The WENO-Schemes and 1D applications CASA Seminar Erwin Karer 25th October 2006 Outline Motivation Reconstruction Outline 1 Motivation 2 Reconstruction 3 WENO 4 Numerical Implementation & Results CASA Seminar WENO-Schemes & Applications Motivation Reconstruction WENO 1D Conservation Law ut (x, t) + fx (u(x, t)) = 0 x∈Ω Domain Ω and one dimensional “cell-structure”: Considered example: f(u(x, t)) = u(x, t)2 . . . Burgers’ equation . 2 CASA Seminar WENO-Schemes & Applications Motivation Reconstruction WENO Finite Volume Method Balance form for cell Ii : Z ut (ξ, t) dξ + f(u(xi+ 1 , t)) − f(u(xi− 1 , t)) = 0 . 2 Ii 2 Idea: Approximate flux f by a numerical flux ˆf: d 1 ˆ u ¯i (t) = − fi+ 1 − ˆfi− 1 , 2 2 dt ∆xi where u ¯i (t) = 1 ∆xi Z xi+ 1 2 u(ξ, t) dξ . xi− 1 2 CASA Seminar WENO-Schemes & Applications Motivation Reconstruction WENO Numerical Flux Numerical flux-functions h(u, v) are the Godunov, the Engquist-Osher or the Lax-Friedrichs flux-function, etc. Approximation: ˆf 1 = h(u− 1 , u+ 1 ). i+ 2 i+ 2 i+ 2 Hence u− and u+ have to be reconstructed. i+ 1 i+ 1 2 2 Figure: sketch of cell boundary CASA Seminar WENO-Schemes & Applications Motivation Reconstruction WENO Numerical Flux Numerical flux-functions h(u, v) are the Godunov, the Engquist-Osher or the Lax-Friedrichs flux-function, etc. Approximation: ˆf 1 = h(u− 1 , u+ 1 ). i+ 2 i+ 2 i+ 2 Hence u− and u+ have to be reconstructed. i+ 1 i+ 1 2 2 Figure: sketch of cell boundary CASA Seminar WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results ENO: A short Repetition Properties of ENO The Task of Reconstruction Given are the cell averages {¯ vi }i=1,...,N with 1 v ¯i := ∆xi Z xi+1/2 v(ξ) dξ , i ∈ {1, . . . , N} xi−1/2 We want: to reconstruct the cell boundary values vi+ 1 and vi− 1 . 2 to achieve a certain accuracy k, i.e. vi± 1 = v(xi± 1 ) + O(∆xk ) i = 1, . . . , N. 2 2 CASA Seminar WENO-Schemes & Applications 2 Reconstruction WENO Numerical Implementation & Results ENO: A short Repetition Properties of ENO ENO Idea For given order k and fixed left-shift r one is able to obtain a reconstruction of v(xi+ 1 ): 2 k−1 X vi+ 1 = crj v ¯i−r+j . 2 j=0 with special crj . (Fixed Stencil Approximation) Newton Divided Differences indicate smoothness! Idea of ENO: start with a two point stencil add in each step another cell depending on the value of the Newton Divided Differences ⇓ Result is the left shift r! CASA Seminar WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results ENO: A short Repetition Properties of ENO Properties of ENO k stencils are considered in the choosing process covering 2k − 1 cells. But only one stencil is used to reconstruct. One could try to get (2k − 1)-th accuracy. Further properties: 1 The stencil might change even by round-off error perturbation. If both Newton Divided Differences are near 0 a small change at the round off level would change the stencil. 2 The numerical flux is not smooth. The stencil pattern might change at neighboring cells. 3 ENO stencil choosing procedure contains many “if” directives which are not efficient on certain vector computers. CASA Seminar WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results Basics Coefficients for Convex Combination Basics Assume a uniform grid, i.e. ∆xi = ∆x, i = 1, . . . , N. Let us define the k candidate stencils Sr (i) = {xi−r , . . . , xi−r+k−1 } r = 0, . . . , k − 1. The reconstruction produces k different approximations to the value v(xi+ 1 ): 2 (r) vi+ 1 2 = k−1 X crj v ¯i−r+j r = 0, . . . , k − 1 j=0 (r) WENO takes a convex combination of the vi+ 1 ’s! 2 CASA Seminar WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results Basics Coefficients for Convex Combination New Approximation We get a new approximation to the value v(xi+ 1 ): 2 vi+ 1 = 2 k−1 X (r) ωr vi+ 1 . 2 r=0 Due to consistency and stability we require ωr ≥ 0 k−1 X ∀r = 0, . . . , k − 1 ωr = 1. r=0 If function v is smooth on all candidate stencils there are constants dr such that one obtains v(xi+ 1 ) = 2 k−1 X (r) dr vi+ 1 + O(∆x2k−1 ). r=0 CASA Seminar 2 WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results Basics Coefficients for Convex Combination Properties Due to constistency we have k−1 X dr = 1. r=0 If v is smooth on all Sr (i) we would like to have ωr = dr + O(∆xk−1 ), which would imply the (2k − 1)-th order accuracy vi+ 1 = 2 k−1 X r=0 (r) ωr vi+ 1 = v(xi+ 1 ) + O(∆x2k−1 ) 2 CASA Seminar 2 WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results Basics Coefficients for Convex Combination Calculation of Coefficients dr Let us define the described reconstruction by qkr (¯ vi−r , . . . , v ¯i−r+k−1 ) := k−1 X ckrj v ¯i−r+j j=0 Let us define the dr via the relation q2k−1 vi−k+1 , . . . , v ¯i+k−1 ) = k−1 (¯ k−1 X r=0 CASA Seminar (k−1−r) dr vi+ 1 . 2 WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results Basics Coefficients for Convex Combination Explanation of the Calculation of the dr Figure: Calculation of the dr Calculation of the dr : Take k − 1 sequential points i ⇒ k − 1 equations P The dr have to fulfill k−1 r=0 dr = 1 ⇒ 1 equation Solve the k equations for the k unknowns. CASA Seminar WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results Basics Coefficients for Convex Combination Coefficients ωr We require the following properties of the coefficients: If v(x) is smooth on Sr (i) we want ωr ≈ dr hence ωr = O(1). If v(x) has a discontinuity inside Sr (i) ωr should be essentially 0, i.e. ωr = O(∆x` ). CASA Seminar WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results Basics Coefficients for Convex Combination Calculation of ωr Combining these aspects we get αr ωr = Pk−1 s=0 αs r = 0, . . . , k − 1 , where αr is given by αr = dr . (ε + βr )2 with ε ∈ [10−5 , 10−7 ] (usually 10− 6). The βr are the so-called “smooth indicators”. CASA Seminar WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results Basics Coefficients for Convex Combination Smooth indicator βr If v(x) is smooth in Sr (i) one wants βr = O(∆x2 ) and for discontinuous v one obtains βr = O(1). One choice for βr which fulfills the requirements is βr = k−1 Z X l=1 xi+ 1 2 2l−1 ∆x xi− 1 ∂ l pr (x) ∂xl 2 dx. 2 CASA Seminar WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results Basics Coefficients for Convex Combination WENO Procedure (r) 1 Calculate the k reconstructed values vi+ 1 for each i. 2 Find the constants dr . 3 Find the smooth indicators βr . 4 Form the weights ωr . 5 The (2k − 1)-th order reconstruction is given by 2 vi+ 1 = 2 k−1 X r=0 (r) ωr vi+ 1 . 2 − Note, that the values vi+ 1 are later denoted by vi+ 1. 2 2 + Additionally one can obtain the values vi− 1 for the cell i using the same considerations. CASA Seminar 2 WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results Basics Coefficients for Convex Combination Comparison of the Reconstruction Figure: Comparison of the fixed-stencil approx.(fixed r = 1) left and the WENO(k = 3) right for ∆x = 0.02. CASA Seminar WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results FVM 1D scalar FDM with WENO-Roe 1D scalar FDM with flux-splitting Burger’s equation using FVM and WENO CASA Seminar WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results FVM 1D scalar FDM with WENO-Roe 1D scalar FDM with flux-splitting FDM approximation We use a conservative approximation to the spatial derivative 1 ˆ dui (t) =− fi+ 1 − ˆfi− 1 . 2 2 dt ∆x ui (t) is the numerical approximation to the point value u(xi , t). We want 1 ˆ fi+ 1 − ˆfi− 1 = fx (u(xi , t)) + O(∆xk ) ∀ i. 2 2 ∆xi This numerical flux is obtained by ENO or WENO reconstruction using the setting v ¯(x) = f(u(x, t)). CASA Seminar WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results FVM 1D scalar FDM with WENO-Roe 1D scalar FDM with flux-splitting Upwinding using the Roe speed We have to define the Roe speed ai+ 1 := ¯ 2 f(ui+1 ) − f(ui ) . ui+1 − ui − + Depending on the sign of the ¯ai+ 1 one uses vi+ for the 1 or v i+ 1 2 numerical flux, i.e. 2 2 if ¯ ai+ 1 ≥ 0 one says that the wind blows from the left, 2 hence one uses v− 1 for the numerical flux ˆf 1 . i+ 2 i+ 2 if ¯ai+ 1 < 0 one says that the wind blows from the right, 2 hence one uses v+ 1 for the numerical flux ˆf 1 . i+ 2 CASA Seminar i+ 2 WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results FVM 1D scalar FDM with WENO-Roe 1D scalar FDM with flux-splitting Burger’s equation using FDM and WENO-Roe CASA Seminar WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results FVM 1D scalar FDM with WENO-Roe 1D scalar FDM with flux-splitting Burger’s equation using FDM and WENO-Roe 2 CASA Seminar WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results FVM 1D scalar FDM with WENO-Roe 1D scalar FDM with flux-splitting Flux-splitting A more stable approach is to use a splitting of the flux, i.e. f(u) = f + (u) + f − (u) where df + (u) df − (u) ≥ 0 and ≤ 0. du du hold. For example the Lax-Friedrich splitting: 1 f ± (u) = (f(u) ± αu) 2 where α is defined by α = max |f 0 (u)| u CASA Seminar WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results FVM 1D scalar FDM with WENO-Roe 1D scalar FDM with flux-splitting Flux splitting procedure The numerical flux is then obtained by the following procedure: 1 Identify v ¯i = f + (u(xi )) and use ENO or WENO − reconstruction procedure to obtain the values vi+ 1. 2 2 Set the positive numerical flux as ˆf + 1 = v− 1 . i+ i+ 2 2 f − (u(xi )) 3 Identify v ¯i = and use ENO or WENO + reconstruction procedure to obtain the values vi+ 1. 4 Set the negative numerical flux as ˆf − 1 = v+ 1 . 2 i+ 2 5 i+ 2 Form the numerical flux as ˆf 1 = ˆf + 1 + ˆf − 1 . i+ i+ i+ 2 CASA Seminar 2 2 WENO-Schemes & Applications Reconstruction WENO Numerical Implementation & Results FVM 1D scalar FDM with WENO-Roe 1D scalar FDM with flux-splitting Burger’s equation using FDM with flux-splitting CASA Seminar WENO-Schemes & Applications

© Copyright 2020 ExploreDoc