Direct Modeling For Computational Fluid Dynamics: Construction And Application Of Unified Gas-kinetic Schemes

Construction and Application of Unified Gas-Kinetic Schemes

Kun Xu

• Publisher: World Scientific Publishing Company
• Year: 2014
• eBook ISBN: 9789814623735
• Topic: Biological Sciences
• Subtopic: Science General
• Index: Biological Sciences

Computational fluid dynamics (CFD) studies the flow motion in a discretized space. Its basic scale resolved is the mesh size and time step. The CFD algorithm can be constructed through a direct modeling of flow motion in such a space. This book presents the principle of direct modeling for the CFD algorithm development, and the construction unified gas-kinetic scheme (UGKS). The UGKS accurately captures the gas evolution from rarefied to continuum flows. Numerically it provides a continuous spectrum of governing equation in the whole flow regimes.

Contents:

• Direct Modeling for Computational Fluid Dynamics
• Introduction to Gas Kinetic Theory
• Introduction to Nonequilibrium Flow Simulations
• Gas Kinetic Scheme
• Unified Gas Kinetic Scheme
• Low Speed Microflow Studies
• High Speed Flow Studies
• Unified Gas Kinetic Scheme for Diatomic Gas
• Conclusion

Readership: Undergraduate and graduate students, researchers and professionals interested in computational fluid dynamics.
Key Features:

• Direct modeling for CFD is self-contained and unified in presentation
• It may be used as an advanced textbook by graduate students and even ambitious undergraduates in computational fluid dynamics
• It is also suitable for experts in CFD who wish to have a new understanding of the fundamental problems in the subject and study alternative approaches in CFD algorithm development and application
• The explanations in the book are detailed enough to capture the interest of the curious reader, and complete enough to provide the necessary background material needed to go further into the subject and explore the research literature

Chapter 1

Direct Modeling for Computational Fluid Dynamics

Computational fluid dynamics (CFD) is a scientific discipline, which aims to capture fluid motion in a discretized space. The description of the flow behavior depends closely on the scales which are used to identify or “see” it. All theoretical equations, such as the Boltzmann equation or the Navier-Stokes equations, are constructed and valid only on their modeling scales, even though these scales cannot be explicitly observed in these equations. The mechanism of these governing equations depends on the physical modeling, such as the constitutive relationship in the stress and strain of the hydrodynamic equations, and the separation of transport and collision of the kinetic equation. The existence of a few distinct governing equations, such as the Boltzmann equation, the Navier-Stokes equations, and the Euler equations, only presents a partial picture about flow physics in their specific modeling scales. The governing equations between these scales have not been fully explored yet due to the tremendous difficulties in the modeling, even the flow variables to be used for the description of a non-equilibrium flow in the scale between the Navier-Stokes and the Boltzmann equation are not clear. However, the CFD provides us an opportunity to present the flow physics in the mesh size scale. With the variation of the mesh size to resolve the flow physics, the direct modeling of CFD may open a new way for the description and simulation of flow motion. This book is mainly about the construction of numerical algorithms through the principle of direct modeling. The ultimate goal of CFD is to construct the discrete flow dynamic equations, the so-called algorithm, in a discretized space. These equations should be able to cover a continuum spectrum of flow dynamics with the variation of the ratio between the mesh size and the particle mean free path.

Instead of direct discretization of existing fluid dynamic equations, the direct modeling of CFD is to study the corresponding flow behavior in the cell size scale. The direct discretization of a well-defined governing equation may not be an appropriate way for CFD research, because the modeling scale of partial differential equations (PDEs) and mesh size scale may not be matched. Here, besides introducing numerical error the mesh size doesn’t actively play any dynamic role in the flow description. For example, the Fourier’s law of heat conduction is valid in a scale where the heat flux is proportional to temperature gradient. How could we imagine that such a law is still applicable on a mesh size scale, which can be freely chosen, such as 1cm, 1m, or even 1km? To avoid this difficulty, one may think to resolve everything through the finest scale, such as the molecular dynamics. But, the use of such a resolution numerically in the simulation is not practical due to the overwhelming computational cost, and it is not necessary at all in real engineering applications, since in most times only macroscopic flow distributions are needed, such as the pressure, stress, and heat flux on the surface of a flying air vehicle. Instead of direct discretization of PDEs or resolving the smallest scale of molecular dynamics, a possible way is to model and capture the flow dynamics in the corresponding mesh size scale, and the choice of the mesh size depends on how much information is sufficient to capture the flow evolution in any specific application. Depending on the flow regimes, there is a wide variation between the cell size and the local particle mean free path. Therefore, a multiple scale modeling is needed in the CFD algorithm development, i.e, the construction of the so-called discrete governing equations.

1.1 Physical Modeling and Numerical Solution of Fluid Dynamic Equations

There are different levels in flow modelings. The theoretical fluid mechanics is to apply physical laws in a certain scale with the modeling of the flux and constitutive relationship. Then, based on the construction of discrete physical law, as the control volume shrinking to zero, and with the assumption of smoothness of flow variables in the scale of control volume, the corresponding PDEs are obtained. For the PDEs, even with a continuous variations of space and time, the applicable regime of these equations is on its modeling scale, such as the scale for the validity of constitutive relationship and the fluxes. For the Boltzmann equation, the modeling scale is the particle mean free path and the particle collision time, where the particle collision and transport in such a scale are separated and modeled in an operator splitting way. For the Navier-Stokes (NS) equations, the scale is the dissipative layer thickness, where the Stokes and Fourier’s laws are valid. In the NS modeling, the accumulating effect from a gigantic amount of particle collision and transport, instead of individual one, is modeled. It is more or less a mean field approximation. For the Euler equations, the scale becomes even coarse and only the advection wave propagation is followed, where the dissipative wave structures are replaced by discontinuities at contact surface and shock layer. The equations constructed for fluid dynamics can be the potential flow, the Euler equations, the Navier-Stokes equations, and the Boltzmann equations with different modeling scales. After the establishment of these PDEs, research work has been concentrated on the mathematical analysis of these equations, or on their numerical solution. If these equations are obtained from models with distinguishable scale, could we use a continuous variation of scale and get the corresponding governing equations as well? This is the question we will try to answer in this book.

With the above theoretical PDEs, the traditional CFD method is to discretize these PDEs without referring to the underlying modeling scale of these equations. The intrinsic modeling scale of the equation has no direct connection with the definition of numerical mesh size, except resolving the specific layer numerically. The traditional CFD concerns the limiting solution of the scheme as the cell size and time step approaching to zero. The order of accuracy is used to judge the quality of the scheme. In order to improve the accuracy of the solution in the computation, the truncation error and modified equation are studied. Then, according to the numerical errors, the discretization can be adjusted in order to improve the accuracy. Certainly, the numerical stability of the scheme has to be considered as well. This numerical principle is passive and the ultimate goal is to remove the mesh size effect. In this approach, the best result we can get is the exact solution of the governing equations. But, even in this case the solution is still limited by the physical modeling scale of the equations. In practice, the mesh size and time step can never take limiting values, and the exact governing equations of the numerical algorithms become unknown. Usually the PDEs become simplified model of a real physical reality, such as the absence of dissipation in the Euler equations. However, in a discretized space with limited resolution, the inclusion of dissipation becomes necessary for a numerical scheme. Therefore, the artificial one, instead of physical one, is implicitly added in the numerical solution. The vast amount of shock capturing schemes for the Euler equations clearly indicate that the inviscid Euler equations with incomplete physical modeling have been modified with a variety of artificial dissipations, and there is no unique solution at all in the mesh size scale for the Euler solution, even though they may have the same limiting solution with zero mesh size and time step.

In the past decades, the equations to be solved numerically cover the potential flow (60s-70s), the Euler equations (70s-80s), the NS equations (90s- now), and the Boltzmann type equations (90s- now). Even though these equations have been routinely used in a wide range of engineering applications, the acceptability of the numerical schemes depend more or less on the Verification and Validation (V&V) process. Due to the absence of a solid foundation in the numerical PDEs, there is no a precise predicability about the outcome of these schemes. Peter Lax clearly states “the theory of difference schemes is much more sophisticated than the theory of differential equations.” Even with great success in the CFD research in the past decades, dynamic flaw in most numerical schemes still exist, such as the numerical shock instability in high Mach number flow simulations. There is still uncertainty about how to design high-order schemes for the capturing of both continuous and discontinuous solutions. Now we are facing an insurmountable difficulty about the numerical principle for the computational fluid dynamics. For any scientific discipline, a fundamental principle is necessary once the subject needs to be promoted from an experience-based study, like the current CFD method, to a principle-based scientific discipline. Now we propose that the principle of CFD is to study flow dynamics in the mesh size scale through the direct physical modeling, the so-called direct construction of discrete governing equations.

1.2 Direct Modeling of Fluid Motion

A gas is composed of molecules, and there are kinematic and dynamic properties for these molecules, such as trajectories of these molecules, molecular mass, mean free path, collision time, and interaction potential between colliding particles. Their representation in a numerical scheme depends on the cell resolution. In a discretized space, in order to identify different flow behavior, a numerical cell size Knudsen number, i.e., Kn_c = l_mfp/Δx, can be defined, which gives a connection between the physical gas property (particle mean free path l_mfp) and the cell resolution (Δx). The cell Knudsen number is a parameter to connect the numerical resolution to the molecular reality. In a flow computation, due to the freedom in choosing the cell size and the vast variation of physical condition, such as the flying vehicle at different altitude, the cell’s Knudsen number can cover a wide range of values continuously. For example, around a space shuttle flying in near space, the mesh size can take the value of the molecular mean free path in the non-equilibrium shock regions and hundreds of mean free path far away from the shuttle. At different locations, different values of cell Knudsen numbers correspond to different flow physics.

CFD algorithm is to model the flow motion in the cell size scale. Certainly, the modeling can be much simplified by taking Kn_c ≈ 1 everywhere, where the mesh size is on the same order as the particle mean free path. However from a computational point of view, both efficiency and accuracy of the schemes have to be considered. In a real application, such as the flow around a spacecraft, the particle mean free path can be changed significantly. Besides the difficulties of identifying the local mean free path beforehand, to resolve this scale everywhere will be prohibitively expensive. So, the choosing of a numerical mesh size needs to compromise among the flow information needed for a certain design, the efficiency to achieve such a solution, and the robustness of the scheme if the mesh size is not chosen properly. The purpose of multiple scale modeling is to efficiently capture flow behavior with a variation of scales.

In order to design such a multi-scale modeling scheme, the flow evolution with different Kn_c numbers has to be captured accurately. In the regions of Kn_c ≥ 1, the same modeling mechanism of deriving the Boltzmann equation needs to be used numerically, such as the modeling of particle transport and collision. In the region of Kn_c

1, the modeling in deriving the NS equations should be used in the numerical process, such as the drifting of an equilibrium state in the hydrodynamic limit. Between these two limits, a local flow evolution model has to be constructed as well, even though there is no a valid governing equation yet. Without valid equations, based on numerical PDE approach, it is impossible to design multiscale method, except the artificial brutal connection of different solvers, such as the use of molecular dynamics and the Euler equations in different domain. The aim of the unified scheme is to model the physics in all regimes and construct the algorithm. Different from numerical PDEs, in the direct modeling scheme, the mesh size will actively participate the flow evolution with the changing of the local value Kn_c. In other words, the mesh size and time step are dynamic quantities, where the flow physics is associated explicitly with their resolution, or the accumulation of particle transport and collision in such a scale. The choice of local resolution depends on the specific application.

Many physical problems do need such a unified approach for a valid description of all Kn_c regimes. In the aerospace engineering, we are interested in designing vehicles in the near space, i.e., flying between 20km and 100km. In the flow field, there will have both non-equilibrium flow region with the requirement of the cell size being on the same order of particle mean free path, and the continuum flow region with the cell size being much larger than the local particle mean free path. The challenge is that there is no any distinct boundary between different flow regions. In plasma physics, one needs to recover “macroscopic” modeling in the quasi-neutral region and “kinetic” modeling in non-quasi-neutral region, and there needs a smooth dynamic transition between these models. In radiation and neutron transport, there is optically thick and thin regions, and the dynamics for the photon or neutron transport is different, when a uniform mesh size scale is adopted everywhere. The goal for the development of the direct modeling method is not to use the domain decomposition to construct different flow solvers in different regions, but provides a framework with the automatic capturing of flow physics in different regimes.

The methodology of the unified scheme is to model different flow physics in different regimes in a uniform way, from the particle transport and collision to macroscopic wave propagation. The recipe here is that a time dependent evolution solution from the kinetic to the hydrodynamic scale will be constructed and used in the scheme construction. A valid evolution solution covering different scale physics is critically important here. Therefore, the unified scheme can be considered as an evolution solution-based modeling scheme. Since the kinetic equation itself is valid in the kinetic scale, but the time evolution solution with the account of both transport and collision can go beyond the kinetic scale. The inclusion of intensive particle collision as time step being much larger than the particle collision time will push the solution to the hydrodynamic regime. Since there is no any analytic evolution solution from the Boltzmann equation in general, the key of the unified scheme depends closely on the construction of such an evolution solution. For example, with the integral solution of the kinetic model equation, each term in the solution has to be modeled numerically with the consideration of mesh size scale. The unified scheme provides a general framework for the construction of multiscale method for the transport phenomena with multiple scales physics involved, such as the non-equilibrium flow study and radiative transport. The full Boltzmann collision term can be used as well in its valid scale in the construction of unified scheme.

1.3 Direct Modeling and Multiscale Coarse-graining Models

The macroscopic Navier-Stokes equations can be formally derived from the Boltzmann equation, which in turn can be derived from the Liouville equation via the BBGKY hierarchy[Cercignani (1988)]. Many physical assumptions are built into these derivations and their validations are warranted under certain physical conditions. In the theoretical framework of fluid dynamics, there are only a few distinct successful governing equations, such as the Boltzmann, the Navier-Stokes, and the Euler equations, which can be used to describe the flow physics in the corresponding scales. Between these equations, we may have also the Burnett, or super-Burnett, and many moment equations. But, these equations can be hardly judged from the physical point of view, because they are not directly obtained from first principle of scale-based direct modeling, rather than derived from mathematical manipulation of the kinetic equation, where even the modeling scales of these equations are not clearly defined. The kinetic equation is valid in the kinetic scale, i.e., mean free path and collision time. If the Boltzmann equation is regarded as a valid model in all scales, it still means that the kinetic scale physics is fully resolved and the solution in other scale is obtained through the accumulation of kinetic scale flow evolution. The Boltzmann equation can indeed derive the NS equations which are valid in the hydrodynamic flow regime, but it can also present many other equations through different asymptotic expansions and these equations may not be so successful as the NS ones. Actually, the construction of the NS equations has nothing to do with the Boltzmann equation, and the NS equations were obtained through rational analysis in the macroscopic level earlier. It may not be appropriate to claim that the Boltzmann equation is valid in all scales except they are resolved down to the particle mean free path everywhere.

On the hydrodynamic scale, the macroscopic equations are very efficient, but are not accurate enough to describe the non-equilibrium flow phenomena. Microscopic equation, on the other hand, may have better accuracy, but it is too expensive to be used in the hydrodynamic scale with the required resolution on the kinetic scale. The traditional multiscale modeling is to construct valid technique which combines the efficiency of macroscale models and the accuracy of microscale description. The constitutive relationship is critically important for the macroscopic equations, which has the corresponding physics in microscopic level. For a complex system, the constitutive relations become quite complicated, where too many parameters need to be fitted, and the physical meaning of these parameters becomes obscure. As a result, the original appeal of s...

Table of contents

• Cover
• Half Title
• Editors
• Title
• Copyright
• Preface
• Contents
• 1. Direct Modeling for Computational Fluid Dynamics
• 2. Introduction to Gas Kinetic Theory
• 3. Introduction to Nonequilibrium Flow Simulations
• 4. Gas Kinetic Scheme
• 5. Unified Gas Kinetic Scheme
• 6. Low Speed Microflow Studies
• 7. High Speed Flow Studies
• 8. Unified Gas Kinetic Scheme for Diatomic Gas
• 9. Conclusion
• Appendix A Non-dimensionlizing Fluid Dynamic Variables
• Appendix B Connection between BGK, Navier Stokes and Euler Equations
• Appendix C Moments of Maxwellian Distribution Function and Expansion Coefficients
• Appendix D Flux Evaluation through Stationary and Moving Cell Interfaces
• Bibliography
• Index