.2 −0.1

0

s

0.1

0.2

0.3

0.4

0.5

(f)

∗ , (b) normal stress p ∗ , (c) normal stress p ∗ (d) normal Fig. 7.8 Variation of (a) shear stress pxy yy xx ∗ , (e) normal heat flux q ∗ , and (f) tangential heat flux q ∗ in the cross stream direction in stress pzz 2 1 force-driven compressible plane Poiseuille flow at Kn = 0.1. Comparison is made for OBurnett equations, Navier–Stokes equations, conventional Burnett equations, and NCCR equations and validation is performed against the DSMC and MD results. Note that stresses are represented as σ ’s. Figures taken from Jadhav et al. [78]

Chapter 8

Overview to Numerical and Experimental Techniques

In the previous chapters, the governing equation and analytical solution of the equation was presented for flows in relatively simple geometries. However, problems encountered in most practical situations are much more difficult, involving complex geometry, unclear boundary conditions, transition from one flow regime to another, and other complications. Obtaining an analytical solution in these cases is virtually not possible. One therefore resorts to either numerical methods for solution of the governing equation or experiments. In this chapter, we briefly comment on these techniques, while details can be found in specialized books and papers on the subject.

8.1 Numerical Approach In this section, we provide a brief overview of the popular DSMC and molecular dynamics techniques. The discussion on these techniques is far from exhaustive.

8.1.1 Direct Simulation Monte Carlo Method The Direct Simulation Monte Carlo (DSMC) method is the preferred choice for numerically simulating flow in microchannels, and as seen in previous chapters, analytical results in the high Knudsen number range are often benchmarked against DSMC data. In fact, it will perhaps not be an overstatement, that DSMC data virtually enjoys the same status as experimental data does in any other field! This statement also simultaneously underscores the absence of high-quality detailed experimental data in rarefied gas flows. DSMC is a particle based probabilistic numerical technique invented by Bird [21] and is particularly suited for simulation © Springer Nature Switzerland AG 2020 A. Agrawal et al., Microscale Flow and Heat Transfer, Mechanical Engineering Series, https://doi.org/10.1007/978-3-030-10662-1_8

305

306

8 Overview to Numerical and Experimental Techniques

Fig. 8.1 The molecular direct simulation Monte Carlo method (DSMC) for low-Reynolds-number flows in the transition regime, Zhao et al. [175]

of dilute gases. Unlike conventional numerical techniques, DSMC does not solve a set of partial differential equations, rather simulates the movement and collision of particles on a grid. The model therefore has some similarity with the molecular nature of matter, with each simulated molecule representing a number of real molecules. The ratio of real to simulated molecules is a simulation variable, and has to be fine-tuned for each simulation case. Similar is the case with the number of cells in which the entire computational domain is divided into. Each cell is further sub-divided into sub cells such that the particles which are nearer to each other and within a distance of one mean free path are chosen for collision, so as to achieve physically consistent simulations while reducing the simulation time for choosing the particle pair for collision. A step by step implementation of DSMC method is shown in Fig. 8.1. Note that the movement and collision of the particles in this method is uncoupled. That is, all particles are moved simultaneously to the next location as per the corresponding particle velocity, and then all collisions happen between the particle pairs selected on probabilistic basis. The time step involved in the computation is a fraction of the mean collision time between particles. Similarly, the cell size should be a fraction (typically one-third) of the mean free path of the gas. The gas–wall interaction can be modelled as either diffuse or specular (or some fraction thereof). The physical variables (density, velocity, temperature) and even higher order moments (stress and heat flux) can be obtained by taking time averaging of the appropriate quantities (i.e., using the equations for these quantities in terms of distribution function introduced in Chap. 5) for steady flows. Unsteady flow simulations however tend to become expensive as ensemble averaging of various quantities is required, for which multiple simulations have to be performed with different (random) initial conditions.

8.2 Experimental Approach

307

It has been shown that DSMC solves the Boltzmann equation in the limit of small time step, small grid size, and large number of simulated molecules. It is further claimed [21] that the method does not have some of the restrictions of the Boltzmann equation, and can therefore yield physical effects that cannot be modelled even within the framework of the Boltzmann equation.

8.1.2 Alternative Techniques Another popular alternative is Molecular Dynamics (MD) method. Unlike DSMC, MD is a deterministic method particularly suited for dense gases. The method involves solution of the Newton laws of motion as applied to gas particles, with the force acting on the particles being determined by the inter-molecular potential between the particles. The governing equations provided in Chaps. 5–7 can be solved using standard numerical simulation techniques of computational fluid dynamics. Since these models involve higher-order terms of Knudsen number, the envelope of Knudsen number over which these equations apply is larger than the Navier-Stokes equations. These numerical results can eventually supplement DSMC data especially in the slip and early transition regimes, regimes where DSMC computations are relatively expensive. On the other hand, DSMC computations are particularly well suited to simulate flows in the Knudsen number range of 0.01–10, and most simulations in the literature are in this range of Knudsen number.

8.2 Experimental Approach As seen earlier, microscale gaseous flows are characterized by relatively high Knudsen numbers. In practice, high Knudsen number can be realized either by reducing the length scale or by increasing the mean free path of the gas. Both these approaches have been followed in the literature while performing experiments on these flows.

8.2.1 Measurement at the Microscale A typical microchannel would typically have a width of 60–200 µm, depth of 15–100 µm, and length of 300–2000 µm. The ability to first fabricate such microchannels and then instrument them are the two important prerequisites for performing measurements at the microscale. Note that in order to find the variation of any quantity along either the depth or width, the spatial resolution of the instrument measuring that quantity should be about an order of magnitude smaller than these dimensions. Further, needless to emphasize, the probe introduced should not disturb/alter the flow.

308

8 Overview to Numerical and Experimental Techniques

Therefore, measurement of mass flow rate, pressure, and temperature is more straightforward as these can be done outside the microchannel, while measurement of local flow velocity or wall shear stress is challenging as the probe has to be introduced into the microchannel in these cases. Again, measurement of pressure or temperature at the inlet and outlet reservoirs is much more straightforward, as compared to measurement of local pressure or temperature in the microchannel. In order to measure the local pressure, taps can be added along the microchannel length as shown in Fig. 8.2a. Figure 2.12c shows the obtained pressure values from six pressure taps introduced by Pong et al. [88]. This data was subsequently employed by numerous researchers to benchmark their numerical and theoretical results. Wholefield pressure in the microchannel has also been measured using pressure sensitive paint by Huang et al. [76]. The arrangement of their setup is shown in Fig. 8.2b and a sample result in Fig. 8.2c. Thermocouples remain the preferred choice for measurement of temperature. Thermocouples are typically mounted in the inlet/outlet reservoirs for obtaining the inlet and outlet temperatures of the fluid, and on the outer wall of the microchannel for measuring the local temperature. Of course, these measurements require extremely fine thermocouples beads; 25 µm bead size thermocouples are now commercially available. Use of infrared camera, which can provide wholefield temperature information, besides being non-intrusive, is another commonly employed measurement technique. The surface whose temperature is to be measured is coated with a paint of known emissivity for obtaining reliable data. Micro-PIV (particle image velocimetry) is the preferred technique for obtaining velocity field in a microchannel. Here, fine oil droplets are added to the flow; these droplets scatter light (usually from twin Nd:YAG pulsed lasers), which is collected by a camera (or CCD). Two such images are collected and correlated to obtain the displacement of the droplets. Finally, knowing the time delay between the images gives the velocity of the droplets. From the assumption that the velocity of the droplet is equal to velocity of the fluid (an assumption that is satisfied when the Stokes number of the flow is small), one obtains the wholefield velocity information. However, although micro-PIV is a well-established experimental technique, there are only a few studies reporting measurement of gas flow in a microchannel, primarily because of the difficulty in carriage of oil-droplets by the flow at these small scales. However, measurement of gas flow rate is much more standard, and involves the use of mass flow controller of appropriate range and accuracy. The microchannels can be fabricated using wet-etching and photolithography processes. An example process flow involved in the fabrication process is shown in Fig. 8.3. The material of the substrate on which microchannels are fabricated varies from thermally conducting materials such as silicon and copper to nonconducting materials such as polydimethylsiloxane (PDMS) and poly(methyl methyl acrylate) (PMMA). Good thermal conductivity of the substrate is obviously required in heat transfer studies. Closing of the microchannel is another aspect which has to be carefully handled. For this, plasma-bonding and PDMS-PDMS bonding are some of the options.

8.2 Experimental Approach Fig. 8.2 (a) A close-up picture of transition section of a tested device with an included angle of 180◦ , Lee et al. [86]. (b) Schematic showing pressure sensitive paint (PSP) measurement, Huang et al. [76]. (c) Pressure map inside a microchannel with pressure ratio 4.74, Huang et al. [76]

309

310

8 Overview to Numerical and Experimental Techniques

Fig. 8.3 Fabrication procedure for microdevices, (a) SU8 spin coating on oxidized silicon wafer, and pre-exposure bake, (b) exposure, and post-exposure bake and (c) development, (d) casting of PDMS on SU8 master, (e) peeling off PDMS chip from SU8 master, (f) spin coating of PDMS on glass slide, Prabhakar et al. [118]

8.2.2 Measurement in Rarefied Gas Flows Working with reduced pressure (to increase the mean free path of the gas) while employing a scaled up facility (flow passage in conventional size range) makes measurements relatively easier [41, 137, 159]. Instruments for measurement of low pressure and low flow rate along with vacuum pumps required in such experiments are well developed and easily available in various ranges. The accuracy of vacuum pressure gauges and mass flow controllers can be very high, which helps in obtaining high-quality repeatable data. Leakage of ambient air into the test section is a concern that has to be carefully addressed in order to obtain reliable data. There are various procedures that can be used to estimate the amount of leakage—including use of helium detectors, monitoring the rate of increase of pressure in the system after shutting off all the valves, etc. The use of pitot tube is a cheaper and more convenient alternative than microPIV, although it is an intrusive and point-wise measurement technique. Pitot tube based measurements basically rely on the Bernoulli equation for determining the velocity from the measured pressures; however, the use of the Bernoulli equation in the case of rarefied gas can lead to substantial error. This is because the relatively

8.2 Experimental Approach

311

Fig. 8.4 (a) Schematic diagram of a pitot tube, Varade et al. [159]. (b) Schematic of a tube-in-tube heat exchanger, Demsis et al. [41]

low Reynolds number of the flow alters the flow pattern around the tube and slip at the walls, which change the pressure field at the tip of the tube. Recently, Varade et al. [159] showed that reliable measurements are possible with appropriate corrections to the Bernoulli equation for these viscous and rarefaction effects. The arrangement of pitot tube employed in this work is shown in Fig. 8.4a. Measurement of local pressure [137] and local temperature [73, 156–158] is possible because of the relatively large passage size. A tube-in-tube counter-flow heat exchanger for measurement of heat transfer coefficient is shown in Fig. 8.4b. Attempts have also been made to obtain the local flow velocity by employing a hot wire. Providing an optical window and undertaking laser based measurements can provide detailed information about the flow, although such measurements have rarely been reported.

312

8 Overview to Numerical and Experimental Techniques

8.3 Summary The purpose of this brief chapter is to provide a quick overview of the available options for performing numerical or physical experiments, with flows in the high Knudsen number regime. We noted that DSMC is a well-established numerical technique, while numerical solution of the higher-order models can further provide another compelling alternative way to study rarefied gas flows in complex situations. There is lot of possibility for improving the instrumentation for local flow measurements with substantially higher spatial resolution. There is also a rich possibility of developing optical based measurement techniques for measurement in low pressure gas flows.

Chapter 9

Summary and Future Research Directions

In the various chapters of this book, we have covered several different aspects of gas flow and heat transfer in the slip and transition regimes. In this last chapter, we present a broad summary of what is well understood and what remains to be better explored. We also identify some important research problems which can help in further development of this subject. We have considered flow in simple (straight) and complex (involving change in cross sectional, flow direction, etc.) passages using both analytical and/or experimental/simulation tools. Therefore, there are four broad categories in which these various scenarios can be classified: slip or transition regime, simple or complex passages, fluid flow or heat transfer aspects, and analytical or experimental/simulation tools. We first summarize what is known (or reasonably well known) about these categories: • Analytical solution for flow in simple passages in slip regime is available. • Experimental and DSMC data for flow in simple passages in slip and transition regimes is available to a reasonable extent. • Analytical solution for heat transfer in simple passages in slip regime is available. However, there is scope for considering several complicating effects together. • Several theoretical models for describing flow and heat transfer in the transition regime are available. However, the reliability of these models has to be thoroughly tested and a final word on the superset of the Navier-Stokes equations has to be proclaimed. We next summarize the major gaps (at least there is enough scope for further work in these areas): • Reliable experimental data for heat transfer in slip regime for simple passages is not available. • There is lot of scope (and need) for obtaining experimental flow and heat transfer data in the slip and transition regimes for complex passages.

© Springer Nature Switzerland AG 2020 A. Agrawal et al., Microscale Flow and Heat Transfer, Mechanical Engineering Series, https://doi.org/10.1007/978-3-030-10662-1_9

313

314

9 Summary and Future Research Directions

We now elaborate upon the above identified gaps by suggesting certain specific problems, which are particularly crucial for further development of this important subject. 1. More data on slip coefficients: As reviewed in Sect. 2.3.4, the value of tangential momentum accommodation coefficient or σv is available for a number of cases. However, such information has added to the need to resolve the dependency of slip coefficients (C1 and C2 ) on various parameters (esp. Knudsen number). 2. Do slip coefficients depend on the gas (and surface material)? This is a particularly crucial question, which has not received the attention that it should. A dependence of slip coefficient on the nature of the gas suggests that the friction factor and Nusselt number will additionally depend on the gas that we are working with, and will therefore substantially complicate the analysis of slip flows. Clearly the question is of fundamental interest as well as relevant from the viewpoint of developing sensors (and other applications) based on this difference. 3. Data for alternate cross sections: Although analysis and data are available for flow in straight passages, such information is limited to relatively simple cross sections (such as circular and rectangular). However, cross section of practical microchannels and microdevices is not limited to these simple shapes, rather may be trapezoidal, triangular, rhombic, and even irregular. Therefore, there is scope for obtaining reliable experimental and DSMC data for other cross sections of the flow passage and with surface modifications (esp. surface roughness). 4. Single effect taken at a time in heat transfer problems: As already stated, there is scope for considering several complicating effects (temperature jump, viscous dissipation, property variation, axial conduction, conjugate effect, end effects) together to make heat transfer analysis more realistic. 5. Paucity of experimental heat transfer data: The need for reliable experimental data is particularly felt in heat transfer because it not a priori obvious which of the multiple effects that co-exist can be neglected. This may be reason for the poor agreement between available theoretical and experimental findings. Further, the overall heat transfer coefficient has to be supplemented with local heat transfer coefficient measurements wherever possible. 6. Development of miniature sensors for local flow and heat transfer measurements: Having reliable local information is advantageous and therefore this requirement is obvious. The spatial resolution of most current flow and temperature sensors is at best several hundreds of microns; the resolution needs to be reduced to a few microns for local measurements to be meaningful. 7. Development of sensors for measurement of additional variables: As mentioned in Chap. 8, the current measurements are limited to overall behavior and a limited number of flow/thermal parameters. The measurements need to be expanded to include variables such as: velocity, wall shear stress, heat flux, etc.

9 Summary and Future Research Directions

315

8. Generation of database of DSMC computations: It would be useful to perform highly accurate DSMC computations for a range of problems and flow conditions. Any proposed new equations should then be tested against this database. This database is also useful because variability in DSMC computations from various groups exists. The theoretical research group can then focus exclusively on deriving new equations. The same requirement goes for generating high quality experimental data (for corresponding to DSMC database) and even other problems. 9. Equations should be benchmarked against more involved problems: Newly proposed equations need to be tested over a wider range of problems, and a wider range of parameter values to establish confidence in them. With the generation of comprehensive database mentioned above in point 8, this should become possible. 10. Establishment of accurate boundary conditions for higher order models: As seen in Chaps. 5–7, all higher order models (other than OBurnett equations) involve higher order derivatives and therefore need more number of boundary conditions than the Navier-Stokes equations. Procedure to obtain additional boundary conditions is therefore required. 11. Dependency of solution on the type of gas needs testing: As mentioned above, a final word on the correct form of the superset of the Navier-Stokes equations still needs to be proclaimed. Moreover a dependence of the form of the governing equation on the gas is felt (through the inter-molecular potential). Therefore the dependence of boundary condition (commented in point 2 above) and governing equation on the gas needs to be unambiguously resolved.

Appendix A

Appendix to Basic Tensor Algebra

Tensor notation greatly helps to represent equations in compact form. In this brief introduction to tensor algebra, we present the basic rules to be followed along with the general operations performed on tensors. Although this section aims to aid understanding of the tensor algebra employed in the book, the section is a standalone introduction to tensors and therefore other readers should also find it useful.

A.1 Introduction to Scalar, Vector, and Tensor A scalar quantity can be considered as a zeroth order tensor that is completely defined by magnitude only and is independent of coordinate system. A vector is a first order tensor that has magnitude and a direction and requires three components along with three specified coordinate directions for its complete description. Generally, it is represented by a boldface symbol, for example, x for position vector, u for bulk velocity, q for heat flux vector, etc. In three-dimensional physical space, a vector x can be represented as x = x1 e1 + x2 e2 + x3 e3 where x1 , x2 , x3 are the components of the vector x and e1 , e2 , e3 are the unit vectors in the direction of three mutually perpendicular axes. In tensorial notation, vector x is represented as xi where i is known as the free index and goes from 1 to 3 in a three-dimensional Cartesian coordinate system. Similarly, a unit vector ei refers to any one of the unit vectors e1 , e2 , or e3 . According to formal definition, a vector is a quantity whose components change like the components of a position vector under the rotation of the coordinate system [85]. To illustrate this, consider a coordinate system in two dimensions, O 12 which is rotated to obtain a new coordinate system O 1 2 as shown in Fig. A.1. Let the © Springer Nature Switzerland AG 2020 A. Agrawal et al., Microscale Flow and Heat Transfer, Mechanical Engineering Series, https://doi.org/10.1007/978-3-030-10662-1

317

318

A Appendix to Basic Tensor Algebra

Fig. A.1 Coordinate system O 12 is rotated to O 1 2

components of a position vector x in the old and new coordinate systems be xi and xi where i goes from 1 to 2. The cosine of the angle β’s between the old i and new j axes is represented by Cij . In matrix form, Cij can be represented as cos β11 cos β12 C11 C12 = = cos βij (A.1) Cij = C21 C22 cos β21 cos β22 The first index of a matrix Cij refers to the old axes while the second index refers to the new axes. Note that Cij is not equal to Cj i . The components in the new coordinate system are related to the components in the old system as x1 = OB = OA + AB = OC cos β11 + PC sin β11

[∵ AB = CD]

= x1 cos β11 + x2 sin β11 = x1 cos β11 + x2 cos β21

[β11 = π/2 − β21 , ∴ sin β11 = cos β21 ]

= x1 C11 + x2 C21

(A.2)

Similarly, for the second component, x2 = x2 cos β22 + x1 cos β12 = x2 C22 + x1 C12

(A.3)

where we have used β11 = β22 = β12 − π/2 and sin β11 = sin(β12 − π/2) = − cos β12 . Generalizing to three dimensions, we can write the three components in the new coordinate system as x1 = x1 C11 + x2 C21 + x3 C31 =

3

Ci1 xi i=1

A.1 Introduction to Scalar, Vector, and Tensor

319 3

x2 = x1 C12 + x2 C22 + x3 C32 =

Ci2 xi i=1

x3

3

= x1 C13 + x2 C23 + x3 C33 =

Ci3 xi i=1

In compact tensorial notation, the above three equations can be represented in a single equation as 3

xj =

Cij xi

(A.4)

i=1

where the index j goes from 1 to 3 and called as free index. The index i is repeated in the same term of the right-hand side and summation operation is carried out for all values of this repeated index. Such an index is called as dummy index and choice of letter for a dummy index is immaterial as long as it is different from a free index. A generally accepted rule is to suppress the summation symbol in a term where an index is appearing twice. Such a rule is called as Einstein summation convention wherein for a repeated index, summation over all three values of that index is implied. With this convention, we can simply write as xj = Cij xi

(A.5)

Equation (A.5) gives transformation for a position vector in a new coordinate system under the rotation operation. In a similar way, any quantity can be called as a Cartesian vector when its components transform like a position vector under the rotation of the coordinate system. Hence, u is a vector if its components transform as uj = Cij ui

(A.6)

A second order tensor has nine components and two directions associated with it. The stress tensor, strain rate tensor, and velocity gradient tensor are frequently encountered second order tensors in fluid mechanics. In matrix form, a second order tensor can be represented as ⎡ τ11 τij = ⎣τ21 τ31

τ12 τ22 τ32

⎤ τ13 τ23 ⎦ τ33

(A.7)

Hence, for complete specification of a second order tensor, all these nine components must be determined. Similar to the definition of a vector, a transformation rule can be defined for a second order tensor as

320

A Appendix to Basic Tensor Algebra

τij = Ckj Clj τkl

(A.8)

Further generalizing for an nth-order tensor, it has 3n components which needs to be completely determined for complete specification of the tensor and n directions associated with it. The transformation rule for an nth order tensor can be written as Ai1 i2 ...in = Cj1 i1 Cj2 i2 . . . Cjn in Aj1 j2 ...jn

(A.9)

Note that in transformation of a second order tensor, cos2 , product of cos and sin, sin2 terms are involved. This is because there are two directions for a second order tensor, and transformation of each direction involves a cos or sin term. Similarly, transformation of an nth-order tensor involves product of n cos and sin terms.

A.2 Kronecker Delta The Kronecker delta δij is defined as 2 δij =

1,

if i = j,

0,

if i = j.

(A.10)

In matrix notation, we can write as ⎡ δ11 δij = ⎣δ21 δ31

δ12 δ22 δ32

⎤ ⎡ 1 δ13 δ23 ⎦ = ⎣0 0 δ33

0 1 0

⎤ 0 0⎦

(A.11)

1

According to the above definition, it can be shown that the components of δij remain unchanged by a rotation of the coordinate system. Hence, δij is a special isotropic tensor in the sense that it is the only isotropic tensor of second order. The Kronecker delta δij establishes the orthonormality property of the unit vectors such that ei ej = δij

(A.12)

The Kronecker delta is applied extensively in the following operation: For a term containing δij with one of its indices being repeated, then we can simply replace that repeated index by the other index of δij . For example, qi δij = q1 δ1j + q2 δ2j + q3 δ3j

(A.13)

The right-hand side is q1 when j = 1, q2 when j = 2, and q3 when j = 3, thus qi δij = qj

(A.14)

A.3 Some Important Results

321

Alternatively, the Kronecker delta can be used to change the index of an expression as ri = rj δij

(A.15)

cik = cij δj k

(A.16)

A.3 Some Important Results Some important results are summarized below for easy reference: 1.

∂xi = δij ∂xj

(A.17)

To prove this identity, we simply expand in the regular matrix notation as ⎡

∂x1 ⎢ ∂x1 ⎢ ∂x ∂xi ⎢ 2 =⎢ ⎢ ∂x1 ∂xj ⎣ ∂x3 ∂x1

∂x1 ∂x2 ∂x2 ∂x2 ∂x3 ∂x2

⎤ ∂x1 ⎡ ∂x3 ⎥ 1 ∂x2 ⎥ ⎥ ⎣ ⎥= 0 ∂x3 ⎥ 0 ∂x3 ⎦

0 1 0

⎤ 0 0⎦ = δij 1

∂x3

where we have utilized the fact that x1 , x2 , and x3 are independent variables so that we can set terms like ∂x1 /∂x3 to 0. 2. δii = 3

(A.18)

Since the index i is repeated, a summation is implied over i as 3

δii =

δii = δ11 + δ22 + δ33 = 1 + 1 + 1 = 3 i=1

where we have used definition of Kronecker delta (Eq. (A.10)). 3. δij δj k = δik

(A.19)

The index j is repeated, so simply replacing the index j by other index k, we get the required expression (i.e. using the result in Eq. (A.14)). 4.

∂x 2 = 2xi ∂xi

(A.20)

322

A Appendix to Basic Tensor Algebra

Introducing k as dummy variable, ∂x 2 ∂xk ∂xk ∂xk xk = = xk + xk = 2xk δik = 2xi ∂xi ∂xi ∂xi ∂xi where we have used results obtained in Eqs. (A.14) and (A.17). The most useful takeaways from this section are the ideas of free index and repeated index. Note that in any term of a tensorial equation, index cannot occur more than twice. If it occurs, that equation is incorrect in tensorial structure. Further, each term of an equation should have exactly the same free indices for that equation to be tensorially consistent.

A.4 Decomposition of Tensors Most of the tensors that we deal with are symmetric and trace-free; we therefore examine the procedure for making tensors symmetric and trace-free in this section. Consider a second order tensor Aij and form a new tensor, say Aj i by interchanging indices i and j . If the components of Aj i are equal to Aij , then Aij is called symmetric in the indices i and j , i.e., Aij = Aj i . Given any nth order tensor, its symmetric part can be obtained as A(i1 i2 ···in ) =

1 Ai1 i2 ···in + Ai2 i1 ···in + · · · all possible permutations of indices n! (A.21)

where rounded brackets denote the symmetric part of the tensor. Accordingly, the symmetric part of a second order tensor can be obtained as A(ij ) =

1 Aij + Aj i 2

(A.22)

For the third order tensor, its symmetric part is A(ij k) =

1 Aij k + Akij + Aj ki + Aikj + Aj ik + Akj i 6

(A.23)

There are nine and twenty seven independent components for a general nonsymmetric second order and third order tensor, respectively. After making symmetric an nth order tensor, there are 12 (n + 1)(n + 2) independent components of the tensor. As such, there are six independent components for a second order tensor, ten independent components for a third order tensor, and fifteen independent components for a fourth order tensor. A second order tensor Aij is called trace-free (or divergence-free) if the sum of the diagonal elements is zero, i.e., Aii = 0. The trace-free part of a symmetric first,

A.4 Decomposition of Tensors

323

second, and third order tensor can be obtained as Ai = Ai

(A.24)

1 Aij = A(ij ) − Akk δij 3 1 Aij k = A(ij k) − (A(ill) δj k + A(j ll) δik + A(kll) δij ) 5

(A.25) (A.26)

where symmetric parts A(ij ) and A(ij k) are given by Eqs. (A.22) and (A.23), respectively. To illustrate in matrix form, a general second order tensor Aij can be represented as ⎡ A11 Aij = ⎣A21 A31

A12 A22 A32

⎤ A13 A23 ⎦ A33

(A.27)

Using Eq. (A.22), its symmetric part A(ij ) can be represented in matrix form as ⎡

A(ij )

A11 = ⎣(A21 + A12 )/2 (A31 + A13 )/2

(A12 + A21 )/2 A22 (A32 + A23 )/2

⎤ (A13 + A31 )/2 (A23 + A32 )/2⎦ A33

(A.28)

Using Eq. (A.25), symmetric and trace-free part Aij can be obtained as ⎤ ⎡ (A12 + A21 )/2 (A13 + A31 )/2 A11 − (A11 + A22 + A33 )/3 ⎥ ⎢ Aij = ⎣ A22 − (A11 + A22 + A33 )/3 (A23 + A32 )/2 (A21 + A12 )/2 ⎦ (A32 + A23 )/2 A33 − (A11 + A22 + A33 )/3 (A31 + A13 )/2 (A.29)

A general second order tensor Aij has nine (3n where n = 2) independent components while its symmetric part A(ij ) has six independent components. Further, making it trace-free, Aij has five independent components, three off-diagonal components, A12 , A13 , and A23 plus two of the three diagonal components. Notice that the third diagonal component can be determined by the relation A11 + A22 + A33 = 0. A third order tensor Aij k has twenty-seven (3n where n = 3) independent components and can be represented in matrix form as

Aij k

⎤⎡ ⎤⎡ ⎤ ⎡ A111 A112 A113 A211 A212 A213 A311 A312 A313 = ⎣A121 A122 A123 ⎦ ⎣A221 A222 A223 ⎦ ⎣A321 A322 A323 ⎦ A331 A332 A333 A231 A232 A233 A131 A132 A133 (A.30)

324

A Appendix to Basic Tensor Algebra

The symmetric part of third order tensor A(ij k) has ten independent components (underlined terms) and can be obtained using Eq. (A.23) as A(111) = A(112) = = A(113) = = A(122) = = A(123) = A(133) = = A(222) = A(223) = = A(233) = = A(333) =

1 (A111 + A111 + A111 + A111 + A111 + A111 ) = A111 6 1 (A112 + A211 + A121 + A121 + A112 + A211 ) 6 1 (A112 + A211 + A121 ) 3 1 (A113 + A311 + A131 + A131 + A113 + A311 ) 6 1 (A113 + A311 + A131 ) 3 1 (A122 + A212 + A221 + A212 + A221 + A122 ) 6 1 (A122 + A212 + A221 ) 3 1 (A123 + A312 + A231 + A132 + A213 + A321 ) 6 1 (A133 + A313 + A331 + A313 + A331 + A133 ) 6 1 (A133 + A313 + A331 ) 3 1 (A222 + A222 + A222 + A222 + A222 + A222 ) = A222 6 1 (A223 + A322 + A232 + A232 + A223 + A322 ) 6 1 (A223 + A322 + A232 ) 3 1 (A233 + A323 + A332 + A323 + A332 + A233 ) 6 1 (A233 + A323 + A332 ) 3 1 (A333 + A333 + A333 + A333 + A333 + A333 ) = A333 6

A.4 Decomposition of Tensors

325

The non-underlined terms (in Eq. (A.30)) can be obtained from the underlined terms as A(112) = A(211) = A(121) A(113) = A(311) = A(131) A(122) = A(212) = A(221) A(123) = A(312) = A(231) = A(132) = A(213) = A(321) A(133) = A(313) = A(331) A(223) = A(322) = A(232) A(233) = A(323) = A(332)

(A.31)

Further to make trace-free, we can perform contraction as i = j , or j = k, or k = i; as such, there will be seven (= 10 − 3) independent components for the trace-free part of third order tensor Aij k . The elements of the three diagonals are related as A111 + A122 + A133 = 0

⇒

A133 = −A111 − A122

(A.32)

A211 + A222 + A233 = 0

⇒

A222 = −A211 − A233

(A.33)

A311 + A322 + A333 = 0

⇒

A333 = −A311 − A322

(A.34)

Hence, the components A133 , A222 , and A333 are not exactly independent since they can be determined from the remaining components. As a result, we have A111 , A112 , A113 , A122 , A123 , A223 , and A233 as seven independent components of trace free third order tensor. Using Eq. (A.26), components of Aij k can be obtained as 1 A(1ll) δ11 + A(1ll) δ11 + A(1ll) δ11 5 3 A(111) + A(122) + A(133) = A(111) − 5 1 2 = A111 − [A122 + A212 + A221 + A133 + A313 + A331 ] 5 5 1 A(1ll) δ12 + A(1ll) δ12 + A(2ll) δ11 = A(112) − 5 1 A(211) + A(222) + A(233) = A(112) − 5 4 1 1 A222 + (A233 + A323 + A332 ) = (A112 + A211 + A121 ) − 15 5 3

A111 = A(111) −

A112

326

A Appendix to Basic Tensor Algebra

1 A(1ll) δ13 + A(1ll) δ13 + A(3ll) δ11 5 1 = A(113) − A(311) + A(322) + A(333) 5 4 1 1 = (A113 + A311 + A131 ) − (A322 + A232 + A223 ) + A333 15 5 3

A113 = A(113) −

1 A(1ll) δ22 + A(2ll) δ12 + A(2ll) δ21 5 1 = A(122) − A(111) + A(122) + A(133) 5 4 1 1 A111 + (A133 + A313 + A331 ) = (A122 + A212 + A221 ) − 15 5 3

A122 = A(122) −

A123 = A(123) −

1 A(1ll) δ23 + A(2ll) δ13 + A(3ll) δ21 5

1 (A123 + A312 + A231 + A132 + A213 + A321 ) 6 1 A(2ll) δ22 + A(2ll) δ22 + A(2ll) δ22 = A(222) − 5 3 = A(222) − A(211) + A(222) + A(233) 5 1 2 = A222 − [A211 + A121 + A112 + A233 + A323 + A332 ] 5 5 1 A(2ll) δ23 + A(2ll) δ23 + A(3ll) δ22 = A(223) − 5 1 = A(223) − A(311) + A(322) + A(333) 5 4 1 1 = (A223 + A322 + A232 ) − (A311 + A131 + A113 ) + A333 15 5 3 =

A222

A223

∂Qij which is formed by taking the gradient of a ∂xk symmetric and trace-free second order tensor Qij . Its symmetric part is obtained as Consider a third order tensor

∂Q(ij 1 = ∂xk) 6

∂Qj k ∂Qj i ∂Qkj ∂Qij ∂Qki ∂Qik + + + + + ∂xk ∂xj ∂xi ∂xj ∂xk ∂xi ∂Qij ∂Qj k ∂Qki 1 [∵ Qij = Qj i ] 2 +2 +2 = 6 ∂xk ∂xj ∂xi ∂Qj k 1 ∂Qij ∂Qki = + + 3 ∂xk ∂xj ∂xi

(A.35)

A.5 Some Examples

327

Its trace-free and symmetric part is then obtained using Eq. (A.26) as

∂Qj k ∂Qij 1 ∂Qil ∂Qki ∂Qil ∂Qll − δj k + + + + ∂xk ∂xj ∂xi 15 ∂xl ∂xl ∂xi ∂Qj l ∂Qj l ∂Qkl ∂Qll ∂Qkl ∂Qll δik + δij + + + + + ∂xl ∂xl ∂xj ∂xl ∂xl ∂xk

∂Qj k ∂Qj l 2 ∂Qil 1 ∂Qij ∂Qki ∂Qkl − = + + δj k + δik + δij 3 ∂xk ∂xj ∂xi 15 ∂xl ∂xl ∂xl (A.36)

∂Qij 1 = ∂xk 3

where we have used Qll = 0 (since Qij is trace-free and symmetric). Likewise, for the fourth order symmetric tensor A(ij kl) , there are fifteen independent components [1/2(n+1)(n+2); n = 4]. For the trace-free part, we can perform contraction as i = j , or i = k or i = l or j = k or j = l, or k = l (six traces); as such, its trace-free part, Aij kl will have nine (15 − 6) independent components.

A.5 Some Examples To illustrate the tensor operations, we consider three examples as below: 1. We first consider expression for viscous dissipation ", (Eq. (2.7)) and using tensorial operations, show that " is always a positive quantity (a result of profound significance!). ∂uk 2 2 " = 2μeij eij − μ 3 ∂xk 1 ∂uk ∂ul = 2μ eij eij − 3 ∂xk ∂xl 1 ∂uk ∂ul 2 ∂uk ∂ul = 2μ eij eij + 3− 9 ∂xk ∂xl 3 ∂xk ∂xl 1 ∂uk ∂ul 2 1 ∂ui ∂ui ∂ul = 2μ eij eij + δij δij − + 9 ∂xk ∂xl 3 2 ∂xi ∂xi ∂xl [∵ δij δij = δii = 3] ∂uj ∂ul 1 ∂uk ∂ul 2 1 ∂ui δij δij − δij + δij = 2μ eij eij + 9 ∂xk ∂xl 3 2 ∂xj ∂xi ∂xl ∂ul 1 ∂uk ∂ul 2 δij δij − eij δij = 2μ eij eij + 9 ∂xk ∂xl 3 ∂xl 2 1 ∂uk δij (A.37) = 2μ eij − 3 ∂xk

328

A Appendix to Basic Tensor Algebra

∂uj ∂ui is strain rate tensor. Note the use of Kronecker where eij = 12 ∂x + ∂xi j delta in changing the index of the term in step 5. For example, term ∂ui /∂xi is transformed to eij by introducing δij . 2. As a next exercise, we expand " in terms of velocity gradients using eij = 1 2

∂ui ∂xj

+

∂uj ∂xi

as

∂uk 2 2 " = 2μeij eij − μ 3 ∂xk ∂uj 1 ∂ui ∂uj 2 ∂uk 2 1 ∂ui − μ = 2μ + + 2 ∂xj ∂xi 2 ∂xj ∂xi 3 ∂xk ∂uj ∂uj ∂ui 2 ∂ui ∂uk 2 1 − μ = μ + + 2 ∂xj ∂xi ∂xj ∂xi 3 ∂xk ∂uj ∂uj 2 ∂ui ∂ui ∂uk 2 1 ∂ui ∂uj − μ = μ +2 + 2 ∂xj ∂xj ∂xj ∂xi ∂x ∂x 3 ∂xk ) i*+ i, i→j ;j →i

∂ui ∂ui ∂ui ∂uj =μ + ∂xj ∂xj ∂xj ∂xi

∂uk 2 2 − μ 3 ∂xk

(A.38)

Note that there are two repeated indices in the first two terms which implies summation over i (with i going from 1 to 3) and summation over j (with j going from 1 to 3) giving a total of nine terms. Performing summation first for j and then for i (the sequence of summation does not matter), 3

3

(a) = i=1 j =1

∂ui ∂ui ∂ui ∂ui ∂ui ∂ui = + + ∂x1 ∂x1 ∂x2 ∂x2 ∂x3 ∂x3 i=1 ∂u1 ∂u1 ∂u2 ∂u2 ∂u3 ∂u3 ∂u2 ∂u2 ∂u3 ∂u3 ∂u1 ∂u1 + + + + + = ∂x1 ∂x1 ∂x1 ∂x1 ∂x1 ∂x1 ∂x2 ∂x2 ∂x2 ∂x2 ∂x2 ∂x2 ∂u1 ∂u1 ∂u2 ∂u2 ∂u3 ∂u3 + + + ∂x3 ∂x3 ∂x3 ∂x3 ∂x3 ∂x3 2 2 ∂u1 ∂u2 ∂u3 2 ∂u1 2 ∂u2 2 ∂u3 2 = + + + + + ∂x1 ∂x1 ∂x1 ∂x2 ∂x2 ∂x2 2 2 2 ∂u1 ∂u2 ∂u3 + + + (A.39) ∂x3 ∂x3 ∂x3 3

∂ui ∂ui ∂xj ∂xj

A.5 Some Examples 3

3

(b) = i=1 j =1

329

∂ui ∂uj ∂xj ∂xi

∂ui ∂u1 ∂ui ∂u2 ∂ui ∂u3 = + + ∂x1 ∂xi ∂x2 ∂xi ∂x3 ∂xi i=1 ∂u1 ∂u2 ∂u2 ∂u1 ∂u3 ∂u1 ∂u2 ∂u2 ∂u3 ∂u2 ∂u1 ∂u1 + + + + + = ∂x1 ∂x1 ∂x1 ∂x2 ∂x1 ∂x3 ∂x2 ∂x1 ∂x2 ∂x2 ∂x2 ∂x3 ∂u1 ∂u3 ∂u2 ∂u3 ∂u3 ∂u3 + + + ∂x3 ∂x1 ∂x3 ∂x2 ∂x3 ∂x3 2 ∂u1 ∂u2 2 ∂u3 ∂u2 ∂u2 ∂u1 ∂u3 ∂u1 ∂u1 ∂u2 = + + + + + ∂x1 ∂x1 ∂x2 ∂x1 ∂x3 ∂x2 ∂x1 ∂x2 ∂x2 ∂x3 2 ∂u3 ∂u1 ∂u3 ∂u2 ∂u3 + + + (A.40) ∂x3 ∂x1 ∂x3 ∂x2 ∂x3 3

3

and (c) = k=1

∂uk ∂xk

2

=

∂u1 ∂u2 ∂u3 + + ∂x1 ∂x2 ∂x3

2 (A.41)

Combining all the terms (a), (b), and (c), and completing squares for some of the terms, one can easily obtain " =2μ

∂u1 ∂x1

2

+

∂u2 ∂x2

2

+

∂u3 ∂x3

2

∂u2 2 ∂u1 + +μ ∂x2 ∂x1

∂u1 ∂u2 ∂u3 ∂u3 2 ∂u1 2 2 ∂u2 ∂u3 2 +μ + +μ + − μ + + ∂x3 ∂x2 ∂x1 ∂x3 3 ∂x1 ∂x2 ∂x3 (A.42) On replacing (u1 , u2 , u3 ) by (u, v, w), and (x1 , x2 , x3 ) by (x, y, z), Eq. (A.42) is same as that of Eq. (2.24). 3. Consider a complex equation (an evolution equation for symmetric and trace-free third order tensor mij k ) in index notation as pij ∂pkl ∂vij kl Dmij k ∂ lnρ 3 ∂Qij −3 − 3pij RT + + Dt ρ ∂xl ∂xk ∂xl 7 ∂xk + 3RT

∂pij ∂uk ∂ul 12 ∂uj 3p mij k + 3mlij + mij k + qi =− ∂xk ∂xl ∂xl 5 ∂xk 2μ (A.43)

330

A Appendix to Basic Tensor Algebra

We list some of the common rules as: a. Each term in an equation must be of the same tensorial order. In Eq. (A.43), all the individual terms are third order tensors. In other words, the three free indices i, j , and k must appear in all the terms of the equation. ∂v b. In the term, ∂xijlkl , l is a dummy index. Since the choice of a dummy index ∂v

∂v

is immaterial, we can write ∂xijlkl = ∂xijrkr . However, note that the choice of dummy index should not clash with the free indices, i.e., dummy index should be selected other than i, j , and k for the above example considered. ∂v c. An index cannot appear more than twice, for example, the term ∂xijlll is incorrect, since index l is appearing thrice in a given term. In order to illustrate the usefulness of the tensor algebra, it is worthwhile mentioning that Eq. (A.43) represents the evolution equations of seven independent components of mij k . Recall that mij k is symmetric and trace-free and therefore has seven independent components: m111 , m112 , m113 , m122 , m123 , m222 , and m223 . With the help of Einstein summation convention and angular bracket notation, many of the terms can be represented in a compact form. For example, Einstein ∂v summation convention is implied in the term ∂xijlkl which upon expansion is actually ∂vij k1 ∂vij k2 ∂vij k3 ∂vij kl = + + ∂xl ∂x1 ∂x2 ∂x3 The usefulness of the angular bracket notation can be illustrated by consid∂Q ering the term ∂xkij . Considering Qij to be symmetric, the symmetric part of the third order tensor obtained by taking gradient of Qij using Eq. (A.23) can be written as ∂Q(ij ∂Qj k ∂Qj i ∂Qkj 1 ∂Qij ∂Qki ∂Qik (A.44) = + + + + + ∂xk) 6 ∂xk ∂xj ∂xi ∂xj ∂xk ∂xi Since second order tensor Qlm is symmetric, we can interchange the indices as Qlm = Qml . Accordingly, we get ∂Q(ij ∂Qij ∂Qj k 1 ∂Qki 2 = +2 +2 ∂xk) 6 ∂xk ∂xj ∂xi ∂Qj k ∂Qki 1 ∂Qij + + = 3 ∂xk ∂xj ∂xi

(A.45)

A.6 General Tensor Operations

331

Its trace-free and symmetric part is then obtained using Eq. (A.26) as

∂Qj k ∂Qij 1 ∂Qil ∂Qki ∂Qil ∂Qll − δj k + + + + ∂xk ∂xj ∂xi 15 ∂xl ∂xl ∂xi ∂Qj l ∂Qj l ∂Qkl ∂Qll ∂Qkl ∂Qll δik + δij + + + + + ∂xl ∂xl ∂xj ∂xl ∂xl ∂xk (A.46)

∂Qij 1 = ∂xk 3

Hence, with this example, it is clear that the Einstein summation convention and angular bracket notation greatly help to represent the equations in a compact form.

A.6 General Tensor Operations In this section, several useful tensor operations are compiled. The most important for this book being the gradient, divergence, and Laplacian operations. The other operations are less frequently employed here and are given for the sake of completeness [116]. 1. Addition rule: Two tensors strictly of the same order can be added or subtracted and the resulting tensor is of the same order. 2. Product rule: A product of mth order tensor with an nth order tensor gives a tensor of order (m + n). For example, a tensor product of a first order tensor qi with a second order tensor Pij gives a third order tensor, say rij k . In terms of components, we can write rij k = qi Pij

(A.47)

3. Contraction: In contraction operation, two indices of a tensor are equated, and a summation is then performed over this repeated index. For example, for a second order tensor Pij , we equate i = j and obtain Pii as Pii = P11 + P22 + P33 The contraction operation decreases the order of the tensor by 2. In the above example, performing contraction of a second order tensor results in a scalar. Similarly, contracting a fourth order tensor gives a second order tensor. 4. Inner product: The inner product between a second order tensor Aij and third order tensor Bj kl results in a third order tensor. In terms of components, Cikl = Aij Bj kl

(A.48)

332

A Appendix to Basic Tensor Algebra

Generalizing the rule, the inner product of an nth order tensor and mth order tensor (n, m ≥ 1) results in a tensor of order n + m − 2. The dot product of two vectors is probably a good example of inner product, where a scalar is obtained after the operation. Here, n = m = 1 and the inner product is of order zero (1 + 1 − 2 = 0). 5. Division: Though addition, subtraction, and product operations are well defined in tensor algebra, there is no tensor operation corresponding to division. 6. Gradient: The gradient operator ‘del’ (a vector operator) is defined as ∇ ≡ ei

∂ ∂xi

(A.49)

The gradient of a tensor, say a second order tensor Aij , is obtained by applying the gradient operator and gives a third order tensor as Bij k =

∂Aij ∂xk

(A.50)

In general, the gradient operation on an nth order tensor gives a tensor of order n+1. It is easy to see that the repeated application of a gradient operation yields still higher derivatives. 7. Divergence: The inner product of a tensor and operator ∇ is termed as divergence. The divergence operation on an nth order tensor gives a tensor of order n − 1. Thus, the divergence operation reduces the order of the tensor by 1. Probably, the best example is the divergence of the velocity vector which is given as ∇ ·u≡

∂uk ∂u1 ∂u2 ∂u3 = + + ∂xk ∂x1 ∂x2 ∂x3

(A.51)

8. Laplacian: The Laplacian is obtained by taking the dot product of operator ∇ with itself as ∇ 2 = ∇ · ∇ = ei

∂ ∂ ∂2 ∂2 ∂2 · ej = ei · ej = δij = ∂xi ∂xj ∂xi ∂xj ∂xi ∂xj ∂xi ∂xi (A.52)

where we used Eqs. (A.14) and (A.12). 9. Gauss’s theorem: This theorem relates a volume integral to a surface integral and is applicable to tensors of any order. According to this theorem, V

∂pij dV = ∂xk

pij nk dA A

(A.53)

A.6 General Tensor Operations

333

where pij is a second order tensor, A is a smooth surface that encloses a volume V , and nk denotes the unit outward-pointing normal on A. When the indices j and k are contracted, we obtain the divergence theorem. 10. Doubly contracted product: Consider a symmetric tensor pij and any tensor Aij . The doubly contracted product which gives a scalar is defined as P ≡ pij Aij

(A.54)

According to the rule, the doubly contracted product of a symmetric tensor pij with any tensor Aij is equal to pij times the symmetric part of Aij , P ≡ pij Aij = pij A(ij )

(A.55)

Appendix B

Appendix to Burnett Equations

B.1 Derivation of f (0) The quantity log f is a summational invariant for collisions. Hence, it should be a linear combination of the three summational invariants ψ (i) = {1, mc, 12 mc2 } and can be written as 1 log f = Σα (i) ψ (i) = α (1) + α (2) .mc − α (3) mc2 2

(B.1)

Since log f is a scalar, α (1) and α (3) are scalars and α (2) is a vector. Since the state of the gas is uniform and steady, all three must be independent of x and t. 1 log f = α (1) + m(αx(2) u + αy(2) v + αz(2) w) − α (3) m(u2 + v 2 + w 2 ) 2

(B.2)

Considering the following terms separately, ⎡ ⎤ (2) (2) 2 (2) 2 α α 1 1 α x x x ⎦ mαx(2) u − α (3) mu2 = − α (3) m ⎣u2 − 2u (3) + − 2 2 α α (3) α (3) ⎡ ⎤ (2) 2 (2) 2 αx 1 (3) ⎣ αx ⎦ =− α m u − (3) − 2 α α (3) Simplifying the other components in a similar way, (B.2) becomes ⎡ ⎤ (2) 2 (2) 2 (2) 2 αy αx 1 αz log f = log α (0) − α (3) m ⎣ u − (3) + v − (3) + w − (3) ⎦ 2 α α α

© Springer Nature Switzerland AG 2020 A. Agrawal et al., Microscale Flow and Heat Transfer, Mechanical Engineering Series, https://doi.org/10.1007/978-3-030-10662-1

335

336

B Appendix to Burnett Equations

where log α (0)

⎡ (2) 2 ⎤ (2) 2 (2) 2 α α α 1 y x z ⎦ = α (1) + α (3) m ⎣ (3) + + 2 α α (3) α (3)

Defining C ≡ c − α (2) /α (3) , we get f = α (0) e−α

(3) 1 mC 2 2

(B.3)

This result was first obtained by Maxwell and the state of the system defined by (B.3) is said to be in the Maxwellian state. The constants α (0) , α (2) , and α (3) need to be expressed in terms of the number density n, the mean velocity c0 , and the absolute temperature T . According to the definition of number density, n=

f dc = α (0)

e−α

(3) 1 mC 2 2

dC

(B.4)

Expressing C in terms of polar coordinates C , θ and φ where θ is the azimuthal angle and φ is the zenith angle,

∞

n = α (0)

C 2 e−α

(3) 1 mC 2 2

dC

0

=α

(0)

π

0

2π mα (3)

2π

sin φdφ

dθ 0

3 2

(B.5)

Also, nu =

α (2) cf dc = + C f dC α (3) α (2) (3) 1 2 = n (3) + α (0) e−α 2 mC C dC α

(B.6)

Since the integrand is an odd function of the components C , the second term vanishes. Hence, we get u = α (2) /α (3) and the parameter C can be identified with the peculiar velocity. Using these results, (B.3) is written as

f =α

(0) −α (3) 21 mC 2

e

mα (3) =n 2π

3 2

e−α

(3) 1 mC 2 2

(B.7)

B.2 Derivation of f (1)

337

Using the definition of temperature, 3 m kB T = 2 2n m = 2 ∴

C 2 f dc

mα (3) 2π

3 3 kB T = 2 2α (3)

3 2

C 2 e−α

⇒ α (3) =

(3) 1 mC 2 2

dC

1 kB T

(B.8)

Hence, the final form of the distribution function in a steady and uniform state known as Maxwell’s distribution function is given as m 3 mC 2 2 (0) f =n (B.9) exp − 2π kT 2kB T For a given density, mean velocity, and the absolute temperature, there appears to be only one possible permanent mode of distribution function. When the system is away from equilibrium, the actual distribution function will tend to approach this mode.

B.2 Derivation of f (1) B.2.1 Evaluation of Time Derivatives In the evaluation of the distribution function at each approximation level, we will encounter the time derivatives of ρ, u, and T . These time derivatives terms need to be replaced using the conservation laws (Eqs. (5.53)–(5.55)) as ∂ρ ∂(ρuk ) =− , ∂t ∂xk ∂ui ∂ui 1 = −uk + Fi − ∂t ∂xk ρ 2 ∂T ∂T 2 = −uk − ∂t ∂xk 3ρR

∂Pij(r) ∂xj ∂ui Pij(r) ∂xj

(r)

−

∂qk ∂xk

3

Note that we have used Pij = pδij +pij in the above equations. The time derivatives of ρ, u, and T are divided into parts as (subscript attached with ∂ symbol is not a tensorial index)

338

B Appendix to Burnett Equations

∂n = ∂t

∂r n , ∂t

∂r uk , ∂t

∂uk = ∂t

∂T = ∂t

∂r T ∂t

(B.10)

The quantities on the right-hand side are not time-derivatives but they are defined as For r = 0,

∂ρuk ∂0 ρ =− ∂t ∂xk

(B.11) (0)

∂0 ui ∂ui 1 ∂Pij = −uk + Fi − , ∂t ∂xk ρ ∂xj 3 2 (0) ∂qk ∂0 T ∂T 2 (0) ∂ui = −uk − − , Pij ∂t ∂xk 3ρR ∂xj ∂xk For r > 0,

∂r ρ = 0, ∂t

(B.12)

(B.13) (B.14)

(r)

∂r ui 1 ∂Pij =− , ∂t ρ ∂xj 3 2 (r) ∂qk 2 ∂r T (r) ∂ui =− − Pij ∂t 3ρR ∂xj ∂xk

(B.15)

(B.16)

At this stage, we have already solved for the zeroth approximation of the distribution function which turns out to be Maxwellian distribution (Eq. (B.9)). Utilizing this form of distribution function, expressions for zeroth approximation of pressure tensor and heat flux vector are obtained as Pij(0) = δij p,

qi(0) = 0

(B.17)

Using these results, we have For r = 0,

∂ρuk ∂0 ρ =− ∂t ∂xk

(B.18)

∂0 ui ∂ui 1 ∂p = −uk + Fi − , ∂t ∂xk ρ ∂xi

(B.19)

∂0 T ∂T 2 ∂ui = −uk − T , ∂t ∂xk 3 ∂xi For r > 0,

∂r ρ = 0, ∂t

[∵ p = ρRT ]

(B.20) (B.21)

(r)

∂r ui 1 ∂Pij =− , ∂t ρ ∂xj

(B.22)

B.3 Expression for D (1)

339

2

2 ∂r T =− ∂t 3ρR

(r) ∂ui Pij ∂xj

(r)

∂q − k ∂xk

3 (B.23)

In terms of material derivative, we can write D0 ρ ∂uk = −ρ , Dt ∂xk Dr ρ = 0, Dt

2T ∂uk D0 T =− (B.24) Dt 3 ∂xk 3 2 (r) ∂qk 2 Dr T (r) ∂ui =− − Pij Dt 3ρR ∂xj ∂xk (B.25)

D0 ui 1 ∂p = Fi − , Dt ρ ∂xi (r)

Dr ui 1 ∂Pij =− , Dt ρ ∂xj

B.3 Expression for D (1) As f (0) is a function of peculiar velocity C, it is more desirable to express D (1) in terms of C. For this, the operator D of Boltzmann equation can be transformed as (see Appendix C.1), D

(1)

∂f (0) D0 f (0) D0 uk ∂f (0) ∂f (0) ∂ui + Ck = + Fk − − Cj , Dt ∂xk Dt ∂Ck ∂Ci ∂xj

Using Eq. (B.24) (b) for the underlined term, and using

∂f0 ∂t

= f (0) ∂ log∂tf

(B.26) (0)

, we have

⎧ ⎫ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨D log f (0) (0) (0) (0) ∂ log f ∂ui ⎬ 1 ∂p ∂ log f ∂ log f 0 (1) (0) , + Ck + − Cj D =f ⎪ ∂xk ρ ∂xk ∂Ck ∂Ci ∂xj ⎪ ⎪) Dt ⎪ *+ , ⎪ ⎪ ) *+ , ) *+ ,) *+ ,⎭ ⎩ (a)

(b)

(c)

(d)

(B.27) Taking the logarithm of the Maxwellian distribution, log f (0) = ⇒

(a) =

m mC 2 3 3 log + log n − log T − 2 2π kB 2 2kB T

∂ log f (0) mCi =− ∂Ci kB T D0 log f (0) D0 log n 3 D0 log T mC 2 D0 = − − Dt Dt 2 Dt 2kB Dt

(B.28) 1 T

(B.29)

340

B Appendix to Burnett Equations

Using Eq. (B.24) for material derivatives (ρ can be replaced by n in Eq. (B.24) (a)) the first two terms cancel out as D0 log n 3 D0 log T 3 −2T ∂uk 1 D0 n 3 D0 T 1 ∂uk − =0 − = − = n Dt 2 Dt n Dt 2T Dt n ∂xk 2T 3 ∂xk ⇒ (a) =

mC 2 D0 T mC 2 ∂uk D0 log f (0) = = − Dt 3kB T ∂xk 2kB T 2 Dt

(B.30)

Thus, the addition of terms (a) and (d) is given by (a) + (d) =

D0 log f (0) ∂ log f (0) ∂ui Cj − Dt ∂Ci ∂xj

∂ui mC 2 ∂ui mCi Cj + 3kB T ∂xi kB T ∂xj

m ∂ui 1 2 ∂ui Ci Cj = − C δij kB T ∂xj 3 ∂xj =−

=

[Using (B.28)]

∂ui m Ci Cj kB T ∂xj

(B.31)

For middle terms, (b) and (c), we simplify as (b) + (c) = Ck

∂ log f (0) 1 ∂p ∂ log f (0) + ∂xk ρ ∂xk ∂Ck

∂ log f (0) ∂p m Ck − [Using (B.28)] ∂xk ρkB T ∂xk ∂ log f (0) ∂ log p 1 p = Ck − ∂xk ρRT ∂xk ∂ log f (0) ∂ log nkB T = Ck − [Using p = ρRT = nkB T ] ∂xk ∂xk ∂ log(f (0) /nkB T ) = Ck (B.32) ∂xk

= Ck

Dividing the Maxwellian distribution by nkB T and taking logarithm log

3 m 5 1 5 mC 2 f (0) = log + log − log T − nkB T 2 2π 2 kB 2 2kB T ) *+ , constants

B.3 Expression for D (1)

∴

341

∂ log(f (0) /nkB T ) ∂xk 1 5 ∂ log T mC 2 ∂ = Ck − − 2 ∂xk 2kB ∂xk T 5 ∂ log T mC 2 ∂ log T = Ck − + 2 ∂xk 2kB T ∂xk 2 5 mC ∂ log T − Ck = 2kB T 2 ∂xk

(b) + (c) = Ck

(B.33)

Hence, the final expression for D (1) can be written as

D (1) = f (0)

∂ui mC 2 ∂ log T m 5 Ck + − Ci Cj 2kB T 2 ∂xk kB T ∂xj

(B.34)

Appendix C

Appendix to Grad Equations

C.1 Transformation of Boltzmann Equation: f (x, c, t) ⇒ f (x, C, t) The particle distribution function in the Boltzmann equation is expressed in terms of molecular velocity as f (x, c, t). Sometimes, it is more convenient to express distribution function f in terms of peculiar velocity [C = c − u]. In such a case, the Boltzmann equation has to be transformed into the new phase space coordinate system f (x, c, t) ⇒ f (x, C(x, t), t). Care has to be taken while transforming the equation since the peculiar velocity C depends on x and t. Starting with the original Boltzmann equation, we have ∂f ∂f ∂f + ck + Fk = J (f, f1 ) ∂t ∂xk ∂ck

(C.1)

The time derivative of distribution function transforms as ∂f ∂ui ∂f ∂f ∂Ci ∂f ∂f − ⇒ + = + [Ci (x, t) = ci − ui (x, t), ] ∂t ∂t ∂Ci ∂t ∂t ∂Ci ∂t ∂Ci ∂ui ∂f ∂ui ∂f Differentiating w.r.t time, − =− (C.2) = ∂t ∂t ∂Ci ∂t ∂t The spatial derivative is transformed as ck

∂f ∂f ∂f ∂Cj ⇒ ck + ck ∂xk ∂xk ∂Cj ∂xk ∂uj ∂f ∂f − = ck + ck ∂xk ∂Cj ∂xk

[Cj (x, t) = cj − uj (x, t), ]

© Springer Nature Switzerland AG 2020 A. Agrawal et al., Microscale Flow and Heat Transfer, Mechanical Engineering Series, https://doi.org/10.1007/978-3-030-10662-1

∂Cj ∂uj ∵ =− ∂xk ∂xk

343

344

C Appendix to Grad Equations

= Ck

∂f ∂f ∂f ∂uj ∂f ∂uj + uk − Ck − uk ∂xk ∂xk ∂Cj ∂xk ∂Cj ∂xk

(C.3)

and finally, the third term transforms as Fk

∂f ∂f ∂Ci ∂f ∂ci ∂f ⇒ Fk = Fk = Fk δik ∂ck ∂Ci ∂ck ∂Ci ∂ck ∂Ci ∂Ci ∂f ∂ci ∵ = Fk = = δik ∂Ck ∂ck ∂ck

(C.4)

By combining all the transformations, we get ∂uj ∂f ∂f ∂f ∂f ∂f ∂uj − + Ck + uk − Ck ∂t ∂t ∂Cj ∂xk ∂xk ∂Cj ∂xk − uk

∂f ∂uj ∂f + Fk = J (f, f1 ) ∂Cj ∂xk ∂Ck

(C.5)

The underlined terms can be represented in terms of material derivative as Df ∂f ∂f ∂uj ∂f ∂f Duj − Ck + Fk = J (f, f1 ) − + Ck Dt ∂Cj Dt ∂xk ∂Cj ∂xk ∂Ck

(C.6)

Rearranging some of the terms, we get the final expression as (dummy index j in the second term is changed to k) ∂f ∂f ∂uj Duk ∂f Df + Ck + Fk − − Ck = J (f, f1 ) Dt ∂xk Dt ∂Ck ∂Cj ∂xk

(C.7)

In the subsequent use of this equation (in Chaps. 5 and 6), the readers will notice that x, C, t are treated as independent variables. This differential treatment of C is referred to as mixed phase space treatment in the literature (see for example, [45]).

C.2 Note on Hermite Polynomials Just like we expand a given vector in terms of a set of orthogonal base vectors, a sufficiently well-behaved function f (x) can be expanded in terms of orthogonal functions like Legendre functions, Bessel functions, etc. The only difficulty lies in the fact that we are working in an infinite dimensional functional space. (n) Grad chose the orthogonal tensorial Hermite polynomials Hi [61] to expand the single particle distribution function because of its desirable mathematical

C.2 Note on Hermite Polynomials

345

(n)

properties. Hi is an n-th order tensor with subscripts, i = (i1 , i2 , . . . , in ) as well as a polynomial of n-th degree. A complete set of orthonormal multi-dimensional Hermite polynomials H can be obtained by using products of such polynomials in a single variable. In terms of Hermite polynomials, any function f (x) can be represented as ∞

N

f (x) = ω1/2

(n)

ai Hi

(n)

(C.8)

n=0 i=1

ω(x) =

where weight function,

2 1 x exp − N/2 2 (2π )

(C.9)

and Einstein summation convention is assumed for the inner summation as N

N

N

=

N

··· i1 =1 i2 =1

i=1

(C.10) in =1

The weight function ω(x) is normalized so that ω(x)dx = 1.

(C.11)

where (. . . )dx represents an N -fold integration over −∞ < xi < +∞. In expanded form, function f (x) is represented as ⎫ ⎧ N N N ⎬ ⎨ (1) (1) (2) (2) (3) (3) f (x) = ω1/2 a (0) H (0) + ai Hi + aij Hij + aij k Hij k + · · · ⎭ ⎩ i=1

i,j =1

i,j,k=1

(C.12) (n)

Since Hi is symmetric in its subscripts, the scalar product of any tensor of order (n) (n) (n) (n) n, ai and Hi is equal to the product of symmetric part of ai and Hi . The orthogonality of Hermite polynomials can be established by proving the condition, (n) (n) (C.13) ωHi Hj dx = δij n where δij n is a 2n-th order tensor. To represent a function f (x) completely, we still need to evaluate the coefficients (n) ai . Just like we proceed in Fourier series expansion, we multiply both sides of (m) Eq. (C.8) by ω(1/2) Hj and integrate as

346

C Appendix to Grad Equations

ω

1/2

(m) Hj f (x)dx

=

∞

N (n)

ai Hi

ω

(n)

(m)

Hj

dx

(C.14)

n=0 i=1

Using orthogonality condition of Hermite polynomials (Eq. (C.13)), only one term ! in summation ∞ n=0 survives where n = m. ∴

ω

1/2

(m) Hj f (x)dx

N (m)

=

ai

ωHi

(m)

(m)

Hj

dx

i=1 N (m) ai

=

ωHi

(m)

(m)

Hj

dx

i=1 N (m)

=

ai

δij m

[Using Eq. (C.13)]

i=1

Choosing one term of the n! contained in δij m and summing over i gives the result (m) aj . Consequently, we can write as

(m)

ω1/2 Hj ∴

(n) aj

1 = n!

(m)

f (x)dx = m!aj (n)

ω1/2 Hj f (x)dx

(C.15)

where dummy index m is changed to n. In the process of generating Hermite polynomials, we need to take the gradients of the weight function frequently as x x ∂ 1 k k exp − N/2 ∂xi 2 (2π ) x x ∂ x x 1 k k k k − = exp − 2 ∂xi 2 (2π )N/2 ∂xk 1 2xk =ω − [Using Eq. (C.9)] 2 ∂xi ∂xk ∵ = δik = −ωxk δik ∂xi

∇i ω =

∴ ∇i ω = −xi ω,

Similarly, we can prove

(C.16)

C.2 Note on Hermite Polynomials

347

1 xi = ∇i ω ω

(C.17)

The Hermite polynomial of order n (n ≥ 0 is an integer) can be generated using the definition, H (n) =

(−1)(n) (n) ∇ ω ω

(C.18)

First few Hermite polynomials are obtained as 1 ω=1 ω −1 −1 ∇i ω = (−xi ω) = xi = ω ω −1 1 1 ∇i ∇j ω = ∇i (−xj ω) = xj ∇i ω + ω∇i xj = ω ω ω −1 xj (−xi ω) + ωδij = ω

H (0) = Hi

(1)

(2)

Hij

= xi xj − δij (3)

−1 −1 1 ∇i ∇j ∇k ω = ∇i ∇j (−xk ω) = ∇i (xk ∇j ω + ω∇j xk ) ω ω ω 1 1 ∇i (−xk xj ω + ωδj k ) = −∇i (xk xj ω) + ∇i (ωδj k ) = ω ω −1 = xk xj ∇i ω + xk ω∇i xj + xj ω∇i xk − δj k ∇i ω ω −1 = xk xj (−xi ω) + xk ωδij + xj ωδik − δj k (−xi ω) ω (C.19) = xi xj xk − xi δj k + xj δik + xk δij

Hij k =

(4)

Similarly, we can obtain an expression for fourth order Hermite polynomial Hij kl as

(4) Hij kl = xi xj xk xl − xi xj δkl + xi xk δj l + xi xl δj k + xj xk δil + xj xl δik + xk xl δij (C.20) + δij δkl + δik δj l + δil δj k

References

1. Acosta R, Muller R, Tobias C (1985) Transport processes in narrow (capillary) channels. AIChE J 31(3):473–482 2. Agarwal RK, Yun KY, Balakrishnan R (2001) Beyond Navier–Stokes: Burnett equations for flows in the continuum-transition regime. Phys Fluids 13(10):3061–3085; see also: Agarwal RK, Yun KY, Balakrishnan R (2002) Erratum: “Beyond Navier–Stokes: Burnett equations for flows in the continuum-transition regime” [Phys. Fluids 13, 3061 (2001)]. Phys Fluids 14:1818 3. Agrawal A (2016) Higher-order continuum equation based heat conduction law. INAE Lett 1(2):35–39 4. Agrawal A, Prabhu SV (2008) Deduction of slip coefficient in slip and transition regimes from existing cylindrical Couette flow data. Exp Thermal Fluid Sci 32(4):991–996 5. Agrawal A, Prabhu SV (2008) Survey on measurement of tangential momentum accommodation coefficient. J Vac Sci Technol A 26(4):634–645 6. Agrawal A, Djenidi L, Agrawal A (2009) Simulation of gas flow in microchannels with a single 90◦ bend. Comput Fluids 38(8):1629–1637 7. Ai DK (1960) Cylindrical Couette flow in a rarefied gas according to Grad’s equations. PhD thesis, California Institute of Technology 8. Akintunde A, Petculescu A (2014) Infrasonic attenuation in the upper mesosphere–lower thermosphere: a comparison between Navier–Stokes and Burnett predictions. J Acoust Soc Am 136(4):1483–1486 9. Albertoni S, Cercignani C, Gotusso L (1963) Numerical evaluation of the slip coefficient. Phys Fluids 6(7):993–996 10. Alexeenko AA, Fedosov DA, Gimelshein SF, Levin DA, Collins RJ (2006) Transient heat transfer and gas flow in a MEMS-based thruster. J Microelectromech Syst 15(1):181–194 11. Arkilic E, Schmidt M, Breuer K (1997) Gaseous slip flow in long microchannels. J Microelectromech Syst 6(2):167–178 12. Arkilic EB, Breuer KS, Schmidt MA (2001) Mass flow and tangential momentum accommodation in silicon micromachined channels. J Fluid Mech 437:29–43 13. Avcı M, Aydın O (2008) Laminar forced convection slip-flow in a micro-annulus between two concentric cylinders. Int J Heat Mass Transf 51(13–14):3460–3467 14. Aydın O, Avcı M (2006) Heat and fluid flow characteristics of gases in micropipes. Int J Heat Mass Transf 49(9–10):1723–1730 15. Aydın O, Avcı M (2007) Analysis of laminar heat transfer in micro-Poiseuille flow. Int J Thermal Sci 46(1):30–37

© Springer Nature Switzerland AG 2020 A. Agrawal et al., Microscale Flow and Heat Transfer, Mechanical Engineering Series, https://doi.org/10.1007/978-3-030-10662-1

349

350

References

16. Balaj M, Roohi E, Mohammadzadeh A (2017) Regulation of anti-Fourier heat transfer for non-equilibrium gas flows through micro/nanochannels. Int J Thermal Sci 118:24–39 17. Balakrishnan R, Agarwal RK (1997) Numerical simulation of Bhatnagar-Gross-KrookBurnett equations for hypersonic flows. J Thermophys Heat Transf 11(3):391–399 18. Barber RW, Emerson DR (2001) A numerical investigation of low Reynolds number gaseous slip flow at the entrance of circular and parallel plate micro-channels. In: ECCOMAS computational fluid dynamics conference, Swansea, Wales 19. Bentz JA, Tompson R, Loyalka S (1997) The spinning rotor gauge: measurements of viscosity, velocity slip coefficients, and tangential momentum accommodation coefficients for N2 and CH4 . Vacuum 48(10):817–824 20. Bhatnagar PL, Gross EP, Krook M (1954) A model for collision processes in gases. I. Small amplitude processes in charged and neutral one-component systems. Phys Rev 94:511–525 21. Bird G (2013) The DSMC method. CreateSpace Independent Publishing Platform, California 22. Bobylev A (1982) The Chapman–Enskog and Grad methods for solving the Boltzmann equation. Akad Nauk SSSR Dokl 262:71–75 23. Bobylev A (2008) Generalized Burnett hydrodynamics. J Stat Phys 132(3):569–580 24. Bobylev A, Bisi M, Cassinari M, Spiga G (2011) Shock wave structure for generalized Burnett equations. Phys Fluids 23(3):030607 25. Brenner H (2005) Navier–Stokes revisited. Phys A Stat Mech Appl 349(1–2):60–132 26. Burnett D (1936) The distribution of molecular velocities and the mean motion in a nonuniform gas. Proc Lond Math Soc 2(1):382–435 27. Cercignani C (1988) The Boltzmann equation and its applications, 1st edn. Applied mathematical sciences, vol 67. Springer, New York 28. Cercignani C (2000) Rarefied gas dynamics: from basic concepts to actual calculations, vol 21. Cambridge University Press, Cambridge 29. Cercignani C, Daneri A (1963) Flow of a rarefied gas between two parallel plates. J Appl Phys 34(12):3509–3513 30. Cercignani C, Lampis M, Lorenzani S (2004) Variational approach to gas flows in microchannels. Phys Fluids 16:3426–3437 31. Chapman S, Cowling TG (1970) The mathematical theory of non-uniform gases: an account of the kinetic theory of viscosity, thermal conduction and diffusion in gases. Cambridge University Press, Cambridge 32. Chen RY (1973) Flow in the entrance region at low Reynolds numbers. J Fluids Eng 95(1):153–158 33. Chen S, Tian Z (2009) Simulation of microchannel flow using the lattice Boltzmann method. Phys A Stat Mech Appl 388(23):4803–4810 34. Choi S (1991) Fluid flow and heat transfer in microtubes. Micromechanical sensors, actuators, and systems. ASME, New York, pp 123–134 35. Colin S, Lalonde P, Caen R (2004) Validation of a second-order slip flow model in rectangular microchannels. Heat Transf Eng 25(3):23–30 36. Comeaux KA, Chapman DR, MacCormack RW (1995) An analysis of the Burnett equations based on the second law of thermodynamics. In: 33rd Aerospace sciences meeting and exhibit, p 415 37. Dadzie SK (2013) A thermo-mechanically consistent Burnett regime continuum flow equation without Chapman–Enskog expansion. J Fluid Mech 716:R6 38. Date AW (2005) Introduction to computational fluid dynamics. Cambridge University Press, Cambridge 39. De Groot SR, Mazur P (2013) Non-equilibrium thermodynamics. Courier Dover Publications, New York 40. Deissler R (1964) An analysis of second-order slip flow and temperature-jump boundary conditions for rarefied gases. Int J Heat Mass Transf 7(6):681–694 41. Demsis A, Verma B, Prabhu SV, Agrawal A (2009) Experimental determination of heat transfer coefficient in the slip regime and its anomalously low value. Phys Rev E 80(1):016311

References

351

42. Demsis A, Prabhu SV, Agrawal A (2010) Influence of wall conditions on friction factor for flow of gases under slip condition. Exp Thermal Fluid Sci 34(8):1448–1455 43. Dombrowski N, Foumeny E, Ookawara S, Riza A (1993) The influence of Reynolds number on the entry length and pressure drop for laminar pipe flow. Can J Chem Eng 71(3):472–476 44. Dongari N, Agrawal A, Agrawal A (2007) Analytical solution of gaseous slip flow in long microchannels. Int J Heat Mass Transf 50(17):3411–3421 45. Dosch A, Zank GP (2016) Transport processes in space physics and astrophysics. Springer, Berlin 46. Duan Z (2012) Second-order gaseous slip flow models in long circular and noncircular microchannels and nanochannels. Microfluid Nanofluid 12(5):805–820 47. Duan Z, Muzychka Y (2010) Slip flow in the hydrodynamic entrance region of circular and noncircular microchannels. J Fluids Eng 132(1):011201 48. Durst F, Ray S, Ünsal B, Bayoumi O (2005) The development lengths of laminar pipe and channel flows. J Fluids Eng 127(6):1154–1160 49. Duryodhan VS, Singh SG, Agrawal A (2013) Liquid flow through a diverging microchannel. Microfluid Nanofluid 14(1–2):53–67 50. Duryodhan VS, Singh A, Singh SG, Agrawal A (2015) Convective heat transfer in diverging and converging microchannels. Int J Heat Mass Transf 80:424–438 51. Duryodhan VS, Singh SG, Agrawal A (2017) Effect of cross aspect ratio on flow in diverging and converging microchannels. J Fluids Eng 139(6):061203 52. Eu CB (1980) A modified moment method and irreversible thermodynamics. J Chem Phys 73(6):2958–2969 53. Ewart T, Perrier P, Graur IA, Méolans JG (2007) Mass flow rate measurements in a microchannel, from hydrodynamic to near free molecular regimes. J Fluid Mech 584:337– 356 54. Ewart T, Perrier P, Graur IA, Méolans JG (2007) Tangential momentum accommodation in microtube. Microfluid Nanofluid 3(6):689–695 55. Fan H, Xue Q (2000) A new analytic solution of the Navier–Stokes equations for microchannel flows. Microscale Thermophys Eng 4(2):125–143 56. Fiscko KA, Chapman DR (1989) Comparison of Burnett, super-Burnett, and Monte Carlo solutions for hypersonic shock structure. Progress in astronautics and aeronautics (AIAA), pp 374–395 57. Gad-el Hak M (1999) The fluid mechanics of microdevices—the Freeman scholar lecture. J Fluids Eng 121(1):5–33 58. Gad-el Hak M (2001) The MEMS handbook. CRC Press, Boca Raton 59. Gavasane A, Sachdev SS, Mittal BK, Bhandarkar UV, Agrawal A (2011) A critical assessment of the Maxwell slip boundary condition for rarefied wall bounded flows. Int J Micro-Nano Scale Transp 2:109–116 60. Gembaroviˇc J, Majernik V (1988) Non-Fourier propagation of heat pulses in finite medium. Int J Heat Mass Transf 31(5):1073–1080 61. Grad H (1949) Note on N -dimensional Hermite polynomials. Commun Pure Appl Math 2(4):325–330 62. Grad H (1949) On the kinetic theory of rarefied gases. Commun Pure Appl Math 2(4):331– 407 63. Grad H (1952) The profile of a steady plane shock wave. Commun Pure Appl Math 5(3):257– 300 64. Grad H (1958) Principles of the kinetic theory of gases. Springer, Berlin, pp 205–294 65. Graur I, Veltzke T, Méolans J, Ho M, Thöming J (2015) The gas flow diode effect: theoretical and experimental analysis of moderately rarefied gas flows through a microchannel with varying cross section. Microfluid Nanofluid 18(3):391–402 66. Greenshields CJ, Reese JM (2007) The structure of shock waves as a test of Brenner’s modifications to the Navier–Stokes equations. J Fluid Mech 580:407–429 67. Gross EP, Jackson EA (1958) Kinetic theory of the impulsive motion of an infinite plane. Phys Fluids 1(4):318–328

352

References

68. Gu XJ, Emerson DR (2009) A high-order moment approach for capturing non-equilibrium phenomena in the transition regime. J Fluid Mech 636:177–216 69. Gu XJ, Emerson DR, Tang GH (2009) Kramer’s problem and the Knudsen minimum: a theoretical analysis using a linearized 26-moment approach. Contin Mech Thermodyn 21(5):345 70. Hemadri V, Varade VV, Agrawal A, Bhandarkar UV (2016) Investigation of rarefied gas flow in microchannels of non-uniform cross section. Phys Fluids 28(2):022007 71. Hemadri V, Varade VV, Agrawal A, Bhandarkar UV (2017) Rarefied gas flow in converging microchannel in slip and early transition regimes. Phys Fluids 29:032002 72. Hemadri V, Agrawal A, Bhandarkar UV (2018) Determination of tangential momentum accommodation coefficient and slip coefficients for rarefied gas flow in a microchannel. S¯adhan¯a 43(10):164 73. Hemadri V, Biradar GS, Shah N, Garg R, Bhandarkar UV, Agrawal A (2018) Experimental study of heat transfer in rarefied gas flow in a circular tube with constant wall temperature. Exp Thermal Fluid Sci 93:326–333 74. Hooman K (2008) A superposition approach to study slip-flow forced convection in straight microchannels of uniform but arbitrary cross-section. Int J Heat Mass Transf 51(15– 16):3753–3762 75. Hsia YT, Domoto G (1983) An experimental investigation of molecular rarefaction effects in gas lubricated bearings at ultra-low clearances. J Lubr Technol 105(1):120–129 76. Huang C, Gregory JW, Sullivan JP (2007) Microchannel pressure measurements using molecular sensors. J Microelectromech Syst 16(4):777–785 77. Inman RM (1964) Laminar slip flow heat transfer in a parallel-plate channel or a round tube with uniform wall heating. NASA Technical note D-2393. National Aeronautics and Space Administration, Washington, DC 78. Jadhav RS, Singh N, Agrawal A (2017) Force-driven compressible plane Poiseuille flow by Onsager-Burnett equations. Phys Fluids 29(10):102002 79. Jang J, Wereley ST (2006) Gaseous slip flow analysis of a micromachined flow sensor for ultra small flow applications. J Micromech Microeng 17(2):229 80. Jie D, Diao X, Cheong KB, Yong LK (2000) Navier–Stokes simulations of gas flow in micro devices. J Micromech Microeng 10(3):372 81. Karniadakis GE, Beskok A, Aluru N (2006) Microflows and nanoflows: fundamentals and simulation, vol 29. Springer, Berlin 82. Kennard EH (1938) Kinetic theory of gases with an introduction to statistical mechanics. McGraw-Hill, New York 83. Kogan MN (1969) Rarefied gas dynamics. Plenum Press, New York 84. Kremer GM (2010) An introduction to the Boltzmann equation and transport processes in gases. Springer, Berlin 85. Kundu PK (1990) Fluid mechanics. Academic Press, New York 86. Lee WY, Wong M, Zohar Y (2002) Microchannels in series connected via a contraction/expansion section. J Fluid Mech 459:187–206 87. Lihnaropoulos J, Valougeorgis D (2011) Unsteady vacuum gas flow in cylindrical tubes. Fusion Eng Des 86(9–11):2139–2142 88. Liu J, Tai YC, Pong CMHC (1995) MEMS for pressure distribution studies of gaseous flows in microchannels. In: Proceedings of the IEEE Micro Electro Mechanical Systems. IEEE, Piscataway, pp 209–215 89. Lockerby DA, Reese JM, Emerson DR, Barber RW (2004) Velocity boundary condition at solid walls in rarefied gas calculations. Phys Rev E 70(1):017303 90. Loyalka S (1996) Theory of the spinning rotor gauge in the slip regime. J Vac Sci Technol A 14(5):2940–2945 91. Mahendra AK, Singh RK (2013) Onsager reciprocity principle for kinetic models and kinetic schemes. arXiv preprint arXiv:13084119 92. Maranzana G, Perry I, Maillet D (2004) Mini-and micro-channels: influence of axial conduction in the walls. Int J Heat Mass Transf 47(17–18):3993–4004

References

353

93. Maurer J, Tabeling P, Joseph P, Willaime H (2003) Second-order slip laws in microchannels for helium and nitrogen. Phys Fluids 15(9):2613–2621 94. Maxwell JC (1879) On stresses in rarified gases arising from inequalities of temperature. Philos Trans R Soc Lond A 170:231–256 95. McLennan JA (1974) Onsager’s theorem and higher-order hydrodynamic equations. Phys Rev A 10(4):1272 96. McNamara S, Gianchandani YB (2005) On-chip vacuum generated by a micromachined Knudsen pump. J Microelectromech Syst 14(4):741–746 97. Mitsuya Y (1993) Modified Reynolds equation for ultra-thin film gas lubrication using 1.5order slip-flow model and considering surface accommodation coefficient. J Tribol 115:289– 289 98. Miyamoto M, Shi W, Katoh Y, Kurima J (2003) Choked flow and heat transfer of low density gas in a narrow parallel-plate channel with uniformly heating walls. Int J Heat Mass Transf 46(14):2685–2693 99. Mohammadzadeh A, Rana AS, Struchtrup H (2015) Thermal stress vs. thermal transpiration: a competition in thermally driven cavity flows. Phys Fluids 27(11):112001 100. Moharana MK, Khandekar S (2013) Effect of aspect ratio of rectangular microchannels on the axial back conduction in its solid substrate. Int J Microscale Nanoscale Thermal Fluid Transp Phenom 4(3/4):211 101. Morini GL, Spiga M, Tartarini P (2004) The rarefaction effect on the friction factor of gas flow in microchannels. Superlattices Microstruct 35(3):587–599 102. Muralidhar K, Biswas G (2005) Advanced engineering fluid mechanics. Narosa Publishing House, New Delhi 103. Myong RS (1999) Thermodynamically consistent hydrodynamic computational models for high-Knudsen-number gas flows. Phys Fluids 11(9):2788–2802 104. Myong R (2011) A full analytical solution for the force-driven compressible Poiseuille gas flow based on a nonlinear coupled constitutive relation. Phys Fluids 23(1):012002 105. Niazmand H, Renksizbulut M, Saeedi E (2008) Developing slip-flow and heat transfer in trapezoidal microchannels. Int J Heat Mass Transf 51(25–26):6126–6135 106. Ohira SI, Toda K (2008) Micro gas analyzers for environmental and medical applications. Anal Chim Acta 619(2):143–156 107. Onsager L (1931) Reciprocal relations in irreversible processes. I. Phys Rev 37(4):405 108. Onsager L (1931) Reciprocal relations in irreversible processes. II. Phys Rev 38(12):2265 109. Özi¸sik M, Vick B (1984) Propagation and reflection of thermal waves in a finite medium. Int J Heat Mass Transf 27(10):1845–1854 110. Panigrahi PK (2016) Transport phenomena in microfluidic systems. Wiley, New York 111. Panigrahi AK, Singh V, Singh SG (2017) Nonlithographic fabrication of plastic-based nanofibers integrated microfluidic biochip for sensitive detection of infectious biomarker. Appl Mater Interfaces 9(46):39994–40005 112. Petculescu A (2016) Acoustic properties in the low and middle atmospheres of Mars and Venus. J Acoust Soc Am 140(2):1439–1446 113. Pfahler J, Harley J, Bau H, Zemel JN (1991) Gas and liquid flow in small channels. Am Soc Mech Eng Dyn Syst Control Div 32:49–60 114. Pollard W, Present RD (1948) On gaseous self-diffusion in long capillary tubes. Phys Rev 73(7):762 115. Pong K, Ho CM, Liu J, Tai YC (1994) Non-linear pressure distribution in uniform microchannels. In: Proceedings of application of microfabrication to fluid mechanics, ASME Winter annual meeting, Chicago, pp 51–56 116. Pope SB (2001) Turbulent flows. IOP Publishing, Bristol 117. Porodnov B, Suetin P, Borisov S, Akinshin V (1974) Experimental investigation of rarefied gas flow in different channels. J Fluid Mech 64(3):417–438 118. Prabhakar A, Kumar YBV, Tripathi S, Agrawal A (2015) A novel, compact and efficient microchannel arrangement with multiple hydrodynamic effects for blood plasma separation. Microfluid Nanofluid 18(5–6):995–1006

354

References

119. Prabhu SV, Mahulikar SP (2014) Effects of density and thermal conductivity variations on entropy generation in gas micro-flows. Int J Heat Mass Transf 79:472–485 120. Rath A, Singh N, Agrawal A (2018) A perturbation-based solution of Burnett equations for gaseous flow in a long microchannel. J Fluid Mech 844:1038–1051 121. Rohsenow WM, Hartnett JP, Cho YI (1998) Handbook of heat transfer, vol 3. McGraw-Hill, New York 122. Romero M, Velasco R (1995) Onsager’s symmetry in the Burnett regime. Phys A Stat Mech Appl 222(1–4):161–172 123. Sambasivam R (2012) Extended Navier–Stokes equations: derivations and applications to fluid flow problems. PhD thesis, Universität Erlangen-Nürnberg 124. Satapathy AK (2010) Slip flow heat transfer in an infinite microtube with axial conduction. Int J Thermal Sci 49(1):153–160 125. Schamberg R (1947) The fundamental differential equations and the boundary conditions for high speed slip-flow, and their application to several specific problems. PhD thesis, California Institute of Technology 126. Schlichting H (1960) Boundary layer theory. McGraw-Hill, New York 127. Sharipov F, Graur I (2014) General approach to transient flows of rarefied gases through long capillaries. Vacuum 100:22–25 128. Sharma A (2016) Introduction to computational fluid dynamics: development, application and analysis. Wiley, New York 129. Shavaliyev MS (1993) Super-Burnett corrections to the stress tensor and the heat flux in a gas of Maxwellian molecules. J Appl Math Mech 57(3):573–576 130. Singh N, Agrawal A (2014) The Burnett equations in cylindrical coordinates and their solution for flow in a microtube. J Fluid Mech 751:121–141 131. Singh N, Agrawal A (2016) Onsager’s-principle-consistent 13-moment transport equations. Phys Rev E 93(6):063111 132. Singh N, Dongari N, Agrawal A (2014) Analytical solution of plane Poiseuille flow within Burnett hydrodynamics. Microfluid Nanofluid 16(1–2):403–412 133. Singh N, Gavasane A, Agrawal A (2014) Analytical solution of plane Couette flow in the transition regime and comparison with direct simulation Monte Carlo data. Comput Fluids 97:177–187 134. Singh SG, Jain A, Sridharan A, Duttagupta S, Agrawal A (2009) Flow map and measurement of void fraction and heat transfer coefficient using an image analysis technique for flow boiling of water in a silicon microchannel. J Micromech Microeng 19(7):075004 135. Singh SG, Bhide R, Duttagupta S, Sridharan A, Agrawal A (2010) Experimental study of water boiling in microchannel. Adv Multiphase Flow Heat Transf 3:246–303 136. Singh N, Jadhav RS, Agrawal A (2017) Derivation of stable Burnett equations for rarefied gas flows. Phys Rev E 96(1):013106 137. Sreekanth A (1969) Slip flow through long circular tubes. In: Trilling L, Wachman HY (eds) Proceedings of the sixth international symposium on rarefied gas dynamics. Academic Press, New York, pp 667–680 138. Steckelmacher W (1986) Knudsen flow 75 years on: the current state of the art for flow of rarefied gases in tubes and systems. Rep Prog Phys 49(10):1083 139. Stemme E, Stemme G (1993) A valveless diffuser/nozzle-based fluid pump. Sens. Actuators A 39(2):159–167 140. Struchtrup H (2004) Stable transport equations for rarefied gases at high orders in the Knudsen number. Phys Fluids 16(11):3921–3934 141. Struchtrup H (2005) Macroscopic transport equations for rarefied gas flows. Springer, Berlin 142. Struchtrup H, Torrilhon M (2003) Regularization of Grad’s 13 moment equations: derivation and linear analysis. Phys Fluids 15(9):2668–2680 143. Suetin P, Porodnov B, Chernjak V, Borisov S (1973) Poiseuille flow at arbitrary Knudsen numbers and tangential momentum accommodation. J Fluid Mech 60(3):581–592 144. Tang G, Li Z, He Y, Tao W (2007) Experimental study of compressibility, roughness and rarefaction influences on microchannel flow. Int J Heat Mass Transf 50(11–12):2282–2295

References

355

145. Tekasakul P, Bentz J, Tompson R, Loyalka S (1996) The spinning rotor gauge: measurements of viscosity, velocity slip coefficients, and tangential momentum accommodation coefficients. J Vac Sci Technol A 14(5):2946–2952 146. Timokhin MY, Struchtrup H, Kokhanchik A, Bondar YA (2017) Different variants of R13 moment equations applied to the shock-wave structure. Phys Fluids 29(3):037,105 147. Torrilhon M (2016) Modeling nonequilibrium gas flow based on moment equations. Ann Rev Fluid Mech 48(1):429–458 148. Torrilhon M, Struchtrup H (2004) Regularized 13-moment equations: shock structure calculations and comparison to Burnett models. J Fluid Mech 513:171–198 149. Truesdell C, Muncaster RG (1980) Fundamentals of Maxwel’s kinetic theory of a simple monatomic gas: treated as a branch of rational mechanics, vol 83. Academic Press, New York 150. Tunc G, Bayazitoglu Y (2001) Heat transfer in microtubes with viscous dissipation. Int J Heat Mass Transf 44(13):2395–2403 151. Turner SE, Sun H, Faghri M, Gregory OJ (1999) Local pressure measurement of gaseous flow through microchannels. ASME-Publications-HTD 364:71–80 152. Tzou D, Chiu K (2001) Temperature-dependent thermal lagging in ultrafast laser heating. Int J Heat Mass Transf 44(9):1725–1734 153. Uribe FJ (2011) The shock wave problem revisited: the Navier–Stokes equations and Brenner’s two velocity hydrodynamics. In: Coping with complexity: model reduction and data analysis. Springer, Berlin, pp 207–229 154. Uribe F, Garcia A (1999) Burnett description for plane Poiseuille flow. Phys Rev E 60(4A):4063–4078 155. Uribe F, Velasco R, Garcia-Colin L (2000) Bobylev’s instability. Phys Rev E 62(4):5835 156. Varade V, Agrawal A, Pradeep AM (2014) Behaviour of rarefied gas flow near the junction of a suddenly expanding tube. J Fluid Mech 739:363–391 157. Varade V, Agrawal A, Pradeep AM (2014) Experimental study of rarefied gas flow near sudden contraction junction of a tube. Phys Fluids 26(6):062002 158. Varade V, Agrawal A, Prabhu SV, Pradeep AM (2015) Early onset of flow separation with rarefied gas flowing in a 90◦ bend tube. Exp Thermal Fluid Sci 66:221–234 159. Varade V, Agrawal A, Prabhu SV, Pradeep AM (2015) Velocity measurement in low Reynolds and low Mach number slip flow through a tube. Exp Thermal Fluid Sci 60:284–289 160. Varade V, Duryodhan VS, Agrawal A, Pradeep AM, Ebrahimi A, Roohi E (2015) Low Mach number slip flow through diverging microchannel. Comput Fluids 111:46–61 161. Veltzke T, Baune M, Thöming J (2012) The contribution of diffusion to gas microflow: an experimental study. Phys Fluids 24(8):082004 162. Verma B, Demsis A, Agrawal A, Prabhu SV (2009) Semiempirical correlation for the friction factor of gas flowing through smooth microtubes. J Vac Sci Technol A 27(3):584–590 163. Wallace MAG, Pleil JD (2018) Evolution of clinical and environmental health applications of exhaled breath research: review of methods and instrumentation for gas-phase, condensate, and aerosols. Anal Chim Acta 1024:18–38 164. Weiss W (1995) Continuous shock structure in extended thermodynamics. Phys Rev E 52:R5760–R5763 165. White C, Borg MK, Scanlon TJ, Reese JM (2013) A DSMC investigation of gas flows in micro-channels with bends. Comput Fluids 71:261–271 166. Woods L (1979) Transport processes in dilute gases over the whole range of Knudsen numbers. Part 1. General theory. J Fluid Mech 93(3):585–607 167. Woods L, Troughton H (1980) Transport processes in dilute gases over the whole range of Knudsen numbers. Part 2. Ultrasonic sound waves. J Fluid Mech 100(2):321–331 168. Wu L (2008) A slip model for rarefied gas flows at arbitrary Knudsen number. Appl Phys Lett 93(25):253103 169. Wu P, Little W (1984) Measurement of the heat transfer characteristics of gas flow in fine channel heat exchangers used for microminiature refrigerators. Cryogenics 24(8):415–420 170. Xu K (2003) Super-Burnett solutions for Poiseuille flow. Phys Fluids 15(7):2077–2080

356

References

171. Yamaguchi H, Hanawa T, Yamamoto O, Matsuda Y, Egami Y, Niimi T (2011) Experimental measurement on tangential momentum accommodation coefficient in a single microtube. Microfluid Nanofluid 11(1):57–64 172. Yang HT (1955) Rayleigh’s problem at low Mach number according to the kinetic theory of gases. Doctoral dissertation, California Institute of Technology 173. Yang R, Chen G, Laroche M, Taur Y (2005) Simulation of nanoscale multidimensional transient heat conduction problems using ballistic-diffusive equations and phonon Boltzmann equation. J Heat Transf 127(3):298–306 174. Yu D (1995) An experimental and theoretical investigation of fluid flow and heat transfer in microtubes. In: ASME/JSME thermal engineering conference, vol 1 175. Zhao S, Povitsky A (2009) A hybrid molecular and continuum method for low-Reynoldsnumber flows. Nonlinear Anal Theory Methods Appl 71(12):e2551–e2564 176. Zheng Y, Garcia AL, Alder BJ (2002) Comparison of kinetic theory and hydrodynamics for Poiseuille flow. J Stat Phys 109(3–4):495–505 177. Zhong X, MacCormack RW, Chapman DR (1993) Stabilization of the Burnett equations and application to hypersonic flows. AIAA J 31(6):1036–1043

Index

A Agrawal and Singh iterative approach additional boundary condition, 179 error vs. Knudsen number, 179, 180 Navier–Stokes equations, 177 procedure, 177, 178 Augmented Burnett equations, 262–263, 266

B Bhatnagar–Gross–Krook (BGK) kinetic model, 149, 150 Bobylev’s instability, 260 Boltzmann equation, 189, 307 conservation laws, 136–139 constitutive relations, 139–141 evolution equations for moments, 134–136 higher order moments, 131–134 macroscopic quantities, distribution function, 127–131 mathematical representation, 127 Maxwell–Boltzmann distribution, 127 Brenner’s hypothesis, 117–119 Brinkman number, 12, 13 Burnett equations, 22–23 Agrawal and Singh iterative approach additional boundary condition, 179 error vs. Knudsen number, 179, 180 Navier–Stokes equations, 177 procedure, 177, 178 augmented Burnett equations, 262–263, 266 BGK-Burnett equations Euler equations, 263 Navier–Stokes equations, 263

O(Kn3 ), 266–269 stress tensor and heat flux vector, 263–265 two-dimensional form, 263 Boltzmann equation conservation laws, 136–139 constitutive relations, 139–141 evolution equations for moments, 134–136 higher order moments, 131–134 macroscopic quantities, distribution function, 127–131 mathematical representation, 127 Maxwell–Boltzmann distribution, 127 bulk velocity, 261 Chapman–Enskog method central idea, 143–144 dimensionless Boltzmann equation, 141–143 distribution function, 144–151 Couette flow, 295–298 cylindrical coordinates Burnett stress and heat flux, 167–174 Navier–Stokes stress and heat flux, 167 stress tensor and heat flux vector, 166 D (1) expression, 339–341 Euler and Navier–Stokes equations, 261 f (0) derivation, 335–337 f (1) derivation, 337–339 force-driven Poiseuille flow basic conservation laws, 297, 299 conventional Burnett equations, 299–300 DSMC, 302 molecular dynamics simulations, 302

© Springer Nature Switzerland AG 2020 A. Agrawal et al., Microscale Flow and Heat Transfer, Mechanical Engineering Series, https://doi.org/10.1007/978-3-030-10662-1

357

358 Burnett equations (cont.) monotonic pressure profile, 302 Navier–Stokes equation, 299 non-dimensionalization, coupled ordinary differential equation, 301–302 normal heat flux, 302, 304 normal stress, 302, 304 OBurnett equations, 300 schematic diagram, 296, 298 shear stress, 302, 304 tangential heat flux, 302, 304 generalized free path theory, 269 heat flux vector, 261 hydrodynamic equations Burnett stress and heat flux terms, 162, 163 conventional Burnett equations, 162 Euler equations, 151–154 Navier–Stokes equations, 154–161 super-Burnett equations, 164–165 velocity gradient tensor, 162 limitations, 260–261 macroscopic parameters, 125 mass density, 261 order of magnitude analysis, 175–176 perturbation based solution cross-stream (y) momentum equation, 181 flow distribution, cross-stream direction, 183, 184 non-dimensional form, 181 pressure distribution, 183 pressure equation, 181 streamwise (x) momentum equation, 181 three boundary conditions, 182 y-momentum equation, 182 pressure tensor, 261 stability analysis constitutive equations, 185 dispersion relation, 186 Maxwellian molecules, 184 real part of Ω vs. wave number, 186, 187 x-momentum and energy, 184 temperature, 261

C Cartesian vector, 319 Cattaneo’s equation, 121, 122 Cattaneo’s model, 254–255 Chapman–Enskog method, 20–21

Index accuracy, 257 central idea, 143–144 closure relation/integral, 257 complexity, 257 constitutive relation, 256–257 convergence properties, 260 dimensionless Boltzmann equation collision integral, 141–142 dimensionless variables, 142 final form, 142 Knudsen number, 143 Strouhal number, 142 transport operator, 142 velocity scale, 142 distribution function, 256 Boltzmann equation, 144 Cauchy’s rule, 145, 146 first approximation, 149–151 transport equations, 144 zeroth approximation, 147, 149 initial and boundary conditions, 255–256 Maxwellian distribution, 165 set of variables, 255 Computational fluid dynamics (CFD), 3 Conjugate effects, 15 Conservation laws conservation of energy final form, 138–139 Lagrangian framework, 139 rate of change, internal energy, 139 transfer equation, 137 conservation of mass, 136 conservation of momentum, 136–137 Couette flow boundary conditions, 45 Cf,slip Ma vs. Re/Ma, 48 schematic diagram, 45 skin friction coefficient, 47 slip velocity, 46, 47 velocity distribution, Knudsen number, 46, 47 volume flow rate, 47 x-momentum equation, 44 Cylindrical Couette flow, Grad equations boundary conditions, 243, 247, 249 1/(CD Ma ), 251, 253 drag coefficient, 248 monatomic gas, Mach number and Reynolds number, 247–248 Navier–Stokes solution, 246, 247 outer stationary cylinder, 244 Prandtl number, 252 Re/Ma variation, 251–253 schematic diagram, 241

Index slip velocity, 248–249 Stanton number, 251 steady state assumption, 241 temperature distribution, 249 temperature jumps, 250 13-moment equations continuity, 242 energy, 242 equation of state, 242 heat flux, 243 linearization of, 244–246 momentum, 242 stresses, 242–243 TMAC, 244 uθ /U variation, 251, 252 velocity profile, 251

D Dimensionless Boltzmann equation collision integral, 141–142 dimensionless variables, 142 final form, 142 Knudsen number, 143 Strouhal number, 142 transport operator, 142 velocity scale, 142 Diodicity, 73 Direct Simulation Monte Carlo (DSMC) method, 2–3, 116, 117, 183, 260, 295, 302, 303, 305–307, 314–315

E Euler equations Cartesian to spherical coordinate system, 151, 152 generalized factorial function, 153 integral evaluation, 152 Maxwellian distribution, pressure tensor, 151 pressure tensor and heat flux vector, 154 standard Gaussian integral, 153

F Force-driven Poiseuille flow basic conservation laws, 297, 299 conventional Burnett equations, 299–300 DSMC, 302 molecular dynamics simulations, 302 monotonic pressure profile, 302 Navier–Stokes equation, 299

359 non-dimensionalization, coupled ordinary differential equation, 301–302 normal heat flux, 302, 304 normal stress, 302, 304 OBurnett equations, 300 schematic diagram, 296, 298 shear stress, 302, 304 tangential heat flux, 302, 304 Friction factor calculation, Wu and Little expression, 78 correlation, 79 value, 78

G Gauss divergence theorem, 196 Generalized free path theory, 269 Grad equations, 23 additional boundary conditions, 271 Boltzmann equation, 189, 191, 192, 343–344 and Chapman–Enskog method accuracy, 257 closure relation/integral, 257 complexity, 257 constitutive relation, 256–257 distribution function, 256 initial and boundary conditions, 255–256 set of variables, 255 convective moments, 190 Couette flow, 295–298 cylindrical Couette flow (see Cylindrical Couette flow, Grad equations) force-driven Poiseuille flow basic conservation laws, 297, 299 conventional Burnett equations, 299–300 DSMC, 302 molecular dynamics simulations, 302 monotonic pressure profile, 302 Navier–Stokes equation, 299 non-dimensionalization, coupled ordinary differential equation, 301–302 normal heat flux, 302, 304 normal stress, 302, 304 OBurnett equations, 300 schematic diagram, 296, 298 shear stress, 302, 304 tangential heat flux, 302, 304 fourth order moment, 190 Hermite polynomials, 344–347 higher order moments, 191

360 Grad equations (cont.) hyperbolic nature of equations, 270 Leibnitz rule, 192 Maxwell-Boltzmann distribution, 192 moment-based method, 189 nth order moment, 190, 191 regularized 13-moment equations (see Regularized 13 (R13) moment equations) regularized 26-moment equations (see Regularized 26 (R26) moment equations) Stokes’s first problem free-molecular regime solution, 240 kinetic theory based solution, 238–239 Navier–Stokes based solution, 227–228 Rayleigh’s problem, 226, 227 significance, 240–241 stress tensor and heat flux vector, 271–272 symmetric traceless second order moment, 190–191 third order moment, 190–191 13-moment approximation (see 13-moment approximation) 20-moment approximation closure, 204 distribution function, 203–204 expressions for variables, 193 flow diagram, 192, 193 Hermite polynomials, 192, 200–203 Pij evolution equation, 194–198 Sij k evolution equation, 198–199 20 scalar equations, 205–207 unphysical negative values, distribution function, 270–271

H Heat transfer problems, 314 Hermite polynomials, 192, 200–203, 344–347 Higher order continuum transport equations, 19 Hydrodynamic equations Burnett stress and heat flux terms, 162, 163 conventional Burnett equations, 162 Euler equations, 151–154 Navier–Stokes equations, 154–161 super-Burnett equations, 164–165 velocity gradient tensor, 162

I Integral approach elemental control volume, 54

Index mass flow driven system, 57 mass flow rate, 56 maximum velocity, streamwise location, 54 momentum balance, streamwise direction, 54, 55 momentum equation, rectangular microchannel, 55 parabolic velocity profile, 53 pressure driven system, 57 pressure field, 56 Reynolds number, 57 shear stress, 54

K Knudsen minima, 59, 67–68 Knudsen number, 8–10, 18, 307

L Lattice Boltzmann method, 21 Leibnitz rule, 192 Linearized Maxwell–Burnett boundary condition, 39

M Mach number, 11 Macroscopic quantities, distribution function absolute temperature, 127 bulk velocity, 127 hydrostatic pressure, 130 ideal gas equation, 131 internal energy density, 130 macroscopic velocity of gas, 128 mass density, 127, 128 number density, 128 peculiar velocity, 129 pressure tensor, 130 stress tensor, 130 Maxwellian distribution, 338 Maxwellian molecules, 207 Maxwell’s model, 38 Maxwell’s slip theory actual velocity, 35 diffuse interaction, 34, 35 “first-order” slip boundary condition, 37 gas–wall interaction, 34–35 specular interaction, 34, 35 tangential momentum, 35–36 TMAC, 35 Micro-annulus constant heat flux boundary condition, 99 energy equation, 99

Index gas flow, 97, 98 momentum equation, 98 non-dimensional radius, 99 non-dimensional variables, 98–102 Nusselt number, 102 parametric variation, 102–106 velocity slip boundary conditions, 98 Microchannel arbitrary cross section effective aspect ratio, 62, 64 gas flow, 62 mass flow rate, 63 modified Knudsen number, 62 slip model, 62 velocity distribution and average velocity, 62 bend, 74–76 comments, 58–59 diverging/converging, 72–74 integral approach, 53–57 microscale heat transfer constant wall temperature case, 89 energy equation, 87 gas flow, uniform wall heat flux, 86 parametric variation, 90–92 uniform heat flux, 87–89 velocity profile, 86 perturbation of Navier–Stokes equations, 49–53 rough, 68–69 sudden expansion/contraction, 70–72 Microdevice blood test blood flow, view, 7 microchannel, plasma separation, 7 micro-sensor, detecting malaria, 7, 8 physiological condition, 6 sample preparation module, 7 fabrication procedure, 308, 310 Micro-nozzles and micro-thruster fluid flow characteristics, 6 force generation, 5 micro/nano satellite technology, 4–5 Micro-PIV, 308 Microscale flows annulus of rotating sphere and cylinder, 64–66 appearance of Knudsen minima, 67–68 applications advantages, 4 breath analyser, 6 electronic device cooling, 4, 5 microdevice, blood test, 6–8 micro-nozzles and micro-thruster, 4–6

361 arbitrary cross section microchannel effective aspect ratio, 62, 64 gas flow, 62 mass flow rate, 63 modified Knudsen number, 62 slip model, 62 velocity distribution and average velocity, 62 boundary condition alternate slip models, 38–40 higher-order slip boundary condition, 37–38 Kundt and Warburg ideas, 34 Maxwell’s slip theory, 34–37 value of slip coefficients, 40–44 Burnett and Grad equations, 2 CFD, 3 characteristics axial conduction, 14 compressibility, 10, 11 conjugate heat transfer, 14–15 property variation, 13, 14 rarefaction, 9–10 thermal creep, 11, 12 viscous dissipation, 12–13 Couette flow boundary conditions, 45 Cf,slipMa vs. Re/Ma, 48 schematic diagram, 45 skin friction coefficient, 47 slip velocity, 46, 47 velocity distribution, Knudsen number, 46, 47 volume flow rate, 47 x-momentum equation, 44 DSMC method, 2–3 governing equations, fluid flow compressible Navier–Stokes equations, 30–31 constitutive relation, 28–30 incompressible Navier–Stokes equations, 31–33 tensorial form, 26–28 mathematical modelling Boltzmann equation, 20 Chapman-Enskog expansion, 20–21 conventional equations and boundary conditions, 16–19 Grad moment method, 20, 21 higher order continuum transport equations, 18 Navier–Stokes equations, 16 non-Newtonian fluids, 18

362 Microscale flows (cont.) microchannel bend, 74–76 comments, 58–59 diverging/converging, 72–74 integral approach, 53–57 perturbation of Navier–Stokes equations, 49–53 rough, 68–69 sudden expansion/contraction, 70–72 microtube, 60–61 regime classification, 8, 9 transient flow in capillary, 69–70 useful empirical correlations friction factor, 78–79 length development, 76–77 Microscale heat transfer analysis through micropipe constant wall temperature case, 94–95 energy equation, 92 gas flow, 92 modified Brinkman number, 93, 94 non-dimensional variables, 93 Nusselt number expression, 94 parametric variation, 95–97 thermal boundary condition, 93 axial conduction, 108–109 boundary condition first order temperature jump condition, 84–85 temperature jump, in wall, 83–84 thermal accommodation coefficient, 83 conduction, substrate, 107–108 conjugate effect, 82 flow work and shear work, 109 governing equation, 82–83 Knudsen pump, 110–112 micro-annulus constant heat flux boundary condition, 99 energy equation, 99 gas flow, 97, 98 momentum equation, 98 non-dimensional radius, 99 non-dimensional variables, 98–102 Nusselt number, 102 parametric variation, 102–106 velocity slip boundary conditions, 98 microchannel constant wall temperature case, 89 energy equation, 87 gas flow, uniform wall heat flux, 86 parametric variation, 90–92 uniform heat flux, 87–89

Index velocity profile, 86 observations from experiments, 110 rarefaction and non-rarefaction effects, 82 useful empirical correlations, 112–113 variation in thermophysical properties, 106–107 Microscale measurement, 307–310 Miniature sensors, 314 Molecular Dynamics (MD) method, 307

N Navier–Stokes equations, 158 Agrawal and Singh iterative approach, 177 Brenner’s hypothesis, 117–119 Couette flow, later slip and early transition regimes, 116 flow, sudden contraction, 116 Grad equations, 189 heat transfer, microchannels and cavities, 117 hydrodynamic equations approximation of pressure tensor, 154 BGK kinetic model, 160 common integrals, kinetic theory of gases, 155 evaluation of integral I3 , 157 expression for q (1) , 158–159 Fourier’s law, 160 fourth order symmetric and isotropic tensor, 156 Gaussian integral, 159 heat flux vector, 160 integral equation for I2 , 156–157 Navier–Stokes–Fourier equations, 161 Prandtl number, 160 pressure tensor, 157, 158 impulsively started flow problem, 116 microscale flows, 16 modified, 119–120 13-moment equations, 254 non-Fourier heat conduction Cattaneo’s equation, 121, 122 Fourier law, 120 hyperbolic heat conduction equation, 121 Vernotte number, 121 plate flow, leading edge, 116 Poiseuille flow, transition regime, 116 shock wave flow problem, 117, 122–123 Stokes’s first problem, 226–228 wave propagation, atmosphere of Mars, 117 Nusselt number, 89, 314

Index O Onsager–Burnett (OBurnett) equations base equation and change of variables, 290 features, 291–292 flowchart, 288 flow field variable derivatives, constants, 289–290 heat flux component, 289 linear stability analysis, 291 normal stress, 288 pressure tensor and heat flux vector expression, 287 shear stress, 289 Onsager consistent approach entropy, 284 first order distribution function, 285 Onsager–Burnett equations, 287–292 Onsager-13 moment equations derivation of, 292 and Grad closure, 294–295 Onsager-consistent distribution function, 292 stress tensor and heat flux vector, 293 phenomenological coefficients, 284 second order distribution function, 286–287 thermodynamic force, 284 P Parametric variation analysis through micropipe non-dimensional temperature profiles, 96 Nusselt number vs. Knudsen number, 96, 97 Nusselt number vs. modified Brinkman number, 96 velocity distribution, 95 micro-annulus gas flow, 102 non-dimensional temperature vs. radial co-ordinate, 103–104 non-dimensional velocity profiles, 103 Nusselt number vs. Brinkman number, 105, 106 Nusselt number vs. Knudsen number, 104–106 microchannel Knudsen number, 90 Nusselt number vs. Brinkman number, 91, 92 Nusselt number vs. Knudsen number, 90, 91 various modified Brinkman number, 90, 91

363 Particle image velocimetry, 308 PDMS-PDMS bonding, 309 Peclet number, 14, 108 Perturbation of Navier–Stokes equations dimensional mass flow rate, 53 governing equations, 51 non-dimensional variables, 49, 50 Reynolds number, 50 two-dimensional flow assumption, 49, 50 velocity profile in dimensional form, 52 x-momentum ˜ equation, 51 zeroth-order pressure distribution, 52 Prandtl number, 160 Pressure sensitive paint (PSP) measurement, 308, 309

R Rarefied gas flow measurement, 310–311 Regularized 13 (R13) moment equations closure expressions, 273 governing equations, 277 regularization method, 277 Chapman–Enskog scaling, 275 deviations for unknowns, 273 non-zero approximations, 273 process of, 274–275 transport equations, 277–278 unknown moments, 274 variable expressions, 273, 274 shock structure problem, 278 Regularized 26 (R26) moment equations Chapman–Enskog scaling, 282 collision constants, 281 constitutive relationships, 278 evolution equations, 278–279, 281 first order approximations, 282–283 functional form of distribution function, 280 non-zero approximations, 280 numerical solution, 283 transport equations, 283 variable expressions, 278, 279 zeroth order approximations, 282

S Slip coefficients, 314 Slip velocity, 11, 12 Sreekanth’s slip model, 38 Super Burnett equations augmented Burnett equations, 266 BGK-Burnett equations, 266–269

364 T Tangential momentum accommodation coefficient (TMAC), 35, 244 Tensorial form continuity equation, 26 equation of motion, 26 heat flux vector, 27 Stokes’ hypothesis, 26, 27 thermal energy equation, 27, 28 Tensor notation addition rule, 331 angular bracket notation, 330 contraction operation, 331 decomposition of tensors, 322–327 divergence operation, 332 division, 332 doubly contracted product, 333 dummy index, 330 Einstein summation, 319, 330, 331 evolution equation, 329 Gauss’s theorem, 332–333 gradient operator, 332 inner product, 331–332 Kronecker delta, 320–321 Laplacian, 332 product rule, 331 regular matrix notation, 321 results, 321–322 scalar quantity, 317 sequence of summation, 328–329 three-dimensional Cartesian coordinate system, 317–319 vector quantity, 317–318 velocity gradients, 328 viscous dissipation, 327 Thermal creep, 11, 12, 111 13-moment approximation boundary conditions, 224–225 Boltzmann equation, 218 distribution function, 219 gas–wall interaction, 218–219 kinetic energy, 223 laws of conservation of momentum and energy, 222, 223 Maxwellian distribution, 219–223 molecular distribution function, 220 tangential momentum, 222 two-dimensional flow problem, 218 closure, 215–216 distribution function, 208, 214–215 expression for variables, 208

Index second order moment, 208 stress tensor and heat flux vector evolution equation, 209 heat flux vector, 210 higher order moment, 212 ideal gas equation, 213 Lagrangian derivative, 210 Leibnitz’s rule, 210 Maxwellian distribution function, 213 Maxwellian molecules, 210, 211 momentum conservation equation, 211 third order moment, 209 third order moment, 208 13-moment equations Cattaneo’s model, 254–255 conservation laws, 216 continuity, 228 energy, 229 flowchart, 216, 217 governing equations, 231 heat, 229–230 initial conditions, 230 Laplace transform, 232 large values of time, 234–235 momentum, 229 Navier–Stokes equations, 254 plate and initial gas temperature difference, 230, 232 significance of, 225–226 small values of time, 233–234, 236–238 state, 229 stress, 229 tangential stress and tangential energy, 237 viscosity, 231 20-moment approximation closure, 204 distribution function, 203–204 expressions for variables, 193 flow diagram, 192, 193 Hermite polynomials, 192, 200–203 Pij evolution equation, 194–198 Sij k evolution equation, 198–199 20 scalar equations, 205–207

U Uniform heat flux axial temperature gradient, 87 bulk mean temperature, 89 constant wall, boundary condition, 87

Index non-dimensional boundary conditions, 88 non-dimensional bulk mean temperature, 89 non-dimensional temperature distribution, 88 non-dimensional variables, 87 Nusselt number, 89

365 V Value of slip coefficients different researchers report, 40, 42 mean TMAC, monatomic gases, 40, 43 relation, TMAC, 40, 41 TMAC vs. Knudsen number, 40, 43 Vernotte number, 121