i ii TRANSPORT PHENOMENA
An Introduction to Advanced Topics
LARRY A. GLASGOW Professor of Chemical Engineering Kansas State University Manhattan, Kansas
iii Copyright © 2010 by John Wiley & Sons, Inc. All rights reserved.
Published by John Wiley & Sons, Inc., Hoboken, New Jersey. Published simultaneously in Canada.
No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, scanning, or otherwise, except as permitted under Section 107 or 108 of the 1976 United States Copyright Act, without either the prior written permission of the Publisher, or authorization through payment of the appropriate per-copy fee to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, (978) 750-8400, fax (978) 750-4470, or on the web at www.copyright.com. Requests to the Publisher for permission should be addressed to the Permissions Department, John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, (201) 748-6011, fax (201) 748-6008, or online at http://www.wiley.com/go/permission.
Limit of Liability/Disclaimer of Warranty: While the publisher, the editors, and the authors have used their best efforts in preparing this book, they make no representations or warranties with respect to the accuracy or completeness of the contents of this book and specifically disclaim any implied warranties of merchantability or fitness for a particular purpose. No warranty may be created or extended by sales representatives or written sales materials. The advice and strategies contained herein may not be suitable for your situation. You should consult with a professional where appropriate. Neither the publisher nor author shall be liable for any loss of profit or any other commercial damages, including but not limited to special, incidental, consequential, or other damages.
For general information on our other products and services or for technical support, please contact our Customer Care Department within the United States at (800) 762-2974, outside the United States at (317) 572-3993 or fax (317) 572-4002.
Wiley also publishes its books in a variety of electronic formats. Some content that appears in print may not be available in electronic formats. For more information about Wiley products, visit our web site at www.wiley.com.
Library of Congress Cataloging-in-Publication Data
Glasgow, Larry A., 1950- Transport phenomena : an introduction to advanced topics / Larry A. Glasgow. p. cm. Includes index. ISBN 978-0-470-38174-8 (cloth) 1. Transport theory–Mathematics. I. Title. TP156.T7G55 2010 530.4’75–dc22 2009052127
Printed in the United States of America
10987654321
iv CONTENTS
Preface ix 3.13 Flows in Open Channels, 41 3.14 Pulsatile Flows in Cylindrical Ducts, 42 1. Introduction and Some Useful Review 1 3.15 Some Concluding Remarks for Incompressible 1.1 A Message for the Student, 1 Viscous Flows, 43 1.2 Differential Equations, 3 References, 44 1.3 Classification of Partial Differential Equations and Boundary Conditions, 7 4. External Laminar Flows and Boundary-Layer 1.4 Numerical Solutions for Partial Differential Theory 46 Equations, 8 4.1 Introduction, 46 1.5 Vectors, Tensors, and the Equation of Motion, 8 4.2 The Flat Plate, 47 1.6 The Men for Whom the Navier-Stokes Equations 4.3 Flow Separation Phenomena About Bluff are Named, 12 Bodies, 50 1.7 Sir Isaac Newton, 13 4.4 Boundary Layer on a Wedge: The Falkner–Skan References, 14 Problem, 52 4.5 The Free Jet, 53 2. Inviscid Flow: Simplified Fluid Motion 15 4.6 Integral Momentum Equations, 54 2.1 Introduction, 15 4.7 Hiemenz Stagnation Flow, 55 2.2 Two-Dimensional Potential Flow, 16 4.8 Flow in the Wake of a Flat Plate at Zero 2.3 Numerical Solution of Potential Flow Problems, 20 Incidence, 56 2.4 Conclusion, 22 4.9 Conclusion, 57 References, 23 References, 58
3. Laminar Flows in Ducts and Enclosures 24 5. Instability, Transition, and Turbulence 59 3.1 Introduction, 24 5.1 Introduction, 59 3.2 Hagen–Poiseuille Flow, 24 5.2 Linearized Hydrodynamic Stability Theory, 60 3.3 Transient Hagen–Poiseuille Flow, 25 5.3 Inviscid Stability: The Rayleigh Equation, 63 3.4 Poiseuille Flow in an Annulus, 26 5.4 Stability of Flow Between Concentric 3.5 Ducts with Other Cross Sections, 27 Cylinders, 64 3.6 Combined Couette and Poiseuille Flows, 28 5.5 Transition, 66 3.7 Couette Flows in Enclosures, 29 5.5.1 Transition in Hagen–Poiseuille 3.8 Generalized Two-Dimensional Fluid Motion in Flow, 66 Ducts, 32 5.5.2 Transition for the Blasius Case, 67 3.9 Some Concerns in Computational Fluid 5.6 Turbulence, 67 Mechanics, 35 5.7 Higher Order Closure Schemes, 71 3.10 Flow in the Entrance of Ducts, 36 5.7.1 Variations, 74 3.11 Creeping Fluid Motions in Ducts and Cavities, 38 5.8 Introduction to the Statistical Theory of 3.12 Microfluidics: Flow in Very Small Channels, 38 Turbulence, 74 3.12.1 Electrokinetic Phenomena, 39 5.9 Conclusion, 79 3.12.2 Gases in Microfluidics, 40 References, 81
v vi CONTENTS
6. Heat Transfer by Conduction 83 8.2 Unsteady Evaporation of Volatile Liquids: The 6.1 Introduction, 83 Arnold Problem, 120 6.2 Steady-State Conduction Problems in 8.3 Diffusion in Rectangular Geometries, 122 Rectangular Coordinates, 84 8.3.1 Diffusion into Quiescent Liquids: 6.3 Transient Conduction Problems in Rectangular Absorption, 122 Coordinates, 86 8.3.2 Absorption with Chemical Reaction, 123 6.4 Steady-State Conduction Problems in Cylindrical 8.3.3 Concentration-Dependent Diffusivity, 124 Coordinates, 88 8.3.4 Diffusion Through a Membrane, 125 6.5 Transient Conduction Problems in Cylindrical 8.3.5 Diffusion Through a Membrane with Coordinates, 89 Variable D, 125 6.6 Steady-State Conduction Problems in Spherical 8.4 Diffusion in Cylindrical Systems, 126 Coordinates, 92 8.4.1 The Porous Cylinder in Solution, 126 6.7 Transient Conduction Problems in Spherical 8.4.2 The Isothermal Cylindrical Catalyst Coordinates, 93 Pellet, 127 6.8 Kelvin’s Estimate of the Age of the Earth, 95 8.4.3 Diffusion in Squat (Small L/d) 6.9 Some Specialized Topics in Conduction, 95 Cylinders, 128 6.9.1 Conduction in Extended Surface Heat 8.4.4 Diffusion Through a Membrane with Edge Transfer, 95 Effects, 128 6.9.2 Anisotropic Materials, 97 8.4.5 Diffusion with Autocatalytic Reaction in a 6.9.3 Composite Spheres, 99 Cylinder, 129 6.10 Conclusion, 100 8.5 Diffusion in Spherical Systems, 130 References, 100 8.5.1 The Spherical Catalyst Pellet with Exothermic Reaction, 132 7. Heat Transfer with Laminar Fluid Motion 101 8.5.2 Sorption into a Sphere from a Solution of 7.1 Introduction, 101 Limited Volume, 133 7.2 Problems in Rectangular Coordinates, 102 8.6 Some Specialized Topics in Diffusion, 133 7.2.1 Couette Flow with Thermal Energy 8.6.1 Diffusion with Moving Boundaries, 133 Production, 103 8.6.2 Diffusion with Impermeable 7.2.2 Viscous Heating with Obstructions, 135 Temperature-Dependent Viscosity, 104 8.6.3 Diffusion in Biological Systems, 135 7.2.3 The Thermal Entrance Region in Rectangular 8.6.4 Controlled Release, 136 Coordinates, 104 8.7 Conclusion, 137 7.2.4 Heat Transfer to Fluid Moving Past a Flat References, 137 Plate, 106 7.3 Problems in Cylindrical Coordinates, 107 9. Mass Transfer in Well-Characterized Flows 139 7.3.1 Thermal Entrance Length in a Tube: The 9.1 Introduction, 139 Graetz Problem, 108 9.2 Convective Mass Transfer in Rectangular 7.4 Natural Convection: Buoyancy-Induced Fluid Coordinates, 140 Motion, 110 9.2.1 Thin Film on a Vertical Wall, 140 7.4.1 Vertical Heated Plate: The Pohlhausen 9.2.2 Convective Transport with Reaction at the Problem, 110 Wall, 141 7.4.2 The Heated Horizontal Cylinder, 111 9.2.3 Mass Transfer Between a Flowing Fluid and 7.4.3 Natural Convection in Enclosures, 112 a Flat Plate, 142 7.4.4 Two-Dimensional Rayleigh–Benard 9.3 Mass Transfer with Laminar Flow in Cylindrical Problem, 114 Systems, 143 7.5 Conclusion, 115 9.3.1 Fully Developed Flow in a Tube, 143 References, 116 9.3.2 Variations for Mass Transfer in a Cylindrical Tube, 144 8. Diffusional Mass Transfer 117 9.3.3 Mass Transfer in an Annulus with Laminar 8.1 Introduction, 117 Flow, 145 8.1.1 Diffusivities in Gases, 118 9.3.4 Homogeneous Reaction in Fully-Developed 8.1.2 Diffusivities in Liquids, 119 Laminar Flow, 146 CONTENTS vii
9.4 Mass Transfer Between a Sphere and a Moving 11.2 Liquid–Liquid Systems, 180 Fluid, 146 11.2.1 Droplet Breakage, 180 9.5 Some Specialized Topics in Convective Mass 11.3 Particle–Fluid Systems, 183 Transfer, 147 11.3.1 Introduction to Coagulation, 183 9.5.1 Using Oscillatory Flows to Enhance 11.3.2 Collision Mechanisms, 183 Interphase Transport, 147 11.3.3 Self-Preserving Size Distributions, 186 9.5.2 Chemical Vapor Deposition in Horizontal 11.3.4 Dynamic Behavior of the Particle Size Reactors, 149 Distribution, 186 9.5.3 Dispersion Effects in Chemical 11.3.5 Other Aspects of Particle Size Distribution Reactors, 150 Modeling, 187 9.5.4 Transient Operation of a Tubular 11.3.6 A Highly Simplified Example, 188 Reactor, 151 11.4 Multicomponent Diffusion in Gases, 189 9.6 Conclusion, 153 11.4.1 The Stefan–Maxwell Equations, 189 References, 153 11.5 Conclusion, 191 References, 192 10. Heat and Mass Transfer in Turbulence 155 10.1 Introduction, 155 Problems to Accompany Transport Phenomena: An 10.2 Solution Through Analogy, 156 Introduction to Advanced Topics 195 10.3 Elementary Closure Processes, 158 10.4 Scalar Transport with Two-Equation Models of Appendix A: Finite Difference Approximations for Turbulence, 161 Derivatives 238 10.5 Turbulent Flows with Chemical Reactions, 162 Appendix B: Additional Notes on Bessel’s Equation and 10.5.1 Simple Closure Schemes, 164 Bessel Functions 241 10.6 An Introduction to pdf Modeling, 165 10.6.1 The Fokker–Planck Equation and pdf Appendix C: Solving Laplace and Poisson (Elliptic) Modeling of Turbulent Reactive Partial Differential Equations 245 Flows, 165 10.6.2 Transported pdf Modeling, 167 Appendix D: Solving Elementary Parabolic Partial 10.7 The Lagrangian View of Turbulent Differential Equations 249 Transport, 168 10.8 Conclusions, 171 Appendix E: Error Function 253 References, 172 Appendix F: Gamma Function 255 11. Topics in Multiphase and Multicomponent Systems 174 Appendix G: Regular Perturbation 257 11.1 Gas–Liquid Systems, 174 11.1.1 Gas Bubbles in Liquids, 174 Appendix H: Solution of Differential Equations by 11.1.2 Bubble Formation at Orifices, 176 Collocation 260 11.1.3 Bubble Oscillations and Mass Transfer, 177 Index 265 viii PREFACE
This book is intended for advanced undergraduates and first- Problem solving in transport phenomena has consumed year graduate students in chemical and mechanical engineer- much of my professional life. The beauty of the field is that ing. Prior formal exposure to transport phenomena or to sep- it matters little whether the focal point is tissue engineering, arate courses in fluid flow and heat transfer is assumed. Our chemical vapor deposition, or merely the production of gaso- objectives are twofold: to learn to apply the principles of line; the principles of transport phenomena apply equally to transport phenomena to unfamiliar problems, and to improve all. The subject is absolutely central to the formal study of our methods of attack upon such problems. This book is suit- chemical and mechanical engineering. Moreover, transport able for both formal coursework and self-study. phenomena are ubiquitous—all aspects of life, commerce, In recent years, much attention has been directed toward and production are touched by this engineering science. I can the perceived “paradigm shift” in chemical engineering ed- only hope that you enjoy the study of this material as much ucation. Some believe we are leaving the era of engineering asIhave. science that blossomed in the 1960s and are entering the age It is impossible to express what is owed to Linda, Andrew, of molecular biology. Proponents of this viewpoint argue that and Hillary, each of whom enriched my life beyond measure. dramatic changes in engineering education are needed. I sus- And many of the best features of the person I am are due to pect that the real defining issues of the next 25–50 years are the formative influences of my mother Betty J. (McQuilkin) not yet clear. It may turn out that the transformation from Glasgow, father Loren G. Glasgow, and sister Barbara J. petroleum-based fuels and economy to perhaps a hydrogen- (Glasgow) Barrett. based economy will require application of engineering skills and talent at an unprecedented intensity. Alternatively, we Larry A. Glasgow may have to marshal our technically trained professionals to stave off disaster from global climate change, or to combat a viral pandemic. What may happen is murky, at best. How- ever, I do expect the engineering sciences to be absolutely Department of Chemical Engineering, Kansas State University, crucial to whatever technological crises emerge. Manhattan, KS
ix x 1 INTRODUCTION AND SOME USEFUL REVIEW
1.1 A MESSAGE FOR THE STUDENT Fundamentals of Momentum, Heat, and Mass Transfer, 4th edition, Welty, Wicks, Wilson, and Rorrer. This is an advanced-level book based on a course sequence Fluid Mechanics for Chemical Engineers, 2nd edition, taught by the author for more than 20 years. Prior exposure Wilkes. to transport phenomena is assumed and familiarity with the Vectors, Tensors, and the Basic Equations of Fluid classic, Transport Phenomena, 2nd edition, by R. B. Bird, Mechanics, Aris. W. E. Stewart, and E. N. Lightfoot (BS&L), will prove par- ticularly advantageous because the notation adopted here is mainly consistent with BS&L. In addition, there are many other more specialized works There are many well-written and useful texts and mono- that treat or touch upon some facet of transport phenom- graphs that treat aspects of transport phenomena. A few of ena. These books can be very useful in proper circumstances the books that I have found to be especially valuable for and they will be clearly indicated in portions of this book engineering problem solving are listed here: to follow. In view of this sea of information, what is the point of yet another book? Let me try to provide my rationale Transport Phenomena, 2nd edition, Bird, Stewart, and below. Lightfoot. I taught transport phenomena for the first time in 1977– 1978. In the 30 years that have passed, I have taught our An Introduction to Fluid Dynamics and An Introduction graduate course sequence, Advanced Transport Phenomena 1 to Mass and Heat Transfer, Middleman. and 2, more than 20 times. These experiences have convinced Elements of Transport Phenomena, Sissom and Pitts. me that no suitable single text exists in this niche, hence, this Transport Analysis, Hershey. book. Analysis of Transport Phenomena, Deen. So, the course of study you are about to begin here is the Transport Phenomena Fundamentals, Plawsky. course sequence I provide for our first-year graduate students. Advanced Transport Phenomena, Slattery. It is important to note that for many of our students, formal exposure to fluid mechanics and heat transfer ends with this Advanced Transport Phenomena: Fluid Mechanics and course sequence. It is imperative that such students leave the Convective Transport Processes, Leal. experience with, at the very least, some cognizance of the The Phenomena of Fluid Motions, Brodkey. breadth of transport phenomena. Of course, this reality has Fundamentals of Heat and Mass Transfer, Incropera and profoundly influenced this text. De Witt. In 1982, I purchased my first IBM PC (personal computer); Fluid Dynamics and Heat Transfer, Knudsen and Katz. by today’s standards it was a kludge with a very low clock rate,
Transport Phenomena: An Introduction to Advanced Topics, By Larry A. Glasgow Copyright © 2010 John Wiley & Sons, Inc.
1 2 INTRODUCTION AND SOME USEFUL REVIEW just 64K memory, and 5.25 (160K) floppy drives. The high- that particular scenario. But, blind acceptance of black-box level language available at that time was interpreted BASIC computations for an untested situation is foolhardy. that had severe limits of its own with respect to execution One of my principal objectives in transport phenomena speed and array size. Nevertheless, it was immediately appar- instruction is to help the student develop physical insight and ent that the decentralization of computing power would spur problem-solving capability simultaneously. This balance is a revolution in engineering problem solving. By necessity I essential because either skill set alone is just about useless. became fairly adept at BASIC programming, first using the In this connection, we would do well to remember G. K. interpreter and later using various BASIC compilers. Since Batchelor’s (1967) admonition: “By one means or another, 1982, the increases in PC capability and the decreases in cost a teacher should show the relation between his analysis and have been astonishing; it now appears that Moore’s “law” (the the behavior of real fluids; fluid dynamics is much less inter- number of transistors on an integrated circuit yielding mini- esting if it is treated largely as an exercise in mathematics.” I mum component cost doubles every 24 months) may continue also feel strongly that how and why this field of study devel- to hold true through several more generations of chip devel- oped is not merely peripheral; one can learn a great deal by opment. In addition, PC hard-drive capacity has exhibited obtaining a historical perspective and in many instances I exponential growth over that time frame and the estimated have tried to provide this. I believe in the adage that you can- cost per G-FLOP has decreased by a factor of about 3 every not know where you are going if you do not know where you year for the past decade. have been. Many of the accompanying problems have been It is not an exaggeration to say that a cheap desktop PC developed to provide a broader view of transport phenom- in 2009 has much more computing power than a typical ena as well; they constitute a unique feature of this book, university mainframe computer of 1970. As a consequence, and many of them require the student to draw upon other problems that were pedagogically impractical are now rou- resources. tine. This computational revolution has changed the way I I have tried to recall questions that arose in my mind approach instruction in transport phenomena and it has made when I was beginning my second course of study of trans- it possible to assign more complex exercises, even embrac- port phenomena. I certainly hope that some of these have ing nonlinear problems, and still maintain expectations of been clearly treated here. For many of the examples used in timely turnaround of student work. It was my intent that this this book, I have provided details that might often be omitted, computational revolution be reflected in this text and in some but this has a price; the resulting work cannot be as broad as of the problems that accompany it. However, I have avoided one might like. There are some important topics in transport significant use of commercial software for problem solutions. phenomena that are not treated in a substantive way in this Many engineering educators have come to the realization book. These omissions include non-Newtonian rheology and that computers (and the microelectronics revolution in gen- energy transport by radiation. Both topics deserve far more eral) are changing the way students learn. The ease with consideration than could be given here; fortunately, both are which complicated information can be obtained and diffi- subjects of numerous specialized monographs. In addition, cult problems can be solved has led to a physical disconnect; both boundary-layer theory and turbulence could easily be students have far fewer opportunities to develop somatic com- taught as separate one- or even two-semester courses. That prehension of problems and problem solving in this new envi- is obviously not possible within our framework. I would like ronment. The reduced opportunity to experience has led to a to conclude this message with five observations: reduced ability to perceive, and with dreadful consequence. Recently, Haim Baruh (2001) observed that the computer rev- 1. Transport phenomena are pervasive and they impact olution has led young people to “think, learn and visualize upon every aspect of life. differently.... Because information can be found so easily 2. Rote learning is ineffective in this subject area because and quickly, students often skip over the basics. For the most the successful application of transport phenomena is part, abstract concepts that require deeper thought aren’t part directly tied to physical understanding. of the equation. I am concerned that unless we use computers wisely, the decline in student performance will continue.” 3. Mastery of this subject will enable you to critically eval- Engineering educators must remember that computers are uate many physical phenomena, processes, and systems merely tools and skillful use of a commercial software pack- across many disciplines. age does not translate to the type of understanding needed 4. Student effort is paramount in graduate education. for the formulation and analysis of engineering problems. In There are many places in this text where outside read- this regard, I normally ask students to be wary of reliance ing and additional study are not merely recommended, upon commercial software for solution of problems in trans- but expected. port phenomena. In certain cases, commercial codes can be 5. Time has not diminished my interest in transport phe- used for comparison of alternative models; this is particularly nomena, and my hope is that through this book I can useful if the software can be verified with known results for share my enthusiasm with students. DIFFERENTIAL EQUATIONS 3
1.2 DIFFERENTIAL EQUATIONS
Students come to this sequence of courses with diverse math- ematical backgrounds. Some do not have the required levels of proficiency, and since these skills are crucial to success, a brief review of some important topics may be useful. Transport phenomena are governed by, and modeled with, differential equations. These equations may arise through mass balances, momentum balances, and energy balances. The main equations of change are second-order partial differ- ential equations that are (too) frequently nonlinear. One of our principal tasks in this course is to find solutions for such equa- tions; we can expect this process to be challenging at times. Let us begin this section with some simple examples of ordinary differential equations (ODEs); consider
dy = c (c is constant) (1.1) dx FIGURE 1.1. Solutions for dy/dx = 1, dy/dx = y, and dy/dx = 2xy. and = = dy depend on the product of a and b.Ifweleta b 1, then = y. (1.2) dx y = tan(x + C1). (1.6) Both are linear, first-order ordinary differential equations. Remember that linearity is determined by the dependent vari- able y. The solutions for (1.1) and (1.2) are Before we press forward, we note that Riccati equations were studied by Euler, Liouville, and the Bernoulli’s (Johann and Daniel), among others. How will the solution change if y = cx + C and y = C exp(x), respectively. (1.3) 1 1 eq. (1.5) is rewritten as Note that if y(x = 0) is specified, then the behavior of y is set dy for all values of x. If the independent variable x were time t, = 1 − y2? (1.7) then the future behavior of the system would be known. This dx is what we mean when we say that a system is deterministic. Now, what happens when we modify (1.2) such that Of course, the equation is still separable, so we can write
dy = 2xy? (1.4) dy dx = x + C1. (1.8) 1 − y2 2 We find that y = C1 exp(x ). These first-order linear ODEs have all been separable, admitting simple solution. We will Show that the solution of (1.8), given that y(0) = 0, is sketch the (three) behaviors for y(x) on the interval 0–2, given y = tanh(x). that y(0) = 1 (Figure 1.1). Match each of the three curves with When a first-order differential equation arises in transport the appropriate equation. phenomena, it is usually by way of a macroscopic balance, Note what happens to y(x) if we continue to add addi- for example, [Rate in] − [Rate out] = [Accumulation]. Con- tional powers of x to the right-hand side of (1.4), allowing y sider a 55-gallon drum (vented) filled with water. At t = 0, to remain. If we add powers of y instead—and make the equa- a small hole is punched through the side near the bottom tion inhomogeneous—we can expect to work a little harder. and the liquid begins to drain from the tank. If we let the Consider this first-order nonlinear ODE: velocity of the fluid through the orifice be represented by Torricelli’s theorem (a frictionless result), a mass balance dy reveals = a + by2. (1.5) dx dh R2 This is a type of Riccati equation (Jacopo Francesco Count =− 0 2 2gh, (1.9) Riccati, 1676–1754) and the nature of the solution will dt RT 4 INTRODUCTION AND SOME USEFUL REVIEW where R0 is the radius of the hole. This equation is easily (D2 + 2D + 1)y (D + 1)(D + 1), (1.16) solved as 2 √ √ g R2 + − = − 0 + 2 + + + 3 5 + 3 5 h 2 t C1 . (1.10) (D 3D 1)y (D )(D ). 2 RT 2 2 (1.17) The drum is initially full, so h(t = 0) = 85 cm and 1/2 C1 = 9.21954 cm . Since the drum diameter is about 56 cm, Now suppose the forcing function f(x) in (1.12)–(1.14) is a RT = 28 cm; if the radius of the hole is 0.5 cm, it will take constant, say 1. What do (1.15)–(1.17) tell you about the about 382 s for half of the liquid to flow out and about 893 s nature of possible solutions? The complex conjugate roots in for 90% of the fluid to escape. If friction is taken into account, (1.15) will result in oscillatory behavior. Note that all three how would (1.9) be changed, and how much more slowly of these second-order differential equations have constant would the drum drain? coefficients and a first derivative term. If eq. (1.14) had been We now contemplate an increase in the order of the dif- developed by force balance (with x replaced by t), the dy/dx ferential equation. Suppose we have (velocity) term might be some kind of frictional resistance. We do not have to expend much effort to find second-order d2y + a = 0, (1.11) ODEs that pose greater challenges. What if you needed a dx2 solution for the nonlinear equation where a is a constant or an elementary function of x. This is d2y a common equation type in transport phenomena for steady- = a + by + cy2 + dy3? (1.18) dx2 state conditions with molecular transport occurring in one direction. We can immediately write Actually, a number of closely related equations have fig- ured prominently in physics. Einstein, in an investigation of dy =− adx+ C , and if a is a constant, planetary motion, was led to consider dx 1 a d2y y =− x2 + C x + C . + y = a + by2. (1.19) 2 1 2 dx2 Give an example of a specific type of problem that produces Duffing, in an investigation of forced vibrations, carried out this solution. One of the striking features of (1.11) is the a study of the equation absence of a first derivative term. You might consider what d2y dy conditions would be needed in, say, a force balance to produce + k + ay + by3 = f (x). (1.20) both first and second derivatives. dx2 dx The simplest second-order ODEs (that include first A limited number of nonlinear, second-order differential derivatives) are linear equations with constant coefficients. equations can be solved with (Jacobian) elliptic functions. Consider For example, Davis (1962) shows that the solution of the d2y dy nonlinear equation + 1 + y = f (x), (1.12) dx2 dx d2y = 6y2 (1.21) dx2 d2y dy + 2 + y = f (x), (1.13) can be written as dx2 dx B and y = A + . 2 (1.22) sn (C(x − x1)) d2y dy + 3 + y = f (x). (1.14) dx2 dx Tabulated values are available for the Jacobi elliptic sine, sn; see pages 175–176 in Davis (1962). The reader desiring Using linear differential operator notation, we rewrite the an introduction to elliptic functions is encouraged to work left-hand side of each and factor the result: problem 1.N in this text, read Chapter 5 in Vaughn (2007), √ √ and consult the extremely useful book by Milne-Thomson 1 3 1 3 (D2 + D + 1)y (D + + i)(D + − i), (1950). 2 2 2 2 The point of the immediately preceding discussion is as (1.15) follows: The elementary functions that are familiar to us, such DIFFERENTIAL EQUATIONS 5 as sine, cosine, exp, ln, etc., are solutions to linear differential Gollub (1990) described this map as having regions where equations. Furthermore, when constants arise in the solution the behavior is chaotic with windows of periodicity. of linear differential equations, they do so linearly; for an Note that the chaotic behavior seen above is attained example, see the solution of eq. (1.11) above. In nonlinear through a series of period doublings (or pitchfork bifurca- differential equations, arbitrary constants appear nonlinearly. tions). Baker and Gollub note that many dynamical systems Nonlinear problems abound in transport phenomena and we exhibit this path to chaos. In 1975, Mitchell Feigenbaum can expect to find analytic solutions only for a very lim- began to look at period doublings for a variety of rather sim- ited number of them. Consequently, most nonlinear problems ple functions. He quickly discovered that all of them had must be solved numerically and this raises a host of other a common characteristic, a universality; that is, the ratio of issues, including existence, uniqueness, and stability. the spacings between successive bifurcations was always the So much of our early mathematical education is bound same: to linearity that it is difficult for most of us to perceive and appreciate the beauty (and beastliness) in nonlinear equa- 4.6692016 ... (Feigenbaum number). tions. We can illustrate some of these concerns by examining the elementary nonlinear difference (logistic) equation, This leads us to hope that a relatively simple system or func- tion might serve as a model (or at least a surrogate) for far Xn+1 = αXn(1 − Xn). (1.23) more complex behavior. We shall complete this part of our discussion by select- Let the parameter α assume an initial value of about 3.5 ing two terms from the x-component of the Navier–Stokes and let X1 = 0.5. Calculate the new value of X and insert equation, it on the right-hand side. As we repeat this procedure, the following sequence emerges: 0.5, 0.875, 0.38281, 0.82693, ∂v ∂v x + v x +···, (1.24) 0.5009, 0.875, 0.38282, 0.82694, .... Now allow α to assume ∂t x ∂x a slightly larger value, say 3.575. Then, the sequence of cal- culated values is 0.5, 0.89375, 0.33949, 0.80164, 0.56847, and writing them in finite difference form, letting i be the 0.87699, 0.38567, 0.84702, 0.46324, 0.88892, 0.353, 0.8165, spatial index and j the temporal one. We can drop the subscript 0.53563, 0.88921, 0.35219, 0.81564, 0.53757, ....Wecan “x” for convenience. One of the possibilities (though not a continue this process and report these results graphically; the very good one) is result is a bifurcation diagram. How would you character- vi,j+1 − vi,j vi+1,j − vi,j ize Figure 1.2? Would you be tempted to use “chaotic” as a + vi,j +···. (1.25) descriptor? The most striking feature of this logistic map is t x that a completely deterministic equation produces behavior We might imagine this being rewritten as an explicit algo- that superficially appears to be random (it is not). Baker and rithm (where we calculate v at the new time, j + 1, using velocities from the jth time step) in the following form:
t v + ≈ v − v (v + − v ) +···. (1.26) i,j 1 i,j x i,j i 1,j i,j
Please make note of the dimensionless quantity tvi,j /x; this is the Courant number, Co, and it will be extremely important to us later. As a computational scheme, eq. (1.26) is generally unworkable, but note the similarity to the logistic equation above. The nonlinear character of the equations that govern fluid motion guarantees that we will see unexpected beauty and maddening complexity, if only we knew where (and how) to look. In this connection, a system that evolves in time can often be usefully studied using phase space analysis, which is an underutilized tool for the study of the dynamics of low- dimension systems. Consider a periodic function such as f(t) = A sin(ωt). The derivative of this function is ωA cos(ωt). If we cross-plot f(t) and df/dt, we will obtain a limit cycle FIGURE 1.2. Bifurcation diagram for the logistic equation with in the shape of an ellipse. That is, the system trajectory in the Verhulst parameter α ranging from 2.9 to 3.9. phase space takes the form of a closed path, which is expected 6 INTRODUCTION AND SOME USEFUL REVIEW
What we see here is the combination of a limited number of periodic functions interacting. Particular points in phase space are revisited fairly regularly. But, if the dynamic behav- ior of a system was truly chaotic, we might see a phase space in which no point is ever revisited. The implications for the behavior of a perturbed complex nonlinear system, such as the global climate, are sobering. Another consequence of nonlinearity is sensitivity to ini- tial conditions; to solve a general fluid flow problem, we would need to consider three components of the Navier– Stokes equation and the continuity relation simultaneously. Imagine an integration scheme forward marching in time. It would be necessary to specify initial values for vx , vy , vz , and p. Suppose that vx had the exact initial value, 5 cm/s, but your computer represented the number as 4.99999...cm/s. Would the integration scheme evolve along the “correct” pathway? Possibly not. Jules-Henri Poincare(who´ was perhaps the last FIGURE 1.3. “Artificial” time-series data constructed from man to understand all of the mathematics of his era) noted sinusoids. in 1908 that “... small differences in the initial conditions produce very great ones in the final phenomena.” In more recent years, this concept has become popularly known as the behavior for a purely periodic function. If, on the other hand, “butterfly effect” in deference to Edward Lorenz (1963) who we had an oscillatory system that was unstable, the ampli- observed that the disturbance caused by a butterfly’s wing tude of the oscillations would grow in time; the resulting might change the weather pattern for an entire hemisphere. phase-plane portrait would be an outward spiral. An attenu- This is an idea that is unfamiliar to most of us; in much of the ated (damped) oscillation would produce an inward spiral. educational process we are conditioned to believe a model This technique can be useful for more complicated func- for a system (a differential equation), taken together with its tions or signals as well. Consider the oscillatory behavior present state, completely set the future behavior of the system. illustrated in Figure 1.3. Let us conclude this section with an appropriate exam- If you look closely at this figure, you can see that the ple; we will explore the Rossler (1976) problem that consists function f(t) does exhibit periodic behavior—many features of the following set of three (deceptively simple) ordinary of the system output appear repeatedly. In phase space, this differential equations: system yields the trajectory shown in Figure 1.4. dX dY =−Y − Z, = X + 0.2Y, and dt dt dZ = 0.2 + Z(X − 5.7). (1.27) dt
Note that there is but one nonlinearity in the set, the prod- uct ZX. The Rossler model is synthetic in the sense that it is an abridgement of the Lorenz model of local climate; conse- quently, it does not have a direct physical basis. But it will reveal some unexpected and important behavior. Our plan is to solve these equations numerically using the initial values of 0, −6.78, and 0.02 for X, Y, and Z, respectively. We will look at the evolution of all three dependent variables with time, and then we will examine a segment or cut from the system trajectory by cross-plotting X and Y. The main point to take from this example is that an elementary, low-dimensional system can exhibit unexpect- edly complicated behavior. The system trajectory seen in Figure 1.5b is a portrait of what is now referred to in the FIGURE 1.4. Phase space portrait of the system dynamics illus- literature as a “strange” attractor. The interested student is trated in Figure 1.3. encouraged to read the papers by Rossler (1976) and Packard CLASSIFICATION OF PARTIAL DIFFERENTIAL EQUATIONS ANDBOUNDARY CONDITIONS 7
FIGURE 1.5. The Rossler model: X(t), Y(t), and Z(t) for 0 < t < 200 (a), and a cut from the system trajectory (Y plotted against X) (b). et al. (1980). The formalized study of chaotic behavior is 1.3 CLASSIFICATION OF PARTIAL still in its infancy, but it has become clear that there are DIFFERENTIAL EQUATIONS AND applications in hydrodynamics, mechanics, chemistry, etc. BOUNDARY CONDITIONS There are additional tools that can be used to determine whether a particular system’s behavior is periodic, aperiodic, We have to be able to recognize and classify partial differen- or chaotic. For example, the rate of divergence of a chaotic tial equations to attack them successfully; a book like Powers trajectory about an attractor is characterized with Lyapunov (1979) can be a valuable ally in this effort. Consider the gen- exponents. Baker and Gollub (1990) describe how the expo- eralized second-order partial differential equation, where φ is nents are computed in Chapter 5 of their book and they the dependent variable and x and y are arbitrary independent include a listing of a BASIC program for this task. The Fourier variables: transform is also invaluable in efforts to identify important ∂2φ ∂2φ ∂2φ ∂φ ∂φ periodicities in the behavior of nonlinear systems. We will A + B + C + D + E + Fφ + G = 0. make extensive use of the Fourier transform in our consider- ∂x2 ∂x∂y ∂y2 ∂x ∂y ation of turbulent flows. (1.28) The student with further interest in this broad subject area A, B, C, D, E, F, and G can be functions of x and y,butnot of is also encouraged to read the recent article by Porter et al. φ. This linear partial differential equation can be classified (2009). This paper treats a historically significant project car- as follows: ried out at Los Alamos by Fermi, Pasta, and Ulam (Report LA-1940). Fermi, Pasta, and Ulam (FPU) investigated a one- B2 − 4AC<0 (elliptic), dimensional mass-and-spring problem in which 16, 32, and B2 − 4AC = 0 (parabolic), 64 masses were interconnected with non-Hookean springs. They experimented (computationally) with cases in which B2 − 4AC>0 (hyperbolic). the restoring force was proportional to displacement raised to the second or third power(s). FPU found that the nonlinear For illustration, we look at the “heat” equation (transient systems did not share energy (in the expected way) with the conduction in one spatial dimension): higher modes of vibration. Instead, energy was exchanged ∂T ∂2T ultimately among just the first few modes, almost period- = α . (1.29) ∂t ∂y2 ically. Since their original intent had been to explore the rate at which the initial energy was distributed among all of You can see that A = α , B = 0, and C = 0; the equation is the higher modes (they referred to this process as “thermal- parabolic. Compare this with the governing (Laplace) equa- ization”), they quickly realized that the nonlinearities were tion for two-dimensional potential flow (ψ is the stream producing quite unexpectedly localized behavior in phase function): space! The work of FPU represents one of the very first cases in which extensive computational experiments were ∂2ψ ∂2ψ + = 0. (1.30) performed for nonlinear systems. ∂x2 ∂y2 8 INTRODUCTION AND SOME USEFUL REVIEW
In this case, A = 1 and C = 1 while B = 0; the equation uses a lumped-parameter model to described the cooling is elliptic. Next, we consider a vibrating string (the wave of an object, mCp(dT/dt) =−hA(T − T∞), then the oft- equation): cited form does produce an exponential decrease in the object’s temperature in accordance with Newton’s own obser- ∂2u ∂2u vation. So, do we have an argument over substance or = s2 . (1.31) ∂t2 ∂y2 merely semantics? Perhaps the solution is to exercise greater care when we refer to q = h(T − T∞); we should prob- Note that A = 1 and C =−s2; therefore, −4AC > 0 and ably call it the defining equation for the heat transfer eq. (1.31) is hyperbolic. In transport phenomena, transient coefficient h and meticulously avoid calling the expression problems with molecular transport only (heat or diffusion a “law.” equations) will have parabolic character. Equilibrium prob- lems such as steady-state diffusion, conduction, or viscous flow in a duct will be elliptic in nature (phenomena governed 1.4 NUMERICAL SOLUTIONS FOR PARTIAL by Laplace- or Poisson-type partial differential equations). DIFFERENTIAL EQUATIONS We will see numerous examples of both in the chapters to come. Hyperbolic equations are common in quantum Many of the examples of numerical solution of partial dif- mechanics and high-speed compressible flows, for example, ferential equations used in this book are based on finite inviscid supersonic flow about an airfoil. The Navier–Stokes difference methods (FDMs). The reader may be aware that equations that will be so important to us later are of mixed the finite element method (FEM) is widely used in commer- character. cial software packages for the same purpose. The FEM is The three most common types of boundary conditions particularly useful for problems with either curved or irregu- used in transport phenomena are Dirichlet, Neumann, and lar boundaries and in cases where localized changes require a Robin’s. For Dirichlet boundary conditions, the field variable smaller scale grid for improved resolution. The actual numer- is specified at the boundary. Two examples: In a conduction ical effort required for solution in the two cases is comparable. problem, the temperature at a surface might be fixed (at y = 0, However, FEM approaches usually employ a separate code T = T0); alternatively, in a viscous fluid flow problem, the (or program) for mesh generation and refinement. I decided velocity at a stationary duct wall would be zero. For Neu- not to devote space here to this topic because my intent mann conditions, the flux is specified; for example, for a was to make the solution procedures as general as possi- conduction problem with an insulated wall located at y = 0, ble and nearly independent of the computing platform and = (∂T/∂y)y=0 0. A Robin’s type boundary condition results software. By taking this approach, the student without access from equating the fluxes; for example, consider the solid– to specialized commercial software can still solve many of fluid interface in a heat transfer problem. On the solid side the problems in the course, in some instances using nothing heat is transferred by conduction (Fourier’s law), but on the more complicated than either a spreadsheet or an elementary fluid side of the interface we might have mixed heat trans- understanding of any available high-level language. fer processes approximately described by Newton’s “law” of cooling: 1.5 VECTORS, TENSORS, AND THE EQUATION ∂T OF MOTION −k = h(T0 − Tf ). (1.32) ∂y y=0 For the discussion that follows, recall that temperature T is We hasten to add that the heat transfer coefficient h that a scalar (zero-order, or rank, tensor), velocity V is a vec- appears in (1.32) is an empirical quantity. The numerical tor (first-order tensor), and stress τ is a second-order tensor. value of h is known only for a small number of cases, usually Tensor is from the Latin “tensus,” meaning to stretch. We those in which molecular transport is dominant. can offer the following, rough, definition of a tensor: It is One might think that Newton’s “law” of cooling could a generalized quantity or mathematical object that in three- not possibly engender controversy. That would be a flawed dimensional space has 3n components (where n is the order, presumption. Bohren (1991) notes that Newton’s own or rank, of the tensor). From an engineering perspective, ten- description of the law as translated from Latin is “if equal sors are defined over a continuum and transform according times of cooling be taken, the degrees of heat will be to certain rules. They figure prominently in mechanics (stress in geometrical proportion, and therefore easily found by and strain) and relativity. tables of logarithms.” It is clear from these words that The del operator (∇) in rectangular coordinates is Newton meant that the cooling process would proceed exponentially. Thus, to simply write q = h(T − T∞), with- ∂ ∂ ∂ δ + δ + δ . (1.33) out qualification, is “incorrect.” On the other hand, if one x ∂x y ∂y z ∂z VECTORS, TENSORS, AND THE EQUATION OF MOTION 9
For a scalar such as T, ∇T is referred to as the gradient (of the extremely useful to us in hydrodynamic calculations because scalar field). So, when we speak of the temperature gradient, in the interior of a homogeneous fluid vorticity is neither we are talking about a vector quantity with both direction and created nor destroyed; it is produced solely at the flow bound- magnitude. aries. Therefore, it often makes sense for us to employ the A scalar product can be formed by applying ∇to the veloc- vorticity transport equation that is obtained by taking the curl ity vector: of the equation of motion. We will return to this point and explore it more thoroughly later. In cylindrical coordinates, ∂v ∂v ∂v ∇× ∇·V = x + y + z , (1.34) V is ∂x ∂y ∂z 1 ∂v ∂v z − θ (1.37a) which is the divergence of the velocity, div(V). The physical r ∂θ ∂z meaning should be clear to you: For an incompressible fluid ∂vr ∂vz = ∇· = ∇×V = − (1.37b) (ρ constant), conservation of mass requires that V 0; ∂z ∂r in 3-space, if vx changes with x, the other velocity vector 1 ∂ 1 ∂vr components must accommodate the change (to prevent a net (rvθ) − (1.37c) outflow). You may recall that a mass balance for an element r ∂r r ∂θ of compressible fluid reveals that the continuity equation is These equations, (1.37a)–(1.37c), correspond to the r, θ, and z ∂ρ ∂ ∂ ∂ components of the vorticity vector, respectively. + (ρv ) + (ρv ) + (ρv ) = 0. (1.35a) The stress tensor τ is a second-order tensor (nine compo- ∂t ∂x x ∂y y ∂z z nents) that includes both tangential and normal stresses. For For a compressible fluid, a net outflow results in a change example, in rectangular coordinates, τ is (decrease) in fluid density. Of course, conservation of mass τ τ τ can be applied in cylindrical and spherical coordinates as xx xy xz well: τyx τyy τyz τzx τzy τzz
∂ρ 1 ∂ 1 ∂ ∂ The normal stresses have the repeated subscripts and they + (ρrvr) + (ρvθ) + (ρvz) = 0 (1.35b) ∂t r ∂r r ∂θ ∂z appear on the diagonal. Please note that the sum of the diag- onal components is the trace of the tensor (A) and is often and written as tr(A). The trace of the stress tensor, τii ,isassumed ∂ρ 1 ∂ 2 1 ∂ to be related to the pressure by + (ρr vr) + (ρvθ sin θ) ∂t r2 ∂r r sin θ ∂θ =−1 + + 1 ∂ p (τxx τyy τzz). (1.38) + (ρvφ) = 0. (1.35c) 3 r sin θ ∂φ Often the pressure in (1.38) is written using the Einstein sum- In fluid flow, rotation of a suspended particle can be caused mation convention as p =−τii/3, where the repeated indices by a variation in velocity, even if every fluid element is trav- imply summation. The shear stresses have differing sub- eling a path parallel to the confining boundaries. Similarly, scripts and the corresponding off-diagonal terms are equal; the interaction of forces can create a moment that is obtained that is, τxy = τyx . This requirement is necessary because with- from the cross product or curl. This tendency toward rotation out it a small element of fluid in a shear field could experience is particularly significant, so let us review the cross product an infinite angular acceleration. Therefore, the stress tensor ∇×V in rectangular coordinates: is symmetric and has just six independent quantities. We will temporarily represent the (shear) stress components by ∂vz − ∂vy (1.36a) ∂v ∂y ∂z τ =−µ i . (1.39) ∂v ∂v ji ∂x ∇×V = x − z (1.36b) j ∂z ∂x ∂vy ∂vx Note that this relationship (Newton’s law of friction) between − (1.36c) ∂x ∂y stress and strain is linear. There is little a priori evidence for its validity; however, known solutions (e.g., for Hagen– Note that the cross product of vectors is a vector; further- Poiseuille flow) are confirmed by physical experience. more, you may recall that (1.36a)–(1.36c), the vorticity vector It is appropriate for us to take a moment to think a little components ωx , ωy , and ωz , are measures of the rate of fluid bit about how a material responds to an applied stress. Strain, rotation about the x, y, and z axes, respectively. Vorticity is denoted by e and referred to as displacement, is often written 10 INTRODUCTION AND SOME USEFUL REVIEW as l/l. It is a second-order tensor, which we will write as eij . We now divide by xyz and take the limits as all three We interpret eyx as a shear strain, dy/dx or y/x. The normal are allowed to approach zero. The result, upon applying the strains, such as exx , are positive for an element of material definition of the first derivative, is that is stretched (extensional strain) and negative for one that is compressed. The summation of the diagonal components, ∂ρvx ∂ ∂ ∂ + ρvxvx + ρvyvx + ρvzvx which we will write as eii , is the volume strain (or dilatation). ∂t ∂x ∂y ∂z Thus, when we speak of the ratio of the volume of an element ∂p ∂τxx ∂τyx ∂τzx V V =− − − − + ρgx. (1.41) (undergoing deformation) to its initial volume, / 0,weare ∂x ∂x ∂y ∂z referring to dilatation. Naturally, dilatation for a real material must lie between zero and infinity. Now consider the response This equation of motion can be written more generally in of specific material types; suppose we apply a fixed stress to a vector form: material that exhibits Hookean behavior (e.g., by applying an extensional force to a spring). The response is immediate, and ∂ (ρv) + [∇·ρvv] =−∇p − [∇·τ] + ρg. (1.41a) when the stress is removed, the material (spring) recovers its ∂t initial size. Contrast this with the response of a Newtonian fluid; under a fixed shear stress, the resulting strain rate is If Newton’s law of friction (1.39) is introduced into (1.41) and constant, and when the stress is removed, the deformation if we take both the fluid density and viscosity to be constant, remains. Of course, if a Newtonian fluid is incompressible, no we obtain the x-component of the Navier–Stokes equation: applied stress can change the fluid element’s volume; that is, the dilatation is zero. Among “real fluids,” there are many that ∂v ∂v ∂v ∂v ρ x + v x + v x + v x exhibit characteristics of both elastic solids and Newtonian ∂t x ∂x y ∂y z ∂z fluids. For example, if a viscoelastic material is subjected to constant shear stress, we see some instantaneous deformation ∂p ∂2v ∂2v ∂2v =− + µ x + x + x + ρg . that is reversible, followed by flow that is not. ∂x ∂x2 ∂y2 ∂z2 x We now sketch the derivation of the equation of motion (1.42) by making a momentum balance upon a cubic volume ele- ment of fluid with sides x, y, and z. We are formulating It is useful to review the assumptions employed by Stokes a vector equation, but it will suffice for us to develop just in his derivation in 1845: (1) the fluid is continuous and the the x-component. The rate at which momentum accumu- stress is no more than a linear function of strain, (2) the fluid lates within the volume should be equal to the rate at which is isotropic, and (3) when the fluid is at rest, it must develop momentum enters minus the rate at which momentum leaves a hydrostatic stress distribution that corresponds to the ther- (plus the sum of forces acting upon the volume element). modynamic pressure. Consider the implications of (3): When Consequently, we write the fluid is in motion, it is not in thermodynamic equilibrium, ∂ yet we still describe the pressure with an equation of state. accumulation xyz (ρv ) = (1.40a) ∂t x Let us explore this further; we can write the stress tensor as Stokes did in 1845: convective transport of x-momentum in the x-, y-, and z- directions ∂vi ∂vj τij =−pδij + µ + + δij λ div V. (1.43) ∂xj ∂xi + | − | yzvx ρvx x yzvx ρvx x+x + | − | Now suppose we consider the three normal stresses; we will xzvy ρvx y xzvy ρvx y+y (1.40b) illustrate with just one, τxx : +xyvzρvx| − xyvz ρvx| + z z z ∂vx molecular transport of x-momentum in the x-, y-, and z- τxx =−p + 2µ + λ div V. (1.44) directions ∂x + | − | We add all three together and then divide by (−)3, resulting yzτxxx yzτxx x+x in +xzτ − xzτ yx y yx y+y (1.40c) + | − | 1 2µ + 3λ xy τzx z xy τzx z+z − (τxx + τyy + τzz) = p − div V. (1.45) 3 3 pressure and gravitational forces If we want the mechanical pressure to be equal to (neg- + | − | + yz(p x p x+x) xyzρgx (1.40d) ative one-third of) the trace of the stress tensor, then either VECTORS, TENSORS, AND THE EQUATION OF MOTION 11 div V = 0, or alternatively, 2 µ +3λ = 0. If the fluid in ques- It is also possible to obtain an energy equation by multiply- tion is incompressible, then the former is of course valid. ing the Navier–Stokes equation by the velocity vector v.We But what about the more general case? If div V = 0, then it employ subscripts here, noting that i and j can assume the would be extremely convenient if 2 µ =−3λ. This is Stokes’ values 1, 2, and 3, corresponding to the x, y, and z directions: hypothesis; it has been the subject of much debate and it is almost certainly wrong except for monotonic gases. Never- ∂ 1 ∂ ∂vi ρvj vivi = (τij vi) − τij . (1.49) theless, it seems prudent to accept the simplification since ∂xj 2 ∂xj ∂xj as Schlichting (1968) notes, “...the working equations have been subjected to an unusually large number of experimental τi.j is the symmetric stress tensor, and we are employing verifications, even under quite extreme conditions.” Landau Stokes’ simplification: and Lifshitz (1959) observe that this second coefficient of viscosity (λ) is different in the sense that it is not merely τij =−pδij + 2µSij . (1.50) a property of the fluid, as it appears to also depend on the frequency (or timescale) of periodic motions (in the fluid). δ is the Kronecker delta (δij = 1ifi = j, and zero otherwise) Landau and Lifshitz also state that if a fluid undergoes expan- and Sij is the strain rate tensor, sion or contraction, then thermodynamic equilibrium must be restored. They note that if this relaxation occurs slowly, then 1 ∂vi ∂vj Sij = + . (1.51) it is possible that λ is large. There is some evidence that λ may 2 ∂xj ∂xi actually be positive for liquids, and the student with deeper interest in Stokes’ hypothesis may wish to consult Truesdell In the literature of fluid mechanics, the strain rate tensor is (1954). often written as it appears in eq. (1.51), but one may also find = + We can use the substantial time derivative to rewrite Sij ∂vi/∂xj ∂vj/∂xi . Symmetric second-order tensors eq. (1.42) more compactly: have three invariants (by invariant, we mean there is no change resulting from rotation of the coordinate system): Dv 2 ρ =−∇p + µ∇ v + ρg. (1.46) = Dt I1(A) tr(A), (1.52)
We should review the meaning of the terms appearing 1 above. On the left-hand side, we have the accumulation of I (A) = (tr(A))2 − tr(A2) (1.53) momentum and the convective transport terms (these are the 2 2 nonlinear inertial terms). On the right-hand side, we have = + + pressure forces, the molecular transport of momentum (vis- (which for a symmetric A is I2 A11A22 A22A33 − 2 − 2 − 2 cous friction), and external body forces such as gravity. Please A11A33 A12 A23 A13), and note that the density and the viscosity are assumed to be I (A) = det(A). (1.54) constant. Consequently, we should identify (1.46) as the 3 Navier–Stokes equation; it is inappropriate to refer to it as The second invariant of the strain rate tensor is particularly the generalized equation of motion. We should also observe useful to us; it is the double dot product of S , which we write that for the arbitrary three-dimensional flow of a nonisother- ij as S S . For rectangular coordinates, we obtain mal, compressible fluid, it would be necessary to solve (1.41), i j ij ji along with the y- and z-components, the equation of continu- ∂v 2 ∂v 2 ∂v 2 ∂v ∂v 2 ity (1.35a), the equation of energy, and an equation of state I = 2 x + y + z + x + y simultaneously. In this type of problem, the six dependent 2 ∂x ∂y ∂z ∂y ∂x variables are vx , vy , vz , p, T, and ρ. 2 2 As noted previously, we can take the curl of the Navier– ∂vx ∂vz ∂vy ∂vz + + + + . Stokes equation and obtain the vorticity transport equation, ∂z ∂x ∂z ∂y which is very useful for the solution of some hydrodynamic (1.55) problems: You may recognize these terms; they are used to compute ∂ω =∇×(v × ω) + ν∇2ω, (1.47) the production of thermal energy by viscous dissipation, and ∂t they can be very important in flow systems with large velocity or alternatively, gradients. We will see them again in Chapter 7. We shall make extensive use of these relationships in this Dω book. This is a good point to summarize the Navier–Stokes = ω·∇v + ν∇2ω. (1.48) Dt equations, so that we can refer to them as needed. 12 INTRODUCTION AND SOME USEFUL REVIEW Rectangular coordinates ∂v ∂v v ∂v v ∂v v v − v 2 cot θ ρ θ + v θ + θ θ + φ θ + r θ φ ∂t r ∂r r ∂θ r sin θ ∂φ r ∂vx + ∂vx + ∂vx + ∂vx ρ vx vy vz 1 ∂p 1 ∂ ∂v 1 ∂ 1 ∂ ∂t ∂x ∂y ∂z =− + µ r2 θ + (v sin θ) r ∂θ r2 ∂r ∂r r2 ∂θ sin θ ∂θ θ 2 2 2 ∂p ∂ vx ∂ vx ∂ vx =− + µ + + + ρgx, 2 ∂x ∂x2 ∂y2 ∂z2 1 ∂ vθ 2 ∂vr 2 cot θ ∂vφ + + − , +ρgθ (1.58b) r2 sin2 θ ∂φ2 r2 ∂θ r2 sin θ ∂φ (1.56a) ∂vy ∂vy ∂vy ∂vy ρ + vx + vy + vz ∂t ∂x ∂y ∂z + 2 2 2 ∂vφ ∂vφ vθ ∂vφ vφ ∂vφ vφvr vθvφ cot θ ∂p ∂ vy ∂ vy ∂ vy ρ + vr + + + =− + µ + + + ρgy, ∂t ∂r r ∂θ r sin θ ∂φ r ∂y ∂x2 ∂y2 ∂z2 1 ∂p 1 ∂ ∂v 1 ∂ 1 ∂ (1.56b) =− + µ r2 φ + (v sin θ) 2 2 φ ∂vz + ∂vz + ∂vz + ∂vz r sin θ ∂φ r ∂r ∂r r ∂θ sin θ ∂θ ρ vx vy vz ∂t ∂x ∂y ∂z 2 1 ∂ vφ 2 ∂vr 2 cot θ ∂vθ + + + + ρgφ (1.58c) ∂p ∂2v ∂2v ∂2v r2 sin2 θ ∂φ2 r2 sin θ ∂φ r2sin θ ∂φ =− + µ z + z + z + ρg . ∂z ∂x2 ∂y2 ∂z2 z These equations have attracted the attention of many (1.56c) eminent mathematicians and physicists; despite more than Cylindrical coordinates 160 years of very intense work, only a handful of solu- tions are known for the Navier–Stokes equation(s). White ∂v ∂v v ∂v ∂v v 2 (1991) puts the number at 80, which is pitifully small com- ρ r + v r + θ r + v r − θ ∂t r ∂r r ∂θ z ∂z r pared to the number of flows we might wish to consider. The Clay Mathematics Institute has observed that “... although 2 2 ∂p ∂ 1 ∂ 1 ∂ vr ∂ vr 2 ∂vθ these equations were written down in the 19th century, our =− + µ (rv + + − ∂r ∂r r ∂r r r2 ∂θ2 ∂z2 r2 ∂θ understanding of them remains minimal. The challenge is to make substantial progress toward a mathematical theory + ρgr, which will unlock the secrets hidden in the Navier–Stokes ∂v ∂v v ∂v ∂v v v (1.57a) equations.” ρ θ + v θ + θ θ + v θ + r θ ∂t r ∂r r ∂θ z ∂z r 2 2 1 ∂p ∂ 1 ∂ 1 ∂ vθ ∂ vθ 2 ∂vr 1.6 THE MEN FOR WHOM THE NAVIER–STOKES =− + µ rvθ + + + r ∂θ ∂r r ∂r r2 ∂θ2 ∂z2 r2 ∂θ EQUATIONS ARE NAMED + ρgθ, The equations of fluid motion given immediately above are ∂v ∂v v ∂v ∂v (1.57b) named after Claude Louis Marie Henri Navier (1785–1836) ρ z + v z + θ z + v z ∂t r ∂r r ∂θ z ∂z and Sir George Gabriel Stokes (1819–1903). There was no 2 2 professional overlap between the two men as Navier died in ∂p 1 ∂ ∂vz 1 ∂ vz ∂ vz =− + µ r + + + ρgz. 1836 when Stokes (a 17-year-old) was in his second year ∂z r ∂r ∂r r2 ∂θ2 ∂z2 at Bristol College. Navier had been taught by Fourier at the (1.57c) Ecole Polytechnique and that clearly had a great influence upon his subsequent interest in mathematical analysis. But Spherical coordinates in the nineteenth century, Navier was known primarily as a bridge designer/builder who made important contributions to 2 2 ∂vr ∂vr vθ ∂vr vφ ∂vr vθ +vφ structural mechanics. His work in fluid mechanics was not as ρ + vr + + − ∂t ∂r r ∂θ r sin θ ∂φ r well known. Anderson (1997) observed that Navier did not 2 understand shear stress and although he did not intend to ∂p 1 ∂ 2 1 ∂ ∂vr =− + µ (r vr) + sin θ derive the equations governing fluid motion with molecular ∂r r2 ∂r2 r2sin θ ∂θ ∂θ friction, he did arrive at the proper form for those equa-
2 tions. Stokes himself displayed talent for mathematics while 1 ∂ vr + + ρgr, at Bristol. He entered Pembroke College at Cambridge in r2 sin2 φ ∂φ2 1837 and was coached in mathematics by William Hopkins; (1.58a) later, Hopkins recommended hydrodynamics to Stokes as an SIR ISAAC NEWTON 13 area ripe for investigation. Stokes set about to account for fric- Certainly Newton had a difficult personality with a tional effects occurring in flowing fluids and again the proper dichotomous nature—he wanted recognition for his devel- form of the equation(s) was discovered (but this time with opments but was so averse to criticism that he was reticent intent). He became aware of Navier’s work after completing about sharing his discoveries through publication. This char- his own derivation. In 1845, Stokes published “On the Theo- acteristic contributed to the acrimony over who should be ries of the Internal Friction of Fluids in Motion” recognizing credited with the development of differential calculus, New- that his development employed different assumptions from ton or Leibniz. Indeed, this debate created a schism between those of Navier. For a better glimpse into the personalities British and continental mathematicians that lasted decades. and lives of Navier and Stokes, see the biographical sketches But two points are absolutely clear: Newton’s development written by O’Connor and Robertson2003 (MacTutor History of the “method of fluxions” predated Liebniz’s work and each of Mathematics). A much richer picture of Stokes the man man used his own, unique, system of notation (suggesting that can be obtained by reading his correspondence (especially the efforts were completely independent). Since differential between Stokes and Mary Susanna Robinson) in Larmor’s calculus ranks arguably as the most important intellectual memoir (1907). accomplishment of the seventeenth century, one can at least comprehend the vitriol of this long-lasting debate. Newton used the Royal Society to “resolve” the question of priority; 1.7 SIR ISAAC NEWTON however, since he wrote the committee’s report anonymously, there can be no claim to impartiality. Much of what we routinely use in the study of transport phe- Newton also had a very contentious relationship with nomena (and, indeed, in all of mathematics and mechanics) John Flamsteed, the first Astronomer Royal. Newton needed is due to Sir Isaac Newton. Newton, according to the con- Flamsteed’s lunar observations to correct the lunar theory he temporary calendar, was born on Christmas Day in 1642; had presented in Principia (Philosophiae Naturalis Principia by modern calendar, his date of birth was January 4, 1643. Mathematica). Flamsteed was clearly reluctant to provide His father (also Isaac Newton) died prior to his son’s birth these data to Newton and in fact demanded Newton’s promise and although the elder Newton was a wealthy landowner, he not to share or further disseminate the results, a restriction that could neither read nor write. His mother, following the death Newton could not tolerate. Newton made repeated efforts to of her second husband, intended for young Isaac to manage obtain Flamsteed’s observations both directly and through the the family estate. However, this was a task for which Isaac influence of Prince George, but without success. Flamsteed had neither the temperament nor the interest. Fortunately, an prevailed; his data were not published until 1725, 6 years uncle, William Ayscough, recognized that the lad’s abilities after his death. were directed elsewhere and was instrumental in getting him There is no area in optics, mathematics, or mechanics entered at Trinity College Cambridge in 1661. that was not at least touched by Newton’s genius. No less Many of Newton’s most important contributions had their a mathematician than Lagrange stated that Newton’s Prin- origins in the plague years of 1665–1667 when the Univer- cipia was the greatest production of the human mind and this sity was closed. While home at Lincolnshire, he developed evaluation was echoed by Laplace, Gauss, and Biot, among the foundation for what he called the “method of fluxions” others. Two anecdotes, though probably unnecessary, can be (differential calculus) and he also perceived that integration used to underscore Newton’s preeminence: In 1696, Johann was the inverse operation to differentiation. As an aside, we Bernoulli put forward the brachistochrone problem (to deter- note that a fluxion, or differential coefficient, is the change in mine the path in the vertical plane by which a weight would one variable brought about by the change in another, related descend most rapidly from higher point A to lower point B). variable. In 1669, Newton assumed the Lucasian chair at Leibniz worked the problem in 6 months; Newton solved it Cambridge (see the information compiled by Robert Bruen overnight according to the biographer, John Conduitt, fin- and also http://www.lucasianchair.org/) following Barrow’s ishing at about 4 the next morning. Other solutions were resignation. Newton lectured on optics in a course that began eventually obtained from Leibniz, l’Hopital, and both Jacob in January 1670 and in 1672 he published a paper on light and and Johann Bernoulli. In a completely unrelated problem, color in the Philosophical Transactions of the Royal Society. Newton was able to determine the path of a ray by (effec- This work was criticized by Robert Hooke and that led to tively) solving a differential equation in 1694; Euler could a scientific feud that did not come to an end until Hooke’s not solve the same problem in 1754. Laplace was able to death in 1703. Indeed, Newton’s famous quote, “If I have solve it, but in 1782. seen further it is by standing on ye shoulders of giants,” which It is, I suppose, curiously comforting to ordinary mortals has often been interpreted as a statement of humility appears to know that truly rare geniuses like Newton always seem to to have actually been intended as an insult to Hooke (who be flawed. His assistant Whiston observed that “Newton was was a short hunchback, becoming increasingly deformed of the most fearful, cautious and suspicious temper that I ever with age). knew.” 14 INTRODUCTION AND SOME USEFUL REVIEW
Furthermore, in the brief glimpse offered here, we have Bird, R. B., Stewart, W. E., and E. N. Lightfoot. Transport Phe- avoided describing Newton’s interests in alchemy, history, nomena, 2nd edition, Wiley, New York (2002). and prophecy, some of which might charitably be charac- Bohren, C. F. Comment on “Newton’s Law of Cooling—A Critical terized as peculiar. It is also true that work he performed Assessment,” by C. T. O’Sullivan. American Journal of Physics, as warden of the Royal Mint does not fit the reclusive 59:1044 (1991). scholar stereotype; as an example, Newton was instrumen- Clay Mathematics Institute, www.claymath.org. tal in having the counterfeiter William Chaloner hanged, Davis, H. T. Introduction to Nonlinear Differential and Integral drawn, and quartered in 1699. Nevertheless, Newton’s legacy Equations, Dover Publications, New York (1962). in mathematical physics is absolutely unique. There is no Fermi, E., Pasta, J., and S. Ulam. Studies of Nonlinear Problems, other case in history where a single man did so much to 1. Report LA-1940 (1955). advance the science of his era so far beyond the level of his Landau, L. D. and E. M. Lifshitz. Fluid Mechanics, Pergamon contemporaries. Press, London (1959). We are fortunate to have so much information available Larmor, J., editor. Memoir and Scientific Correspondence of the regarding Newton’s life and work through both his own writ- Late Sir George Gabriel Stokes, Cambridge University Press, ing and exchanges of correspondence with others. A select New York (1907). number of valuable references used in the preparation of this Lorenz, E. N. Deterministic Nonperiodic Flow. Journal of the account are provided immediately below. Atmospheric Sciences, 20:130 (1963). Milne-Thomson, L. M. Jacobian Elliptic Function Tables: A Guide The Correspondence of Isaac Newton, edited by H. W. to Practical Computation with Elliptic Functions and Integrals, Dover, New York (1950). Turnbull, FRS, University Press, Cambridge (1961). O’Connor, J. J. and E. F. Robertson. MacTutor History of Mathe- The Newton Handbook, Derek Gjertsen, Routledge & matics, www.history.mcs.st-andrews.ac.uk (2003). Kegan Paul, London (1986). Packard, N. H., Crutchfield, J. P., Farmer, J. D., and R. S. Shaw. Memoirs of Sir Isaac Newton, Sir David Brewster, Geometry from a Time Series. Physical Review Letters, 45:712 reprinted from the Edinburgh Edition of 1855, Johnson (1980). Reprint Corporation, New York (1965). Porter, M. A., Zabusky, N. J., Hu, B., and D. K. Campbell. A Short Account of the History of Mathematics, 6th edi- Fermi, Pasta, Ulam and the Birth of Experimental Mathematics. tion, W. W. Rouse Ball, Macmillan, London (1915). American Scientist, 97:214 (2009). Powers, D, L. Boundary Value Problems, 2nd edition, Academic See also http://www-groups.dcs.st-and.ac.uk and http:// Press, New York (1979). www.newton.cam.ac.uk. Rossler, O. E. An Equation for Continuous Chaos. Physics Letters, 57A:397 (1976). Schlichting, H. Boundary-Layer Theory, 6th edition, McGraw-Hill, New York (1968). REFERENCES Stokes, G. G. On the Theories of the Internal Friction of Fluids in Motion. Transactions of the Cambridge Philosophical Society, Anderson, J. D. A History of Aerodynamics, Cambridge University 8:287 (1845). Press, New York (1997). Truesdell, C. The Present Status of the Controversy Regarding the Baker, G. L. and J. P. Gollub. Chaotic Dynamics, Cambridge Bulk Viscosity of Liquids. Proceedings of the Royal Society of University Press, Cambridge (1990). London, A226:1 (1954). Baruh, H. Are Computers Hurting Education? ASEE Prism,p.64 Vaughn, M. T. Introduction to Mathematical Physics, Wiley-VCH, (October 2001). Weinheim (2007). Batchelor, G. K. An Introduction to Fluid Dynamics, Cambridge White, F. M. Viscous Fluid Flow, 2nd edition, McGraw-Hill, New University Press, Cambridge (1967). York (1991). 2
INVISCID FLOW: SIMPLIFIED FLUID MOTION
2.1 INTRODUCTION that direction; for example,
In the early years of the twentieth century, Prandtl (1904) ∂φ v = . (2.3) proposed that for flow over objects the effects of viscous x ∂x friction would be confined to a thin region of fluid very close to the solid surface. Consequently, for incompressible flows These steps allow us to rewrite the Euler equation as follows: in which the fluid is accelerating, viscosity should be unim- 2 portant for much of the flow field. This hypothesis might (in ∂ φ ∂vx ∂vy ∂vz 1 ∂p ∂ + vx + vy + vz =− + , (2.4) fact, did) allow workers in fluid mechanics to successfully ∂t∂x ∂x ∂x ∂x ρ ∂x ∂x treat some difficult problems in an approximate way. Con- sider the consequences of setting viscosity µ equal to zero in where is a potential energy function. Of course, this result the x-component of the Navier–Stokes equation: can be integrated with respect to x:
2 v2 v2 ∂v ∂v ∂v ∂v ∂p ∂φ + vx + y + z + p − = ρ x + v x + v x + v x =− + ρg . F1. (2.5) ∂t x ∂x y ∂y z ∂z ∂x x ∂t 2 2 2 ρ
(2.1) Note that F1cannot be a function of x. The very same pro- cess sketched above can also be carried out for the y- and The result is the x-component of the Euler equation and you z-components of the Euler equation; when the three results can see that the order of the equation has been reduced from are combined, we get the Bernoulli equation: 2 to 1. Of course, this automatically means a loss of informa- tion; we can no longer enforce the no-slip condition. We will ∂φ 1 p also require that the flow be irrotational so that ∇×V = 0; + |V |2 + + gZ = F(t). (2.6) consequently, ∂t 2 ρ This is an inviscid energy balance; it can be very useful in ∂v ∂v ∂v ∂v x = z and x = y . (2.2) the preliminary analysis of flow problems. For example, one ∂z ∂x ∂y ∂x could use the equation to qualitatively explain the operation of an airfoil or a FrisbeeTM flying disk. For the latter, consider Now we introduce the velocity potential φ. We can obtain the a flying disk with a diameter of 22.86 cm and mass of 80.6 g, fluid velocity in a given direction by differentiation of φ in given an initial velocity of 6.5 m/s. The airflow across the
Transport Phenomena: An Introduction to Advanced Topics, By Larry A. Glasgow Copyright © 2010 John Wiley & Sons, Inc.
15 16 INVISCID FLOW: SIMPLIFIED FLUID MOTION top of the disk (along a center path) must travel about 26 cm, These are the Cauchy–Riemann relations and they guaran- corresponding to an approximate velocity of 740 cm/s. This tee the existence of a complex potential, a mapping between increased velocity over the top gives rise to a pressure differ- the φ–ψ plane (or flow net) and the x–y plane. This simply ence of about 75 dyn/cm2, generating enough lift to partially means that any analytic function of z (z = x + iy) corresponds offset the effect of gravity. to the solution of some potential flow problem. This branch We emphasize that the Bernoulli equation does not of mathematics is called conformal mapping and there are account for dissipative processes, so we cannot expect quan- compilations of conformal representations that can be used titative results for systems with significant friction. We are, to “solve” potential flow problems; see Kober (1952), for however, going to make direct use of potential flow theory example. Alternatively, we can simply assume a form for the a little later when we begin our consideration of boundary- complex potential; suppose we let layer flows. W(z) = z + z3 = (x + iy) + (x + iy)3; (2.12)
2.2 TWO-DIMENSIONAL POTENTIAL FLOW therefore,
We now turn our attention to two-dimensional, inviscid, φ + iψ = x + iy + x3 + 3ix2y − 3xy2 − iy3 irrotational, incompressible (potential) flows. The descrip- tor “potential” comes from analogy with electrostatics. In and fact, Streeter and Wylie (1975) note that the flow net for a set of fixed boundaries can be obtained with a voltmeter ψ = y + x2y − y3. using a nonconducting surface and a properly bounded elec- 3 (2.13) trolyte solution. The student seeking additional background and detail for inviscid fluid motions should consult Lamb What does this flow look like? It is illustrated in Figure 2.1. (1945) and Milne-Thomson (1958). The continuity equation Note that the general form of the complex potential for = π/θ for these two-dimensional flows is flow in a corner is W(z) Vh(z/h) , where θ is the included angle. Therefore, for a 45◦ corner (taking the refer- ∂v ∂v ence length to be 1), θ = π/4 and W(z) = Vz4. x + y = 0. (2.7) ∂x ∂y Let us now consider the vortex, whose complex potential is given by Using the velocity potential φ to represent velocity vector components in eq. (2.7), we obtain the Laplace equation: i φ + iψ = ln(x + iy), (2.14) 2π ∂2φ ∂2φ + = 0, or simply ∇2φ = 0. (2.8) ∂x2 ∂y2
We define the stream function such that
∂ψ ∂ψ v =− and v = . (2.9) x ∂y y ∂x
This choice means that for a case in which ψ increases in the vertical (y) direction, flow with respect to the x-axis will be right-to-left. We can reverse the signs in (2.9) if we prefer the flow to be left-to-right. If we couple (2.9) with the irrotational requirement (2.2), we find
∂2ψ ∂2ψ + = 0. (2.10) ∂x2 ∂y2
Note that the velocity potential and stream function must be related by the equations
∂φ ∂ψ ∂φ ∂ψ FIGURE 2.1. Variation of flow in a corner obtained from the com- =− and = . (2.11) ∂x ∂y ∂y ∂x plex potential W(z) = z + z3. TWO-DIMENSIONAL POTENTIAL FLOW 17 where is the circulation around a closed path. It is conve- nient in such cases to write the complex number in polar form, that is, x + iy = reiθ . The stream function and the velocity potential can then be written as
θ ψ = ln r and φ =− . (2.15) 2π 2π Note that the stream function assumes very large negative values as the center of the vortex is approached. What does this tell you about velocity at the center of an ideal vortex? Many interesting flows can be constructed by simple com- FIGURE 2.3. Potential flow past a circular cylinder. Note the fore- bination. For example, if we take uniform flow, and-aft symmetry, which of course means that there is no form drag. This feature of potential flow is the source of d’Alembert’s para- + = + φ iψ V (x iy), (2.16) dox and it was an enormous setback to fluid mechanics since many hydrodynamicists of the era concluded that the Euler equation(s) and combine it with a source, was incorrect. Q φ + iψ = ln(x + iy), (2.17) 2π This stream function is plotted in Figure 2.3. Note that there we can get the stream function for flow about a two- is no difference in the flow between the upstream and down- dimensional half-body: stream sides. In fact, the pressure distribution at the cylinder’s surface is perfectly symmetric: Q ψ = Vr sin θ − θ. (2.18) 1 2 2 2π p − p∞ = ρV∞(1 − 4 sin θ). (2.21) 2 This is illustrated in Figure 2.2. The radius of the body at the leading edge, or nose, is Q/(2πV). Make sure you understand how this result is obtained using = − 1 2 The complex potential for flow around a cylinder is eq. (2.6)! At θ 0, p p∞ is the dynamic head, 2 ρV∞. Note ◦ 1 2 also that the pressure at 90 corresponds to −3( ρV∞) 2 ◦ a2 and that the recovery is complete as one moves on to 180 . W(z) =−V z + , (2.19) z Experimental measurements of pressure on the surface of circular cylinders show that the minimum is usually attained and the stream function is at about 70◦ or 75◦ and the pressure recovery on the down- stream side is far from complete. The potential flow solution a2y =∼ ◦ ψ =−V y − . (2.20) gives a reasonable result only to about θ 60 for large x2 + y2 Reynolds numbers. This is evident from the pressure dis- tributions shown in Figure 2.4. If we combine a uniform flow with a doublet (a source and a sink combined with zero separation) and a vortex, we obtain flow around a cylinder with circulation (by circulation we mean the integral of the tangential component of velocity around a closed path): R2 ψ = V sin θ r − + ln r. (2.22) r 2π The pressure at the surface of the cylinder is ρV 2 2 p = 1 − 2 sin θ + . (2.23) 2 2πRV
Obviously, since this is inviscid flow there is no frictional FIGURE 2.2. Two-dimensional potential flow around a half-body. drag, but might we have form drag? That is, is there a net The flow is symmetric about the x-axis, so only the upper half is force in the direction of the uniform flow? Consult Figure 2.5; shown. note that the flow is symmetric fore and aft (upstream and 18 INVISCID FLOW: SIMPLIFIED FLUID MOTION
Vθ is 3927 cm/s and the cylinder is generating a lift of 2.22 × 106 dyn per cm of length. This phenomenon is famil- iar to anyone who has played a sport in which sidespin and translation are simultaneously imparted to a ball; soccer, ten- nis, golf, and baseball come immediately to mind. Schlichting (1968) points out that an attempt was made to utilize the effect commercially with the Flettner “rotor” ship in the 1920s. More details regarding these efforts are provided by Ahlborn (1930). The first full-scale efforts to exploit the phenomenon were carried out with the steamship Buckau. This vessel made 7.85 knots in trials with 134 hp using its screw propeller; under favorable conditions in early 1925, it attained 8.2 knots using only 33.4 hp to turn the rotors (no propeller). Ahlborn noted that although wind tunnel tests indicated that the rotors might be considerably more efficient than canvas sails of com- parable surface area, the Flettner rotor was a nautical and
1 2 economic failure. In more recent years, spinning cylinders − ∞ FIGURE 2.4. Dimensionless pressure (p p )/( 2 ρV ) distribu- tions for flow over a cylinder; the potential flow case is clearly have been incorporated into experimental airfoils to promote labeled and the experimental data points are from Fage and Falkner lift and control the boundary layer; see Chapter 5 in Chang (1931) for Re = 108,000, 170,000, and 217,000. (1976). A modern computational study of steady, uniform flow past rotating cylinders has been carried out by Padrino and Joseph (2006). downstream). Of course, this means that there is no net force Among other particularly interesting complex potentials in the horizontal direction, and hence, no drag. But suppose are the infinite row of vortices and the von Karman vortex we look at the vertical component, that is, −p sin θ. When this street. For the former, quantity is integrated over the surface, the result is not zero; the rotating cylinder is generating lift. This phenomenon is πz W(z) = iκ ln sin (2.24) known as the Magnus effect. a The lift being generated by the cylinder is ρV, which is equivalent to 2πρRVVθ. For example, suppose air is and approaching a circular cylinder (from the left) at 30 m/s. The cylinder is rotating in the clockwise direction at κ 1 2πy 2πx ψ = ln cosh − cos . (2.25) 1500 rpm (157 rad/s). If the cylinder diameter is 50 cm, then 2 2 a a
The row of vortices is illustrated in Figure 2.6. For the von Karman vortex street, the complex potential is π ib W(z) = iκ ln sin z − a 2 π a ib −iκ ln sin z − + , (2.26) a 2 2
FIGURE 2.5. Two-dimensional potential flow about a cylinder with circulation. Note how the fluid is wrapped up and around the rotating cylinder. This generates lift since the pressure is larger across the bottom of the cylinder than across the top; the (Mag- nus) effect is significant for rotating bodies with large translational FIGURE 2.6. An infinite row of vortices each with the same velocities. strength and spaced a distance a apart. TWO-DIMENSIONAL POTENTIAL FLOW 19
FIGURE 2.7. von Karman vortex street. and the corresponding stream function is ψ 1 cosh(2π(y/a − k/2)) − cos(2πx/a) = ln , κ 2 cosh(2π(y/a + k/2) − cos(2π(x/a − 1/2)) (2.27) where k = b/a. This flow field is illustrated in Figure 2.7. Many other interesting potential flows have been compiled by Kirchhoff (1985). Complex potentials are also known for a variety of airfoils, including flat plate and Joukowski type (with and without camber), at different angles of attack; see Currie (1993) for additional examples. The complex potentials for these flows FIGURE 2.8. Concentric circles that map into confocal ellipses. are linked to the z-plane through the Joukowski transforma- tion; the Joukowski transformation between the z-plane and the ξ-plane is generally written as We can make use of the identity sin2 λ + cos2 λ = 1 to obtain L2 z = ξ + , (2.28) x 2 y 2 ξ + = 1. (2.29c) α + L2/α α − L2/α where L is a real constant. One of the features of this choice is that for very large ξ, z =∼ ξ. Consequently, points that are If we let α = 3 and L = 2, this equation produces an ellipse far from the origin are unaffected by the mapping. Let us and the right half of this conic section is shown in Figure 2.9. now illustrate how this works. Consider concentric circles located at the origin of the ξ-plane. Since the distance from the origin (O) to the point P1 is a constant, then for the z- plane, SP + HP = constant. Accordingly, circles (with their centers at the origin) in the ξ-plane will map into confocal ellipses in the z-plane as demonstrated by Milne-Thomson (1958) and illustrated in Figure 2.8. We should explore this process with an example. We take (2.28) and substitute ξ = α eiλ; therefore,
L2 L2 L2 z = αeiλ + = α + cos λ + i α − sin λ. αeiλ α α (2.29a) This yields L2 L2 x = α + cos λ and y = α − sin λ. α α (2.29b) FIGURE 2.9. An ellipse constructed with eq. 2.29c. 20 INVISCID FLOW: SIMPLIFIED FLUID MOTION
TABLE 2.1. Streamline Identification for Joukowski Airfoil; ν = 0.5 and ξ = ρ exp(iν)
ψ/(Vc) ρ1 ρ2 ρ3 1.00 0.021654 0.66468 1.23728 1.25 0.031686 0.33862 2.07478 1.50 0.048249 0.20397 2.71431
As an exercise, you may wish to verify the values provided in Table 2.1, obtain some additional sets, and then employ (2.31) to transform them to the physical plane.
2.3 NUMERICAL SOLUTION OF POTENTIAL FLOW PROBLEMS
Although hundreds of complex potentials (conformal map- pings) have been developed over the years, we are not limited to flows that have been cataloged for us. Recall that both the velocity potential and the stream function satisfy the Laplace equation in ideal flows. We now employ a simple numerical FIGURE 2.10. Mapping of an “off-center” circle. procedure that will allow us to examine inviscid, irrotational, incompressible flows about nearly any object of our choice. We begin by writing the Laplace equation In contrast, if we start with a circle whose center is on the real axis to the right of the origin as illustrated in Figure 2.10, ∇2ψ = 0 (2.33) we should get a map that lies between that of the concentric circles (with centers at the origin). in finite difference form using second-order central differ- The “off-center” circle maps into the z-plane as a sym- ences: metric shape with a blunt nose on the right and a point (cusp) ψ + − 2ψ + ψ − ψ + − 2ψ + ψ − on the left. This technique can be used to generate potential i 1,j i,j i 1,j + i,j 1 i,j i,j 1 =∼ 2 2 0. flows about shapes that approximate a rudder or airfoil. For (x) (y) an airfoil with a chord of 4 and a thickness of 0.48, we can (2.34) start with the complex potential The index i refers to the x-direction and j to the y-direction. a2 Now we assume a square mesh such that x = y; we isolate F(ξ) = V (ξ + m) + , (2.30) ξ + m the term with the largest coefficient, which is ψi,j . Conse- quently, we obtain a simple algorithm for computation of the where a = l/4 + 0.77tc/l and m = 0.77tc/l. Note that l and central nodal point: t are 4 and 0.48, such that the thickness (ratio) of the airfoil is 12%. The transformation—as above—is given by 1 ψi,j = (ψi+ ,j + ψi− ,j + ψi,j+ + ψi,j− ). (2.35) 4 1 1 1 1 c2 z = ξ + , (2.31) The solution of such a problem is easy, in principle. We can ξ apply (2.35) at every interior nodal point and solve the result- and the dimensionless equation for streamlines is ing system of equations iteratively, or we can solve the set of simultaneous algebraic equations directly using an elimina- ψ ρ(1 + e)2sin ν tion scheme (if the number of nodal points is not too large). = ρ sin ν + . (2.32a) We now illustrate the numerical procedure for flow over a Vc ρ2 + e2 + 2ρe cos ν reverse step; we will use the very simple Gauss–Seidel iter- For the chosen parameters, e = 0.0924; if we take ν = 0.5, ative method. The principal parts of the computation are as then the dimensionless streamlines are given by follows: r ψ 0.57212ρ initialize ψ throughout the flow field and on the bound- = 0.47943ρ + . (2.32b) Vc ρ2 + 0.16218ρ + 0.008538 ary; NUMERICAL SOLUTION OF POTENTIAL FLOW PROBLEMS 21
FIGURE 2.13. Confined potential flow about a triangular wedge placed at the centerline.
FIGURE 2.11. Potential flow over a reverse step where the flow area doubles. in the appendices; in essence, the size of the change made by one Gauss–Seidel iteration is increased by (typically) about r perform iterative computation row-by-row in the inte- 80%. In well-conditioned problems, the number of iterations rior using the latest computed values as soon as they are can be reduced by a factor of roughly 10–100. available; This method can also be used to compute the flow fields r test for convergence; around arbitrary shapes; for example, consider a triangular r output results to a suitable file. wedge placed in the center of a confined flow. The stagnation streamline is incident upon the leading vertex and the flow is exactly split by the wedge. The iterative solution appears The result of the computation is shown in Figure 2.11. as shown in Figure 2.13. Note how the flow accelerates to Note that the result in Figure 2.11 is not what one would the position of maximum thickness and then adheres to the expect for a similar flow with a viscous fluid; the decrease wedge during deceleration at the trailing edge (a region of in velocity as the fluid comes off the step is accompanied by increasing pressure). an increase in pressure. This situation usually results in the We conclude this chapter with an example in which flow formation of a region of recirculation (a vortex) at the bottom about an airfoil is computed with the technique described of the step. There are several illustrations of this phenomenon immediately above. This case will illustrate two very impor- in Van Dyke (1982); see pages 13–15. tant complications that one must take into account while A closely related problem is flow over an overhang and solving such problems. An airfoil, with an angle of attack computed results are shown in Figure 2.12; again the resulting of 14◦, is placed in a uniform potential flow. Because of the streamlines do not correspond to what one would expect from shape of the object, the nodal points of a square mesh will not the flow of a viscous fluid. necessarily coincide with the airfoil surface. We have a few For larger problems, the rate of convergence of the Gauss– options in computational fluid dynamics (CFD) for dealing Seidel method can be increased significantly through use of with this problem: We might use an adaptive mesh generating successive over-relaxation (SOR). SOR is also known as the program (if available), a transformed coordinate system that extrapolated Liebmann method and it is described in detail conforms to the surface of the body (if one could be found), or a node-by-node approximation to compute mesh points near (but not on) the surface. The latter was employed here. Now consider the computed result shown in Figure 2.14. Pay particular attention to the stagnation streamline at the leading edge of the airfoil; now find the stagnation streamline that leaves the body. This will require that the fluid flowing underneath the airfoil turns sharply at the trailing edge and flows up the surface. This is untenable because the required fluid velocities at the trailing edge would be enormous; cer- tainly, no viscous fluid can behave this way, although the phenomenon can be reproduced with a Hele-Shaw apparatus (see Van Dyke (1982), p. 10). It is necessary that the stag- nation streamline leaving the upper surface in Figure 2.14 FIGURE 2.12. Numerical solution (Gauss–Seidel) for potential actually leaves the body smoothly at the trailing edge. A cir- flow over an overhang. culation about the airfoil is required to satisfy this criterion 22 INVISCID FLOW: SIMPLIFIED FLUID MOTION
visualization for flow over a NACA 64A015 airfoil at a 5◦ angle of attack. The photograph clearly shows that separa- tion (where the boundary layer is detached from the airfoil surface) will occur at a position corresponding to x/L ≈ 0.5.
2.4 CONCLUSION
We referred earlier to the schism that developed between practical fluid mechanics (hydraulics) and theoretical fluid mechanics (hydrodynamics). Since potential flow around any symmetric bluff body looks exactly the same fore and aft (see Figure 2.3), there are no pressure differences. And with- out pressure differences, there can be no form drag. This, FIGURE 2.14. Computed inviscid flow about an airfoil with an angle of attack of 14◦ and no circulation. Note the nasty turn in the of course, is contrary to common physical experience (i.e., flow underneath the wing at the trailing edge. d’Alembert’s paradox). A student of fluid mechanics might therefore conclude (based on a cursory examination of the subject) that potential flow is a mere curiosity, a footnote to be (the Kutta–Joukowski condition). Therefore, the stagnation appended to the history of fluid mechanics. That is an unwar- streamline value must be adjusted such that the computed ranted characterization. There is a wonderful unattributed flow appears as shown in Figure 2.15. You will note at once quote in de Nevers (1991) that clearly captures the situ- that the flow over the upper surface of the airfoil is now much ation: “Hydrodynamicists calculate that which cannot be faster; that is, through the addition of circulation, the flow observed; hydraulicians observe that which cannot be cal- about the airfoil is generating lift. This phenomenon has an culated.” At the very least, potential flow theory allows us interesting consequence: When circulation about the airfoil to think rationally about complicated flows that cannot be is established, a strong vortex with opposing circulation is easily calculated. generated by—and shed from—the wing. Such vortices can In reality, there are many types of problems where viscous be persistent (due to conservation of angular momentum) friction is quite unimportant, including flow through orifices and they can pose control problems for other aircraft that are and nozzles and flows into channel entrances. Another sig- unlucky enough to encounter them. nificant example is the behavior of waves on the surface of Once again it is important that we make the essential dis- deep water. Indeed, this is a case where potential flow the- tinction between the ideal flow shown in Figure 2.15 and ory is reasonably accurate. Lamb (1945) devotes an entire the movement of a real, viscous fluid past the same shape. chapter (IX) to this type of problem. For the case of “standing” For example, Van Dyke (1982) provides an example of flow waves in two dimensions, he notes that the velocity potential is governed by
∂2φ ∂2φ + = 0. (2.36) ∂x2 ∂y2
The y-coordinate is measured from the (resting) free surface upward, and the bottom is located at y =−h.Ifwetakeφ = P(y)cos(kx)e1(σt+ε), then the amplitude function P is found from (2.36) to be
P = A exp(−ky) + B exp(+ky). (2.37)
Since there can be no vertical motion at the bottom, ∂φ/∂y = 0aty =−h. Consequently, we have
i(σt+ε) FIGURE 2.15. Computed flow about the same airfoil with circu- φ = C cosh[k(y + h)]cos(kx)e . (2.38) lation. The flow leaves the trailing edge of the wing smoothly and a significant difference in local velocities now exists between the top At the free surface, the vertical velocity vy must be related and bottom surfaces. The reduced pressure on top, relative to the to the rate of change of the position of the surface: ∂φ/∂y = pressure acting upon the bottom, produces lift. ∂η/∂t, where η is the surface elevation (and a function of x REFERENCES 23 and t). If the pressure above the water surface is constant, de Nevers, N. Fluid Mechanics for Chemical Engineers, 2nd edition, then the Bernoulli equation can be used to close the set of McGraw-Hill, New York (1991). equations at the free surface. Lamb shows that the stream Fage, A. and V. M. Falkner. Further Experiments on the Flow function ψ for the standing waves is given by Around a Circular Cylinder. British Aeronautical Research Com- mission, R&M, 1369 (1931). gα sinh[k(y + h)] ψ = sin(kx)cos(σt + ε), (2.39) Kirchhoff, R. H. Potential Flows, Marcel Dekker, Inc., New York σ cosh(kh) (1985). Kober, H. Dictionary of Conformal Representations, Dover Publi- where α is the vertical amplitude of the wave. The reader is cations, New York (1952). invited to plot some streamlines for this example and then Lamb, H. Hydrodynamics, 6th edition, Dover Publications, New observe how ∂ψ/∂y behaves with increasing depth. You will York (1945). note immediately that the motion is rapidly attenuated in Milne-Thomson, L. M. Theoretical Aerodynamics, 4th edition, the negative y-direction; this is one case where the model Dover Publications, New York (1958). obtained from potential flow theory corresponds nicely with Padrino, J. C. and D. D. Joseph. Numerical Study of the Steady- physical experience. State Uniform Flow Past a Rotating Cylinder. Journal of Fluid Mechanics, 557:191 (2006). Prandtl, L. Uber Flussigkeitsbewgung bei sehr kleiner Reibung. REFERENCES Proceedings of the 3rd International Mathematics Congress, Heidelberg (1904). Ahlborn, F. The Magnus Effect in Theory and in Reality, NACA Schlichting, H. Boundary-Layer Theory, 6th edition, McGraw-Hill, Technical Memorandum 567 (1930). New York (1968). Chang, P. K. Control of Flow Separation, Hemisphere Publishing, Streeter, V. L. and E. B. Wylie. Fluid Mechanics, 6th edition, Washington, DC (1976). McGraw-Hill, New York (1975). Currie, I. G. Fundamental Mechanics of Fluids, 2nd edition, Van Dyke, M. An Album of Fluid Motion, Parabolic Press, Stanford, McGraw-Hill, New York (1993). CA (1982). 3
LAMINAR FLOWS IN DUCTS AND ENCLOSURES
3.1 INTRODUCTION The appropriate Navier–Stokes equation for the steady flow case is Laminar fluid motion is atypical; it is a very highly ordered ∂p 1 ∂ ∂v phenomenon in which viscous forces are dominant and 0 =− + µ r z . (3.1) momentum is transported by molecular friction. Disturbances ∂z r ∂r ∂r that arise in, or are imposed upon, stable laminar flows are We should recognize that the entire left-hand side of the rapidly damped by viscosity. One can see some of the essen- z-component (Navier–Stokes) equation has been reduced to tial differences between laminar and turbulent flows with 0. This means that there are no inertial forces. Consequently, simple experiments; please examine Figure 3.1. the Reynolds number There are a couple of important inferences that can be drawn from these images: d
Transport Phenomena: An Introduction to Advanced Topics, By Larry A. Glasgow Copyright © 2010 John Wiley & Sons, Inc.
24 TRANSIENT HAGEN–POISEUILLE FLOW 25
FIGURE 3.1. Digital images (using a short-duration flash) of water jets obtained at low (a) and high speed (b). Note the distorted surface of the high-speed jet. integration across the cross section, (e.g., sample withdrawal or additive injection). The governing equation is 1 dp R π dp Q = 2πr(r2 − R2)dr =− R4, (3.5) 4µ dz 0 8µ dz ∂v ∂p 1 ∂ ∂v ρ z =− + µ r z . (3.7) ∂t ∂z r ∂r ∂r and the average velocity vz is then simply − (p0 pL) 2 This problem has been solved by Szymanski (1932); it is a vz = R . (3.6) 8µL worthwhile exercise to reproduce the analysis. We begin by eliminating the inhomogeneity (dp/dz); let the fluid velocity Thus, if water is to be pushed through a 1 cm diameter be represented by the sum of transient and steady functions: tube at 20 cm/s, we would need a pressure drop of about 2 6.4 dyn/cm per cm. If the tube was 100 m long, then = + ∼ 2 vz V1 vzSS, (3.8) p0 − pL = 64,000 dyn/cm , which is equivalent to a head of about 65 cm of water (not a very large p for a tube of such length). where vzSS is the steady-state velocity distribution for the Hagen–Poiseuille flow (3.4). This ensures that V1 → 0as t →∞. The result of this substitution is 3.3 TRANSIENT HAGEN–POISEUILLE FLOW ∂V ∂2V 1 ∂V 1 = ν 1 + 1 . (3.9) The unsteady variant of the preceding example has some ∂t ∂r2 r ∂r important practical implications. Consider a viscous fluid, initially at rest, in a cylindrical tube. At t = 0, a fixed pressure The operator on the right-hand side is an indicator; we can gradient (dp/dz) is imposed and the fluid begins to move in expect to see some form of Bessel’s differential equation here. the z-direction. How long will the fluid take to attain, say, 50 Using the product method, with V = f(r)g(t), we confirm that or 90% of its ultimate centerline velocity? You can see imme- 1 diately that such questions are crucial to process dynamics 2 and control—especially in situations with intermittent flow V1 = A exp(−νλ t)J0(λr). (3.10) 26 LAMINAR FLOWS IN DUCTS AND ENCLOSURES
Since V1 must disappear at the wall, J0(λR) = 0. There are an Contrast this result with the case in which glycol is at rest infinite number of λ’s that can satisfy this relation; therefore, in a 1 cm diameter tube; again, a pressure drop is imposed at t = 0. The time required for the centerline velocity to reach ∞ 65% of the ultimate value is only about 0.29 s. = − 2 V1 An exp( νλnt)J0(λnr). (3.11) n=1 3.4 POISEUILLE FLOW IN AN ANNULUS Now one must impose the initial condition so that An ’s that cause the series to converge properly can be identified. Note The annulus is often employed in engineering applications that at t = 0, and it warrants special attention. The governing equation for the pressure-driven flow in an annulus is V =−v . (3.12) 1 zSS ∂v ∂p 1 ∂ ∂v ρ z =− + µ r z . (3.14) The interested reader should complete this analysis by ∂t ∂z r ∂r ∂r demonstrating that Let the cylindrical surfaces be located at r = R1 (inner) and v r2 = z = − r R2 (outer). For the steady laminar flow, the velocity dis- 1 2 Vmax R tribution is given by eq. (3.3): ∞ 4J2(λnR) 2 1 dp 2 − exp(−νλ t)J0(λnr). v = r + C ln r + C , (3.15) (λ R)2J2(λ R) n z 1 2 n=1 n 1 n 4µ dz
(3.13) but unlike the Hagen–Poiseuille case (where C1 = 0),
2 − 2 The results are displayed in Figure 3.2. We should explore (1/4µ)(dp/dz)(R2 R1) C1 =− . (3.16) some examples to get a better sense of the duration of the ln(R2/R1) start-up, or acceleration, period. Consider water initially at rest in a 10 cm diameter tube. The second constant of integration is found by applying the At t = 0, a pressure gradient is imposed and the fluid begins no-slip condition at either R1 or R2. Accordingly, we find to move. When will the water at the centerline achieve 65% 1 dp of its ultimate value? C =− R2 − C ln R . (3.17) 2 4µ dz 2 1 2 νt (25)(0.2) =∼ 0.2, therefore t =∼ = 500 s. R2 (0.01) Note that the location of maximum velocity corresponds to 2 − 2 (R2 R1) Rmax = . (3.18) 2ln(R2/R1)
Therefore, if the inner and outer radii are 1 and 2, respec- tively, the position of maximum velocity is 1.47107—closer to the inner surface than the outer. As the radii become larger (with diminishing annular gap), the location of maximum velocity moves toward the center of the annulus. However, we must add some amplification to this remark; eq. (3.18) has been tested experimentally by Rothfus et al. (1955), who found that the radial position of maximum velocity devi- ates from eq. (3.18) for the Reynolds numbers (defined as) = 2 − 2 Re (2(R2 Rmax) vz )/νR2 between about 700 and 9000. This discrepancy is actually greatest at Re ≈ 2500. Suppose we consider an example (Figure 3.3) in which water is initially at rest in an annulus with R1 and R2 equal to 1 and 2 cm, respectively. At t = 0, a pressure gradient of 2 FIGURE 3.2. Start-up flow in a tube. The five curves correspond −0.1 dyn/cm per cm is imposed and the fluid begins to to the values of the parameter, νt/R2, of 0.05, 0.1, 0.2, 0.4, and 0.8. move in the z-direction. This problem requires solution of These data were obtained by computation. eq. (3.14); the reader is encouraged to explore the alternatives. DUCTS WITH OTHER CROSS SECTIONS 27
It is not surprising to find that the polynomial
2 2 a0 + a1x + a2y + a3x + a4y + a5xy (3.22)
can satisfy eq. (3.21). If we wish to apply the product method (separation of variables) to eq. (3.21), we must eliminate the inhomogeneity. Suppose we let V ∗ = V + y2/2? The result is
∂2V ∗ ∂2V ∗ + = 0. (3.23) ∂x∗2 ∂y∗2
In the usual fashion, we let V ∗ = f (x)g(y), substitute it into (3.18), and then divide by fg. The result is two ordinary differential equations:
f − λ2f = 0 and g + λ2g = 0. (3.24) FIGURE 3.3. Velocitydistributions for the example problem, start- up flow in an annulus, at t = 5, 10, 20, and 40 s. Note that the 50, Since we choose to place the origin at the center of the duct, 70, and 90% velocities will be attained in about 8, 14, and 28 s, the solutions for (3.24) must be written in terms of even respectively. functions. Consequently,
y2 V =− + B cos λy cosh λx. (3.25) How long does it take for the velocity to approach Vmax?In 2 particular, when will the velocity at R attain 50, 70, and max =± = 90% of its ultimate value? Of course, when y h, V 0, so ∞ h2 y2 nπy nπx V = − + Bncos cosh . (3.26) 2 2 2h 2h 3.5 DUCTS WITH OTHER CROSS SECTIONS n=1,3,5,...
We turn our attention to the steady pressure-driven flow in V must also disappear for x =±w: the z-direction in a generalized duct. The governing equation ∞ is 1 2 2 nπy nπw (y − h ) = Bncos cosh . (3.27) 2 2h 2h 2 2 n=1,3,5,... 1 dp = ∂ vz + ∂ vz 2 2 . (3.19) µ dz ∂x ∂y The leading coefficients can now be determined by Fourier theorem: This is a Poisson (elliptic) partial differential equation; since the Newtonian no-slip condition is to be applied every- h 1 (y2 − h2) nπy where at the duct boundary, the problem posed is of the Bn = cos dy. (3.28) Dirichlet type. As one might expect, some analytic solutions h cosh(nπw/2h) 2h 0 are known; this group includes rectangular ducts, eccentric annuli, elliptical ducts, circular sectors, and equilateral trian- You should verify that gles. White (1991) and Berker (1963) summarize solutions for these cross sections and others. We shall review the steps 16h2 sin(nπ/2) Bn =− . (3.29) one might take to find an analytic solution for this type of n3π3 cosh(nπw/2h) problem in the case of a rectangular duct. Let An illustration of the computed velocity distribution is shown ∗ ∗ −µv = = x = x/h, y = y/h, and V = z , (3.20) in Figure 3.4 for the case h 1 and w 2h. h2(dp/dz) The pressure-driven duct flows described by the ellip- tic partial differential equation (3.19) are also easily solved which when applied to (3.19) results in numerically either by iteration or by direct elimination. To illustrate this, we rewrite eq. (3.19) using the second-order ∇2V =−1. (3.21) central differences for the second derivatives; let the indices 28 LAMINAR FLOWS IN DUCTS AND ENCLOSURES
above that is of interest. Observe that the shear stress at the wall τw is not constant on the perimeter. In fact, it is clear that the maximum value occurs at the midpoints of the sides in both cases. What about the magnitude of τw at the vertices? We see that our conventional definition of the friction factor
F 1 2 F = AKf or = τ = ρ vz f, (3.31) A w 2 is no longer applicable. Obviously, f defined in this man- ner would be position dependent. One remedy is to use the mean shear stress in eq. (3.31), obtaining it either by integra- tion around the perimeter or from the pressure drop by force FIGURE 3.4. Velocity distribution for the steady flow in a rectan- balance. gular duct obtained from the analytic solution (3.26), with h = 1 and w = 2h.
3.6 COMBINED COUETTE AND POISEUILLE i and j correspond to the x- and y-directions, respectively. For FLOWS the sake of legibility, we shall replace vz with V: There are many physical situations in which fluid motion is Vi+ ,j − 2Vi,j + Vi− ,j Vi,j+ − 2Vi,j + Vi,j− 1 dp driven simultaneously by both a moving surface and a pres- 1 1 + 1 1 =∼ . (x)2 (y)2 µ dz sure gradient. There are important lubrication problems of this type and we can also encounter such flows in coating and (3.30) extrusion processes. We begin by examining a viscous fluid We shall apply this technique to a duct with a cross section contained between parallel planar surfaces. The upper surface + in the form of an isosceles triangle where the base is 15 cm will move to the right ( z-direction) at constant velocity V and the height is 7.5 cm. This means that the flow area is and then dp/dz will be given a range of values (both negative 56.25 cm2. The resulting velocity distribution is shown in and positive). Obviously, a negative dp/dz will support (aug- Figure 3.5. ment) the Couette flow and a positive dp/dz will oppose it. As one might expect, the vertices have a pronounced effect The appropriate equation is upon the velocity distribution in a duct of this shape. If ∂p ∂2v the same p was applied to water in a cylindrical tube of =− + z 0 µ 2 . (3.32) equal flow area, the average velocity would be 3.55 cm/s ∂z ∂y and the Reynolds number 2530. That is, for the Hagen– We choose to place the origin at the bottom plate and locate Poiseuille flow in a tube with R = 4.23 cm, the average the top (moving) plate at y = b. Equation (3.32) can be inte- velocity v would be about 75% greater than in the trian- z grated twice to yield gular duct illustrated in Figure 3.5. There is another feature of both the rectangular and the triangular ducts illustrated 1 dp v = y2 + C y + C . (3.33) z 2µ dz 1 2
Of course, C2 = 0 by application of the no-slip condition at y = 0. At y = b, vz = V,so
1 dp 2 V vz = (y − by) + y. (3.34) 2µ dz b
It is convenient to rewrite the equation as follows: v b2 dp y2 y y z = − + . (3.35) V 2µV dz b2 b b FIGURE 3.5. Computed velocity distribution for the steady lami- nar flow in a triangular duct; the fluid is water with dp/dz set equal What kinds of profiles can be represented by this velocity dis- to −0.0159 dyn/cm2 per cm. The computed average velocity for this tribution? Depending upon the sign and magnitude of dp/dz, example is 2.03 cm/s. we can get a variety of forms, as illustrated in Figure 3.6; COUETTE FLOWS IN ENCLOSURES 29
and the shear stress τrz is 1 dp C1 τrz =−µ r + . (3.41) 2µ dz r
Once again, dp/dz could be adjusted to produce zero net flow; the reader might wish to develop the criterion as an exercise.
3.7 COUETTE FLOWS IN ENCLOSURES
Shear flows driven solely by a moving surface are common in lubrication and viscometry. There is an important difference between this class of flows and the Poiseuille flows we exam- ined previously. Consider a steady Couette flow between parallel planar surfaces—one plane is stationary and the other moves with constant velocity in the z-direction: FIGURE 3.6. Velocity distributions for the combined Couette– Poiseuille flow occurring between parallel planes separated by 2 d vz a distance b. The upper surface moves to the right (positive z- = = + 0 2 , resulting in vz C1y C2. (3.42) direction) at constant velocity V. dy Note that the velocity distribution is independent of viscosity. A closely related problem, and one that is considerably more in fact, we can adjust the pressure gradient to obtain zero net practical, is the Couette flow between concentric cylinders. flow: The general arrangement is shown in Figure 3.7. In this scenario, one (or both) cylinder(s) rotates and the b flow occurs in the θ- (tangential) direction. Flows of this type 1 dp V Q = (y2 − by) + y Wdy = 0. (3.36) were extensively studied by Rayleigh, Couette, Mallock, and 2µ dz b others in the late nineteenth century; work continued through- 0 out the twentieth century, and indeed there is still an active Consequently, if dp/dz has the positive value of research interest in the case in which the flow is dominated by the rotation of the inner cylinder. This particular flow continues to attract attention because the transition process dp 6µV = , (3.37) is evolutionary, that is, as the rate of rotation of the inner dz b2 there will be no net flow in the duct. The very same problem can arise in cylindrical coordinates when a rod or wire is coated by drawing it through a die (cylindrical cavity) containing a viscous fluid. We have ∂p 1 ∂ ∂v 0 =− + µ r z . (3.38) ∂z r ∂r ∂r
Accordingly,
1 dp 2 vz = r + C ln r + C . (3.39) 4µ dz 1 2
The boundary conditions are vz = V at r = R1 and vz = 0at r = R2, therefore 1 dp FIGURE 3.7. The standard Couette flow geometry for concentric C = V − (R2 − R2) /ln(R /R ) (3.40) 1 4µ dz 1 2 1 2 cylinders. 30 LAMINAR FLOWS IN DUCTS AND ENCLOSURES cylinder is increased, a sequence of stable secondary flows Now we turn our attention back to the more general prob- develops in which the annular gap is filled with Taylor vor- lem as described by eq. (3.43); we assume that the fluid in tices rotating in opposite directions. We will examine this the annular space is initially at rest. At t = 0, the outer cylin- phenomenon in greater detail in Chapter 5. For present pur- der begins to rotate with some constant angular velocity. The poses, we will write down the governing equation for the governing equation looks like a candidate for separation of Couette flow between concentric cylinders: variables, so we will try ∂v ∂ 1 ∂ vθ = f (r)g(t). (3.46) ρ θ = µ (rv ) . (3.43) ∂t ∂r r ∂r θ We find
For the steady flow case, g f + (1/r)f − (1/r2)f = =−λ2, (3.47) C νg f v = C r + 2 . (3.44) θ 1 r resulting in If the outer cylinder is rotating at a constant angular velocity = − 2 = + ω and the inner cylinder is at rest, then g Cexp( νλ t) and f AJ1(λr) BY1(λr). (3.48) ωR2R2 1 r v = 1 2 − . (3.45) θ 2 − 2 2 It clearly makes sense for us to combine the steady-state R1 R2 r R1 solution with this result: The shear stress for this flow is given by τrθ = 2 2 2 2 2 C2 2 −µr(∂/∂r)(vθ/r) = (2µωR R /R − R )(1/r ). Consider v = C r + + C exp(−νλ t)[AJ (λr) + BY (λr)]. 1 2 1 2 θ 1 r 1 1 the case in which the radii R1 and R2 are 2 and 8 cm (a very (3.49) wide annular gap), respectively, and the outer cylinder rotates Noting that our boundary conditions at 30 rad/s. The resulting velocity distribution is illustrated in Figure 3.8. r = R1,vθ = 0 and r = R2,vθ = ωR2 (3.50) Note the deviation from linearity apparent in Figure 3.8. If a Couette apparatus has large radii but a small gap, the must be satisfied by the steady-state solution, it is necessary velocity distribution can be accurately approximated with a that straight line. In the case of the example above with the radii −1 of 2 and 8 cm, τrθ /µ will range from about −34 to −64 s 0 = AJ1(λR2) + BY1(λR2) and if ω = 30 rad/s. 0 = AJ1(λR1) + BY1(λR1). (3.51)
Consequently,
0 = J1(λR1)Y1(λR2) − J1(λR2)Y1(λR1). (3.52)
This transcendental equation has an infinite number of roots and it allows us to identify the λn ’s that are required for the series solution. However, we are still confronted with the constants A and B in eq. (3.49). There is a little trick that has been used by Bird and Curtiss (1959), among others, that allows us to proceed. We define a new function
Z1 = J1(λnr)Y1(λnR2) − J1(λnR2)Y1(λnr) (3.53)
that automatically satisfies the boundary conditions. We can now rewrite the solution for this problem as
∞ C FIGURE 3.8. Velocitydistribution in a concentric cylinder Couette v = C r + 2 + A exp(−νλ2t)Z (λ r). (3.54) θ 1 r n n 1 n device with a wide gap. n=1 COUETTE FLOWS IN ENCLOSURES 31
FIGURE 3.9. The helical Couette flow resulting from the rotation of the outer cylinder and the imposition of a small axial pressure FIGURE 3.10. A square duct with upper surface sliding horizon- = = gradient. For this case, Ta 245 and Rez 18 (photo courtesy of tally (in the z-direction) at a constant velocity. the author).
Reynolds numbers between 700 and 2200. Glasgow and The solution is completed by using the initial condition (with Luecke (1977) added rotation of the outer cylinder to the orthogonality) to find the An ’s: pressure-driven axial flow and discovered that the Reynolds number for the transition could be as low as about 350 for R2 (−C r − (C /r))Z (λ r)rdr ≈ R1 1 2 1 n Ta 200. An = . (3.55) R2 2 Of course, the Couette flows can also be generated in Z (λnr)rdr R1 1 rectangular ducts. For example, suppose we have a square There is another important variation of Couette flow in the duct in which the top surface slides forward in the z-direction concentric cylinder apparatus; if an axial pressure gradient (Figure 3.10). is added to the rotation, a helical flow results from the com- The governing Laplace equation for this flow is bination of the θ- and z-components. If the rotation of the 2 2 outer cylinder is dominant relative to the axial flow, one ∂ vz ∂ vz 0 = + . (3.58) can use dye injection to reveal the flow pattern shown in ∂x2 ∂y2 Figure 3.9. The rotational motion is characterized with the Taylor We place the origin at the lower left corner and allow the number; for the case illustrated here (outer cylinder rotating), square duct to have a width and height of 1. The no-slip it is defined as condition applies at the sides and the bottom and the top surface has a constant velocity of 1 in the z-direction. This − − = ωR2(R2 R1) R2 R1 problem is readily solved with the separation of variables Ta . (3.56) = ν R2 by letting vz f(x)g(y); the resulting ordinary differential equations are The axial component of the flow is driven by dp/dz and the resulting velocity distribution was given previously by (3.15). f + λ2f = 0 and g − λ2g = 0. (3.59) The rotational motion is described by (3.44). The resultant point velocity is obtained from Due to our choice of location for the origin, the solution can only be constructed from odd functions. Therefore, = 2 + 2 1/2 V (r) (vθ vz) . (3.57) ∞ For the Poiseuille flow in plain annuli, Prengle and Rothfus vz = An sin nπx sinh nπy. (3.60) (1955) found that the transition would occur at the axial n=1 32 LAMINAR FLOWS IN DUCTS AND ENCLOSURES
FIGURE 3.12. Flow over a rectangular obstruction in a duct.
and ∂v ∂v ∂v 1 ∂p ∂2v ∂2v y + v y + v y =− + ν y + y . ∂t x ∂x y ∂y ρ ∂y ∂x2 ∂y2 (3.64)
FIGURE 3.11. A laminar flow in a square duct with the upper You will note immediately that there are three dependent surface sliding in the z-direction at a constant velocity of 1. variables: vx , vy , and p. Of course we can add the continuity equation to close the system, but we now recognize a common dilemma in computational fluid dynamics (CFD). We cannot Of course, at y = 1,vz = 1, so compute the correct velocity field without the correct pres- sure distribution p(x,y,t). Let us examine an approach that will ∞ allow us to circumvent this difficulty. We cross-differentiate 1 = An sin nπx sinh nπ. (3.61) eqs. (3.63) and (3.64) and subtract one from the other, elim- n=1 inating pressure from the problem. We also note that for this two-dimensional flow, the vorticity vector component is This is a Fourier series, so the leading coefficients can be determined by integration: ∂vy ∂vx ω = − . (3.65) z ∂x ∂y 2(1 − cos nπ) An = . (3.62) The stream function is defined such that continuity is auto- nπ sinh nπ matically satisfied:
The solution is computed using eq. (3.60) and the result is ∂ψ ∂ψ shown in Figure 3.11. v = and v =− . (3.66) x ∂y y ∂x
We can show that the result of this exercise is the vorticity 3.8 GENERALIZED TWO-DIMENSIONAL FLUID transport equation (you may remember its introduction in MOTION IN DUCTS Chapter 1): We now turn our attention to a very common problem in ∂ω ∂ω ∂ω ∂2ω ∂2ω + v + v = ν + . (3.67) which fluid motion occurs in two directions simultaneously. ∂t x ∂x y ∂y ∂x2 ∂y2 In a duct, this could result from a change in cross section, for example, flow over a step or obstacle. The conduit is In addition, the stream function and the vorticity are related assumed to be very wide in the z-direction such that the through a Poisson-type equation: x- and y-components of the velocity vector are dominant. ∂2ψ ∂2ψ A typical problem type is illustrated in Figure 3.12. −ω = + . (3.68) For the most general case, the governing equations are ∂x2 ∂y2