(Galerkin) Finite element approximations

The finite element method (FEM): special choice for the shape functions .

Ne = 5 Subdivide into elements e:
Ne
[
= e
e=1
1 2 3 4 5 e1 e2 =
x=a x=b
Approximate u on each element separately by a polynomial of some degree p, for
example by Lagrangian interpolation (using p + 1 nodal points per element). The
end points of an element must be nodal points.
 Example: linear elements

Global shape functions:

Ne = 5 n
X
1 uh(x) = uii(x) = T (x)u
3(x)
i=1

1 2 3 4 5 n: number of global nodal points.

x1 = a x2 x3 x4 x5 x6 = b

Local element shape functions:

ueh(x) = ue11(x)+ue22(x) = T (x)ue

1(x) 2(x)
1 1
with T = [1(x), 2(x)] and

x xe2 x xe1
xe1 xe2 xe1 xe2 1(x) = e , 2(x) = e
x1 xe2 x2 xe1
 Example: quadratic elements
Local element shape functions:

1(x) 2 (x) 3(x)

1 1 1

xe1 xe2 xe3 xe1 xe2 xe3 xe1 xe2 xe3

ueh(x) = ue11(x) + ue22(x) = T (x)ue

with T = [1(x), 2(x), 3(x)] and

(x xe2)(x xe3) (x xe1)(x xe3)
1(x) = e , 2(x) = e ,
(x1 xe2)(xe1 xe3) (x2 xe1)(xe2 xe3)
(x xe1)(x xe2)
3(x) = e
(x3 xe1)(xe3 xe2)
 Example: higher-order elements

General polynomials of order p:

p+1
X
ueh(x) = ueii(x) = T (x)ue
i=1

with T = [1, 2, . . . , p+1].

Various expansions possible:

B Gauss-Lobatto integration points (includes end points) and Lagrangian

interpolation (Spectral elements).

B Hierarchical base functions: end points are nodes but internal shape functions
have no nodes (similar to Legrende polynomials). (hp-FEM).

B Legrendre polynomials in discontinuous Galerkin methods.

 Global numbering

4
X
Ne = 3 uh(x) uh(x) = uii(x) = T (x)u
i=1
u2 u3
u1 u4 u1 1(x)
1 2 3 u2 2(x)
u=

, (x) =
x1 = a x2 x3 x4 = b u3
3(x)
1 u4 4(x)
Ne = 3 3(x)

1 2 3
x1 = a x2 x3 x4 = b
 Local numbering in elements

Weak form: Find uh in Sh such that

dvh duh
( ,A ) + vh(b)hb = (vh, f ) for all vh Vh
dx dx

Split:
XNe XNe
dvh duh
( ,A ) + vh(b)hb = (vh, f ) for all vh Vh
e=1
dx dx e=1
Write in each element e:

2
X 2
X
ueh(x) = ueii(x) = T (x)ue, vhe (x) = viei(x) = T (x)v e
i=1 i=1

where uTe = (ue1, ue2) and T (x) = (1(x), 2(x)).

 Element matrix and vector

We get:
Ne
X Ne
X
(v Te K eue) + vnhb = v Te f e
e=1
e=1

where

 d dT  Z
d dT
Ke = ,A = A dx
dx dx e e dx dx
Z
fe = (, f )e = f dx
e

are the element matrix K e and element vector f e.

 Local global (assembling)

The local vectors ue and v e are part of the global vectors:

ue = P eu, v e = P ev ,

So we get

Ne
X Ne
X X
Ne 
v Te K eue = v T
P Te K eP e u
=v T
K e u
| {z }
e=1
e=1 e=1
K e

XNe XNe X
Ne 
v Te f e = v T P Te f e = v T f e
e=1
e=1
|
{z } e=1
f e

 Weak form

Substitution into the weak form:

vT K u = vT f for all v

or
Ku = f

with
Ne
X
K= K e
e=1


0
Ne
X 0
.
f= f e +
.

e=1 0
hb
 Example (1)

For example:
    u1
u2 0 1 0 0 u2 = P 2u
u2 = =
u3 0 0 1 0 u3
| {z }
P2 u4

0 0   
2
1 0 K11 K12 2
0 1 0 0
K 2 = P T2 K 2P 2 =
0 1 K21 2 2
K22 0 0 1 0
0 0

0 0 0 0
0 K11 2 2
K12 0
=
0 K21 K22 0
2 2

0 0 0 0
 Example (2)

Assembly of K :

1 1

K11 K12 0 0 0 0 0 0 0 0 0 0
K211
K 1
0 0 0 K11 2
K122
0 0
K= 22 + + 0 0 0
0 0 0 0 0 K21 2
K222
0 0 3 3
0 K11 K12
3 3
0 0 0 0 0 0 0 0 0 0 K21 K22
1 1

K11 K12 0 0
K211 1
K22 2
+ K11 2
K12 0
= 2 2 3

3
0 K21 K22 + K11 K12
3 3
0 0 K21 K22

Similar for assembly of f .

 Exercise 5

The bandwidth of K for linear shape functions is 3. How large is the bandwidth

for K when using polynomial shape functions of order k, k 1?

 Dirichlet conditions

Renumber and split vector and matrix (u unknown, p prescribed)

       
uu vu K uu K up fu
u= , v= v , K= , f=
u p p K pu K pp f p

Note: up prescribed values, v p = 0. Weak form

vT K u = vT f for all v

with v p = 0 leads to

v Tu (K uuuu + K upup) = v Tu f u for all v u

or
K uuuu = f u K upup

 1D convection-diffusion-reaction equation

1D convection-diffusion-reaction Eq.: find u(x) such that for x (a, b)

u u u
+ a (A ) + bu = f
t x x x
and

u = ua(t), at x = a, t > 0 (D )
du
A = hb(t) at x = b, t > 0 (N )
dx
Notes:

B Strong form; Classical (strong) solution u(x, t)

B f (x, t) C 0(a, b) (continuous) then u C 2(a, b) (twice continuously
differentiable)
 Oldroyd-B/UCM viscoelastic model

5
+ = 2D
where
5
= L LT
or

+ ~u L LT + = 2 D
t
|{z} | {z } | {z } |
{z }
u a u bu f
t x

Diffusion is missing (A = 0).

 Dimensionless form (1)

Scaling:

t = tct tc : characteristic time

u = U u U : characteristic value solution
x = Lx L : characteristic length scale

Dimensionless variables: O(1).

U u aU u AU 2u

+
2 2
+ bU u =f
tc t L x L x
Relative to convection:
L u u 1 2u bL L
+ + u = f
atc t x Pe x2 a aU
aL
Pe = : Peclet number, convection/diffusion.
A
 Dimensionless form (2)
Time scales:

L
B convection:
a
2
L
B diffusion:
A
1
B source: (relaxation time)
b
B time scales in b.c.
B externally or internally generated frequencies (von Karman vortex)

We choose tc = L/a (convection) and get

u u 1 2u
+ + b u = f
t x Pe x2

bL L
with b = and f = f . Pe > 1 : convection dominated
a aU
 Exercise 6

L2
Assume b = 0 and f = 0. We take for the typical time scale tc = (diffusion
A
time scale). Show that we now have the non-dimensional form

u u 2u

+ Pe 2 = 0
t x x

When will this non-dimensional form be preferable over the one on the previous
slide?
 Steady state 1D convection-diffusion-reaction equation

steady state: u/t = 0

du d du
L(u) = a (A ) + bu = f
dx dx dx

L: linear operator, with b.c.

u = ua, at x = a (D )
du
A = hb at x = b (N )
dx
 Weak form

Multiply with test function v and integrate:

(v, Lu f ) = 0 for all v

Partial integration of the diffusion term and inserting b.c. we get: Find u S
such that

dv du du
( , A ) + (v, a ) + (v, bu) + v(b)hb = (v, f ) for all v V
dx dx dx

where S and V are appropriate spaces.

 Galerkin FEM approximations (1)

Approximation spaces Sh and Vh:

n
X
uh(x) = uii(x) = T (x)u
i=1
n
X
vh(x) = vii(x) = T (x)v
i=1

where (x) are global shape functions. Substituting this into the weak form gives:
Find uh Sh such that

dvh duh duh

( ,A ) + (vh, a ) + (vh, buh) + vh(b)hb = (vh, f ) for all vh Vh
dx dx dx
or
vT K u = vT f for all v

 Galerkin FEM approximations (2)

and thus
Ku = f

with

d dT di dj
K = ( , A ) or Kij = ( ,A )
dx dx dx dx
dT dj
+ (, a ) + (i, a )
dx dx
+ (, bT ) + (i, bj )

f = (, f ) hb(b) fi = (i, f ) hbi(b)

 Galerkin FEM approximations (3)

Build from element matrices. For example with constant coefficients and linear
element shape functions:
 
d dT 1 1 1
( , )e = stiffness matrix
dx dx h 1 1
 
dT 1 1 1
(, )e = convection matrix
dx 2 1 1
 
h 2 1
(, T )e = mass matrix
6 1 2

where h is the element length.

 Global stiffness matrix

Global stiffness matrix (uniform element size):

1 1 0 . . . . . . . . . . . . . 0
1 2 1 0 . . . . . . . . 0

0 1 2 1 0 . . . 0
. ..
.
1
.. ..
h
.. ..


0 ... 0 1 2 1 0

0 . . . . . . . . 0 1 2 1
0 . . . . . . . . . . . . . 0 1 1

Finite difference scheme: (divide by h)

ui1 + 2ui ui+1 d2u 2

= (x i ) + O(h )
h2 dx2
 Global convection matrix

Global convection matrix (uniform element size):

1 1 0 . . . . . . . . . . . . . 0
1 0 1 0 . . . . . . . . 0

0 1 0 1 0 ... 0
. ..
.
1
.. ..
2
.. ..


0 . . . 0 1 0 1 0

0 . . . . . . 0 1 0 1
0 . . . . . . . . . . . 0 1 1

Finite difference scheme: (divide by h)

ui1 + ui+1 du
= (xi) + O(h2)
2h dx
 Global mass matrix

Global mass matrix (uniform element size):

2 1 0 ......... 0
1 4 1 0 . . . . . . 0

0 1 4 1 0 . . . 0
. ..
.
h .. ..
6
.. ..


0 . . . 0 1 4 1 0

0 . . . . . . 0 1 4 1
0 ......... 0 1 2

Finite difference scheme: (divide by h)

ui1 + 4ui + ui+1
= u(xi) + O(h2)
6
 Global right-hand side

Sometimes the following approximation is applied:

X
f (x) fk k (x) fk : nodal point values of f
k

In that case: X
fi = (i, f ) = (i, k )fk
k
with (i, k ) the mass matrix.
Finite difference scheme: (divide by h)

fi1 + 4fi + fi+1

= f (xi) + O(h2)
6

Note that the usual approach is to use numerical integration.

 Exercise 7

a. Verify the element stiffness, convection and mass matrix by performing the
integrations.

b. Assume a non-equidistant grid and denote the element sizes by h1, h2,. . . ,hNe .
How does the global stiffness matrix now look like? Has the global convection
matrix still a zero on the diagonal?
 Example: 1D steady convection-diffusion (1)

Equation:
du 1 d2u
=0 on (0, 1)
dx Pe dx2
with b.c.
u(0) = 0, u(1) = 1
Exact solution:
1 exp(Pex)
u(x) =
1 exp(Pe)
 Example: 1D steady convection-diffusion (2)

Ne = 10; Pe = 1, 10, 100.

1
Pe=1 exact
Pe=1 GFEM
0.8 Pe=10 exact
Pe=10 GFEM
0.6 Pe=100 exact
Pe=100 GFEM
0.4

0.2
u(x)

-0.2

-0.4

-0.6

-0.8
0 0.2 0.4 0.6 0.8 1
x
 Example: 1D steady convection-diffusion (2)

Ne = 10, 20; Pe = 25

1
Pe=25 exact
Pe=25 Ne=10
Pe=25 Ne=20
0.8

0.6
u(x)

0.4

0.2

-0.2
0 0.2 0.4 0.6 0.8 1
x
 Observations (1)

1
B Boundary layer of width .
Pe
B Global upstream wiggles appear when the boundary layer is not resolved by
the mesh.
B Governed by the mesh Peclet number Peh:

ah 1h
Peh = = Pe
2A 2 L
where wiggles appear when

h 2
Peh > 1 or >
L Pe
or wiggle-free when
h 2
<
L Pe
 Observations (2)

The problem is twofold:

1. Rapid change in the solution over small lengths (large gradients)

2. Mesh size h is too large to resolve the length scale

As long as the length scale is resolved GFEM works well for convection-diffusion
problems.
 Example: 1D steady convection with source term (1)

Equation (Pe ):
du
=f on (0, 1)
dx
with
1 (x 12 )2
u(0) = 0, f = exp[ ]
2
Exact solution:
1h x 21 1 i
u(x) = erf( ) + erf( )
2 2
Notes:

B length scale
B for 0
1
u(x) = H(x ) (Heaviside)
2
du 1
= f (x) = (x )
dx 2
 Example: 1D steady convection with source term (2)

Ne = 20; = 0.45, 0.15, 0.05.

1
=0.45 exact
=0.45 GFEM
=0.15 exact
0.8 =0.15 GFEM
=0.05 exact
=0.05 GFEM
0.6
u(x)

0.4

0.2

-0.2
0 0.2 0.4 0.6 0.8 1
x
 Example: 1D steady convection with source term (3)

Ne = 11, 20, 21; = 0.05

1
=0.05 exact
=0.05 Ne=11
=0.05 Ne=20
0.8 =0.05 Ne=21

0.6
u(x)

0.4

0.2

-0.2
0 0.2 0.4 0.6 0.8 1
x
 Observations

B Inaccurate solutions and upstream wiggles when layer with small length scale
is not resolved by the mesh.

B By experimentation we find that

h 1
<
2

to get good smooth solutions with GFEM.

B Strong mesh dependence when solution is underresolved.

Problem is again lack of resolution for rapidly changing solutions (large gradients).
 Summary of the examples

B GFEM works well for smooth solutions and the mesh resolves rapidly changing
solutions.

B For non-smooth solutions and/or a too coarse mesh wiggles can easily appear.

B The GFEM is not robust and local errors can easily be amplified throughout
the domain.

Some might consider the latter a good thing (Dont suppress wiggles they tell
you something). However:

B In many practical problems resolving all large gradients is not an option and
would require too much computer resources.

B In viscoelastic flow problems the GFEM method often performs very poorly.

Stabilization necessary
 Analysis of the discrete equations
Equidistant mesh (mesh size h), inner node i:

ui+1 ui1 ui+1 2ui + ui1 fi1 + 4fi + fi+1

a A 2
=
2h h 6
Notes:

B central differencing

B consistent averaging of source term

B for f = 0:
1
ui+1 ui1 (ui+1 2ui + ui1) = 0
Peh
or
1 + Peh
ui+1 ui = (ui ui1)
1 Peh
or

sign(ui+1 ui) = sign(ui ui1) for Peh > 1 Wiggles!

B for Peh (no diffusion): Simpson rule for integration, which is O(h4).

B for Peh (no diffusion), matrix:

1 1 0 . . . . . . . . . . . . . 0
1 0 1 0 . . . . . . . . 0

0 1 0 1 0 . . . 0
. ..
.

a . .
. . = ...
2h . ..
.
0 . . . 0 1 0 1 0

0 . . . . . . 0 1 0 1
0 . . . . . . . . . . . 0 1 1

zeros on the diagonal, leap frogging.

 Artificial diffusion (1)

ah
For Peh = = 1 we have a diffusivity of:
2A
ah
A = A =
2

d2 u
Adding A 2 to the equations, or
dx

du d2u
a (A + A) 2 = f
dx dx
removes wiggles, but we solve a modified equation. New mesh Peclet:

h= ah Peh
Pe ah
= <1
2(A + 2 ) 1 + Peh
 Artificial diffusion (2)
Discretized equations with equidistant mesh (mesh size h), inner node i:

ui+1 ui1 ah ui+1 2ui + ui1 fi1 + 4fi + fi+1

a (A + ) 2
=
2h 2 h 6
or
ui ui1 ui+1 2ui + ui1 fi1 + 4fi + fi+1
a A 2
=
| {z }h h 6
upwind differencing

ui ui+1
ui1

xi1 xi xi+1
h h

ui ui1 du
= (xi) + O(h)
h dx
 Artificial diffusion (2)
In general artificial diffusivity:
ah
A =
2
with examples:

B = 1, upwind differencing
h = 1 for Peh > 1:
B Pe
(
0 for Peh 1
=
1 Pe1 for Peh > 1
h

B Exact nodal point values (optimal value) for

du d2u
a A 2 = constant
dx dx
gives:
1
= coth(Peh)
Peh
 (Peh)

1
coth(x)-1/x
1-1/x
x/3

0.8

0.6

0.4

0.2

0
0 2 4 6 8 10
 Exercise 8

Show that for 6= 0 the first-order derivative can be interpreted as being

discretized by the following finite difference scheme

1 ui+1 ui 1 ui ui1 du
2 (1 )
h
+ 2 (1 + )
h
= (xi) + O(h)
dx

i.e. a linear combination of downwind and upwind differencing.

 Example: 1D steady convection-diffusion

Ne = 10; Pe = 25 (Peh = 1.25)

1
Pe=25 exact
GFEM
0.8 =1
=0 for Peh1, 1-1/Peh for Peh>1
=coth(Peh)-1/Peh
0.6
u(x)

0.4

0.2

-0.2
0 0.2 0.4 0.6 0.8 1
x
 Observations

B Upwind differencing rather inaccurate (first-order) near large gradients.

B Wiggles are gone. As expected the boundary layer is not resolved.

B Optimal (Peh) can reduce error significantly (weighted upwind/downwind).

B Away from the boundary layer the solution seems accurate.

 Example: 1D steady convection with source term

Ne = 20, 40; = 0.05; = 1

1.2
=0.05 exact
GFEM Ne=20
1 GFEM Ne=40
AD =1 Ne=20
0.8 AD =1 Ne=40

0.6
u(x)

0.4

0.2

-0.2
0 0.2 0.4 0.6 0.8 1
x
 Example: 1D steady convection with source term

Error in solution. Ne = 100 (Galerkin), 3000 ( = 1); = 0.05

0.002
AD =1 Ne=3000
0.0015 GFEM Ne=100

0.001
maxk|uk-uexact|

0.0005

-0.0005

-0.001

-0.0015

-0.002
0 0.2 0.4 0.6 0.8 1
x
 Observations

B GFEM: global wiggles and inaccuracy for Ne = 20 but very accurate for
Ne = 40 (fast convergence)

B Upwind: no wiggles (robust) but convergence is very slow near large gradients.
Seems to be accurate outside the layer with large gradients.

Question: Can we combine the accuracy (fast convergence) of GFEM with the
robustness of upwind?
 Set of equations (flow of a visco-elastic fluid)

Rewrite: Find (~u, p, ) such that,

~u
( + ~u ~u) (2sD) + p = ~b, in
t
~u = 0, in

( + ~u L LT ) + = 2D, in
t
 Weak form of modified equation by AD

Artificial diffusion solves the modified equation:

du d du  ah
a (A + A) + bu = f, with A =
dx dx dx 2
where we assume also the modified boundary condition:

u = ua, at x = a (D )
du
(A + A) = hb at x = b (N )
dx
For the weak form we get: Find u S such that

dv du du
( , (A + A) ) + (v, a ) + (v, bu) + v(b)hb = (v, f ) for all v V
dx dx dx
 Streamline upwinding (SU) (1)

Compared to standard weak form we have added the term

dv du dv ah du h dv du
( , A ) = ( , )=( ,a )
dx dx dx 2 dx 2 dx dx
This suggest another form of the modified weak form:
dv du h dv du
( , A ) + (v + , a ) + (v, bu) + v(b)hb = (v, f ) for all v V
dx dx 2 dx dx
the convection term has been weighted with a modified test function
h dv
v = v +
2 dx
streamline upwinding (SU). For constant a, uniform size h, 1D equivalent to
AD (artificial diffusion). Very inaccurate (overly diffusive) with time-dependent,
reaction and source terms (similar to AD).
 Streamline upwinding (SU) (2)

For the Galerkin FEM:

h dvh T h d
vh = vh + = v with = ( + )
2 dx 2 dx
where h may vary from element to element.

xi1 xi xi+1
2

hi hi+1
 Consistent weighting

Hughes & Brooks (1982): problem with SU in inconsistency : The exact solution
does not satisfy the weak form. Their solution is to modify the weighting of the
differential equation from the standard form

(v, Lu f ) = 0 for all v

to
h dv
(v + , Lu f ) = 0 for all v
2 dx
where
du d du
L(u) = a (A ) + bu
dx dx dx
The space of the weight functions (not V !) is in effect different from the space
of shape functions when applying Vh = Uh like in the Galerkin method (
Petrov-Galerkin).
 Interpretation of second-derivatives

Interpretation of Hughes & Brooks (1982). Split:

h dv h dv
(v + , Lu f ) = (v, Lu f ) + ( , Lu f ) = 0 for all v
2 dx | {z } | 2 dx {z }
I II

B Term I: partial integration of diffusion terms with fluxes on the boundary

(standard procedure)

B Term II: interprete element wise:

h dv X h dv
( , Lu f ) = ( , Lu f )e =
2 dx e
2 dx
 Streamline upwind Petrov-Galerkin (SUPG)

This leads to the SUPG method for the 1D steady convection-diffusion-reaction

equation: Find u S such that

dv du du
( , A ) + (v, a ) + (v, bu) + v(b)hb
dx dx dx
X  h dv du d du 
+ , a (A ) + bu f
2 dx dx dx dx e
e |{z}
a
= (v, f ) for all v V

Notes:
P
B A = 0 (no diffusion): e can be removed (viscoelastic CE)

B consistency
 h
B usual definition: a = with time scale
2

h1 ah 1 1
= = = A
2a 2 a2 a2

d du
B constant A, linear elements: (A ) = 0 in the interior of the element. If
dx dx
reaction term bu and source term f are also absent (convection-diffusion only)
SU=SUPG!

B optimal values of (Peh) neccessary to improve accuracy of diffusion terms.

 Example: 1D steady convection with source term (1)

Ne = 20; = 0.05

1.2
=0.05 exact
GFEM Ne=20
1 AD =1 Ne=20
SUPG =1 Ne=20
0.8

0.6
u(x)

0.4

0.2

-0.2
0 0.2 0.4 0.6 0.8 1
x
 Example: 1D steady convection with source term (2)

= 0.05; maximum nodal error as a function of number of elements Ne.

1
AD
SUPG
GFEM

0.1
maxk|uk-uexact|

0.01

0.001
10 100
Ne
 Example: 1D steady convection with source term (3)

= 0.05; maximum nodal error as a function of number of elements Ne.

1
AD
SUPG
GFEM
7/x
10/x2

0.1
maxk|uk-uexact|

0.01

0.001
10 100
Ne
 Observations

B Compared to GFEM: SUPG greatly enhances the stability/robustness for coarse

grids in convection(-dominated) problems where solutions are non-smooth or
underresolved (no global wiggles).

B Compared to SU: SUPG is much more accurate because convergence is faster

than linear.
 Extension to 3D: convection-diffusion-reaction equation

Strong solution u(~x, t):

u
+ ~a u (Au) + bu = f on
t | {z }
Lu

with boundary and initial

conditions
D
u(~x, t = 0) = u0(~x) in
N
u = uD on D
~n
u
A = A~n u = hN on N ~n: outside normal
n
 Weak form (1)

Multiply by testfunction v:

u
(v, + Lu f ) = 0 for all v
t
where the standard inner product on L2() is
Z
(a, b) = ab dx

With


Au v = A(u)v) Au v
and the divergence theorem (Gauss):
Z Z
~a dx = ~n ~a d

 Weak form (2)

we get the weak form: Find u S such that

u
(v, ) + (v, ~a u) + (v, Au) + (v, bu) + (v, hN )N = (v, f )
t

for all v V . Notes:

B v = 0 on D , u = uD on D
B we have silently introduced:
Z
(~a, ~b) = ~a ~b d

Z
(a, b) = ab d

 GFEM (1)
= ee, e f = for e 6= f

1(~x) 3(~x)
30 2(~x)
1

1
0 1
0
1 2

03
1 03
1 03
1
1
0 1
0 1
0
1 1 1

1
0 1
0 1
0
2 2 2
 GFEM (2)

Approximation spaces Sh and Vh:

n
X
uh(~x, t) = ui(t)i(~x) = T (~x)u(t)
i=1
n
X
vh(~x, t) = vi(t)i(~x) = T (~x)v (t)
i=1

where (~x) are global shape functions. Substituting into the weak form:

v T M u + v T K u = v T f for all v

or
M u + K u = f

 GFEM (2)

The mass matrix M is given by

M = (, T ), Mij = (i, j )

and the stiffnes matrix by

K = (, AT ) or Kij = (i, Aj )
T
+ (, ~a ) + (i, ~a j )
T

+ (, b ) + (i, bj )

f = (, f ) (, hN )N fi = (i, f ) (i, hN )N

 Example: 2D steady convection with source term (1)

|~a| = 1

u=0

u=0

u u
ax + ay =f on (0, 1) (0, 1)
x y
1
1
with (ax, ay ) = ( 2 2, 2 2) and

1 (x + y 1)2
u(x, 0) = 0, u(0, y) = 0, f = exp[ 2
]
2

Exact solution goes from 0 for x + y < 1 to 1 for x + y > 1 over a length scale
of .
 Example: 2D steady convection with source term (2)

Ne = 15 15; = 0.05; GFEM

1.03E+00

7.61E01

4.92E01

2.23E01

4.57E02

3.15E01

max: 1.03E+00
min:3.15E01
 Example: 2D steady convection with source term (3)

Ne = 30 30; = 0.05; GFEM

1.02E+00

8.08E01

5.99E01

3.89E01

1.79E01

3.08E02

max: 1.02E+00
min:3.08E02
 Example: 2D steady convection with source term (4)

Ne = 45 45; = 0.05; GFEM

1.00E+00

7.99E01

5.95E01

3.91E01

1.88E01

1.61E02

max: 1.00E+00
min:1.61E02
 Example: 2D steady convection with source term (5)

Ne = 100 100; = 0.05

1.00E+00

8.01E01

6.00E01

3.99E01

1.98E01

3.47E03

max: 1.00E+00
min:3.47E03
Example: 2D steady convection with discontinuous inflow (1)

|~a| = 1
u=1

u=0

u u
ax + ay =0 on (0, 1) (0, 1)
x y

with (ax, ay ) = ( 12 2, 12 2) and

u(x, 0) = 0, u(0, y) = 1

Exact solution is: (

1 for y > x
0 for y < x
Example: 2D steady convection with discontinuous inflow (2)

Ne = 15 15; GFEM

1.06E+00

8.38E01

6.13E01

3.88E01

1.63E01

6.22E02

max: 1.06E+00
min:6.22E02
Example: 2D steady convection with discontinuous inflow (3)

Ne = 30 30; GFEM

1.06E+00

8.38E01

6.13E01

3.88E01

1.63E01

6.22E02

max: 1.06E+00
min:6.22E02
 Observations for GFEM in 2D

B Inaccurate solutions and global upstream wiggles when layer with small length
scale is not resolved by the mesh.

B Almost wiggle free, very localized cross-stream jumps possible even for coarse
meshes.

B Strong mesh dependence when solution is underresolved.

 Artificial diffusion (1)

1D added diffusion:
d2 u ah
A 2 with A =
dx 2
which stabilizes
du
a
dx
3D added diffusion:
|a|h
(Au) with A =
2
which stabilizes
~a u
however, in all directions. This is too much added diffusion. Only diffusion in the
direction of ~a is needed since
u ~a
~a u = |~a|~e u = |~a| with ~e =
e |~a|
 Artificial diffusion (2)

Add anisotropic diffusion using a tensorial diffusion coefficient, with a component

in the direction of ~e (i.e. the streamline direction) only:

(A~e~e u)

or
~a~a
(A u) with A = A~e~e = A 2
|~a|
In the weak form we get an extra term:

(v, A u)
 SU (1)

Diffusion added in AD method:

~a~a
(v, A u) = (v, A 2 u)
|~a|
Z
~a~a
= v A 2 u) d
|~a|
Z
A
= 2
(~a v)(~a u) d
|~
a|
A
=( 2
~a v, ~a u)
|~a|
= (~a v, ~a u)

with
A h
= 2=
|~a| 2|~a|
 SU (2)

The weak form now becomes

+ v + ~a v, ~a u +

i.e. the convection term is multiplied by a modified weighting function

v = v + ~a v
 SUPG

SU SUPG: consistent weighting and interpret extra terms elementwise:

 u 
v + ~a v, + Lu f = 0
t

or (with elementwise interpretation):

 u  X u 
v, + Lu f + ~a v, + Lu f =0
t e
t e

First term: partial integrate to standard weak form.

 Notes on SUPG (1)

A
B = 2 , with A the amount of streamline diffusion. When the real diffusion
|~a|
ah
A 6= 0, we need to generalize A = (Peh) to more dimensions. There are
2
various ways to do this. For example for a bilinear quadrilateral:
~h1 = h1~e1 ~a0
~h2 = h2~e2

|~a ~h1| |~a ~h2|

~e2 A = (Peh1) + (Peh2)
2 2
h2
~e1 with

|~a ~h1| |~a ~h2|

Peh1 = , Peh2 =
h1 2A 2A

B For higher-order elements a constant is non-optimal.

 Notes on SUPG (2)

B The choice of is somewhat arbitrary, in particular when diffusion is absent

(A = 0). Note, that
A h
= 2=
|~a| 2|~a|
In viscoelastic flow it is common to use:

h
=
2U

where h is a characteristic element size and U is a characteristic velocity.

There are various possible choices for h and U (see separate notes).
A bit of experimentation with the 1D Gaussian distribution in the right-hand
side example (source term, no diffusion) seems to suggest that the solution
is rather insensitive to the exact value of in this case, as long as the scaling
with h is there.
 Example: 2D steady convection with source term (1)

|~a| = 1

u=0

u=0

u u
ax + ay =f on (0, 1) (0, 1)
x y
1
1
with (ax, ay ) = ( 2 2, 2 2) and

1 (x + y 1)2
u(x, 0) = 0, u(0, y) = 0, f = exp[ 2
]
2

Exact solution goes from 0 for x + y < 1 to 1 for x + y > 1 over a length scale
of .
 Example: 2D steady convection with source term (2)

Ne = 15 15; = 0.05; SUPG

1.02E+00

8.14E01

6.08E01

4.03E01

1.98E01

7.67E03

max: 1.02E+00
min:7.67E03
 Example: 2D steady convection with source term (3)

Ne = 30 30; = 0.05; SUPG

1.00E+00

8.03E01

6.02E01

4.01E01

2.00E01

4.83E04

max: 1.00E+00
min:4.83E04
Example: 2D steady convection with discontinuous inflow (1)

|~a| = 1
u=1

u=0

u u
ax + ay =0 on (0, 1) (0, 1)
x y

with (ax, ay ) = ( 12 2, 12 2) and

u(x, 0) = 0, u(0, y) = 1

Exact solution is: (

1 for y > x
0 for y < x
Example: 2D steady convection with discontinuous inflow (2)

Ne = 15 15; SUPG

1.05E+00

8.27E01

6.09E01

3.91E01

1.73E01

4.47E02

max: 1.05E+00
min:4.47E02
Example: 2D steady convection with discontinuous inflow (3)

Ne = 30 30; SUPG

1.05E+00

8.30E01

6.11E01

3.91E01

1.72E01

4.70E02

max: 1.05E+00
min:4.70E02
Example: 2D steady convection with discontinuous inflow (4)

Ne = 30 30; AD method with isotropic diffusion of |a|h/2

1.00E+00

8.00E01

6.00E01

4.00E01

2.00E01

0.00E+00

max: 1.00E+00
min: 0.00E+00
Example: 2D steady convection with discontinuous inflow (5)

Ne = 100 100; AD method with isotropic diffusion of |a|h/2

1.00E+00

8.00E01

6.00E01

4.00E01

2.00E01

0.00E+00

max: 1.00E+00
min: 0.00E+00
 Observations for SUPG in 2D

B In streamline direction: SUPG greatly enhances the stability/robustness for

coarse grids in convection(-dominated) problems where solutions are non-
smooth or underresolved (no global wiggles).

B Normal to the streamline direction: SUPG shows a small diffusion-like

smoothing that decreases with mesh size. The diffusion is much less than the
diffusion in the AD method (when added isotropically).