- Home
- Documents
*Discretization and Preconditioning Algorithms for the ... Discretization and Preconditioning...*

prev

next

out of 56

View

5Download

0

Embed Size (px)

Discretization and Preconditioning Algorithms for

the Euler and Navier-Stokes Equations on

Unstructured Meshes

Tim Barth

NASA Ames R.C.

Moffett Field, California USA

https://ntrs.nasa.gov/search.jsp?R=20020023595 2020-03-25T06:28:20+00:00Z

Contents

Symmetrization of Systems of Conservation Laws 3 1.1 A Brief Review of Entropy Symmetrization Theory .............. 4

1.2 Symmetrization and Eigenvector Scaling ................... 5

1.3 Example: Compressible Navier-Stokes Equations ............... 6

1.3.1 Entropy Scaled Eigenvectors for the 3-D Euler Equations ..... 8

1.4 Example: Quasi-Conservative MHD Equations ................ 9

1.4.1 Powell's Quasi-Conservative Formulation of the MHD Equations . . 10

1.4.2 Symmetrization of Powell's Quasi-Conservative MHD Equations . . 11

1.4.3 Eigensystem of Powell's Quasi-Conservative MHD Equations .... 12

1.4.4 Entropy Scaled Eigensystem of Powell's Quasi-Conservative MHD

Equations ................................. 12

Maximum Principles for Numerical Discretizations on Triangulated Do- mains 15

2.1 Discrete Maximum Principles for Elliptic Equations ............. 16

2.1.1 Laplace's Equation on Structured Meshes ............... 16

2.1.2 Monotone Matrices ............................ 17

2.1.3 Laplace's Equation on Unstructured Meshes .............. 17

2.2 Discrete Total Variation and Maximum Principles for Hyperbolic Equations 21

2.2.1 Maximum Principles and Monotonicity Preserving Schemes on Mul- tidimensional Structured Meshes .................... 25

2.2.2 Maximum Principles and Monotonicity Preserving Schemes on Un- structured Meshes ............................ 26

Upwind Finite Volume Schemes 34

3.1 Reconstruction Sctmmes for Upwind Finite Volume Schemes ......... 34

3.1.1 Green-Gauss Reconstruction ...................... 35

3.1.2 Linear Least-Squares Reconstruction .................. 36

3.1.3 Monotonicity Enforcement ........................ 38 3.2 Numerical Solution of the Euler and Navier-Stokes Equations Using Upwind

Finite Volume Approximation .......................... 42

3.2.1 Extension of Scalar Advection Schemes to Systems of Equations . . 42

3.2.2 Example: Supersonic Oblique Shock Reflections ............ 43

3.2.3 Example: rlYansonic Airfoil Flow .................... 44

3.2.4 Example: Navier-Stokes Flow with Turbulence ............ 44

4 Simplified GLS in Symmetric Form 48 4.1 Congruence Approximation ........................... 48

4.2 Simplified Galerkin Least-Squares in Symmetric Form ............ 49

4.2.1 A Simplified Lea.st-Squares Operator in Symmetric Form ...... 49

4.2.2 Example: MHD Flow for Perturbed Prandtl-Meyer Fan ....... 50

2

Chapter 1

Symmetrization of Systems of

Conservation Laws

This lecture briefly reviews several related topics associated with the symmetrization of

systems of conservation laws and quasi-conservation laws:

1. Basic Entropy Symmetrization Theory

2. Symmetrization and eigenvector scaling

3. Symmetrization of the compressible Navier-Stokes equations

4. Symmetrization of the quasi-conservative form of the MHD equations

There are many motivations, some theoretical, some practical, for recasting conservation

law equations into symmetric form. Three motivations are listed below. Tile first motivation

is widely recognized while the remaining two are less often appreciated.

1. Energy Considerations. Consider the compressible Navier-Stokes equations in quasi- linear form with u the vector of conserved variables, fi the flux vectors, and M the

viscosity matrix

u,t + _u u,x, = (Mi3u,x_),x,. (i.i)

In this form, the inviscid coefficient matrices f/,u are not symmetric and tile viscosity

matrix M is neither symmetric nor positive semi-definite. This makes energy analysis

almost impossible. When recast in symmetric form, tim inviscid coefficient matrices

are symmetric and the viscosity matrix is symmetric positive senti-definite. The energy

analysis associated with Friedrichs systems of this type is well-known.

. Dimensional consistency. As a representative example, consider the time derivative

term from (1.1). The weak variational statement associated with this equation requires

the integration of terms such as - f w Ttu dxdt. When w and u reside in the same

space of functions, the inner product quantity wTu is dimensionMly inconsistent.

Consequently, errors made in a computation would depend fundamentally on how

the equations have been non-dimensionalized. When recast in symmetric form, the

inner product wTu is dimensionally consistent with units of entrol)y density per unit

volume.

. Eigenvector scaling. Apart from degenerate scalings, aIw scaling of eigenvectors satis-

fies the eigenvalue problem. Unfortunately, numerical discretization techniques some-

times place additional demands on the form of right eigenvectors. As noted by Balsara

[Ba194] in his study of high order Godunov methods, several of the schenms he stud-

ied that interpolate "characteristic" data (see for example Harten et al. [HOEC87])

showed accuracy degradation that depended on the specific scaling of the eigenvectors.

In the characteristic interpolation approach, the solution data is projected onto the lo-

cal right eigenvectors of the flux Jacobian, interpolated between cells, and finally trans-

formed back. The interpolant thus depends on the eigenvector form. Symmetrization

provides some additional insight into the scaling of eigenvectors. Let R(n) denote the

matrix of right eigenvectors associated with the generalized flux Jacobian in the direc-

tion n. Using results from entropy symmetrization theory, Sec. 1.2 describes a scaling

of right eigenvectors such that the product R(n)RT(n) is independent of the vector n.

This result is used in Sec. 1.4 which discusses the ideal magnetohydrodynamic (MHD)

equations. The right eigenvectors associated with these equations exhibit notoriously

poor scaling properties, especially near a triple umbilic point where fast, slow and

Allen wave speeds coincide, [BW88]. The entropy symmetrization scaling provides

a systematic approach to scaling eigenvectors which is unique in the sense described above.

1.1 A Brief Review of Entropy Symmetrization Theory

Consider a system of m coupled first order differential equations in d space coordinates and

time which represents a conservation law process. Let u(x, t) : R d × R + _ R m denote the

dependent solution variables and f(u) : R m _-_ R m×a the flux vector

u,t + f/,z, = 0 (1.2)

with implied summation on the index i. Additionally, this system is assumed to possess the

following properties:

1. Hyperbolicity. Tim linear combination

f. (n) =

has m real eigenvalues and a complete set of eigenvalues for all n E R a.

2. Entropy Inequality. Existence of a convex entropy pair U(u), F(u) : R m ,-+ R such

that in addition to (1.2) the following inequality holds

u, + _

then the matrices U,v: H,v,v, fvL= 5_,v,v

are symmetric. To insure positive-definiteness of U,v so that mappings are one-to-one,

convexity of H(v) is imposed. Since v is not yet; known, little progress has been made but

introducing the Legendre transform

followed by differentiation

U(u) = uTv - U(v)

V,u -- v T + uTv,u --/A/,vV,u -----vT

yields an explicit expression for the entropy variables v in terms of the entropy function

U(u). Symmetrization and generalized entropy functions are intimately linked via the

following two theorems:

Theorem 1.1.1 Godunov [God61] If a hyperbolic system is symmetrized via change of

variables, then there exists a generalized entropy pair for the system.

Theorem 1.1.2 Mock [Moc80] If a hyperbolic system is equipped with a generalized en-

tropy pair U, F/, then the system is symmetrized under the change of variables v T = U,u.

For many physical systems, entropy inequalities of the form (1.3) can be derived by ap-

pending directly to the conservation law system and the second law of thermodynamics.

Using this strategy, specific entropy functions for the Navier-Stokes and MHD equations

are considered in Sees. 1.3, 1.4 respectively.

1.2 Symmetrization and Eigenvector Scaling

In this section, an important property of right (or left) symmetrizable systems is given.

Simplifying upon tile previous notation, let A ° -- U,v, A i = _v and rewrite (1.4)

A_°v,t + _Ai A ° v,_, = 0. (1.5) SPD Symm

The following theorem states an important property of the symmetric matrix products AiA °

symmetrized via the symmetric positive definite matrix A °.

Theorem 1.2.1 (Eigenveetor Sealing) Let A E R _×_ be an arbitrary diagonMizable

matrix and S the set of all right symmetrizers:

S={BcR '_xn ] B SPD, AB symmetric}.

Further, let R E R n×n denote the right eigenvector matrix which diagonalizes A

A = RAR-1

with r distinct eigenvalues, A = Diag(A1 [m_xm_, A21m2x,n.,,..., A_Im_xm_). Then for each

B C S there exists a symmetric block diagonM matrix T = Diag(Tml xml, T, n2xm2, •• •, Tmr xm,.)

that block sca