lax-wendroff convergence theorems for a class of overset...

42
Lax-Wendroff Convergence Theorems for a Class of Overset Mesh Methods Based on Higher Order Flux Reconstruction Dr. Robert Tramel Kord Technologies 13th Symposium on Overset Composite Grids and Solution Technology October 23, 2016

Upload: others

Post on 22-Jul-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

  • Lax-Wendroff Convergence Theorems for a Class of Overset Mesh Methods Based on

    Higher Order Flux Reconstruction

    Dr. Robert Tramel Kord Technologies

    13th Symposium on Overset Composite Grids and Solution Technology

    October 23, 2016

  • Conservation Law Based Update Schemes • In this talk convergence to a weak solution under grid refinement is demonstrated for a class of overset grid schemes based on higher order Flux Reconstruction (FR) as well as MUSCL schemes • The method is based on enforcing the conservation laws on a cell-wise basis rather than the traditional approaches involving interpolation • Sources of conservation error are clearly identified • Consider the following conservation law

    2

    0)( =∂

    ∂+

    ∂∂

    xqf

    tq

    2016 Overset Grid Syposium

  • Conservation Law Based Update Schemes

    3

    • In order to solve this equation numerically on a single grid, the domain is partitioned into discrete cells

    • The variable q is averaged over cell j as follows

    •The value of the cell averaged variable in cell Ω𝑗 is updated as follows

    ( ) ( )( ) , ,

    111

    jjjjnj

    nj qrqlFqrqlFx

    tqq −∆∆

    −= +++

    jq

    ∆+

    ∆−=

    2,

    2xxxxj jj

    etc. ;t n ; ) ,(1 2

    2

    ∆=∆

    = ∫∆

    +

    ∆−

    nn

    xx

    xx

    nj tdxtxqx

    qj

    j

    2016 Overset Grid Syposium

  • MUSCL Reconstruction • Here ql, qr are the values of q reconstructed from the cell averages and evaluated at the cell interfaces using the MUSCL procedure and F is a numerical flux function (Roe, Lax–Friedrichs, etc.)

    • Extension to higher order in multiple dimensions requires large

    stencil sizes. This is impractical for overset grids. 4

    njq

    (Barth and Ohlberger, 2004)

    2016 Overset Grid Syposium

  • Flux Reconstruction1,2

    2016 Overset Grid Syposium 5

    Discontinuous Solution

    • The solution in each cell is approximated by K pieces of data • A K-1 order polynomial is used to fit the data. • and are defined by extrapolating the polynomials to the cell faces

    1. Huynh, H. T. "A flux reconstruction approach to high-order schemes including discontinuous Galerkin methods." AIAA paper 4079 (2007): 2007. 2. A Novel Approach for Shock Capturing in Unstructured High-Order Methods, by Abhishek Sheshadri, Stanford University

    1+kql

    1+kqr

    1+kql 1+kqr

    http://www.nas.nasa.gov/publications/ams/2016/07-07-16.html

  • Flux Reconstruction

    2016 Overset Grid Syposium 6

    Discontinuous Flux

    • A K-1 order polynomial is used to approximate the flux function

  • Flux Reconstruction

    2016 Overset Grid Syposium 7

    ( )11, ++ kk qrqlF

    Discontinuous Flux

    • and are used to define an upwind flux at the face 1+kql 1+kqr

  • Flux Reconstruction

    2016 Overset Grid Syposium 8

    Continuous Flux

    • Correction polynomials are used to create a continuous Kth order flux approximation F

    𝑑𝑞𝑗,𝑘𝑑𝑑 = 𝐹𝑥 𝑗,𝑘

    • The continuous flux polynomial is used to time advance the variables

  • Conservation Law Based Update Schemes • What is the “best” way to extend these approaches to cases of multiple overlapping grids?

    •Rather than interpolate variables or fluxes seek alternate formulation

    •Use data from both grids to reconstruct interface states •Consider two grids with some small amount of overlap

    1 2 3 4

    np-3 np-2 np-1 np

    9

    OVERLAP REGION GRID 1 GRID 2 2016 Overset Grid Syposium

  • Conservation Law Based Update Schemes •Let the value of the cell averaged variable in cell j of grid 1 be denoted as , and the cell centroid value of cell j be denoted as etc

    •Let the left and right interface states at face k of grid 1 be denoted as and respectively •Let the face centroid of face k of grid 1 be denoted as respectively

    q j1

    1kql

    1kqr

    1kql

    1kxf

    1kqr

    1kxf

    1jxc

    1jxc

    10 2016 Overset Grid Syposium

  • Conservation Law Based Update Schemes • Consider computing the cell interface states and as follows

    • The states are computed by constructing a least squares monotonically limited approximation to the solution gradient in cell np from mesh 1 using the cells below and then extrapolating q to the cell faces using this gradient

    np 1 np-1 ( )( )12 1112 1

    111

    1111

    npnpnp

    npnpnpnpnp

    xcxfqqqlxcxfqqql

    −•∇+=−•∇+= ++

    11+npql

    21ql

    11+npql

    21ql

    11 2016 Overset Grid Syposium

  • np 1 2

    Conservation Law Based Update Schemes • Consider computing the cell interface states and as follows

    • The states are computed by constructing a least squares monotonically limited approximation to the solution gradient in cell 1 from mesh 2 using the cells below and then extrapolating q to the cell faces using this gradient

    ( )( )212 121212 1

    21

    11

    21

    21

    11

    xcxfqqqrxcxfqqqr npnp

    −•∇+=−•∇+= ++

    11+npqr

    21qr

    11+npqr

    21qr

    12 2016 Overset Grid Syposium

  • Data Exchange for Flux Reconstruction

    2016 Overset Grid Syposium 13

    1. Galbraith, M. C., A Discontinuous Galerkin Overset Solver, Ph.D. thesis, University of Cincinnati, 2013.

    2. Crabill, J., Jameson, A. and Sitaraman, J.: A High-Order Overset Method on Moving and Deforming Grids. AIAA 2016-3225, AIAA Aviation,

    AIAA Modeling and Simulation Technologies Conference, 13-17 June 2016, Washington, DC.

    11+npqr

    21ql

    1. Extrapolate from boundary cell solution polynomial 2. Interpolate from donor cell solution polynomial 3. Compute upwind flux

    21qr

    11+npql

    11+npql

    11+npqr

    ( )1 11 1, ++ npnp qrqlF

  • Data Exchange for Flux Reconstruction

    2016 Overset Grid Syposium 14

    1. Galbraith, M. C., A Discontinuous Galerkin Overset Solver, Ph.D. thesis, University of Cincinnati, 2013.

    2. Crabill, J., Jameson, A. and Sitaraman, J.: A High-Order Overset Method on Moving and Deforming Grids. AIAA 2016-3225, AIAA Aviation,

    AIAA Modeling and Simulation Technologies Conference, 13-17 June 2016, Washington, DC.

    11+npqr

    21ql

    1. Extrapolate from boundary cell solution polynomial 2. Interpolate from donor cell solution polynomial 3. Compute upwind flux

    21qr

    11+npql

    11+npql

    11+npqr

    ( )1 11 1, ++ npnp qrqlF

  • •A consistent and convergent scheme should be achieved as grid refinement is performed if the distance between the cells centers used in the reconstructions approach zero under grid refinement, i.e.

    •Here TV is the total variation

    •A similar argument holds for meshes with small gaps between the grids

    •The basic argument remains unchanged in the case of general overlap

    Conservation Law Based Update Schemes

    ( ) ( )etc.

    )(

    thatin0 as and

    21

    11

    21

    21

    11

    21

    2 1

    1 1

    2 1

    1 1

    2 1

    1 1

    xqTVxfxfqxfxfqqrqr

    xqrqrqlql

    npnpnp

    npnp

    ∆≤−∇≤−•∇≅−

    →∆→→

    +++

    ++

    15 2016 Overset Grid Syposium

  • Generalized Lax-Wendroff Theorem

    • Recall that a function q(x,t) is considered to be a weak solution to the conservation law

    If for any compact, differentiable function φ(x,t) the following integrals are satisfied1

    0)( =∂

    ∂+

    ∂∂

    xqf

    tq

    ( )[ ] ( ) ( )dxxqxdtdxqfq xt ∫∫∫∞

    ∞−

    ∞−

    −=+ 0,0, 0

    φφφ

    2016 Overset Grid Syposium 16

    1. LeVeque, Randall J., and Randall J. Leveque. Numerical methods for conservation laws. Vol. 132. Basel: Birkhäuser, 1992.

  • ( ) ( ) ( )( )

    , , ),( -),( 1 1 1 1

    1 1

    1,11,11jjjjnj

    nj

    njnj qrqlFqrqlFtxx

    tqqtx −∆∆

    −= +++ φφ

    { ( ) ( )

    ( ) ( )( )

    ( ) ( )( ) }

    , , ),(

    , , ),(

    -

    ),( -

    ),(

    2 2 2 1

    2 12

    01

    1 1 1 1

    1 11

    0

    ,21,22

    01

    ,11,11

    0

    xqrqlFqrqlF

    tx

    xqrqlFqrqlF

    tx

    tqq

    txtqq

    txxt

    jjjjnj

    nj

    jjjjnj

    n

    np

    j

    nj

    nj

    njnj

    nj

    nj

    njn

    np

    j

    −−

    −−

    =∆

    +∆

    ∆∆

    ++∞

    =

    =

    ++∞

    =−∞=

    +∞

    =

    =

    +∞

    =−∞=

    ∑∑

    ∑∑

    ∑∑∑∑

    φ

    φ

    φφ

    Generalized Lax-Wendroff Theorem

    • In order to demonstrate that these integrals are satisfied for the proposed method, first multiply the cell average update equation by for mesh 1, etc.

    • Next sum over j and n for each mesh as follows

    ( )nj tx ,1φ

    2016 Overset Grid Syposium 17

  • Generalized Lax-Wendroff Theorem

    18

    • The following summation by parts formula is useful for the proving convergence of the proposed scheme to a weak solution

    ( ) ( ) SPB 1

    10010

    1 ∑∑=

    −+=

    + −−−=−N

    jjjjNN

    N

    jjjj aabbababba

    2016 Overset Grid Syposium

  • Generalized Lax-Wendroff Theorem

    19

    • For the LHS use the SPB formula on the n sum

    ),(-),(

    ),(-),(

    ),( - ),(

    1

    12 2

    ,211 1

    ,1

    1

    1

    0,2 0

    2 0,1 0

    1

    ∆∆+

    ∆∆∆−

    ∆∆−

    ∑∑∑

    ∑∑∞

    =

    −−

    −∞=

    =

    =−∞=

    j

    njnjnj

    njnjnp

    j

    nj

    n

    jjj

    np

    jjj

    ttxtx

    qxt

    txtxqxt

    qtxxqtxx

    φφφφ

    φφ

    • Note that these terms are discrete approximations to the following integrals

    dxdtqqdxxqxdxxqx tt

    +−−− ∫ ∫∫∫ ∫

    ++

    ∞−

    ∞∞

    ∞−

    ∞1

    1np

    21

    11np

    21

    xf

    xf0

    xf

    xf

    )0,()0,()0,()0,( φφφφ

    2016 Overset Grid Syposium

  • ( ) ( )

    ( ) ( ) ),(, ),(,

    ),(),(,

    ),(),(, t

    0

    212

    12

    1

    11

    11

    1

    2

    21

    22 2

    11

    11 1

    0

    ∑ ∑∑∞

    =++

    −∞=

    =

    −−∞

    =

    ∆−

    ∆∆∆−

    −∆+

    −∆∆

    n

    nnnpnpnp

    np

    j j

    njnjjj

    njnjjj

    n

    xtxqrqlF

    xtx

    qrqlFxt

    xtxtx

    qrqlFxx

    txtxqrqlFx

    φφ

    φφφφ

    Generalized Lax-Wendroff Theorem

    20

    • For the RHS use the SPB formula on the j sum

    End Terms from SBP Formula

    2016 Overset Grid Syposium

  • Generalized Lax-Wendroff Theorem

    21

    • The end terms at the grid interface become

    ( ) ( )

    ( )

    ( ) ( )1 11 12 12 112

    0

    112

    1212

    12

    1

    0

    212

    12

    1

    11

    11

    1

    , , where

    ),(),(),( ,

    ),( , ),(

    ,

    ++

    =

    =++

    −=∆

    ∆+

    −∆∆

    =

    ∆−

    ∆∆∆−

    npnp

    n

    nnpnnpn

    n

    nnnpnpnp

    qrqlFqrqlFF

    xtxF

    xtxtx

    qrqlFxt

    xtxqrqlF

    xtx

    qrqlFxt

    φφφ

    φφ

    2016 Overset Grid Syposium

  • Generalized Lax-Wendroff Theorem

    22

    • We can show that ΔF12 → 0 as Δx → 0 as follows

    ),()( ),( 1112 nnpn

    nnp txxqTVCtxF φφ ∆≤∆

    0 as 0 ⇒∆⇒ x

    ( ) ( ) etc. ,)( 211 121211 1212 1 1 1 xqTVxfxfqxfxfqqrqr npnpnp ∆≤−∇≤−•∇≅− +++

    ( ) ( )( ) ( )

    ) , max(

    )( ),( ,

    , ,

    2 1

    1 1

    2 1

    1 112

    2 1

    1 1

    2 1

    2 1

    1 1

    2 1

    2 1

    2 112

    1 1

    1 1

    2 1

    2 112

    qrqrqlqlCF

    qrqrqrqlqlqlFqrqlFF

    qrqlFqrqlFF

    npnp

    npnp

    npnp

    −−≤∆

    −+−+−=∆

    −=∆

    ++

    ++

    ++

    • Given that

    it follows that

    F is Lipschitz Continuous

    2016 Overset Grid Syposium

  • Generalized Lax-Wendroff Theorem

    23

    • Note that these terms are discrete approximations to the following integrals

    dxdtqfdxdtqf xxf

    xf

    x

    np

    φφ ∫∫∫∫∞∞

    ∞−

    ++

    21

    11

    )()(00

    • If then terms like 0 as 0211 1 →∆→−+ xxfxfnp

    dxdtqqdxxqxdxxqx tt

    +−−− ∫ ∫∫∫ ∫

    ++

    ∞−

    ∞∞

    ∞−

    ∞1

    1np

    21

    11np

    21

    xf

    xf0

    xf

    xf

    )0,()0,()0,()0,( φφφφ

    dxdtqdxxqx t∫∫∫∞

    ∞−

    ∞∞

    ∞−

    −− φφ0

    )0,()0,( become

    and the integral conservation law is satisfied under the same caveats invoked for the single grid case

    2016 Overset Grid Syposium

  • Observations Based on Lax-Wendroff Theorem (1)

    24

    • If standard linear interpolation methods are used then no GENERAL bounds may be established a priori for the ΔF12 term

    • We end up with a source term in the integral conservation law which MAY not vanish under grid refinement in the presence of discontinuous solutions.

    • These source terms become ODEs along characteristics and may globally pollute the solution Corrupt convergence rates Generate spurious waves

    ( ) ( )dttxtF R ,10 12 ϕ∫∞∆≈

    2016 Overset Grid Syposium

  • Observations Based on Lax-Wendroff Theorem (2)

    25

    •These terms can and will corrupt solutions when: • Severe mismatches in cell sizes exist • Strong discontinuities exist in the grid interface region • Slow moving discontinuities exist in the grid interface region • All three exist simultaneously!

    2016 Overset Grid Syposium

  • •Shock capturing schemes needed to control TV(q)

    • Follow Zang and Shu.1

    • Let 𝑞 𝐾𝑁 be the solution points in cell Ω𝑁 •Use a discontinuity detector θ based on cell average values to control oscillations as follows: • Here we use Nichols discontinuity dector2 • One can also use the TVB filters of Engquist, et al3

    FR Shock Capturing Considerations

    26 2016 Overset Grid Syposium

    𝑞 𝐾𝑁 → 𝜃 𝑞 𝐾𝑁 − 𝑞� + 𝑞�

    1. Zhang, Xiangxiong, and Chi-Wang Shu. "On maximum-principle-satisfying high order schemes for scalar conservation laws." Journal of Computational Physics 229.9 (2010): 3091-3120.

    2. Tramel, Robert W., Robert H. Nichols, and Pieter G. Buning. "Addition of improved shock-capturing schemes to OVERFLOW 2.1." AIAA Paper 3988 (2009): 2009. 3. Engquist, Björn, Per Lötstedt, and Björn Sjögreen. "Nonlinear filters for efficient shock computation." Mathematics of Computation 52.186 (1989): 509-537.

    𝑞 𝐾𝑁 → 𝜃 𝑞 𝐾𝑁 − 𝑞� 𝐾𝑁 − 𝑞� 𝐾𝑁

  • Sample Cases

    2016 Overset Grid Syposium 27

    • Shock Wave Formation in Burgers Equation • Shock Propagation in Burgers Equation • K=5 • Cells overlap 1/8 ΔX • Shock location lies in the overset region • Use global Lax-Friedrichs flux • Use Gauss points and weights and Radau

    correction polynomials • “Exact” solution for shock formation is a 4000 point

    solution using a 5th order WENO-RBF method

    𝑞 𝑥, 0 = sin (𝑥)

    𝑞 𝑥, 0 = H (− 𝑥 − 𝑥𝑜 )

  • No Shock Treatment

    2016 Overset Grid Syposium 28

    Internal Cell Variables Cell Averages

  • No Shock Treatment

    2016 Overset Grid Syposium 29

    Comparison With Exact Solution

  • Dissipation Switch

    2016 Overset Grid Syposium 30

    Internal Cell Variables Cell Averages

  • Dissipation Switch

    2016 Overset Grid Syposium 31

    Comparison With Exact Solution

  • Discontinuity Detector

    2016 Overset Grid Syposium 32

    Cell Number

    θ

  • TVB Filter

    2016 Overset Grid Syposium 33

    Internal Cell Variables Cell Averages

  • TVB Filter

    2016 Overset Grid Syposium 34

    Comparison With Exact Solution

  • No Shock Treatment

    2016 Overset Grid Syposium 35

    Internal Cell Variables Cell Averages

  • Dissipation Switch

    2016 Overset Grid Syposium 36

    Internal Cell Variables Cell Averages

  • TVB Filter

    2016 Overset Grid Syposium 37

    Internal Cell Variables Cell Averages

  • Discontinuity Detector

    2016 Overset Grid Syposium 38

    Cell Number

    θ

  • Conclusions

    39

    • Within the context of this framework strict lack of conservation becomes merely another source of numerical error to be removed by grid refinement

    • Convergence to a weak solution is guaranteed as grid independence is achieved

    • MUSCL/WENO/RBF schemes can be mixed with Flux Reconstruction schemes

    • Framework seamlessly blends block matching/overset grid grids

    2016 Overset Grid Syposium

  • ACKNOWLEDGEMENTS

    40

    • Thanks to • Dr. Randy Leveque (U Wash) • Dr. John Benek (AFRL/RBAC)

    for discussions and careful reading of earlier versions of this presentation

    2016 Overset Grid Syposium

  • That’s All Folks

    41

    • Questions • Constructive Criticism

    2016 Overset Grid Syposium

  • That’s All Folks

    42

    • Questions • Constructive Criticism

    2016 Overset Grid Syposium

    Lax-Wendroff Convergence Theorems for a Class of Overset Mesh Methods Based on Higher Order Flux ReconstructionConservation Law Based Update SchemesConservation Law Based Update SchemesMUSCL ReconstructionFlux Reconstruction1,2Flux ReconstructionFlux ReconstructionFlux ReconstructionConservation Law Based Update SchemesConservation Law Based Update SchemesConservation Law Based Update SchemesConservation Law Based Update SchemesData Exchange for Flux ReconstructionData Exchange for Flux ReconstructionConservation Law Based Update SchemesGeneralized Lax-Wendroff TheoremGeneralized Lax-Wendroff TheoremGeneralized Lax-Wendroff TheoremGeneralized Lax-Wendroff TheoremGeneralized Lax-Wendroff TheoremGeneralized Lax-Wendroff TheoremGeneralized Lax-Wendroff TheoremGeneralized Lax-Wendroff TheoremObservations Based on Lax-Wendroff Theorem (1)Observations Based on Lax-Wendroff Theorem (2)FR Shock Capturing ConsiderationsSample CasesNo Shock TreatmentNo Shock TreatmentDissipation SwitchDissipation SwitchDiscontinuity DetectorTVB FilterTVB FilterNo Shock TreatmentDissipation SwitchTVB FilterDiscontinuity DetectorConclusionsACKNOWLEDGEMENTSThat’s All FolksThat’s All Folks