Download - Vander Vegt
-
8/8/2019 Vander Vegt
1/82
-
8/8/2019 Vander Vegt
2/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 1
Introduction
Motivation of research:
Many important flow phenomena are modelled by nonconservative (hyperbolic)partial differential equations, e.g. dispersed multiphase flows.
Our aim is to develop space-time discontinuous Galerkin discretizations which aresuitable for both conservative and nonconservative partial differential equations
-
8/8/2019 Vander Vegt
3/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 2
Introduction
Motivation of research:
Nonconservative hyperbolic partial differential equations contain nonconservativeproducts
x
u + A(u)x
u = 0
The essential feature of nonconservative products is that A = Df, hence A isnot the Jacobian matrix of a flux function f.
This causes problems once the solution becomes discontinuous, because the weaksolution in the classical sense of distributions then does not exist.
This also complicates the derivation of discontinuous Galerkin discretizations sincethere is no direct link with a Riemann problem.
-
8/8/2019 Vander Vegt
4/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 3
Alternative: use the theory for nonconservative products from Dal Maso, LeFlochand Murat (DLM)
-
8/8/2019 Vander Vegt
5/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 4
Overview of Presentation
Overview of main results of the theory of Dal Maso, LeFloch and Murat fornonconservative products
Space-time DG discretization of nonconservative hyperbolic partial differentialequations
Extension to the compressible Navier-Stokes equations
Multigrid techniques for space-time DG discretizations
Conclusions
-
8/8/2019 Vander Vegt
6/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 5
Nonconservative Products
Consider the function u(x)
u(x) = uL + H(x xd)(uR uL), x, xd ]a, b[,
with H : R
R the Heaviside function.
For any smooth function g : Rm Rm the product g(u)xu is not defined atx = xd since here |xu| .
Introduce a smooth regularization u of u. If the total variation of u remainsuniformly bounded with respect to then Dal Maso, LeFloch and Murat (DLM)
showed that
g(u)dudx lim
0 gududxgives a sense to the nonconservative product as a bounded measure.
-
8/8/2019 Vander Vegt
7/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 6
Effect of Path on Nonconservative Product
The limit of the regularized nonconservative product depends in general on the path
used in the regularization.
Introduce a Lipschitz continuous path : [0, 1] Rm, satisfying (0) = uLand (1) = uR, connecting uL and uR in R
m.
The following regularization u for u then emerges:
u(x) =
uL, if x ]a, xd [,(
xxd+2 ), if x ]xd , xd + [, > 0.
uR, if x ]xd + , b[
-
8/8/2019 Vander Vegt
8/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 7
When tends to zero, then:
g(u)du
dx Cxd, with C =
10
g(())d
d() d,
weakly in the sense of measures on ]a, b[, where xd is the Dirac measure at xd.
The limit of g(u)xu depends on the path .
There is one exception, namely if an q : Rm R exists with g = uq. In thiscase C = q(uR) q(uL).
-
8/8/2019 Vander Vegt
9/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 8
DLM Theory
Dal Maso, LeFloch and Murat provided a general theory for nonconservative
hyperbolic pdes.
Introduce the Lipschitz continuous maps : [0, 1]
R
m
R
m
R
m which
satisfy the following properties:
(H1) (0; uL, uR) = uL, (1; uL, uR) = uR,
(H2) (; uL, uL) = uL,
(H3)
(; uL, uR)
K|uL uR|, a.e. in [0, 1].
-
8/8/2019 Vander Vegt
10/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 9
Theorem (DLM). Let u :]a, b[ Rm be a function of bounded variation andg : Rm Rm a continuous function. Then, there exists a unique real-valuedbounded Borel measure on ]a, b[ with:
1. If u is continuous on a Borel set B ]a, b[, then
(B) = B g(u)du
dx
2. If u is discontinuous at a point xd of ]a, b[, then
({xd}) =
10
g((; uL, uR))
(; uL, uR) d.
By definition, this measure is the nonconservative product of g(u) by xu anddenoted by =
g(u)dudx
.
-
8/8/2019 Vander Vegt
11/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 10
Rankine-Hugoniot Relations
For conservative hyperbolic system of pdes, xu + xf(u) = 0 theRankine-Hugoniot relations across a jump with uL and uR and velocity v are
equal to
v(uR uL) + f(uR) f(uL) = 0.
For a nonconservative hyperbolic pde xu + A(u)xu = 0 the Rankine-Hugoniotrelations in the DLM theory are equal to
v(uR uL) +1
0
A(D(s, uL
, uR
))sD(s; uL
, uR
)ds = 0
with D a Lipschitz continuous path satisfying D(0; uL, uR) = uL andD(1; uL, uR) = uR.
The Rankine-Hugoniot relations are essential for the definition of the NCP fluxused in the DG discretization.
-
8/8/2019 Vander Vegt
12/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 11
Space-Time Discontinuous Galerkin Method
I nt n+1
tn
t0
Qn
n
n+1
y
x
t
K
A time-dependent problem is considered directly in four dimensional space, withtime as the fourth dimension
-
8/8/2019 Vander Vegt
13/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 12
Key Features of Space-Time Discontinuous Galerkin Methods
Simultaneous discretization in space and time: time is considered as a fourthdimension.
Discontinuous basis functions, both in space and time, with only a weak couplingacross element faces resulting in an extremely local, element based discretization.
The space-time DG method is closely related to the Arbitrary Lagrangian Eulerian(ALE) method.
-
8/8/2019 Vander Vegt
14/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 13
Benefits of Discontinuous Galerkin Methods
Due to the extremely local discretization DG methods provide optimal flexibility for
achieving higher order accuracy on unstructured meshes
hp-mesh adaptation
unstructured meshes containing different types of elements, such as tetrahedra,
hexahedra and prisms
parallel computing
-
8/8/2019 Vander Vegt
15/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 14
Benefits of Space-Time Discontinuous Galerkin Methods
A conservative discretization is obtained on moving and deforming meshes.
No data interpolation or extrapolation is necessary on dynamic meshes, at freeboundaries and after mesh adaptation.
-
8/8/2019 Vander Vegt
16/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 15
Disadvantages of Space-(Time) Discontinuous Galerkin Methods
Algorithms are generally rather complicated, in particular for elliptic and parabolicpartial differential equations
On structured meshes DG methods are computationally more expensive than finitedifference and finite volume methods.
-
8/8/2019 Vander Vegt
17/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 16
Space-Time Domain
Consider an open domain: E Rd. The flow domain (t) at time t is defined as:
(t) := {x E | x0 = t, t0 < t < T}
The space-time domain boundary E consists of the hypersurfaces:
(t0) :={x E | x0 = t0},(T) :={x E | x0 = T},
Q :={x E | t0 < x0 < T}.
The space-time domain is covered with a tessellation Th consisting of space-timeelements K.
-
8/8/2019 Vander Vegt
18/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 17
Discontinuous Finite Element Approximation
The finite element space associated with the tessellation Th is given by:
Wh :=
W (L2(Eh))m : W|K GK (Pk(K))m, K Th
The jump of f at an internal face S SnI in the direction k of a Cartesiancoordinate system is defined as:
[[f]]k = fLnLk + f
RnRk ,
with nRk = nLk .
The average of f at S Sn
I is defined as:
{{f}} = 12(fL
+ fR
).
-
8/8/2019 Vander Vegt
19/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 18
Space-Time DG Formulation of Nonconservative HyperbolicPDEs
Consider the nonlinear hyperbolic system of partial differential equations innonconservative form in multi-dimensions:
Ui
t +
Fik
xk + GikrUr
xk = 0, x Rq
, t > 0,
with U Rm, F Rm Rq, G Rm Rq Rm
These equations model for instance bubbly flows, granular flows, shallow waterequations and many other physical systems.
-
8/8/2019 Vander Vegt
20/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 19
Weak formulation for nonconservative hyperbolic system:
Find a U Vh, such that for V Vh the following relation is satisfiedKTh
K
Vi
Ui,0 + Fik,k + GikrUr,k
dK
+
KTh
K(tn+1)
Vi(URi ULi ) dK K(t+n )
Vi(URi ULi ) dK+
SSI
S
Vi 10
Gikr((; UL
, UR
))r
(; U
L, U
R) d n
Lk
dS
SSI
SVi[[Fik vkUi]]k dS = 0
-
8/8/2019 Vander Vegt
21/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 20
Relation with Space-Time DG Formulation of ConservativeHyperbolic PDEs
Theorem 2. If the numerical flux V for the test function V is defined as:
V = {{V}} at S SI,
0 at K(tn) h(tn) n 0,then the DG formulation will reduce to the conservative space-time DG formulation
when there exists a Q, such that Gikr = Qik/Ur.
-
8/8/2019 Vander Vegt
22/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 21
After the introduction of the numerical flux V we obtain the weak formulation:KTh
K
Vi,0Ui Vi,kFik + ViGikrUr,k dK+
KTh
K(tn+1)
VLi ULi dK
K(t+n )
VLi ULi dK
+
SSI
S
(VL
i VRi ){{Fik vkUi}}nLk dS
+
SSB
S
VLi (FL
ik vkULi )nLk dS
+ SSI
S{{Vi}} 1
0 Gikr((; U
L
, UR
))
r
(; UL
, UR
) d nLk dS = 0
-
8/8/2019 Vander Vegt
23/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 22
Numerical Fluxes
The fluxes at the element faces do not contain any stabilizing terms yet, both forthe conservative and nonconservative part
At the time faces, the numerical flux is selected such that causality in time isensured U = UL at K(tn+1)
UR at K(t+n ).
The space-time DG formulation is stabilized using the NCP (Non-ConservativeProduct) flux
Pnc
i = {{Fik vkUi}} + PiknLk
-
8/8/2019 Vander Vegt
24/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 23
Nonconservative Product Flux
E
T
dd
dd
dd
dd
dd
dd
ddd
xL xR
x
t
SL SRSM
T
1
(U= UL)
4
(U= UR)
2
(U= UL)
3
(U= UR)
Wave pattern of the solution for the Riemann problem
-
8/8/2019 Vander Vegt
25/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 24
NCP Flux
Main steps in derivation of NCP flux:
Consider the nonconservative hyperbolic system:
tU + xF(U) + G(U)xU = 0,
Introduce the averaged exact solution ULR(T) as:
ULR(T) =
1
T(SR SL)T SR
T SL
U(x, T) dx.
Apply the Gauss theorem over each subdomain 1, ,4 and connect eachsubdomain using the generalized Rankine-Hugoniot relations.
-
8/8/2019 Vander Vegt
26/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 25
The NCP-flux is then given by:
Pnci (UL, UR, v, n
L) =
FLiknLk 12
10 Gikr((; UL, UR))
r
(; UL, UR) dnLk
if SL > v,
{{Fik}}nLk +
12
(SR v)Ui + (SL v)Ui SLULi SRURi )
if SL < v < SR,
FRiknLk + 12 10 Gikr((; UL, UR))r (; UL, UR) dnLkif SR < v,
Note, if G is the Jacobian of some flux function Q, then Pnc(UL, UR, v, nL) isexactly the HLL flux derived for moving grids in van der Vegt and van der Ven
(2002).
-
8/8/2019 Vander Vegt
27/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 26
Efficient Solution of Nonlinear Algebraic System
The space-time DG discretization results in a large system of nonlinear algebraicequations:
L(Un; Un1) = 0
This system is solved by marching to steady state using pseudo-time integration
and multigrid techniques:
U
= 1
tL(U; Un1)
-
8/8/2019 Vander Vegt
28/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 27
Depth averaged two-fluid model
The dimensionless depth-averaged two fluid model of Pitman and Le, ignoringsource terms for simplicity, can be written as:
tU + xF + GxU = 0,
where:
U =
h(1 )
hhv
hu(1 )b
, F =
h(1 )uhv
hv2 + 12(1 )xxgh2
hu2 + 12gh2
0
G(U) =
0 0 0 0 00 0 0 0 0
gh gh 0 0 (1)xxgh + gh
2u21 u
2 gh gh u2 u( 1) u 2u1 (1 )gh0 0 0 0 0
.
-
8/8/2019 Vander Vegt
29/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 28
0 5 10 15 200
0.5
1
1.5
2
x
h
(1)+b,h
+b,h
+b,b
h+b
h(1)+b
h+b
b
Steady-state solution for a subcritical two-phase flow (320 cells).
Total flow height h + b, flow height due to the fluid phase h(1 ), flow height dueto solids phase h and the topography b.
-
8/8/2019 Vander Vegt
30/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 29
STDGFEM
h(1) + b h + b
Ncells L2 error p Lmax error p L2 error p Lmax error p
40 0.8171 103 - 0.2308 102 - 0.1404 102 - 0.4194 102 -80 0.2025 103 2.0 0.5584 103 2.0 0.3537 103 2.0 0.9903 103 2.1
160 0.4871 104 2.1 0.1322 103 2.1 0.8511 104 2.1 0.2306 103 2.1320 0.9789 105 2.3 0.2651 104 2.3 0.1712 104 2.3 0.4597 104 2.3
hu(1 ) hv()
Ncells L2 error p Lmax error p L2 error p Lmax error p
40 0.3672 104 - 0.1442 103 - 0.1212 104 - 0.3409 104 -80 0.5911 105 2.6 0.3448 104 2.1 0.1791 105 2.8 0.8054 105 2.1
160 0.1049 105 2.5 0.8471 105 2.0 0.3807 106 2.2 0.2048 105 2.0320 0.1723 106 2.6 0.2078 105 2.0 0.5115 107 2.9 0.4861 106 2.1
Error in h(1) + b, h+ b, hu(1) and hv for subcritical flow over a bump.
-
8/8/2019 Vander Vegt
31/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 30
Two-phase dam break problem
0 0.5 10
0.5
1
1.5
2
2.5
3
3.5
x
h(1),h,h
h(1) 10000
h 10000
h 10000
h(1) 128
h 128
h 128
(a) Solution of h(1 ), h, band h.
0 0.5 10.5
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
x
hu(1),hv
hu(1) 10000
hv 10000
hu(1) 128
hv 128
(b) Solution ofhu(1) and hv.
0 0.2 0.4 0.6 0.8 10
0.2
0.4
0.6
0.8
1
x
10000
128
(c) Solution of.
Two-phase dam break problem at time t = 0.175; mesh with 128 elements comparedto mesh with 10000 elements.
-
8/8/2019 Vander Vegt
32/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 31
Effect of Path
0 0.2 0.4 0.6 0.8 10.5
0
0.5
1
1.5
2
2.5
3
3.5
x
h(1),h,b,h
h(1) 1
h 1
h 1
h(1) t1
h t1
h t1
h(1) t2
h t2
h t2
h(1) t3
h t3
h t3
h(1) t4
h t4
h t4
h(1) t5
h t5
h t5
b
(d) The solution on the whole domain.
0.095 0.1 0.105 0.11
1.5
2
2.5
3
x
h
(1),h,h
h(1) 1
h 1
h 1
h(1) t1
h t1
h t1
h(1) t2
h t2
h t2
h(1) t3
h t3
h t3h(1) t4
h t4
h t4
h(1) t5
h t5
h t5
(e) The solution zoomed in on the left shock wave.
Solution of h(1
), h, b and h at time t = 0.175 calculated on a mesh with
1024 elements using the paths defined by Toumi.
-
8/8/2019 Vander Vegt
33/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 32
Flow Through a Contraction
x1
0
5
10
15
x2
-0.5
0
0.5
h1
Y
X
Z
h: 0.4 0 .5 0.6 0 .7 0.8 0 .9 1 1.1 1 .2 1.3 1 .4 1.5
Flow depth h of water-sand mixture in a contraction
h/L = 0.01, f/s = 0.5, slope 10.
-
8/8/2019 Vander Vegt
34/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 33
Compressible Navier-Stokes Equations
Compressible Navier-Stokes equations in space-time domain E:
Ui
x0+Fek (U)
xk F
ek (U,U)xk
= 0
Conservative variables U R5 and inviscid fluxes Fe R53
U =
ujE
, Fek = ukujuk +pjk
huk
-
8/8/2019 Vander Vegt
35/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 34
Compressible Navier-Stokes Equations
Viscous flux Fv R53
Fv
k =
0jkkj uj qk
with the total stress tensor is defined as:
jk = u
ixi
jk + (u
jxk
+u
kxj
)
and heat flux vector q is defined as:
qk = T
xk
-
8/8/2019 Vander Vegt
36/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 35
Compressible Navier-Stokes Equations
The viscous flux Fv is homogeneous with respect to the gradient of theconservative variables U:
Fvik(U,U) = Aikrs(U)Ur
xs
with the homogeneity tensor A R
5
3
5
3
defined as:
Aikrs(U) :=Fvik(U,U)
(U)
The system is closed using the equations of state for an ideal gas.
-
8/8/2019 Vander Vegt
37/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 36
First Order System
Rewrite the compressible Navier-Stokes equations as a first-order system using theauxiliary variable :
Ui
x0+Feik(U)
xk ik(U)
xk= 0,
ik(U) Aikrs(U)Urxs
= 0.
-
8/8/2019 Vander Vegt
38/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 37
Weak Formulation
Weak formulation for the compressible Navier-Stokes equations
Find a U Wh, Vh, such that for all W Wh and V Vh, the followingholds:
KTh
K
Wi
x0Ui +
Wi
xk(Feik ik)
dK
+
KTh
K
WL
i (Ui + Feik ik)nLk d(K) = 0,
KTn
h
K
Vikik dK =
KTnh
K
VikAikrsUr
xsdK
+ KTn
h
Q
VLik ALikrs(Ur ULr )nLs dQ
-
8/8/2019 Vander Vegt
39/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 38
Transformation to Arbitrary Lagrangian Eulerian form
The space-time normal vector on a grid moving with velocity v is:
n =
(1, 0, 0, 0)T at K(tn+1),
(1, 0, 0, 0)T at K(t+n ),(vknk, n)T at Qn.
The boundary integral then transforms into:KTh
K
WL
i (Ui + Feik ik)nLk d(K)
=
KTh
K(tn+1)
WL
iUi dK +
K(t+n )
WL
iUi dK
+ KTh
Q
WLi (Feik Uivk ik)nLk dQ
-
8/8/2019 Vander Vegt
40/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 39
Numerical Fluxes
The numerical flux U at K(tn+1) and K(t+n ) is defined as an upwind flux toensure causality in time:
U = UL at K(tn+1),
UR at K(t+n
),
At the space-time faces Q we introduce the HLLC approximate Riemann solver asnumerical flux:
nk(
Feik
Uivk)(UL, UR) = H
HLLCi (U
L, UR, v, n)
-
8/8/2019 Vander Vegt
41/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 40
ALE Weak Formulation
The ALE flux formulation of the compressible Navier-Stokes equations transformsnow into:
Find a U Wh, such that for all W Wh, the following holds:
KTn
h
KWix0
Ui + Wixk
(Feik ik) dK+
KTnh
K(tn+1)
WL
i ULi dK
K(t+n )
WL
i URi dK
+ KTnhQ WLi (HHLLCi (UL, UR, v, n) iknLk ) dQ = 0.
-
8/8/2019 Vander Vegt
42/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 41
Lifting Operator
Introduce the global lifting operator R R53, defined in a weak sense as:Find an R Vh, such that for all V Vh:
KTnh
K
VikRik dK =
SSnI
S
{{VikAikrs}}[[Ur]]s dS
+ SSn
B
S
VLik ALikrs(ULr Ubr )nLs dS.
The weak formulation for the auxiliary variable is now transformed into
KTnh KVikik dK = KTnh K
Vik(AikrsUr
xsRik) dK, V Vh.
-
8/8/2019 Vander Vegt
43/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 42
Equation
The primal formulation can be obtained by eliminating the auxiliary variable using
ik = AikrsUr
xsRik, a.e. in Enh
Note, this is possible since hWh Vh
-
8/8/2019 Vander Vegt
44/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 43
ALE Weak Formulation for Primal Variables
Recall the ALE flux formulation of the compressible Navier-Stokes equations:
Find a U Wh, such that for all W Wh, the following holds:
KTnh K Wi
x0Ui +
Wi
xk(F
eik
ik) dK
+
KTnh
K(tn+1)
WLi ULi dK
K(t+n )
WLi URi dK
+
KTnh
Q
WLi (HHLLCi (U
L, UR, v, n)
ikn
Lk ) dQ = 0.
-
8/8/2019 Vander Vegt
45/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 44
Numerical Fluxes for
The numerical flux in the primary equation is defined following Brezzi as acentral flux = {{}}:
ik(U
L, UR) =
{{Aikrs
Urxs
RSik}} for internal faces,Abikrs
Ubrxs
RSik for boundary faces,
The local lifting operator RS R53 is defined as follows:Find an RS Vh, such that for all V Vh:
KTnh K VikRSik dK =
S
{{VikAikrs}}[[Ur]]s dS for internal faces,
S
VLik ALikrs(U
Lr Ubr )ns dS for external faces.
-
8/8/2019 Vander Vegt
46/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 45
Delta Wing Simulations
Simulations of viscous flow about a delta wing with 85 sweep angle.
Conditions
Mach number M = 0.3
Reynolds number Re = 40.000 Angle of attack = 12.5.
Fine grid mesh 1.600.000 elements, 40.000.000 degrees of freedom
Adapted mesh, initial mesh 208.896 elements, after four adaptations 286.416
elements
-
8/8/2019 Vander Vegt
47/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 46
Delta Wing Simulations
Streaklines and vorticity contours in various cross-sections
-
8/8/2019 Vander Vegt
48/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 47
Delta Wing Simulations
Impression of the vorticity based mesh adaptation
-
8/8/2019 Vander Vegt
49/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 48
Delta Wing Simulations
Large eddy simulation of turbulent flow about a delta wing
-
8/8/2019 Vander Vegt
50/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 49
Computational Efficiency
Computational efficiency is the key factor limiting industrial applications of higherorder accurate discontinuous Galerkin methods in computational fluid dynamics.
Multigrid methods are good candidates to increase computational efficiency, butneed significant improvements for higher order accurate DG discretizations.
-
8/8/2019 Vander Vegt
51/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 50
Objectives
Perform a theoretical analysis of multigrid performance for advection dominatedflows, in particular for higher order accurate DG discretizations.
Improve multigrid performance using theoretical analysis tools.
Test multigrid performance on realistic problems.
-
8/8/2019 Vander Vegt
52/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 51
Main Components of h-Multigrid Algorithm
Consider a finite sequence Nc of increasingly coarser meshes Mnh,n {1, , Nc}
Define operators to connect data on the different meshes:
restriction operators
Rmhnh : Mnh Mmh, 1 n < m Nc,
prolongation operators
Pnh
mh : Mmh Mnh, 1 n < m Nc.
-
8/8/2019 Vander Vegt
53/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 52
Use iterative solvers Snh to approximately solve the system of algebraic equationson the various grid levels
Lnhvh = fnh on Mnh
Since, the main eff
ect of the multigrid algorithm should be the damping of highfrequency error components Snh is called a smoothing operator.
Choose a cycling strategy between the different meshes, e.g. V- or W-cycle.
-
8/8/2019 Vander Vegt
54/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 53
Multigrid error transformation operator for linear problems
In order to understand the performance of the multigrid algorithm we need toinvestigate the multigrid error transformation operator.
The multigrid error transformation operator shows how much the error in theiterative solution of the algebraic system is reduced by one full multigrid cycle.
We analyze three-level multigrid algorithms for 2D problems to obtain betterestimates for the convergence rate.
-
8/8/2019 Vander Vegt
55/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 54
Given an initial error eAh , the error eDh after one full multigrid cycle with three gridlevels is given by the relation
eDh = M
3gh e
Ah
with
M3gh = S2h (Ih Ph2h(I2h Mc2h)L12h R2hh Lh)S1hand
M2h = S42h(I2h P2h4h L14h R4h2hL2h)S
32h.
The properties of the multigrid error transformation operator are analyzed usingdiscrete Fourier analysis.
-
8/8/2019 Vander Vegt
56/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 55
Three-grid Fourier analysis
/2 /2 0
0
/2
/2
2
1
/4
/4
/4
/4
Low, medium and high frequencies Fourier modes in three-level multigrid
-
8/8/2019 Vander Vegt
57/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 56
Multigrid Error Transformation Operator
The discrete Fourier transform of the error transformation operator for a three-levelmultigrid cycle M3gh () C16m16m which is equal to
M3gh () =
S3gh ()
2
I3g
P3gh ()
U3g(; c)
Q3gh ()
R3gh ()
L3gh ()
S3gh ()
1
4h \3g
The matrix U3g(; c) is equal toU3g(; c) = I2g M2g2h(2)c.
with
M2g2h(2) = S2g2h(2)4I2gP2g2h (2)L14h (400)R2g2h(2)L2g2h(2)S2g2h(2)3
-
8/8/2019 Vander Vegt
58/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 57
Asymptotic Convergence Rate
The asymptotic convergence factor per cycle is defined as
= limm
sup
e(0)
h =0
e(m)h 2(Gh)
e
(0)h
2(G
h)
1m
and can be rewritten into
= (Mngh ).
-
8/8/2019 Vander Vegt
59/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 58
Optimization of multigrid performance
The spectral radius of the error transformation operator can be used to optimizethe multigrid algorithm.
In particular, the smoother is a good candidate for optimization since in general itcontains a number of free parameters.
We will consider a Runge-Kutta type smoother.
-
8/8/2019 Vander Vegt
60/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 59
Analysis and Optimization of Runge-Kutta Smoothers for theAdvection-Diffusion Equation
The advection-diffusion equation in the domain is defined as
u
t+ a u = (Au))
The equations are discretized with a higher order accurate space-timediscontinuous Galerkin finite element method.
The space-time discretizations are solved using a multigrid algorithm with aRunge-Kutta type smoother.
-
8/8/2019 Vander Vegt
61/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 60
Pseudo-time Runge-Kutta smoothers
Let the system of algebraic equations be denoted as
L(un, un1) = 0.
A pseudo time derivative is added to the system and integrated to steady-state in
pseudo-time
Mu
= L(u, un1),
At steady state un = u.
For the pseudo-time integration we consider 4- and 5-stage Runge-Kutta methods.
-
8/8/2019 Vander Vegt
62/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 61
A four-stage Runge-Kutta scheme is given by:
(1 + 1I)V1
=V0 1
xy21M
1L(V0) + 1V0
(1 + 2I)V2
=V0 M
1
xy
31L(V
0) + 32L(V
1)
+ 2V1
(1 + 3I)V3 =V0 M
1
xy41L(V0) + 42L(V1) + 43L(V2) + 3V2
(1 + 4I)V4
=V0 M
1
xy
51L(V
0) + 52L(V
1) + 53L(V
2) + 54L(V
3)
+4V3
We require from the 4- and 5-stage Runge-Kutta schemes that they are secondorder accurate
-
8/8/2019 Vander Vegt
63/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 62
Optimized Runge-Kutta smoothers for space-time DGFEM
Optimization of the smoother coefficients for three-level multigrid algorithms
Space-time DG discretization of the 2D advection-diffusion equation usingquadratic basis-functions
For the optimization process, we fix the Reynolds numbers Rex = Rey = 100,the flow angle flow = /4 and the aspect ratio AR = 1.
For steady flows we fix CF Lt = 100, while for unsteady flows CF Lt = 1.
-
8/8/2019 Vander Vegt
64/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 63
The optimization process has a big impact on the Runge-Kutta coefficients whichgreatly differ per case.
The Runge-Kutta coefficients are very different from commonly used schemes sincethey are optimized for fast multigrid convergence and not for time accuracy.
Finding a good balance between optimization and more generally applicablealgorithms is still an open question.
In practice, the Runge-Kutta smoothers use local time stepping and this gives theopportunity to apply locally the best smoother for the actual flow state.
-
8/8/2019 Vander Vegt
65/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 64
Optimized coeffi
cients for dRK5 and fRK5 smoothers for 3-level multigrid (steady flow).dRK5 p = 1 fRK5 p = 1 dRK5 p = 2 fRK5 p = 2
21 0.05768995298 0.0578331573 0.04865009589 0.0487743632531 - -0.0002051554736 - -0.000218834843832 0.1405960888 0.1403808301 0.130316854 0.130090612241 - 0.0003953470071 - 2.608884832e-0542 - -0.001195029164 - 2.444376496e-0543 0.267958213 0.2681810517 0.2729621396 0.273480570551 - 0.0001441249202 - -0.00125038548752 - -0.0002608610327 - -0.0007838720635
53 - -0.0003368070181 - -0.000489088771254 0.5 0.8473374098 0.5 4.41213936761 - 0.4115573097 - 0.809721735862 - -0.003144851878 - 0.0843508900963 - -0.0001096455683 - -0.0198679900764 - 0.001555741114 - 0.0135981547665 1.0 0.5901414466 1.0 0.1121972094CFL 0.8 0.8 0.4 0.4
S 0.98812 0.98914 0.98974 0.9896
M G 0.89151 0.81762 0.90049 0.89903
EX IMG 167.06 - 124.02 -
-
8/8/2019 Vander Vegt
66/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 65
10 8 6 4 2 0 25
4
3
2
1
0
1
2
3
4
5
Stability domain of smoother
0.1
0.1
0.1
0.1
0.1
0.1
0.2
0.2
0.2
0.20.2
0.2
0.3
0.3
0.3
0.3
0.3
0.3
0.4
0.40
.4
0.4
0.4
0.4
0.5
0.50
.5
0.5
0.5
0.5
0.6
0.6
0.6
0.6
0.6
0.6
0.7
0.7
0.7
0.7
0.7
0.7
0.7
0.8
0.8
0.8
0.8
0.8
0.8
0.8
0.9
0.90.9
0.9
0.9
0.9
0.9
1
11
1
1
1
1
Re(z)
Im(z)
(a) Stability domain dRK5.
10 8 6 4 2 0 25
4
3
2
1
0
1
2
3
4
5
Re(z)
Im(z)
Spectrum of L and stability domain of smoother
low
high
(b) Spectrum of Lh and stability domain dRK5.
Stability domain of dRK5 smoother and spectrum of a space-time DG discretization of
the 2D advection-diffusion equation using quadratic basis functions.
-
8/8/2019 Vander Vegt
67/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 66
10 8 6 4 2 0 25
4
3
2
1
0
1
2
3
4
5
Stability domain of smoother
0.1
0.1
0.1
0.1
0.1
0.1
0.2
0.2
0.2
0.2
0.20.2
0.3
0.3
0.3
0.3
0.3
0.3
0.4
0.40
.4
0.4
0.4
0.4
0.5
0.5
0.5
0.5
0.5
0.5
0.6
0.6
0.6
0.6
0.6
0.6
0.7
0.7
0.7
0.7
0.70.7
0.8
0.8
0.8
0.
8
0.80.8
0.8
0.9
0.90.9
0.9
0.90.9
0.9
1
11
1
11
1
Re(z)
Im(z)
(a) Stability domain fRK5.
10 8 6 4 2 0 25
4
3
2
1
0
1
2
3
4
5
Re(z)
Im(z)
Spectrum of L and stability domain of smoother
low
high
(b) Spectrum ofLh and stability domain fRK5.
Stability domain of the fRK5 smoother and spectrum of a space-time DG discretization
of the 2D advection-diffusion equation using quadratic basis functions.
-
8/8/2019 Vander Vegt
68/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 67
1 0.5 0 0.5 1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
Re
Im
Spectrum of smoother
low
high
(a) dRK5 smoother.
1 0.5 0 0.5 1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
Re
Im
Spectrum of smoother
low
high
(b) fRK5 smoother.
1 0.5 0 0.5 1
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
Re
Im
Spectrum of threelevel operator
(c) 3-level MG with dRK5.
1 0.5 0 0.5 1
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
Re
Im
Spectrum of threelevel operator
(d) 3-level MG with fRK5.
Eigenvalue spectra for space-time DG discretizations of the 2D advection-diffusion
equation using quadratic basis functions.
-
8/8/2019 Vander Vegt
69/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 68
Testing multigrid performance
In order to demonstrate the performance of the optimized algorithms we considerthe 2D advection-diffusion equation on a square.
The exact steady state solution is
u(x1, x2) =1
2
exp(a1/1) exp(a1x1/1)exp(a1/1) 1
+exp(a2/2) exp(a2x2/2)
exp(a2/2) 1
.
-
8/8/2019 Vander Vegt
70/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 69
The advection-diffusion equation is discretized using the space-time discontinuousGalerkin discretization with quadratic basis functions.
In the discretization we use a Shishkin mesh which is suitable for dealing withboundary layers.
In pseudo-time we apply a rescaling to reduce the effect of grid stretching.
-
8/8/2019 Vander Vegt
71/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 70
The parameters in the tests are:
t = 100, a =
2, x = y = 0.01, N1 = N2 = 32.
Flow angle flow = /4.
Depending on the stability of the smoother, we use different CF L and V N
numbers.
The EXV scheme was used when Rei 1 and the EXI scheme was usedotherwise.
For the multigrid computations, 1 = 2 = 3 = 4 = 1, C = 4 and = 1.
-
8/8/2019 Vander Vegt
72/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 71
Work units
L
/L0
500 1000 1500 2000 2500 300010
-4
10-3
10-2
10-1
dRK5 coarse a pprox.
fRK5 coarse approx.EXI/EXV coarse approx.
dRK5 coarse exact
fRK5 coarse exactEXI/EXV coarse exact
Convergence of three-level multigrid with different Runge-Kutta smoothers for secondorder accurate space-time DG discretization.
-
8/8/2019 Vander Vegt
73/82
-
8/8/2019 Vander Vegt
74/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 73
We can draw the following conclusions:
In all cases using the optimized Runge-Kutta smoothers a big improvement isobtained over the original EXI-EXV smoother.
For linear basis functions the number of multigrid work units to obtain 4 orders of
reduction in the residual is reduced from 3300 to 380.
For quadratic basis functions the number of multigrid work units reduces from22000 to 184.
For linear basis functions the difference in convergence rate between Runge-Kuttasmoothers with only non-zero diagonal terms versus Runge-Kutta smoothers with
all coefficients non-zero is negligible.
-
8/8/2019 Vander Vegt
75/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 74
For quadratic basis functions this difference is, however, significant.
Using more Runge-Kutta coefficients enlarges the possibilities to optimize thesmoother, but the optimization process requires a significantly larger computing
time.
In order to speed up the optimization process the coefficients of Runge-Kuttaschemes with only non-zero diagonal terms are used as initial values.
-
8/8/2019 Vander Vegt
76/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 75
The effect of solving the equations on the coarsest mesh is very large.
In particular, for nonlinear problems it is tempting to solve the algebraic system onthe coarsest mesh only approximately, but the effect is non-negligible.
The flow angle has a small effect on the convergence rate. In general if the flow
angle is close to one of the mesh lines the convergence rate is the slowest.
It is important not to extrapolate the results for the advection-diffusion equationdirectly to the Euler and Navier-Stokes equations.
-
8/8/2019 Vander Vegt
77/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 76
Euler Equations
Amount of work per multigrid cycle differs single grid, p-, h-, and hp-multigrid:
work per cycle =
gpbp, SG,
(gpbp + gp1bp1)(p1 + p2 ) + g
p2bp2pC, pMG,
gpbp
(ch + ch1)(h1 + h2 ) + c
h2hC
, hMG,
ch(gpbp + gp1bp1)(p1 + p2 )+
gp2bp2((ch + ch1)(h1 + h2 ) + ch2hC), hpMG,
with gp and bp, respectively, the number of Gauss quadrature points and basis
functions in an element, and ch a weighting for the number of cells depending on the
grid-level h.
Coefficients: gp = 9, gp1 = 4, gp2 = 1, bp = 6, bp1 = 4, bp2 = 1, ch = 1,ch1 = 1/4 and ch2 = 1/16.
U i i f T Ch i N i l A l i d C i l M h i
-
8/8/2019 Vander Vegt
78/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 77
p-levels h-levels CFL work
1 1 1.0 0.99010 75006
3 1 1.0 0.74318 33370
3 3 1.0 0.73347 32091
1 3 1.0 0.66350 24098
1 3 0.5 0.80364 45360
Spectral radii and multigrid work units of different multigrid strategies.
U i i f T Ch i N i l A l i d C i l M h i 78
-
8/8/2019 Vander Vegt
79/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 78
1e-10
1e-09
1e-08
1e-07
1e-06
1e-05
0.0001
0.001
0 200 400 600 800 1000 1200 1400 1600
L2
total
Work Units
SGhMG
hMG exactpMG
hpMG
Convergence history of different multigrid techniques for space-time DG discretization
for inviscid flow around a NACA0012 airfoil (MTC 1 test case, = 2, M a = 0.5,448 64 elements).
U i it f T t Ch i N i l A l i d C t ti l M h i 79
-
8/8/2019 Vander Vegt
80/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 79
Conclusions
A space-time DG discretization for nonconservative hyperbolic pdes using theDLM theory and the compressible Navier-Stokes has been developed.
A new numerical flux for nonconservative hyperbolic pdes has been developed,which reduces to the HLLC flux for conservative pdes.
The effect of the choice of the path in phase space is in practice for nearly all casesnegligible.
The algorithm has been successfully tested on a depth averaged two-phase flowmodel and compressible flow simulations.
U i it f T t Ch i N i l A l i d C t ti l M h i 80
-
8/8/2019 Vander Vegt
81/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 80
Conclusions
A detailed three-level multigrid analysis has been conducted for h-, p- andhp-multigrid algorithms for the advection-diffusion and the linearized Euler
equations.
The analysis provides the asymptotic convergence rate of the multigrid algorithms
which is used to optimize the multigrid smoothers.
For the advection-diffusion equation the optimization results in a significantimprovement in the convergence rate in actual computations using the h-multigrid
algorithm.
University of Twente Chair Numerical Analysis and Computational Mechanics 81
-
8/8/2019 Vander Vegt
82/82
University of Twente - Chair Numerical Analysis and Computational Mechanics 81
For the Euler equations the optimized Runge-Kutta smoothers for h-, p- andhp-multigrid show initially a significant improvement in convergence rate for
inviscid flow about a NACA 0012 airfoil, but asymptotically most algorithms have
the same convergence rate.
Using an exact solution of the algebraic system on the coarsest mesh has a major
impact ion the convergence rate for the advection-diffusion equation, but not forthe Euler equations.