L D-R35 900 GN NBMIERRII(CHHOLF F FL1U9I8D3 DNY0N0A0M1I4C-S7(5U-)C -0H5A9R6V RD UNIV CAIBRIDGE kR 1/If N U NC L ASSIFf l EG 20/ N L1. 111101 1; 25~ IN NABLTU IOOES AN 2A I ID2S19 NUMERICAL FLUID DYNAMICS By Garrett Birkhoff Harvard University 1983 V1 -Contract N00014-75-C-0596 kiL-A A ii C, .'.CA PREFACE These lecture notes are intended to provide a helpful overview of progress in numerical fluid dynamics, since its conception in the mind of von Neumann 40 years ago. Their main purpose, however, is not to serve as a historical record; still less are they intended *to present a "state of the art" survey in a rapidly changing field. Instead, their purpose is to supply a thoughtful historical perspective which may help readers to assess future possibilities. Qualitatively, everyone is aware of the enormous ii.crease in the computing power that has become available, and specialists are familiar with the great variety of ingenious algorithms that have been proposed for solving mathematical problems originating in fluid mechanics. But reliable quantitative assessments of what can and what cannot be accomplished in a cost-effective manner are much harder to make. Indeed, experts agree that one can at best hope to make reliable order-of-magnitude estimates. Only for a few very specific problems are cost estimates provided in these notes. Instead, a critical review is provided 7of the main achievement and limitations of various analytical and numerical models that have been proposed for predicting and/or simulating fluid motions. For historical reasons and because they are more fundamental, analytical models are reviewed first, in Chapters 1 and 2. Chapter 3 attempts to bring to life von Neumann's original brilliant insights, as seen in the light of later developments. Chapters 4 and 5 are concerned with simple mathematical models of fluid flow that have given useful quantitative insights into reality: those of potential flow and sound waves. These models are amenable to rigorous mathematical analysis, and they provide 'Readers trying to decide whether to read these notes might look at my von Neumann lecture, published in SIAM Review 25 (1983), 1-34. This covers related material much more briefly, but in the same spirit. 2See the chart on p. 3, which is the key to the viewpoint stressed in this book. . . ? .- . :- -,-.- , -- -.. *-.- . -2- a good testing ground for the dream of Euler, Poincare, and Hilbert: of making fluid mechanics into a mathematical science, like geometry. Von Neumann, who seems to have considered the Laplace and wave equations dealt with in Chapters 4 and 5 as well understood, was fascinated by one-dimensional nonlinear waves in a compressible fluid. This may have been because they are not only relatively accessible mathematically, but were also accessible computationally even in his lifetime. Chapter 6 surveys some of the great theoretical and computational progress that has been made in treating them since that time, focussing on a few unsolved problems that should be solvable with moderate resources. Chapter 7 deals with (mathematical) flows of an (idealized) incompressible viscous fluid. Experts are agreed that if we could integrate the Navier-Stokes equations which govern these, analytically or numerically, then a vast variety of important practical problems (pipe flow, airplane resistance at speeds U < 200 miles/hr) could be effectively treated without recourse to experiment. However, the phenomena of flow separation and turbulence have so far required extensive empirical data before becoming amenable to reliable computer simulation. A complex interplay of numerical and physical empiricism permeates the technical literature as a result of this situation; to compare . the 'cost effectiveness' of different computational procedures -S N in this area is therefore very difficult (and problem-dependent). In their present form, these lecture notes were largely prepared while I was ONR Research Professor at the Naval Post- graduate School in Monterey, California. I wish to thank the Office of Naval Research for its continuing support of the effort required to understand an extraordinarily complex and many-sided subject, and to give a reasonably coherent survey of significant parts of it. Garrett Birkhoff A -. ..,. . . . . . .. . ::i .,;- ;. :, .- '.;. - - .' ' ... . TABLE OF CONTENTS 1. DYNAMICS OF IDEAL FLUIDS - Bibliography [All - [Al2] 1. Models of Fluids (pp. 1-5) 2. Euler's Equations (pp. 5-7) 3. Ideal Fluids (pp. 7-12) 4. Potential Flows (pp. 12-16) 5. Plane Potential Flows (pp. 16-18) 6. Gravity Waves (pp. 19-22) 7. Inertial Similarity (pp. 22-26) 8. Fluid Resistance (pp. 26-29) 9. Free Streamlines; Wakes (pp. 30-34) 10. Plane Vortex Flows (pp. 34-39) 11. Airfoil Theory (pp. 39-44) 12. Convection of Vorticity (pp. 44-48) 2. COMPRESSIBILITY AND VISCOSITY, Bibliography (Bl] - [B12] 1. Introduction (pp. 1-3) 2. Sound Waves (pp. 3-8) 3. Helmholtz Equation (pp. 8-10) 4. Equations of State (pp. 11-15) 5. Thermodynamic Effects (pp. 16-19) 6. Mach Numbers; Adiabatic Flow (pp. 20-22) 7. Shock Formation (pp. 22-25) 8. Viscosity (pp. 25-30) 9. Reynolds Number (pp. 30-34) 10. Boundary Layer Theory (pp. 35-37) 11. Turbulence (pp. 37-39) 12. Analytical Fluid Dynamics in 1940 (pp. 39-41) . 3.- VON NEUMANN'S INFLUENCEJ Bibliography (ClI C1]7 - 11 . Background (pp. 1-4) 2. Progress Before 1930 (pp. 4-6) 3. Stability Conditions (pp. 6-8) 4. Southwell and 'Relaxation' Methods (pp. 6-11) 5. Von Neumann's Vision (pp. 11-15) 6. Von Neumann's Influence (pp. 15-18) 7. Von Neumann's Legacy I (pp. 18-22) 8. Von Neumann's Stability Test (pp. 23-25) 9. The Wave Equation (pp. 25-27) 10. Von Neumann's Legacy II (pp. 28-32) 11. Molecular Models of Fluids (pp. 32-36) .ic Cop'I 4. POTENTIAL FLOWS., Bibliography [DlI - [D141 1. Introduction (pp. 1-3) 2. Inverse Methods (pp. 4-5) 3. A Potential Flow Problem (pp. 5-8) 4. Corner Singularity (pp. 8-12) 5. Added Mass (pp. 13-16) 6. Container Effect (pp. 16-19) 7.. Two-Dimensional Airfoils (pp. 20-23) . 8. Free Surfaces (pp. 23-25) 9. Ship Wave Resistance (pp. 25-28) * 40. Interfacial Instability (pp. 28-31) 11. Free Streamlines (pp. 31-32) 5. SOUND WAVES Bibliography' [El] - [E161 1. Introduction (pp. 1-4) 2. Boundary Conditions (pp. 4-6) 3. Dispersion Analysis: Plane Waves (pp. 6-9) 4. Second-order Accuracy: Plane Waves (pp. 10-14) 5. Second-order Accuracy: Cylindrical Waves (pp. 14-20) 6. NONLINEAR ONE-DIMENSIONAL WAVES ' Bibliography [F1] - [F28] 1. Introduction (pp. 1-3) 2. Isentropic Flow (pp. 4-7) 3. Shocks: Von Neumann-Richtmyer Scheme (pp. 7-10) 4. Lax-Wendroff Scheme: Shock 'Capturing' (pp. 11-15) 5. Artificial Viscosity; Recent Developments (pp. 16-18) 6. Shock Fitting; Moretti's Program (pp. 18-22) ,, . 7. Conservation Law Form (pp. 22-27) 7. INCOMPRESSIBLE VISCOUS FLOWS) Bibliography [Gl] - [GI0] 1. Introduction (pp. 1-3) 2. Parallel Flows (pp. 4-9) 3. Nearly Parallel Flows (pp. 10-13) 4. Two Time-Dependent Examples (pp. 14-18) 5. Vorticity Transport (pp. 19-22) 6. Stokes Flows (pp. 23-25) 7. Finite Element Methods (pp. 26-28) 8. Boundary Layers (pp. 29-31) 8. APPENDIX A . Lagrangian Dynamical Systems' [All - [A5] *- " 9. APPENDIX B - Conformal Maps and Potential Flows'Bl] - [B9] l0 APPENDIX C -Fourier Analysis [Cl] - [C3] i. APPENDIX D K'Navier-Stokes Equations; {Dl] - [D61 12.. APPENDIX E Molecular Models of MatterIiEl] - [E61 13 .APPENDIX F 'Courant' Stability Conditions and Amplication Matrices,[Fl] - [F51 li(. APPENDIX G ITwo Dimensional Airfoil Theory, (G1] - [G41 %'L / W\ * .*.*..- **** *--~**,.. *.*,* . . . . . . . CHAP. 1. DYNAMICS of IDEAL FLUIDS 1. Models of fluids. Fluids (i.e., gases and liquids) differ from solids in their physical inability to withstand shear stress without deforming. As a result, fluid motions or "flows" are characterized by the large deformations which blobs of matter can undergo. Many models have been proposed for deriving their behavior from first principles, within the framework of Newtonian mechanics. This chapter will be concerned with the simplest model: that of a so-called ideal fluid. Such an "ideal" fluid is defined physically by two properties: (i) it is incompressible (volume is conserved), and (ii) it is subject to zero shear stress even when moving. These assumptions (within the general framework of Newtonian continuum mechanics) will be formulated mathematically in §§2-3. ". But before discussing in detail the flows of such "ideal" fluids, we will list for contrast some other important mathematical models of fluids that have been studied in depth. Eight of these are listed in tabular form in Table 1; we will consider all of them in this book. In this chapter, we will des- cribe some successful uses of the first two of these models, both of which attempt to rationalize (i.e., predict quantitatively) the behavior of nearly ideal fluids that are very slightly compressed, and whose shear stresses are very much less than their pressure stresses. In practice, mathematical treatments of fluid motions, whether analytical or numerical, ordinarily neglect most of the following: viscosity, compressibility, external gravity, and the effects of variations in temperature and entropy (e.g., natural convection). They often also neglect the effects of lateral or vertical velocity components, or their squares, as in the theories of sound waves and shallow water waves and in boundary layer theory. Indeed, many of the most successful mathematical models K"* -. *.*:--* -o.-*. *-*... --. **>,.***-.". j - • .• . • e. " . " . .. .- ..° . .° .." . . . . '- 1-2 of fluid flows are only asymptotic, in the sense that they depend on a parameter (e.g., the Reynolds number or its reciprocal), and are only accurate when this parameter is very small or very large. Eight standard models. Among the many different initial- boundary value problems that have been proposed as realistic mathematical models for fluid flows, eight are especially signi- ficant. Each of these models is mathematically self-contained, in the sense that its basic equations can be regarded as an axiom system from which the behavior of various kinds of fluid motions can be deduced mathematically, in somewhat the same way that Euclid deduced geometrical theorems from his axioms, and that Newton and his successors (especially Laplace in his Mecanigue Celeste) derived the orbits of the planets and their moons from three force laws and the law of universal gravitation.1 Table 1 also lists (in the second column) the chapters in Lamb's classic treatise [A6] which summarize what was known about each model mathematically as of 1932, and the names of some key concepts which are associated with it. Table 1 can be viewed as a guide through Chaps. 1-3 below, which explain in more detail the significance and typical applications of these concepts. Molecular effects. Diffusion, boiling, condensation, latent heat, and many other physical phenomena can only be understood, even qualitatively, in terms of the molecular structure of matter. Therefore, we have included kinetic theory in our list of analytical models, even though it is based on ordinary and not on partial DE's, and is very different for liquids than for gases. We will discuss this model in Chap. 3. One may well ask: what is the purpose of studying mathe- matical models that are known to neglect physical variables such as compressibility and viscosity? There are three different answers to this question. iA ctually, even celestial mechanics is not exact: it neglects relativistic effects and the asphericity of the sun and planets! ,i bI