galerkin-characteristics finite element methods for flow … · 2013. 11. 14. · 4 7 finite...
TRANSCRIPT
1
Galerkin-Characteristics Finite Element Methods for Flow
Problems, I
Masahisa TabataWaseda University, Tokyo, Japan
1
The 9th Japanese-German International Workshop on Mathematical Fluid Dynamics, November 5-8, 2013, Waseda University, Tokyo, Japan
2
Contents
Galerkin-characteristics FEM and numerical analysis of flow problems
Convergence analysis of
① the scheme for the convection-diffusion equation
② the scheme for the Oseen equations
③ the scheme for the Navier-Stokes equations
④ 2nd order schemes in time and a stabilized scheme for the Navier-Stokes equations
2
Convection-dominated phenomenon
3
4
An example of convection-diffusion problem
Find : such that u R
where
0.1, 0.01, 0.001 Pe 10, 100, 1000
w u u f
0u
(0,0)
(1,1)
(1,0), 1w f 1
i.e., 1u
ux
1 2 1, i.e., ,u x x x
2: :given functionw
, wu v u v dxua
0w
3
5
GK
00.25
0.50.75
1
0
0.25
0.5
0.75
1
0
0.25
0.5
0.75
1
0 250.5
0.751
210
00.25
0.50.75
1
0
0.25
0.5
0.75
1
0
0.25
0.5
0.75
1
0 250.5
0.751
0
0
0.25
0.5
0.75
1 0
1
2
110 310
6
T[1977], Pironneau-T[2010]
00.25
0.50.75
1
0
0.25
0.5
0.75
1
0
0.25
0.5
0.75
1
0 250.5
0.751
210
00.25
0.50.75
1
0
0.25
0.5
0.75
1
0
0.25
0.5
0.75
1
0 250.5
0.751
310
00.25
0.50.75
1
0
0.25
0.5
0.75
1
0
0.25
0.5
0.75
1
0 250.5
0.751
110
4
7
Finite Element Schemes for Flow ProblemsUpwind approximation:
Characteristic method:
upwind element choice approximati T[1on, 977]mass-conservative upwind approximation, Baba-T[1981]streamline upwind Petrov/Galerkin approximation
first order in t
second order in t
Brooks-Hughes[1982], Johnson, Franca,...
Ewing-Russel[1981]Boukir-Maday-Metivet-Razafindrakoto[1997]
multi-step method:
single-step method: Rui-T[2002]
dicscontinuous Galerkin methodLesaint-Raviart[1974], Johson, Brezzi-Marini-Sulli,...
Notsu-T[2, 009]
Notsta su-bilize T[2d: 013], Douglas-Russel[19Pironne 82],au[1982] Suli[1988]
Pironneau-monot T[20one: 10],
Galerkin-Characteristics FEM Method
8
5
Galerkin Method
9
s.t.: Find f x
0 x 0
f
10; 1, ,h jV j N h V H
Find s.t.h hV
h h h h hdx f dx V
Note. Ritz-Galerkin method, Petrov-Galerkin method
Method of Characteristics
10
Find : 0, ,d T
0, 0 , ,0du f x tt
0where : 0, , : 0, , : .d d d du T f T
Find ( , ) : 0, ,dX T
0 0; , ;X t x t t x
, , , 0dX d
u X t f X t tdt dt
00 00 , 0X x x
0x
0;x X t xt
characteristics
6
Approximation of the Material Derivative
11
: (0, ) dX T R
( ), ( ( ), )D d
X t t X t tDt dt
Material derivative:
( ( ), ) ( ( ), )( ( ), )
D X t t X t t t tX t t
Dt t
• Approximation of material derivative:
( ) ( ( ), ) ,dX
t u X t tdt
),0( Tt
Du
Dt t
Position of a fluid particle:
t X tx
Idea of the Galerkin-Characteristics FEM
12
( ( ; ), ), , ,
D X t t x t t
Dt t
Du
Dt t
, ,dX
u X s s tds
( ; )X t x x
wheret
;X s x
xs
t t
7
13
The pros and cons The matrices are symmetric and independent of time
step for convection-diffusion, Oseen, and Navier-Stokes problems; positive definite for CD.
The schemes are unconditionally stable and convergent for CD and Os, conditionally stable and convergent for NS.
The schemes are robust for convection-dominated problems. c of -estimate is independent of ν for CD.
The attention should be paid for the computation of composite function terms, 1
1n nh hiX dx
1 is not smooth on !n n
h hix t u x x K K
2L
I. Convection-diffusion equation
14
8
15
Convection-Diffusion Equation
0
in (0, )
0 on (0, )
at 0 in
u f Tt
T
t
such that ),0(: Find R T
0T( 2,3), boundedd d R
1, 2 0 20 , ,
du W f L L
where
16
Weak Formulation
0
, , , ,
0
Da f V
Dt
Find : (0, ) such that T V 1
0V H
,f g f g dx
uD Du
Dt Dt t
: material derivative
where
, ,a
9
1st order scheme in time
17
1
0 0
, ( , ) ( , ),n n n
n nh hh h h h h h h
Ph h
Xa f V
t
where ( ) ( )n nX x x u x t
such that ,,...,1 , Find Thnh NnV
tn
tn 1
x
xX n1
tO
:time increment, /Tt N T t : FE-spacehV V
composite function
Note. Backward Euler method
1 11Note. ( ) ( )n n n n
h hX x x u x t
Note. : Poisson projectionPh
18
Theorem
-elementkP
2 2 2, k
h hL Lc t h
where
max ; 0,..., ,nh h TX X
n N
1st order scheme in time (cont.)
: Poisson projection of order hV k
1,
1
W
tu
2
1/22
1 TN n
h hnX Xt
, , 0c u T
10
Framework for the convection-diffusion equation
19
10 , : bdd. , 2,3dV H d
Hilbert space.
FEM space.
Bilinear for ,m. :a V V
,a dx
,h hV QV Q
0:regular, inverse ineq.h h
T 0, , diam, h KK Kh T
: radius of the inscribe ballK 2 12 2, 0, , , diamh c h K c hc c h K T
Poisson projection
20
,:Ph hV V ,ˆP
h h
ˆ , , , hh h hha a V
Ex. : -element, 1h kV P k
1ˆ, 0, k
kh HV
c k c h
1 ,. .,
k
P kh H V
i ce I h
L
11
Plan of the proof
21
Poisson projection
Transformation :nX
Discrete Gronwall's inequality
Error equation in Ph h he
Ph h he I
Transformation
22
n nX x x u x t :nX
Lemma
X x x w x t 1,
0
dw W
1, 1 1 : , 1to 1, ontoW
w t X
1, 2 , 0,1 ,W
w t 1
11
X
x
Ref. Rui-T[Prop. 1, 2002]
12
23
, dist ,x d x x
1, 1 1 : , 1to 1,ontoW
w t X
X x x w x t
w x w y t x
y
X x x w x w y t
x y w t 1,W
x y w t d x X x
2 i i iX x x w x t , , i j ij i jX w x t
, , det deti j ij i j
XX w x t
x
d x
,x X x
A tool for the proof
24
Lemma X x x w x t 1,
0 ,d
w W
221n
LLX c t 0,c
1,0,1 ,W
w t
2
22n n
LX X x dx
2 nxy dy y X x
y
21 c t y dy
2
21
Lc t
13
Error equation in
25
1
satisfies TNn
h ne
1
, , , , n n n
n nh hh h h h h h h
e e Xa e R V
t
0 0he
1 2n n nh h hR R R
1
2 ,n n n
n h hh
XR
t
where
ˆ : 0,h h T
1
1 ,n n n n
nh
D XR
Dt t
Lemma
ˆ ˆ, Ph h h h he
26
0 0 0 0 0ˆ ˆ 0Ph h h h he
1
, , ,n n n
n nh hh h h h
Xa f
t
ˆ, , ,n
n nh h h
Da f
Dt
: solution of CD, Poisson projection
1ˆ ˆn n n nnh
D XR
Dt t
1
1 : truncation errorn n n n
nh
D XR
Dt t
1
2 :projection errorn n n
nh hh
XR
t
1
, , ,n n n
n nh hh h h h h
e e Xa e R
t
14
Estimates of the remainders
27
1 01 ,n
hR c u t
2 02 ,n k
hR c u h
Lemma
28
1
1
n n nnnh
x X xDR x x
Dt t
11
0
1 11 , 1n n n n
ns
x X x x s u x t t s tt t
( ) ( )n nX x x u x t
1
01 , 1n n
t nsu x x s u x t t s t ds
1 ,nh t nR x u x t
1
01 , 1n n
t nsu x x s u x t t s t ds
1
11
1 10 01 1 , 1 1n n
t ns su x x s s u x t t s s t ds
1
1 1
, , ,0 01 2n n n
j k jk j jt tts st s ds u x u x u x
1 1 11 1 , 1 1nnx s s u x t t s s t ds
1 0,n
hR c u t
15
29
1
2
n n nn h hh
XR
t
1
20
11 , 1n n
h h ns
R x x s u x t t s tt
1
01 , 1n n
h t h nsu x x s u x t t s t ds
1/21 2
2 0 01 , 1n n n
h h t h nsR dx u x x s u x t t s t ds
12 2
2 01 , 1n n n
h h t h nsR x u x x s u x t t s t ds
1/21 2
01 , 1n n
h t h nsds u x x s u x t t s t dx
1/2
1 2
0, 1n
h t h ns
xds u x y t s t dy
y
1 ny x s u x t 1
1/22 2
0 0
n
n
t
h t htc u dt
1nt t s t
, kc u h
Discrete energy inequality in
30
0
nhe
2 2
0 0
n nt h hD e e 2 21 2 2
2 20 0, ,n n k
h he c u e c u h t
Lemma
1, , Tn N 0 0he
1
, , , , n n n
n nh hh h h h h h h
e e Xa e R V
t
Substitute into .nh he
1
, , , , 1, , n n n
n n n n nh hh h h h h T
e e Xe a e e R e n N
t
2 2 211 20 0 0 0 0 0
1
2n n n n n n nh h h h h he e X e R R e
t
Proof.
2 2 2 21 2 21 20 0 0 0
2 , ,n n n n kt h h h hD e e e c u e c u h t
16
Discrete Gronwall’s inequality
31
0 1 1 , 1, ,t n n n n nD x y a x a x b n N 0 10 1 1
, , , , , , , 0N N N
n n n n n nn n nx y b a a x y b
Lemma
1 :backward differencen nt n
x xD x
t
0
1(0, ]
2t
a
0 1201 1
, 1, ,n na a n t
n j jj jx t y e x t b n N
Proof.1
0 1 1 , 1, ,n nn n n n
x xy a x a x b n N
t
0 1 11 1 , 1, ,n n n na t x y t a t x b t n N
The result is proved by induction.
Proof of Theorem
32
2 2
0 0
n nt h hD e e
1, , Tn N 0 0he
0
Discrte energy inequality :
Apply the discrte Gronwall's inequality for
1,
1 1
2W
tu
2 2 2
2 2 2 2, ,n kh hl L l L
e e c u T h t
2 2 2, ,n k
h hl L l Le e c u T h t
2 21 2 22 20 0
, ,n n kh he c u e c u h t
1
Galerkin-Characteristics Finite Element Methods for Flow
Problems, II
Masahisa TabataWaseda University, Tokyo, Japan
1
The 9th Japanese-German International Workshop on Mathematical Fluid Dynamics, November 5-8, 2013, Waseda University, Tokyo, Japan
II. Oseen Equations
2
2
3
Oseen Equations
0
( ) in (0, )
0 in (0, )
0 on (0, )
at 0 in
uw u u p f T
tu T
u T
u u t
such that ),0(:, Find RR dTpu
0T( 2,3), boundedd d R
where 1, 2 0 1 0
0 0, , , 0d d d
w W f L u H u
1, 2 , ,
2ji
ijj i
vva u v D u D v D v
x x
4
Weak formulation
0
, , , , ,
, =0,
0
Duv a u v b v p f v v V
Dt
b u q q Q
u u
Find , : (0, ) such that u p T V Q 1 2
0 0,d
V H Q L
, ,b u q u q
where
wDu D u u
w uDt Dt t
3
5
1st order scheme in time
1
0 0
1
, , , , ,
, =0,
,0
n n nn n nh h
h h h h h h h h h
nh h h h
Sh h
u u Xv a u v b v p f v v V
t
b u q q Q
u u
, : FE-spaceh hV V Q Q
:time increment, /Tt N T t
Find , , 1,..., , such that n nh h h h Tu p V Q n N
( ) ( )n nX x x w x t where
6
Theorem
1,
1
W
tw
where
2
1/22
1 TN n
h hnX Xv t v
, , , , 0,c T w u p
/ : Stokes projection of order h hV Q k
max ; 0,..., ,nh h TX X
v v n N
1st order scheme in time (cont.)
1 2 2, k
h hH Lu u p p c t h
4
Framework for the Oseen equations
7
10 , 2,3, : bdd. dd
V H d 0
2 ; 0q L q x dxQ L
Hilbert spaces.
FEM spaces.
0:regular, inverse ineq.h h
T
Bilinear forms. ,: :a V V b V Q , ,
1,
2ij i j j iD v v v
, ·b v q q dv x
, 2 :a u v D u D v dx
,h hV QV Q
strain rate tensor
0, , diam, h KK Kh T : radius of the inscribe ballK
2 12 2, 0, , , diamh c h K c hc c h K T
Stokes projection
8
,:Sh h hQ QV V ˆ ˆ, ,S
h h hu p u p
ˆ ˆ, , , , ,h h h hh h hha u v b v p a u vb Vv v p ˆ , , , h hh h hb u q b u q q Q
Ex. 1 / : 2 / 1, Taylor-Hood element, 2h hV Q P P k
1ˆ ˆ, ,, 0, k k
kh h V Q H H
u u p pk c uc p h
1 ,. .,
k k
S kh H H V Q
i hce I
L
Ex. 2 / : 1 bubble / 1, MINI element, 1h hV Q P P k Note. / : 1 / 0 does not work. h hV Q P P
5
Stokes projection (cont.)
9
Key condition for the proof: Inf-sup condition.
0 0
,0, inf sup
h h h h
h hq Q v V
h hQ V
b v q
q v
Taylor-Hood element satisfies this condition.
Reference.[1]Brezzi, F., Fortin, M., Mixed and hypbrid finite element methods, Springer, New York, 1991.
MINI element satisfies this condition.
[2]Tabata, M., Numerical analyis of partial differential equations, Iwanami, Tokyo, 2010.
Stokes problem (d=2)
:),( hhhh QVpu spaceselement finite
2:),( Find hhh pu jhh j ,3,2,1;max )(
hh
jK
jh
6
P2/P1 space
hhhhh QVpuu ),,( 21
2d
1hu 2hu hp
2221
2121|2|1 ,,,,,1, xxxxxxuu
jj KhKh 21| ,,1 xxpjKh
spaceelement finite:
Note. Taylor-Hood element, 2k
P1+/P1 space
hhhhh QVpuu ),,( 21
2d
1hu 2hu hp
32121|2|1 ,,,1, xxuujj KhKh 21| ,,1 xxp
jKh
spaceelement finite:
1 2 3Note. MINI element, 1; : bubble functionk
7
Plan of the proof
13
Stokes projection
Discrete Gronwall's inequality
Transformation :nX
, , ,Sh h h h hu u p p e I u p
Error equation in , , ,Sh h h h he u p u p
Transformation
14
n nX x x w x t
:nX
Lemma
X x x w x t 1,0 ,
dw W
1, 1 1 : , 1to 1, ontoW
w t X
1, 2 , 0,1 ,W
w t 1
11
X
x
1
0 01 ,
LX t w H
1,
0 0,
LX t w W
8
Tools for the proof
15
Korn's inequality
Poincare's inequality
1
0 100, ,
dc D v v c v v H
100 0
0, ,c v c v v H
11 2 1 2 01 10, 0, ,
dc c c v D v c v v H
Error equation 1 in
16
ˆ ˆ ˆ ˆ, , , , , , satisfies Sh h h h h h h h he u u p p u p u p
,h he
1
, , , , , n n n
n n nh hh h h h h h h h h
e e Xv a e v b v R v v V
t
, =0, , 1, ,nh h h h Tb e q q Q n N
0 0
10,S
h he p
1 2,n n nh h hR R R
1
1 ,n n n n
nh
Du u u XR
Dt t
1
2 ,n n n
n h hh
XR
t
where
ˆ : 0, ,dh hu u T n nX x x wt x
9
17
ˆ, = , , =0 -0=0.n n nh h h h h hb e q b u q b u q
0 0 0 0 0 0 0
1 1 1ˆ ,0 , 0,S S S
h h h h h he u u u u p p
1
, , , ,n n n
n n nh hh h h h h h
u u Xv a u v b v p f v
t
ˆ ˆ, , , ,n
n n nh h h h
Duv a u v b v p f v
Dt
, : solution of NS, Stokes projectionu p
1ˆ ˆn n n nnh
Du u u XR
Dt t
1
1 : truncation errorn n n n
nh
Du u u XR
Dt t
1
2 :projection errorn n n
nh hh
XR
t
Error equation 2 in
18
ˆ ˆ ˆ ˆ, , , , , , satisfies Sh h h h h h h h he u u p p u p u p
,h he
1
3, , , , , n n
n n n nh hh h h h h h h h h h
e ev a e v b v R R v v V
t
, =0, , 1, ,nh h h h Tb e q q Q n N
0 0
10,S
h he p
1 2 ,n n nh h hR R R
1
1 ,n n n n
nh
Du u u XR
Dt t
1
2 ,n n n
n h hh
XR
t
where
ˆ : 0, ,dh hu u T n nX x x wt x
1 1
3
n n nn h hh
e e XR
t
10
Estimates of the remainders
19
1 0,n
hR c w u t
2 0, , ,n k
hR c w u p h
Lemma
13 0 0
n nh hR c w e
20
1
1
n n nnnh
u x u X xDuR x x
Dt t
11
0
1 11 , 1n n n n
ns
u x u X x u x s w x t t s tt t
( ) ( )n nX x x w x t 1
01 , 1n n
t nsw x u u x s w x t t s t ds
1 ,nh t nR x w u u x t
1
01 , 1n n
t nsw x u u x s w x t t s t ds
1
11
1 10 01 1 , 1 1n n
t ns sw x u u x s s w x t t s s t ds
1 1 11 1 , 1 1nnx s s w x t t s s t ds
1 0,n
hR c w u t
1
1 1
, , ,0 01 2n n n
j k jk j jt tts st s ds w x w x u w x u u
11
21
1
2 2 :projection errorn n n
n nh hh h
XR R
t
1
20
11 , 1n n
h h ns
R x x s w x t t s tt
1
01 , 1n n
h t h nsw x x s w x t t s t ds
1/21 2
2 0 01 , 1n n n
h h t h nsR dx w x x s w x t t s t ds
12 2
2 01 , 1n n n
h h t h nsR x w x x s w x t t s t ds
1/21 2
01 , 1n n
h t h nsds w x x s w x t t s t dx
1
1/22 2
0 0
n
n
t
h t htc w dt
1nt t s t , , , kc w u p h
1/2
1 2
0, 1n
h t h ns
xds w x y t s t dy
y
1 ny x s w x t
1 1
3 ,n n n
n h hh
e e XR
t
1
3 0 0L
n n nh hR c w e
Discrete energy inequality in
22
0
nhD e
Lemma
1, , Tn N
2 2
00
1
2n n
t h t hD D e D e
21 2 2
1 20
, , ,n khc w D e c w u p h t
0 0
0
kh k
D e c p h
12
23
Substitute into in the error equation 2. nt h hD e v
3, , ,n n n n n n nt h t h h t h h h t hD e D e a e D e R R D e
Proof.
ˆ, , , 0 0 0, 0n n n n n nh h h h h hb e b u b u n
22
1 2 30 0 0 0 00
n n n n n nt h t h h h h t hD e D D e R R R D e
, 1n
2 22 1 2 21 20 0 0
1, , ,
2n n n k
t h t h hD e D D e c w D e c w u p h t
0 0
10,S
h he p 0 0
1
kh k
e c p h
0
10,S
h I p
0 0
0
kh k
D e c p h
Proof of Theorem
24
1, , Tn N
Discrte energy inequality :
Apply the discrte Gronwall's inequality.
2 22
2 2 2 21, , , ,
2n k
h t h l Ll LD e D e c w u p T h t
2 2
00
1
2n n
t h t hD D e D e
21 2 2
1 20
, , ,n khc w D e c w u p h t
0
0
khD e c p h
2 22
1, , , ,
2n k
h t h l Ll LD e D e c w u p T h t
13
25
3
1sup , , ,
h h
n n n nv V h h h t h h h h
h V
c R R v D e v a e vv
Estimate of the pressure.
0
,sup
h h
nh hn
h v Vh V
b vc
v
30 0 0 02n n n n
h h t h hc R R D e D e
1
0 0 0, , , 2k n n n
h t h hc w u p h t e D e D e
0, , , k n
t hc w u p h t D e
2 2 2 2, , , , k n
h t hL Lc w u p T h t D e
, , , , kc w u p T h t
1
Galerkin-Characteristics Finite Element Methods for Flow
Problems, III
Masahisa TabataWaseda University, Tokyo, Japan
1
The 9th Japanese-German International Workshop on Mathematical Fluid Dynamics, November 5-8, 2013, Waseda University, Tokyo, Japan
III. Navier-Stokes Equations
2
2
3
Navier-Stokes Equations
in 0at
),0(on 0
),0(in 0
),0(in )(
0 tuu
Tu
Tu
Tfpuuut
u
such that ),0(:, Find RR dTpu
0T( 2,3), boundedd d R
where 2 0 1, 00, , 0
d df L u W u
1, 2 , ,
2ji
ijj i
vva u v D u D v D v
x x
4
Weak formulation
0
, , , , ,
, =0,
0
Duv a u v b v p f v v V
Dt
b u q q Q
u u
Find , : (0, ) such that u p T V Q 1 2
0 0,d
V H Q L
, ,b u q u q
where
,u wDu D u D u u
w uDt Dt Dt t
3
5
1st order scheme in time
1 1
0 0
1
, , , , ,
, =0,
,0
n n nn n nh h h
h h h h h h h h h
nh h h h
Sh h
u u Xv a u v b v p f v v V
t
b u q q Q
u u
, : FE-spaceh hV V Q Q
:time increment, /Tt N T t
Find , , 1,..., , such that n nh h h h Tu p V Q n N
1 1( ) ( )n nh hX x x u x t
where
Note. Pironneau[1982]
6
Theorem/4
0 0 0 0, 0, , dh c h h t c h
where
2
1/22
1 TN n
h hnX Xv t v
Note. Süli[1988]
, , , 0,c T u p
/ : Stokes projection of order h hV Q k
max ; 0,..., ,nh h TX X
v v n N
1st order scheme in time (cont.)
1 2 2, k
h hH Lu u p p c t h
4
Framework for the Navier-Stokes equations
7
10 , 2,3, : bdd. dd
V H d
02 ; 0q L q x dxQ L
Hilbert spaces.
FEM spaces.
0:regular, inverse ineq.h h
T
Bilinear forms. ,: :a V V b V Q , ,
1,
2ij i j j iD v v v
, ·b v q q dv x
, 2 :a u v D u D v dx
,h hV QV Q
strain rate tensor
0, , diam, h KK Kh T : radius of the inscribe ballK
2 12 2, 0, , , diamh c h K c hc c h K T
Stokes projection
8
,:Sh h hQ QV V ˆ ˆ, ,S
h h hu p u p
ˆ ˆ, , , , ,h h h hh h hha u v b v p a u vb Vv v p ˆ , , , h hh h hb u q b u q q Q
Ex. 1 / : 2 / 1, Taylor-Hood element, 2h hV Q P P k
1ˆ ˆ, ,, 0, k k
kh h V Q H H
u u p pk c uc p h
1 ,. .,
k k
S kh H H V Q
i hce I
L
Ex. 2 / : 1 bubble / 1, MINI element, 1h hV Q P P k
5
Plan of the proof
9
Stokes projection
Error equation in , , ,Sh h h h he u p u p
Discrete Gronwall's inequality
1,
1Induction is employed to evaluate and nh W
t u
1 at each step. n
h Lu
, , ,Sh h h h hu u p p e I u p
New terms to be evaluated
10
1The transformation depends on the function .nhu
1 1= n nh hX x x u xt
1,
1The estimate is requied. nh W
t u
1 appears as coefficients in the error equation.nhu
1The estimate is requied. nh L
u
6
Tools for the proof
11
Korn's inequality
Poincare's inequality 1
0 100, ,
dc D v v c v v H
100 0
0, ,c v c v v H 1
1 2 1 2 01 10, 0, ,
dc c c v D v c v v H
Inverse inequality / /, 0, ,q p
d q d ph h h hL L
c p q v ch v v V 1 p q
Sobolev's imbedding ,
,0, ,q m p
m p
L Wc v c v v W ,
dpq
d mp
Lagrang interpolation operator :h hV V 1
10, ,k
k kh V H
c v v ch v v H
1, 1,
1,0, , ,h W Wc v c v v W
,h L L
v v v L
Error equation1 in
12
ˆ ˆ ˆ ˆ, , , , , , satisfies Sh h h h h h h h he u u p p u p u p
,h he
1 1
, , , , , n n n
n n nh h hh h h h h h h h h
e e Xv a e v b v R v v V
t
0 0
10,S
h he p
1 2 3,n n n nh h h hR R R R
1 1
1
n n n nnh
Du u u XR
Dt t
1 1
2 ,n n n
n h h hh
XR
t
1 1 1 1
3
n n n nn hh
u X u XR
t
where
ˆ : 0, ,dh hu u T 1 1n nX x x u xt
, =0, , 1, ,nh h h h Tb e q q Q n N
7
13
ˆ, = , , =0 -0=0.n n nh h h h h hb e q b u q b u q
0 0 0 0 0 0 0
1 1 1ˆ ,0 , 0,S S S
h h h h h he u u u u p p
1 1
, , , ,n n n
n n nh h hh h h h h h
u u Xv a u v b v p f v
t
ˆ ˆ, , , ,n
n n nh h h h
Duv a u v b v p f v
Dt
, : solution of NS, Stokes projectionu p
1 1ˆ ˆn n n nn hh
Du u u XR
Dt t
1 1
1 : truncation errorn n n n
nh
Du u u XR
Dt t
1 1
2 :projection errorn n n
nh h hh
XR
t
1 1 1 1
3 :perturbation errorn n n n
nhh
u X u XR
t
Error equation2 in
14
ˆ ˆ ˆ ˆ, , , , , , satisfies Sh h h h h h h h he u u p p u p u p
,h he
1
4, , , , , n n
n n n nh hh h h h h h h h h h
e ev a e v b v R R v v V
t
, =0, , 1, ,nh h h h Tb e q q Q n N
0 0
10,S
h he p
1 2 3,n n n nh h h hR R R R
1 1
1
n n n nnh
Du u u XR
Dt t
1 1
2 ,n n n
n h h hh
XR
t
where
ˆ : 0, ,dh hu u T 1 1 n nX x x ut x
1 1 1
4
n n nn h h hh
e e XR
t
1 1 1 1
3
n n n nn hh
u X u XR
t
8
Estimates of the remainders
15
1 0
nhR c u t
13 0 0
,n n kh hR c u p e h
Lemma
12 0
, , 1n n kh h L
R c u p u h
1 14 0 0
L
n n nh h hR c u e
1,
1 1nh W
u t
16
1 1
1
n n nnnh
u x u X xDuR x x
Dt t
1
1 1 1
0
1 11 , 1n n n n
ns
u x u X x u x s u x t t s tt t
1 1( ) ( )n nX x x u x t
1 1 1
01 , 1n n
t nsu x u u x s u x t t s t ds
112 ,n n
h t nR x u x u u x t
1 1 1
01 , 1n n
t nsu x u u x s u x t t s t ds
1
11 1 11 10 0
1 1 , 1 1n nt ns s
u x u u x s s u x t t s s t ds
1
1 1 1 1 1, , ,0 0
1 2n n nj k jk j jt tts s
t s ds u x u x u u x u u
11 1 11 1 , 1 1n
nx s s u x t t s s t ds 12 0
nhR c u t
11, ,n n n
h nR x u x u x u x t 1 11 12n n nh h hR x R x R x
11 0
nhR c u t
9
17
1 1
2 2 :projection errorn n n
n nh h hh h
XR R
t
11
20
11 , 1n n
h h h ns
R x x s u x t t s tt
1 1 1
01 , 1n n
h h t h h nsu x x s u x t t s t ds
1/21 21 12 0 0
1 , 1n n nh h h t h h ns
R dx u x x s u x t t s t ds
12 21 12 0
1 , 1n n nh h h t h h ns
R x u x x s u x t t s t ds
1/21 21 1
01 , 1n n
h h t h h nsds u x x s u x t t s t dx
1/2
1 21
0, 1n
h h t h ns
xds u x y t s t dy
y
11 nhy x s u x t
1
1/22 21
0 0
n
n
tnh h t hL t
c u dt
1nt t s t
1,S
t h h t tt I u p t 1, , 1n kh L
c u p u h
18
1 1 13 0 0
L
n n n nh hR c u u u
1 1 1 1
3 3 :perturbation errorn n n n
n nhh h
u X u XR R
t
1 1 1 1 1
0 0ˆ ˆ
L
n n n n nh h hc u u u u u
1
0, k n
hc u p h e
1 1
4 4 :convection termn n n
n nh h hh h
e e XR R
t
1 14 0 0
L
n n nh h hR c u e
10
Discrete energy inequality in
19
0
nhD e
Lemma
2 2
00
1
2n n
t h t hD D e D e
03
02 k
hD e c p h
21 1 1 2 2
1 20
, , , , , ,n n n kh h hL L
c u p u D e c u p u h t
1,
11 1, , 1nh TW
t u n N
20
1
4, , , , , n n
n n n nh hh h h h h h h h h h
e ev a e v b v R R v v V
t
, =0, , 1, ,nh h h h Tb e q q Q n N
Substitute into .nt h hD e v
4, , ,n n n n n n nt h t h h t h h h t hD e D e a e D e R R D e
Proof of (1).
ˆ, , , 0 0 0, 0n n n n n nh h h h h hb e b u b u n
22
1 2 3 40 0 0 0 0 00
n n n n n n nt h t h h h h h t hD e D D e R R R R D e
, 1n
22 2 2 2 2
1 2 3 40 0 0 0 00
1
2n n n n n n
t h t h h h h hD e D D e R R R R
Error equation2:
11
21
1 1 1
0 0 0'n n n
h h he c e c D e
22 2 2 2 2
1 2 3 40 0 0 0 00
1
2n n n n n n
t h t h h h h hD e D D e R R R R
2 22 1 2 1 2
0, , 1 ,n k n k
h hLc u t c u p u h c u p e h
2 21 1
0
n nh hL
c u e
21 1 1 2 2
1 20
, , , , ,n n n kh h hL L
c u p u D e c u p u h t
21 1 1 2 2
1 20
, , , , , ,n n n kh h hL L
c u p u D e c u p u h t
0 0
1k
kh H
e c p h
0 0 0
1 12 0, 0,S S
h h he p I p
0 0
0, k
h kD e c p h
Proof of Theorem
22
We show the following by induction, 0, , :Tn N
From the discrete energy inequality,
2 21 1 1 2 21 21 1
; , , ; , ,n n n n kt h h h hL L
D e c u u p e c u u p h t
2 20 2 21 21 1
1 exp 1 1n kh hL L L Ln
e c u n t e c u h t n t
1,
1 1, , 1 implies nh TW
t u n N
1 2where , :monotone increasing w.r.t. the first variable. c c
1,2 n
hn Wt u
3 1nh L Ln L
u u
/40 0 0 0Choose small , 0. , .dh c h h t c h
12
23
1 1 1
General step, 1. Assume 1 , 2 and 3 are valid.n n n
n
1
1 From the discrete energy inequality and 3 , we obtain n n
1,
1 implies the existence of , .n n nh h hW
t u u p
2 21 1 1 2 21 21 1
; , , ; , ,n n n n kt h h h hL L
D e c u u p e c u u p h t
21 2 21 21
1 1 .n khL L
c u e c u h t
Applying the discrete Gronwall's inequality, we obtain
2 20 2 21 21 1
exp 1 1 .n kh hL L L L
e c u n t e c u h t n t
Initial step, 0.n
01 LHS=RHS
1,
0
02 . Proved similarly to that of (2) .h nW
t u
0
03 1. Proved similarly to that of (3) .h nL LL
u u
24
1, 1, 1,2 n n n n
h h h hn W W Wu u u u
1 1,
/220 21
d n n nh h H W
c h u u c u
1,
/220 211 1 1
ˆ ˆd n n n n n n nh h h h W
c h u u u u u u c u
1,
/220 3 22 23 21
d k k k n
Wc h c t h c h c h c u
1,
/224 21
d k n
Wc h t h c u
1, 1,
/224 21
n d k nh W W
t u t c h t h c u
1,
/4 /40 24 0 24 21
k d d n
Wc c c c h c h u
1,
/2 2 /224 21
d k d n
Wc h t h t c t u
/4
0dt c h
1,
/4 /40 24 0 24 0 21 0
k d d n
Wc c c c h c h u
13
25
3 n n n nh h h hn L L L
u u u u
6
/630
d n n nh h L L
c h u u u
/630 1 1 1
ˆ ˆd n n n n n n nh h h h L
c h u u u u u u u
/630 3 31 32
d k k k n
Lc h c t h c h c h u
/6 /634
d k d n
Lc h t h u
/12 /634 0
d k d n
Lc c h h u
1 n
Lu
/40
dt c h
/12 /634 0 0 0
d k d n
Lc c h h u
26
2 20 2 21 21 1
1 exp 1 1n kh hL L L Ln
e c u n t e c u h t n t
1,2 n
hn Wt u
3 1nh L Ln L
u u
The induction, 0, , ,Tn N
has been completed.
31, , , ,n k
he c T u p h t n
1 3 , , , .kh H
e c T u p h t
14
27
Now the discrete energy inequaliy
2 2
00
1
2n n
t h t hD D e D e
21 1 1 2 2
1 20
, , , , , ,n n n kh h hL L
c u p u D e c u p u h t
can be written as
2 2
00
1
2n n
t h t hD D e D e
21 2 2
4 50
, , , , , , .n khc u p T D e c u p T h t
Applying the discrete Gronwall's inequaliy, we obtain
2 2 6 , , , .kt h L
D e c u p T h t
0Estimate of n
t hD e
28
4
1sup , , ,
h h
n n n nv V h h h t h h h h
h V
c R R v D e v a e vv
Estimate of the pressure.
0
,sup
h h
nh hn
h v Vh V
b vc
v
40 0 0 02n n n n
h h t h hc R R D e D e
1 1
0 0 0, , , 2n k n n n
h h t h hLc u p u h t e D e D e
0, , k n
t hc u p h t D e
2 2 2 2, , , k n
h t hL Lc u p T h t D e
, , , kc u p T h t
1
Galerkin-Characteristics Finite Element Methods for Flow
Problems, IV
Masahisa TabataWaseda University, Tokyo, Japan
1
The 9th Japanese-German International Workshop on Mathematical Fluid Dynamics, November 5-8, 2013, Waseda University, Tokyo, Japan
IV. 2nd Order Schemes in Time and a Stabilized Scheme
2
2
2nd Order Schemes in Time
3
2nd order Galerkin-characteristics schemes
4
2step scheme
Singlestep scheme
1 2, 2, ,n n nh h h Tu u u n N 1 0, : initial functionsh hu u
1 2, ,n nh h Tu u n N 0 : initial functionhu
23 ( ) 4 ( ) ( 2 )f t f t t f t t dft O t
t dt
2( ) ( )
2
f t f t t df tt O t
t dt
Note. ( ) ( )f t f t t dft O t
t dt
3
5
Idea of 2nd order two-step scheme
1 2
21 23 4( )
2
nn n n n nX Xx u x O t
t t
1 ( ) ( )n nX x x u x t where
2 ( ) 2 ( )n nX x x u x t tn
tn 1
x
2n t 1 ( )nX x
2 ( )nX x
1 2*( ) 2 ( ) ( )n nu x u x u x
or 1 ( ) *( )nX x x u x t
2 ( ) 2 *( )nX x x u x t
Ewing-Russell[1981]
2Note. * ( ) ( ) ( )nu x u x O t
6
Navier-Stokes Equations
in 0at
),0(on 0
),0(in 0
),0(in )(
0 tuu
Tu
Tu
Tfpuuut
u
such that ),0(:, Find RR dTpu
0T( 2,3), boundedd d R
where 2 0 1, 00, , 0
d df L u W u
4
1, 2 , ,
2ji
ijj i
vva u v D u D v D v
x x
7
Weak formulation
0
, , , , ,
, =0,
0
Duv a u v b v p f v v V
Dt
b u q q Q
u u
Find , : (0, ) such that u p T V Q 1 2
0 0,d
V H Q L
, ,b u q u q
where
,u wDu D u D u u
w uDt Dt Dt t
8
2nd order two-step scheme
1 1 2 11 2
1
0 0
1
3 4, , , , ,
2
, =0,
: *
,0
n n n n nn n nh h h h h
h h h h h h h h h
nh h h h
h
Sh h
u u X u Xv a u v b v p f v v V
t
b u q q Q
u
u u
,h hV V Q Q :time increment, /Tt N T t
Find , , 2,..., , such that n nh h h h Tu p V Q n N
1 * * 1 21 ( ) ( ) , 2n n nh h h h hX x x u x t u u u
where
1 *2 ( ) 2 ( )n
h hX x x u x t
1 1 2Note. should be found by another method. khu u O t h
tn
tn 1
x
2n t 1
1 ( )nhX x
12 ( )n
hX x
5
9
Theorem/6
0 0 0 0, 0, , dh c h h t c h
where
2
1/22
1 TN n
h hnX Xv t v
Note. Boukir et al.[1997]
, , , 0,c T u p
/ : Stokes projection of order h hV Q k
max ; 0,..., ,nh h TX X
v v n N
2nd order two-step scheme (cont.)
1 2 2
2, kh hH L
u u p p c t h
Error equation1 in
10
ˆ ˆ ˆ ˆ, , , , , , satisfies Sh h h h h h h h he u u p p u p u p
,h he
1 1 2 1
1 23 4, , , , ,
n n n n nn n nh h h h h
h h h h h h h h h
e e X e Xv a e v b v R v v V
t
, =0, , 1, ,nh h h h Tb e q q Q n N
0 0
10,S
h he p
1 2 3,n n n nh h h hR R R R
1 1 2 11 2
1
3 4n n n n n nnh
Du u u X u XR
Dt t
1 1 2 1
1 22
3 4n n n n nn h h h h hh
X XR
t
1 1 1 1 2 1 2 11 1 2 2
3 4n n n n n n n n
n h hh
u X u X u X u XR
t t
where
ˆ : 0, ,dh hu u T
1 2 khe O t h
1 * * 1 21 ( ) ( ) , 2n n nX x x u x t u u u
1 *2 ( ) 2 ( )nX x x u x t
6
11
ˆ, = , , =0 -0=0.n n nh h h h h hb e q b u q b u q
0 0 0 0 0 0 0
1 1 1ˆ ,0 , 0,S S S
h h h h h he u u u u p p
1 1 2 1
1 23 4, , , ,
n n n n nn n nh h h h h
h h h h h h
u u X u Xv a u v b v p f v
t
ˆ ˆ, , , ,n
n n nh h h h
Duv a u v b v p f v
Dt
, : solution of NS, Stokes projectionu p
1 1 2 11 2ˆ ˆ ˆ3 4n n n n n n
n h hh
Du u u X u XR
Dt t
1 1 2 1
1 21
3 4: truncation error
n n n n n nnh
Du u u X u XR
Dt t
1 1 2 1
1 22
3 4:projection error
n n n n nnh h h h hh
X XR
t
1 1 1 1 2 1 2 1
1 1 2 234 :perturbation error
n n n n n n n nnh hh
u X u X u X u XR
t t
Error equation2 in
12
ˆ ˆ ˆ ˆ, , , , , , satisfies Sh h h h h h h h he u u p p u p u p
,h he
1 2
4
3 4, , , , ,
n n nn n n nh h h
h h h h h h h h h h
e e ev a e v b v R R v v V
t
, =0, , 1, ,nh h h h Tb e q q Q n N
0 0
10,S
h he p
1 2 3,n n n nh h h hR R R R
1 1 2 11 2
1
3 4n n n n n nnh
Du u u X u XR
Dt t
1 1 2 1
1 22
3 4n n n n nn h h h h hh
X XR
t
1 1 1 1 2 1 2 11 1 2 2
3 4n n n n n n n n
n h hh
u X u X u X u XR
t t
where
1 2 khe O t h
1 1 1 2 2 11 2
4 4n n n n n n
n h h h h h hh
e e X e e XR
t t
7
Estimates of the remainders
13
21 0
nhR c u t
13 0 0
,n n kh hR c u p e h
Lemma
12 0
, , 1n n kh h L
R c u p u h
1 14 0 0
L
n n nh h hR c u e
1,
1 1nh W
u t
14
1, 1, 1,2 n n n n
h h h hn W W Wu u u u
1 1,
/220 21
d n n nh h H W
c h u u c u
1,
/220 211 1 1
ˆ ˆd n n n n n n nh h h h W
c h u u u u u u c u
1,
/2 220 3 22 23 21
d k k k n
Wc h c t h c h c h c u
1,
/2 224 21
d k n
Wc h t h c u
1, 1,
/2 224 21
n d k nh W W
t u t c h t h c u
1,
2 /3 /60 24 0 24 21
k d d n
Wc c c c h c h u
1,
/2 3 /224 21
d k d n
Wc h t h t c t u
/6
0dt c h
1,
2 /3 /60 24 0 24 0 21 0
k d d n
Wc c c c h c h u
for the 2nd order scheme1,
Estimate of nh W
u
8
15
3 n n n nh h h hn L L L
u u u u
6
/630
d n n nh h L L
c h u u u
/630 1 1 1
ˆ ˆd n n n n n n nh h h h L
c h u u u u u u u
/6 230 3 31 32
d k k k n
Lc h c t h c h c h u
/6 2 /634
d k d n
Lc h t h u
2 /6 /634 0
d k d n
Lc c h h u
1 n
Lu
/60
dt c h
2 /6 /634 0 0 0
d k d n
Lc c h h u
for the 2nd order scheme
Estimate of nh L
u
16
2nd order one-step scheme
tn1
2n t
x
1n t
, 0, , 0f x t T xt
10, , ,f H
t
1
1/2 1/2 10, , ,
n nn nf H
t
1
1 11, , ,
2 2
n nn n n nh h
h h h h h h hf f Vt
2Crank-Nicolson scheme: t
9
17
Convection-Diffusion Equation
0
in (0, )
0 on (0, )
at 0 in
u f Tt
T
t
such that ),0(: Find R T
0T( 2,3), boundedd d R
1, 2 0 20 , ,
du W f L L
where
18
Pure Crank-Nicholson approximation
1
1 12 1, , ,
2 2
n n nn n n nh h
h h h h h h h
Xf f
t
where
hh V
tt
xuxuxxX nnn
2)()( 2/1
2
00 hh
such that ,,...,1 , Find Thnh NnV
tn
tn 1
x
xX n2
12 ( ) ( )
2n n n nt
X x x u x u x tu x
Heun method
or 2nd order Runge-Kutta method
10
Pure Crank-Nicholson approximation (cont.)
19
)(2
)(2 tOxxXx n
The total accuracy is ! )( tO
1 1/ 2 2
2n n nx x O t
22/11
2
1tOxfxff nnn
1/ 21
22 2 ( )( ) ( )
2
nn n n nX x X xx u O t
t t
However, tn
tn 1x xX n
2
x
tO
This scheme is not of .2( ) O t
2O t
20
A Second Order Characteristic FEM (Rui-T[2002])
where
j
ni
ijn
x
uJ
][
hh V
txuxxX nn )()(1
tt
xuxuxxX nnn
2)()( 2/1
2
00 hh
such that ,,...,1 , Find Thnh NnV
tn
tn 1
x
xX n2 xX n
1
2tO
12 ( ) ( )
2n n n nt
X x x u x u x tu x
or
112
1
1 11
11
, ,2
, div ,2
1,
2
n n nn n nh h
h h h h
n n n n nh h h h
n n nh h h
XX
t
tJ X u
f f X
11
A Second Order Characteristic FEM (Cont.)
21
element-kP
where
2 1
2
1/221' 1
1
2T
n n nN
H nL
Xt
Theorem (Rui-T[2002])
k
HhLh htc
2'
122 ,
2 2 max ; 0,..., ,n
h h TL Ln N
, , 0c u T
Key Expression
22
11
1( )
2n nX
1
1 21
, 1
1
2 2
n ndjn n
i j i j i
utX O t
x x x
1 1/ 2 21 1
1 1+( )
2 2n n n n nX x X O t
tn
tn 1
x
xX n2 xX n
1
2tO
12
23
Example 1 (rotating Gaussian hill)
T ,)5.0,5.0()5.0,5.0(
2 10, ,u u x xt
2
1 2
| ( ) |( , , ) exp
4 4cx t x
x x tt t
3444 100.1,100.5,105.2,1025.1
)cossin,sincos()( 2121 txtxtxtxtx 0.01),0,25.0( cx
Exact solution:
Rotating Gaussian Hill
24
1x
2x
, : 0t t
( 0.5, 0.5)
(0.5,0.5)
13
25
Example 1 (cont.)
)5/(2 ,/2 NTtNh
Pure Crank-Nicolson scheme:
hn
hn
hhnh
nhh
nnh
nh ff
t
X ,
2
1,
2, 112
1
triangles2 ofunion a into divided is 2N
P1 element
O h t O t
2nd order scheme:
2h t , 32,48,64,80N
2 2O h t O t
112
1
1 11 1
, ,2
1, ,
2 2
n n nn n nh h
h h h h
n n n n n nh h h h h
XX
t
tJ X f f X
26
41025.1 4105.2 4100.5 3100.1
2 -error vs.L t : Pure Crank-Nicholson scheme: 2nd order characteristics scheme
2E2E
2E2E
t t t t
2Lh
14
27
Navier-Stokes Equations
in 0at
),0(on 0
),0(in 0
),0(in )(
0 tuu
Tu
Tu
Tfpuuut
u
such that ),0(:, Find RR dTpu
0T( 2,3), boundedd d R
where 2 0 1, 00, , 0
d df L u W u
28
2nd order single-step scheme, Notsu-T[2009]
1 11 12
1 1
1 1 1 11
( , ) 1, , ,
2
1 1, , ,
2 2
, =0,
n n n n nn n n n n nh h h h
h h h h h h h
T Tn n n n n n nh h h h h h h h h h
nh h
u u X u uv D u D u X D v v p p X
t
t J u J u p J u J v f f X v v V
b u q
0 0
h h
h h
q Q
u u
,h hV V Q Q :time increment, /Tt N T t
Find , , 1,..., , such that n nh h h h Tu p V Q n N
11 ( ) ( ) ,n n
hX x x u x t
where
2 , ( )2
n tX w u x x w x u x u x
tn
tn 1
x
xX n2 xX n
1
15
29
2nd order single-step scheme (cont.)
1 11 1
12
1 1
1 1 1 11
( , ) 1, , ,
2
1 1, , ,
2 2
, =0,
n n nn n n nh h
h h h h h
T Tn n n n n n nh h h h h h
hh h
h h h h
hh
hu X uv D D u X D v v p X
t
t J u J u p J u J v f f X v v V
b q
ww
wr
w
0 0
h h
h h
q Q
u u
0 1Internal iteration: nh hw u
Find , , 1,..., , such that l lh h h h Tw r V Q N
1, , ,n nh h h h hw w r u p
1Note. This scheme can be used to get in two-step method.hu
A stabilized Galerkin-characteristics scheme for the Navier-Stokes Equations
30
16
Stabilized P1/P1 scheme
31
P1/P1 element for NS equations.
Stabilizing term is necessary.P1/P1 element does not satisfy the inf-sup condition.
Especially useful for 3D computation.
Stabilized P1/P1 Galerkin-characteristics FE scheme.symmetric matrix, small DOF
less computation timefiner subdivision for 3D problems
32
Navier-Stokes Equations
in 0at
),0(on 0
),0(in 0
),0(in )(
0 tuu
Tu
Tu
Tfpuuut
u
such that ),0(:, Find RR dTpu
0T( 2,3), boundedd d R
where 2 0 1, 00, , 0
d df L u W u
17
1, 2 , ,
2ji
ijj i
vva u v D u D v D v
x x
33
Weak formulation
0
, , , , ,
, =0,
0
Duv a u v b v p f v v V
Dt
b u q q Q
u u
Find , : (0, ) such that u p T V Q 1 2
0 0,d
V H Q L
, ,b u q u q
where
,u wDu D u D u u
w uDt Dt Dt t
34
P1/P1 stabilized scheme
1 1
0 0
1
, , , , ,
, , =0,
,0
n n nn n nh h h
h h h h h h h h h
n nh h h h h h h
Sh h
u u Xv a u v b v p f v v V
t
b u q p q q Q
u u
-C
, : P1 FEspaceh hV V Q Q :time increment, /Tt N T t
Find , , 1,..., , such that n nh h h h Tu p V Q n N
1 1( ) ( )n nh hX x x u x t
where
Note. Notsu-T[2008]
20, , = ,h
h K Kkp q h p q
CT
18
35
Theorem/4
0 0 0 0, 0, , dh c h h t c h
where
2
1/22
1 TN n
h hnX Xv t v
Ref. Notsu-T[2013]
0, , , , 0,c T u p
/ : P1/P1 elementh hV Q
max ; 0,..., ,nh h TX X
v v n N
P1/P1 stabilized scheme (cont.)
1 2 2,h hH L
u u p p c t h
Framework for the Navier-Stokes equations
36
10 , 2,3, : bdd. dd
V H d 0
2 ; 0q L q x dxQ L
Hilbert spaces.
FEM spaces.
0:regular, inverse ineq.h h
T
Bilinear forms. : :, : , hVV b QV QQa C
, ·b v q q dv x
, 2 :a u v D u D v dx
,h hV QV Q
0, , diam, h KK Kh T : radius of the inscribe ballK
2 12 2, 0, , , diamh c h K c hc c h K T
20, , = ,h
h K Kkp q h p q
CT
19
Stokes projection
37
,:Sh h hQ QV V ˆ ˆ, ,S
h h hu p u p
ˆ ˆ, , , , ,h h h hh h hha u v b v p a u vb Vv v p ˆ ˆ, , , , ,h h h h h h h hh hb u q p q b u qp Qq q -C -C
Ref. Brezzi-Douglas[1988]
2 1ˆ0, ˆ, ,h h V Q H Hu u p p p hc uc
2 1 ,. ., S
h H H V Qi e cI h
L
Stokes projection (cont.)
38
Key condition for the proof: Stability inequality.
0 0,
, 0,
, , , ,inf sup
, ,h h h h h h h h
h h h h h h h h hu p V Q v q V Q
h h h hV Q V Q
a u v b v p b u q p q
u p v q
-C
Ref. Brezzi-Douglas[1988], Franca-Stenberg[1991]
Note.
0 0
,0, inf sup
h h h h
h hq Q v V
h hQ V
b v q
q v
Inf-sup condition,
does not hold for the P1/P1 element.
20
Outline of the proof
39
Stokes projection of the stabilized type
Error equation in , , ,Sh h h h he u p u p
Discrete Gronwall's inequality
1,
1Induction is employed to evaluate and nh W
t u
1 at each step. n
h Lu
, , ,Sh h h h hu u p p e I u p
Refer to Notsu-T[2013] for the details.
Estimates of the term h C
Schemes free from numerical integration error
Mass lumping technique:
Pironneau-T[2010]
Finite difference method:
Notsu-Rui-T[2013]
Exact integration for approximated velocity:
T-Uchiumi[2013]
40
esitmateL
2discrete esitmateL
21
Galerkin-characteristics FEM of lumped mass type
41
11
, ( , ) ( , ),n n n
h h h n nh h h h h
I XI f
t
where
hh V
1 ( ) ( )n nX x x u x t
0 0h hI
such that ,,...,1 , Find Thnh NnV
:time increment, /Tt N T t : 1 FEM spacehV V P
: , , : node, interpolationh h hI C V I v P v P P
2: , , : node, lumpingh h h PV L v x v P x D P
A First Order Characteristic FEM (cont.)
42
Theorem
1-element, weakly acute type triangulationP
2
,h h L
hI c h t
t
Note. max ; 0,...,nh h TX X
n N
0,1 , 2; 0, 3d d
1, ,h h L
I c h t h
Pironneau-T[2010]
22
Application to a Two-Fluid Flow Problem
43
44
Multiphase flow problems with interface tension(1)
NS eqns
Interface condition: 0,i
u
0 t BC slip
1
2
TD u u u
Fluids 0,1, ..,m
,i i
0 0,
i t
n
2i
ipn D u n n
: coefficienti: curvature
or 0u
1,...,i m
1 t
m t i t
Note. i t 1,...,i m
23
45
Multiphase flow problems with interface tension(2)
fpIuDuut
ukkk
2)(
0 u
, satisfies NS eqns. in each domain , 0,...,ku p t k m
0Interface conditions on i it t t
,0u ,,0on conditionsBoundary T
0, 0 or 0u n D u n n u
0at conditions Initial t,0uu 00 , 0,...,k k k m
2i
ipn D u n n
Evolution of , ; [0,1)i it s t s
, , 1, ,iiu t i m
t
t�0.00000
46
Different density fluids in an “Hourglass”
32,N 1
8t
0.5,0
0.5, 2
60T
1
0nonslip BC
1 1 1, , 100,0.2,0.1
0 0, 1,1
0
2f
2 2 2 2, , 120,0.2,0.1
Ref. T[2010]
24
References M. Tabata. A finite element approximation corresponding to the upwind finite differencing.
Memoirs of Numerical Mathematics, 4(1977), 47-63.
R. E. Ewing and T. F. Russell. Multistep Galerkin methods along characteristics for convection-diffusion problems. In R. Vichnevetsky and R. S. Stepleman, editors, Advances in Computer Methods for Partial Differential Equations, 4, 28-36, IMACS, 1981.
O. Pironneau. On the transport-diffusion algorithm and its applications to the Navier-Stokes equations. Numerische Mathematik, 38(1982), 309-332.
F. Brezzi and J. Jr. Douglas. Stabilized mixed methods for the Stokes problem. NumerischeMathematik, 53(1988), 225-235.
E. Suli. Convergence and nonlinear stability of the Lagrange-Galerkin method for the Navier-Stokes equations. Numerische Mathematik, 53(1988), 459-483.
L. P. Franca and R. Stenberg. Error analysis of some Galerkin least squares methods for the elasticity equations. SIAM Journal on Numerical Analysis, 28(1991), 1680-1697.
K. Boukir, Y. Maday, B. Metivet, and E. Razafindrakoto. A high-order characteristics/finite element method for the incompressible Navier-Stokes equations. International Journal for Numerical Methods in Fluids, 25(1997), 1421-1454.
H. Rui and M. Tabata. A second order characteristic finite element scheme for convection-diffusion problems. Numerische Mathematik, 92(2002), 161-177.
H. Notsu and M. Tabata. A single-step characteristic-curve finite element scheme of second order in time for the incompressible Navier-Stokes equations. Journal of Scientific Computing, 38(2009), 1-14.
47
O. Pironneau and M. Tabata. Stability and convergence of a Galerkin-characteristics finite element scheme of lumped mass type. International Journal for Numerical Methods in Fluids, 64(2010), 1240-1253.
M. Tabata. Numerical simulation of fluid movement in an hourglass by an energy-stable finite element scheme. In M. N. Hafez, K. Oshima, and D. Kwak, editors, Computational Fluid Dynamics Review 2010, 29-50. World Scientific, Singapore, 2010.
H. Notsu, H. Rui, and M. Tabata. Development and L2-analysis of a single-step characteristics finite difference scheme of second order in time for convection-diffusion problems. Journal of Algorithms & Technology, 7(2013), 343-380.
H. Notsu and M. Tabata. Error estimates of a pressure-stabilized characteristics finite element scheme for the Oseen equations. WIAS-DP-2013-001, Waseda Univ., 2013.
H. Notsu and M. Tabata. Error estimates of a pressure-stabilized characteristics finite element scheme for the Navier-Stokes equations. WIAS-DP-2013-002, Waseda Univ., 2013.
48