Professional Documents
Culture Documents
< ,
what is (the maximum) r such that
u u
h
ch
r
u
, (8)
where
=
L
2, and
. (9)
We note, that in aerodynamics the functional J() might represent important quantities like
aerodynamical force coecients (drag, lift or moment coecients).
Some error estimates like the L
2
-estimate in the case of Poissons equation and the error estimates
with respect to target functionals J() require the use of duality arguments including the solutions
to appropriately dened dual or adjoint problems. Therefore, we continue the above list as follows:
Adjoint consistency: Given the primal problem (1) and a target functional
J(u) =
_
udx +
_
uds, (10)
with j
L
2
() and j
L
2
(), we dene the adjoint problem
L
z = j
in , B
z = j
on . (11)
where L
and B
u v dx
_
n uv ds =
_
fv dx.
Due to v H
1
0
() the boundary integral vanishes. Thereby, the weak form of (15) is given as
follows: nd u V := H
1
0
() such that
B(u, v) = F(v) v V, (16)
where
B(u, v) =
_
u v dx, F(v) =
_
fv dx. (17)
Theorem 2.1 (Lax-Milgram theorem (Existence and Uniqueness)) Let V be a
Hilbert space, i.e. a complete space with scalar product (, ). Let the linear form F : V R be
continuous, i.e. there is a C
F
> 0 such that
F(v) C
F
v
V
v V. (18)
Let the bilinear form B : V V R be continuous, i.e. there is a constant C
B
> 0 such that
B(u, v) C
B
u
V
v
V
, u, v V. (19)
Furthermore, let B be V -coercive (or V -elliptic), i.e. there is a constant > 0, such that
B(v, v) v
2
V
, v V. (20)
Then, there is a unique solution u V such that
B(u, v) = F(v) v V. (21)
We say, Problem (21) is well-posed.
8
In the following we want to employ the Lax-Milgram Theorem 2.1 to show that there exists a
unique solution to problem (16). Using the Cauchy-Schwarz inequality, we nd that
|F(v)| = |
_
fv dx| f
L
2
()
v
L
2
()
f
L
2
()
v
H
1
()
,
|B(u, v)| = |
_
uv dx| u
L
2
()
v
L
2
()
u
H
1
()
v
H
1
()
.
Hence, the linear and bilinear functionals F and B are continuous. Furthermore, we see that B is
coercive with respect to the H
1
-seminorm:
B(v, v) =
_
|v|
2
dx = |v|
2
H
1
()
. (22)
Using the Poincare inequality 24 we obtain
B(v, v) =
_
|v|
2
dx = |v|
2
H
1
()
_
u
2
L
2
()
+|v|
2
H
1
()
_
= v
2
H
1
()
, (23)
with = 1/(C
p
+1), i.e. B(, ) is H
1
-coercive. Existence and uniqueness of a solution to (16) now
follows by application of the Lax-Milgram theorem 2.1.
Theorem 2.2 (Poincare inequality) Let R
d
be contained in a d-dimensional (hyper)cube
with edge length s. Then, there is a constant C
p
= s > 0 such that
v
L
2
()
C
p
|v|
H
1
()
v H
1
0
(). (24)
Proof: Any book on linear functional analysis or nite element methods.
2.1.2 The inhomogeneous Dirichlet problem
Let us now consider the case of an inhomogeneous Dirichlet problem,
u = f in , u = g
D
on
D
, (25)
with g
D
L
2
(
D
) on
D
= , and g
D
0 . Assume that there is a u
D
H
1
() with u
D
= g
D
on
D
. Then, the weak formulation to (25) is given by: nd u = u
D
+u
0
with u
0
H
1
0
() such that
_
u v dx =
_
fv dx v H
1
0
(). (26)
We deduce the existence and uniqueness of a solution to this problem similar to the homogeneous
case by rewriting the problem as follows: nd u
0
H
1
0
() such that
_
u
0
v dx =
_
fv dx
_
u
D
v dx v H
1
0
().
For B(u
0
, v) and F(v) denoting the left and right hand side of this equation, we nd that
F(v) f
L
2
()
v
L
2
()
+|u
D
|
H
1
()
|v|
H
1
()
Cv
H
1
()
.
Furthermore, it has been shown in the previous section that B(, ) is a continuous and H
1
-coercive
bilinear functional. Well-posedness, i.e. existence and uniqueness of a solution to (26) then follows
through the application of the Lax-Milgram theorem 2.1.
9
2.1.3 The Neumann problem
Finally, we consider the Neumann problem,
u = f in , n u = g
N
on
N
, (27)
with
N
= and g
N
L
2
(
N
) is a given function. As we do not have Dirichlet boundary
conditions to impose we consider the space H
1
() instead of H
1
0
(). We multiply (27) by a test
function v H
1
(), integrate over , and integrate by parts, to obtain
_
u v dx
_
n uv ds =
_
fv dx.
In view of (27) we obtain following weak formulation: nd u V := H
1
() such that
B(u, v) = F(v), v V,
with
B(u, v)
_
u v dx, F(v) =
_
fv dx +
_
N
g
N
v ds. (28)
Remark 2.3 (Well-posedness of the Neumann problem) We observe that
B(1, v) = B(u, 1) = 0 (29)
for all u, v H
1
(), i.e. ker B = span{1}. As B(1, 1) = 0, the Lax-Milgram theorem can not be
applied. However, considering H
1
()/ ker B which is the set of all equivalent classes
H
1
() := H
1
()/ ker B
= {u H
1
() :
_
u dx = 0}, (30)
we employ a Poincare inequality for v
H
1
(), [40], and arrive at
|B(v, v)| =
_
|v|
2
dx = |v|
2
H
1
()
v
2
H
1
()
v
H
1
(). (31)
Using a trace inequality v
L
2
()
Cv
H
1
()
for all v H
1
(), see Theorem 2.5, we obtain
|F(v)| |
_
fv dx| +|
_
g
N
v ds| f
L
2
()
v
L
2
()
+g
N
L
2
()
v
L
2
()
Cv
H
1
()
.
Using a generalization to the Lax-Milgram theorem, we obtain existence and uniqueness in
H
1
()
(i.e. uniqueness up to constants) of a solution, provided following compatibility condition is satised
F(1) =
_
f dx +
_
g
N
ds = 0. (32)
Remark 2.4 In order to avoid the theoretical diculties of the pure Neumann problem, we usually
consider the mixed Dirichlet-Neumann problem (14), with
D
= . In fact, imposing Dirichlet
boundary conditions at a single point only, i.e.
D
= {p} , is sucient to obtain a solution
which is unique, and not only unique up to a constant.
10
Theorem 2.5 (Trace theorem) Let R
d
be an open bounded domain with piecewise smooth
boundary. Furthermore, satises a cone condition. Then there is a unique continuous linear map
: H
1
() L
2
(),
and a constant C > 0 such that
(u)
L
2
()
Cu
H
1
()
u H
1
(), (33)
and
(u) = u|
u H
1
() C
0
(
),
where (u) denotes the trace of u on and the trace operator. Note, that usually the trace
operator is omitted and a notation u|
N
= and
D
N
= . Furthermore, we assume
that
D
= , see Remark 2.4. As described in the previous section, this problem is rewritten in a
weak formulation: nd u V such that
B(u, v) = F(v) v V, (35)
where V is an appropriately chosen function space with H
1
0
() V H
1
().
The nite element method generates approximate solutions to (35). To this end, let T
h
= {}
be a geometric discretization of consisting of elements , where h denotes the maximum diameter
of all . Let
j
(x) V, 0 j < N
h
, be N
h
linearly independent functions in V and u
j
, 0 j < N
h
,
real numbers. Then
u
h
(x) =
0j<N
h
u
j
j
(x), (36)
is a discrete function in the discrete function space V
h
dened by
V
h
= span{
j
(x)}
N
h
j=1
V.
We note, that V
h
V , where V is the continuous (and innite-dimensional) function space the
exact solution u to (35) is to be sought in.
Denition 2.6 For p 1 we dene the space of continuous piecewise polynomials of degree p by
V
c
h,p
= {v
h
C
0
() :v
h
|
Q
p
( ) if is the unit hypercube, and
v
h
|
P
p
( ) if is the unit simplex, T
h
},
(37)
where P
p
and Q
p
are the spaces of polynomials and tensor product polynomials of degree p. While
dealing with continuous nite element discretizations, we use the short notation V
h
:= V
c
h,p
.
Replacing u and v in (35) by discrete functions u
h
, v
h
V
h
, the discrete problem is given by:
nd u
h
V
h
such that
B(u
h
, v
h
) = F(v
h
) v
h
V
h
. (38)
11
We note that the bilinear and linear forms in the discrete problem (38) are the same as in the weak
formulation (35). Therefore, in this section we do not need to introduce notations B
h
(, ) and F
h
()
as we did in the general case in Section 1.3.
Due to the trial (ansatz) and test functions u
h
and v
h
taken from the same discrete function
space V
h
this is a so-called Galerkin nite element discretization. There are also so-called Petrov-
Galerkin nite element discretizations where the trial and test functions, u
h
U
h
and v
h
V
h
,
belong to dierent discrete function spaces U
h
= V
h
.
Denition 2.7 Finite elements based on discrete functions u
h
V
h
being contained in the contin-
uous function space V , i.e. V
h
V , are called conforming nite elements.
Finite elements methods where the discrete functions u
h
V
h
do not belong to V , i.e. V
h
V ,
are called non-conforming nite elements.
Note, that continuous Galerkin nite elements, as discussed in this section, are conforming nite
elements. In contrast to that, the discontinuous Galerkin nite elements being discussed in Section
4 onwards are non-conforming as the discontinuous discrete functions spaces are generally not
subspaces of the classic continuous function spaces V .
2.2.1 Consistency
From the weak formulation (35) we see that the discretization (38) is consistent, i.e. the exact
(weak) solution u V to (34) satises
B(u, v) = F(v) v V. (39)
From consistency we immediately deduce following important property of Galerkin nite element
methods, the so-called Galerkin-orthogonality:
B(u u
h
, v
h
) = 0 v
h
V
h
, (40)
which we obtain by subtracting (38) from (39) for v
h
V
h
V and using linearity of B
h
with
respect to its rst argument. This means that the error e = u u
h
is orthogonal (with respect to
the bilinear form B
h
) to the discrete function space V
h
.
2.2.2 Existence and uniqueness of discrete solutions
In Section 2.1 we have shown that the linear and bilinear functionals F : V R and B : V V R
in the weak problem: nd u such that
B(u, v) = F(v) v V (41)
are continuous. Furthermore, we have shown that B is V -coercive. We then applied the Lax-
Milgram theorem to show that a solution u to (41) exists and that this solution is unique. As
discussed above, standard nite elements are conforming nite elements, i.e. V
h
V . Therefore,
F and B are continuous also on V
h
. Furthermore, B is coercive also on V
h
. Again by using the
Lax-Milgram theorem we deduce that the discrete problem: nd u
h
V
h
such that
B(u
h
, v
h
) = F(v
h
) v
h
V
h
, (42)
has a unique solution, i.e. (42) is well-posed.
Remark 2.8 We see that, for all weak problems (41) where we are able to show well-posedness
by using the Lax-Milgram theorem we immediately obtain well-posedness of any discrete problem
(42) provided we have V
h
V , i.e. provided we use conforming nite elements. In contrast to
that, discontinuous Galerkin methods are non-conforming, V
h
V . Therefore, well-posedness of
discrete problems originating from discontinuous Galerkin discretizations is not immediate, but
must be shown for each discrete function space V
h
under consideration.
12
2.2.3 Best approximation property
Lemma 2.9 (Cea Lemma) Let the bilinear form B be continuous and V -coercive, with H
1
0
()
V H
1
(). Furthermore, let u V and u
h
V
h
V be the solutions to
B(u, v) = F(v), v V,
and
B(u
h
, v
h
) = F(v
h
), v
h
V
h
,
respectively. Then,
u u
h
H
1
()
C
B
inf
v
h
V
h
u v
h
H
1
()
. (43)
Proof: Let v
h
V
h
. First we use coercivity of B, then we use the Galerkin orthogonality, B(u
u
h
, w
h
) = 0 for w
h
= v
h
u
h
V
h
, and nally we use continuity of B to obtain
u u
h
2
H
1
()
B(u u
h
, u u
h
)
= B(u u
h
, u v
h
) +B(u u
h
, v
h
u
h
)
= B(u u
h
, u v
h
)
C
B
u u
h
H
1
()
u v
h
H
1
()
Dividing by u u
h
H
1
()
we obtain (43).
In (43) we see that apart from a constant the approximation error e = u u
h
is bounded by
the dierence v v
h
for any discrete function v
h
V
h
. This is the so-called best approximation
property. As we are free to choose any v
h
we can for example take an interpolation v
h
= I
h
u V
h
of u to obtain
u u
h
H
1
()
C
B
u I
h
u
H
1
()
. (44)
Hence, the discretization error e = u u
h
can be bounded by the interpolation error u I
h
u. In
particular, the order of the discretization is the same as the order of an interpolation into V
h
.
2.2.4 Interpolation estimates
In this section we recall some standard interpolation estimates.
Denition 2.10 (Interpolation operator I
c
h,p
onto V
c
h,p
) For p 1 let
i
, 0 i < N
h
, be a
nodal basis of V
c
h,p
with
i
(x
(j)
) =
ij
0 i, j < N
h
,
where x
(j)
, 0 j < N
h
, are the nodal points of V
c
h,p
and N
h
:= #(V
c
h,p
) is the dimension of the
discrete space V
c
h,p
, i.e. the number of degrees of freedom in V
c
h,p
. Given a function u H
2
(), we
dene I
c
h,p
u V
c
h,p
to be the interpolation of u onto V
c
h,p
given by
I
c
h,p
u(x) =
0i<N
h
u(x
i
)
i
(x). (45)
Possible nodal basis functions are the Lagrange interpolation polynomials
i
(x) := L
(p)
i
(x). In the
following we use the short notation I
h
u instead of I
c
h,p
u when it is clear that an interpolation into
V
c
h,p
is meant.
13
Theorem 2.11 (Interpolation estimate) Let T
h
be a shape regular mesh of R
d
, 1 d 3.
Let p 1 and I
h
be an interpolation operator onto V
c
h,p
. Furthermore, let 0 m p + 1. Then
there is a constant C, independent of h, such that for all u H
p+1
() we have
u I
h
u
H
m
()
Ch
p+1m
|u|
H
p+1
()
. (46)
Proof: See e.g. [40].
Example 2.12 In particular, for u H
p+1
() and m = 0, 1, the estimate (46) reduces to
u I
h
u
L
2
()
Ch
p+1
|u|
H
p+1
()
, (47)
and to
u I
h
u
H
1
()
Ch
p
|u|
H
p+1
()
. (48)
I.e. the interpolation error is of O(h
p+1
) in the L
2
()-norm and of O(h
p
) in the H
1
()-norm.
We note that Theorem 2.11 requires the solution u to be in H
p+1
(). However, for the case that
u H
s+1
() with s < p we obtain only a reduced interpolation order.
Corollary 2.13 (Interpolation estimate) Let T
h
be a shape regular mesh of R
d
, 1 d 3.
Let p 1 and I
h
be the interpolation operator onto V
c
h,p
as dened in Denition 2.10. Then there
is a constant C, independent of h, such that for all u H
s+1
() we have
u I
h
u
H
m
()
Ch
t+1m
|u|
H
t+1
()
. (49)
where t = min{s, p} and 0 m t + 1.
We note that for suciently smooth functions u H
p+1
(), i.e. s p, this estimate reduces to
(46), i.e. we have
u I
h
u
H
m
()
Ch
p+1m
|u|
H
p+1
()
, (50)
while for functions with a lower smoothness, i.e. u H
s+1
() with s < p, we have
u I
h
u
H
m
()
Ch
s+1m
|u|
H
s+1
()
, (51)
for m s + 1. The interpolation estimates given above can be combined with a trace theorem for
obtaining interpolation estimates in the L
2
()-norm:
Theorem 2.14 (Interpolation estimate in the L
2
()-norm) Let T
h
be a shape regular mesh
of R
d
, 1 d 3. Let p 1 and I
h
be the interpolation operator onto V
c
h,p
as dened in
Denition 2.10. Then there is a constant C, independent of h, such that for all u H
p+1
() we
have
u I
h
u
L
2
()
Ch
p+1/2
|u|
H
p+1
()
. (52)
2.2.5 A priori error estimates in the H
1
- and L
2
-norm
In this section we derive error estimates in the H
1
()- and the L
2
()-norm for the discretization
error of the standard nite element discretization of Poissons equation.
Corollary 2.15 (H
1
-error estimate) Let u H
p+1
() and u
h
V
h
= V
c
h,p
be the solutions to
(35) and (38), respectively. Then
u u
h
H
1
()
Ch
p
|u|
H
p+1
()
. (53)
14
Proof: Using (44) and (46) for m = 1 we obtain
u u
h
H
1
()
C
B
u I
h
u
H
1
()
Ch
p
|u|
H
p+1
()
.
H
1
()
Ch|u|
H
2
()
. (54)
In the following we derive an error estimate in the L
2
-norm by using a duality argument.
Theorem 2.17 (L
2
-error estimate (Aubin-Nitsche)) Let u H
p+1
() and u
h
V
h
= V
c
h,p
be the solutions to (35) and (38), respectively. Then
u u
h
L
2
()
Ch
p+1
|u|
H
p+1
()
. (55)
Proof: For simplicity, we assume homogeneous Dirichlet boundary conditions. For v L
2
() let
z be the solution to following dual (or adjoint) problem: nd z H
1
0
() such that
B(w, z) =
_
wv dx w H
1
0
().
We assume that z H
1
0
()H
2
() and z
H
2 Cv
L
2 which is satised if is a convex polygon,
for example. Now choosing v = w = e = u u
h
H
1
0
() yields
e
2
L
2
()
=
_
e
2
dx = B(e, z) = B(e, z z
h
) Cu u
h
H
1
()
z z
h
H
1
()
, (56)
where we used Galerkin-orthogonality B(e, z
h
) = 0 for any z
h
V
h
. Choosing z
h
= I
h
z V
h
and
using the interpolation estimate (51) for s = m = 1 we obtain
e
2
L
2
()
Cu u
h
H
1
()
hz
H
2
()
Chu u
h
H
1
()
e
L
2
()
. (57)
Division by e
L
2
()
, and using the H
1
-error estimate (53) we get (55).
Example 2.18 For u H
2
() and u
h
V
c
h,1
we obtain the standard result:
u u
h
L
2
()
Ch
2
|u|
H
2
()
. (58)
In summary, we have seen that the standard (continuous) nite element discretization of Pois-
sons equation based on piecewise polynomials of degree p is of order p + 1 in the L
2
()-norm
provided the exact solution u is suciently smooth.
15
3 Higher order continuous FE methods for the linear advection
equation
In this section we consider standard (continuous) nite element discretizations for the linear advec-
tion equation. In particular, we recall some standard results including a priori error estimates in
the L
2
-norm for the standard Galerkin method and in the H
1,b
-norm for the streamline diusion
nite element method (SDFEM).
3.1 The linear advection equation
For R
d
, d 1, we consider the linear advection equation
Lu := (bu) +cu = f in , u = g on
, (59)
where f L
2
(), b [C
1
()]
d
, c L
() and g L
2
(
), where
() and a number
0
> 0 such that
c(x) +
1
2
b(x) = c
2
0
(x)
0
> 0. (61)
This condition is required for ensuring stability below.
Remark 3.1 In order to demonstrate the similarities with the compressible Euler equations, see
Section 8, we consider the linear advection equation (59) in conservative form. We note, that
problem (59) is equivalent to the linear advection equation in non-conservative form
b u + cu = f (62)
with the hypothesis c
1
2
b = c
2
0
and c = c + b.
In the following we derive the variational formulation of the linear advection equation and dene
the proper function space the solution is to be sought in. To this end, we multiply (59) by a test
function v L
2
() and integrate over the domain ,
_
( (bu) +cu) v dx =
_
fv dx v L
2
(), (63)
where the function space the solution is to be searched in will be dened in the following.
First we note, that for the integral on the left hand side to exist we require (bu)+cu L
2
(),
i.e. we consider the function space
H
1,b
() = {u L
2
() : Lu = (bu) +cu L
2
()}. (64)
Then, it remains to incorporate boundary conditions in (63). There are two ways of doing so,
rst by a so-called strong and second by a so-called weak imposition of boundary conditions, which
both are discussed in the following two subsections.
16
3.1.1 Variational formulation with strong boundary conditions
We recall that for the Dirichlet problem of Poissons equation we have used H
1
0
() instead of
H
1
() for realizing Dirichlet boundary conditions on = . For the linear advection equation
(59) imposition of boundary conditions is allowed only on the inow boundary part
of the
boundary. Consequently, the function space to search the solution in is
H
1,b
() = {u L
2
() : Lu L
2
(), b nu = 0 on
} H
1,b
(). (65)
Then the variational formulation of the linear advection with homogeneous inow boundary con-
ditions, i.e. (59) with g 0 on
, is given by: nd u H
1,b
() such that
_
( (bu) +cu) v dx =
_
fv dx v L
2
(). (66)
The realization of inhomogeneous inow boundary conditions is similar to the imposition of inho-
mogeneous Dirichlet boundary conditions described for Poissons equation in Section 2.1.2. Assume
that there is a u
g
H
1,b
() with u
g
= g on
+u
g
with u
H
1,b
() such that
_
( (bu) +cu) v dx =
_
fv dx v L
2
(). (67)
In summary, strong boundary conditions are realized by considering an appropriate function (sub)space
H
1,b
() H
1,b
() which incorporates boundary conditions on
.
3.1.2 Variational formulation with weak boundary conditions
In the following we derive a variational formulation which imposes boundary conditions in a weak
sense. To this end, we multiply (59) by a test function v H
1,b
(), integrate over the domain ,
integrate by parts and replace u by g on
which gives
(bu) v dx +
_
cuv dx +
_
+
b nuv ds =
_
fv dx
_
b ngv ds,
where
+
= \
( (bu) +cu) v dx
_
b nuv ds =
_
fv dx
_
b ngv ds v H
1,b
(). (68)
Note, that here the boundary condition u = g on the inow boundary
is weakly imposed.
Furthermore, here u is sought in the (full) function space H
1,b
() which is in contrast to u
H
1,b
() in the case of strongly imposed boundary conditions in Section 3.1.1. The transport
equation (59) has a unique weak solution u H
1,b
() given by (68) and the boundary condition
is satised as an equality in L
2
(
), see [34].
3.2 The standard Galerkin method with weak boundary conditions
Starting from the variational formulation (68) the standard Galerkin method with weak boundary
conditions is given as follows: nd u
h
V
h
:= V
c
h,p
such that
B(u
h
, v
h
) = F(v
h
) v V
h
, (69)
17
where
B(u, v) =
_
( (bu) +cu) v dx
_
b nuv ds,
F(v) =
_
fv dx
_
b ngv ds.
(70)
From (68) we see that the discretization is consistent, i.e.
B(u, v) = F(v), v V, (71)
holds for the exact solution u. By subtracting (69) from (71) for v
h
V
h
V we obtain the
Galerkin orthogonality
B(u u
h
, v
h
) = 0, v V
h
. (72)
Lemma 3.2 For any v H
1,b
() we have
B(v, v)
0
v
2
L
2
()
+
1
2
_
|b n|v
2
ds. (73)
Proof: From (70) we have
B(v, v) =
_
(bv) v +cv
2
dx
_
b nv
2
ds.
Noting that
(bv) v = ( b) v
2
+ (b v) v = ( b) v
2
+
1
2
b v
2
= ( b) v
2
+
1
2
_
bv
2
_
1
2
( b) v
2
=
1
2
( b) v
2
+
1
2
_
bv
2
_
,
(74)
we obtain by using the divergence theorem and hypothesis (61)
B(v, v) =
_
_
1
2
b +c
_
v
2
dx +
1
2
_
_
bv
2
_
dx
_
b nv
2
ds
=
_
c
2
0
v
2
dx +
1
2
_
b nv
2
ds
_
b nv
2
ds
0
_
v
2
dx +
1
2
_
+
b nv
2
ds
1
2
_
b nv
2
ds
=
0
_
v
2
dx +
1
2
_
|b n|v
2
ds,
(75)
as b n < 0 on
and b n 0 on
+
.
Hence B(, ) is coercive with respect to
v
2
L
2
()
+
_
|b n|v
2
ds. (76)
However, this is not a norm on H
1,b
(). In fact, there are functions v L
2
() with v
2
L
2
()
+
_
|b
n|v
2
ds < but with unbounded directional derivatives, i.e. b v = , hence v H
1,b
().
We note, that the lack of H
1,b
-coercivity of the standard Galerkin method for the linear ad-
vection equation has direct consequences on the stability of the method. In fact, it is well known
that the standard Galerkin method for the linear advection is unstable. This instability may lead
to highly oscillating numerical solutions.
18
Theorem 3.3 For p 1 let u H
p+1
() be the solution to (59) and u
h
V
c
h,p
the solution to
(69). Then,
u u
h
L
2
()
+
__
|b n| |u u
h
|
2
ds
_
1/2
Ch
p
|u|
H
p+1
()
. (77)
Proof: Let e = u u
h
= with = u I
h
u and = u
h
I
h
u where I
h
: H
2
() V
c
h,p
is the
interpolation operator as dened in Denition 2.10. Then by using triangle inequality
e
2
+
_
|b n| |e|
2
ds
2
+
2
+
_
|b n| ||
2
ds +
_
|b n| ||
2
ds, (78)
where we write =
L
2
()
for short. For the rst term and third term on the right hand
side we can use interpolation estimates given in Section 2.2.4. Furthermore, by using (73) we can
bound the second and fourth term in (78) as follows
2
+
1
2
_
|b n|
2
ds B(, ) = B( e, ) = B(, ) B(e, ) = B(, ), (79)
where we used the Galerkin orthogonality property B(e, ) = 0 for V
c
h,p
, see (72). Using the
Denition (70) of B(, ), and the Youngs inequality, ab
4
a
2
+
1
b
2
, we obtain
2
+
1
2
_
|b n|
2
ds B(, ) =
_
( (b) +c) dx
_
b n ds
0
4
2
+
1
0
b
2
+
0
4
2
+
c
2
2
+
1
4
_
|b n|
2
ds +
_
|b n|
2
ds,
where c = b +c
L
()
. By subtracting all terms of the right hand side we obtain
0
2
2
+
1
4
_
|b n|
2
ds C
_
b
2
+
2
+
_
|b n|
2
ds
_
.
Hence, together with (78) we have
e
2
+
_
|b n| |e|
2
ds C
_
b
2
+
2
+
_
|b n|
2
ds
_
. (80)
Using the interpolation estimates (47), (48) and (52):
L
2
()
Ch
p+1
|u|
H
p+1
()
,
b
L
2
()
C
H
1
()
Ch
p
|u|
H
p+1
()
,
__
|b n|
2
ds
_
1/2
C
L
2
()
Ch
p+1/2
|u|
H
p+1
()
,
(81)
and using (a +b)
2
2(a
2
+b
2
) we nally obtain (77).
The estimate (77) shows that if the exact solution u to problem (59) happens to be smooth so
that u
H
p+1
()
is nite, then the standard Galerkin method (69) will converge at the rate O(h
p
).
Although this rate is one power of h from being optimal, it shows that the standard Galerkin
method will perform satisfactorily in this case. However, in general u will not be smooth and in
this case the standard Galerkin method gives poor results. In fact, for u H
s+1
() with s < p
estimate (77) reduces to
u u
h
L
2
()
+
__
|b n| |u u
h
|
2
ds
_
1/2
Ch
s
|u|
H
s+1
()
. (82)
In particular, for u H
1
() the solution to the standard Galerkin method does not converge as
h 0. Instead, the solution might become oscillatory as there is no control on b u
h
.
19
3.3 The streamline diusion method with weak boundary conditions
We have seen in the previous section that the standard Galerkin discretization for the linear ad-
vection equation is unstable. For obtaining a stable discretization scheme we need to add some
articial diusion to the scheme. However, we do not require a stabilizing eect in all directions.
As we will see in the following, for obtaining a stable scheme it is sucient to add diusion in
streamline direction, only. Starting from the variational formulation (68) we replace u by a discrete
function u
h
V
h
, and the test function v by v
h
+hb v
h
on with > 0, and by v
h
on . Then,
the streamline diusion method is given by
B
h
(u
h
, v
h
) = F
h
(v
h
), v
h
V
h
, (83)
where
B
h
(u, v) =
_
b nuv ds,
F
h
(v) =
_
f (v +hb v) dx
_
b ngv ds.
(84)
We note, that here trial and test functions u
h
, v
h
are taken from dierent discrete function spaces,
u
h
V
h
and v
h
V
h
:= {v
h
= w
h
+ hb w
h
, with w
h
V
h
},
V
h
= V
h
+ hb V
h
. Thereby,
this is a so-called Petrov-Galerkin discretization which is in contrast to the standard Galerkin
discretizations discussed so far, where the ansatz and test functions are taken from the same
discrete function space u
h
, v
h
V
h
. First, we note that (83) is consistent, i.e. we have
B
h
(u, v) = F
h
(v) v V,
for the exact solution u. Furthermore, we see that
B
h
(u, v) = B(u, v) +
_
(b u)(b v) dx (86)
which represents articial viscosity (diusion) in streamline direction b. In fact, integrating by
parts we see, that this term corresponds to
2
b
u where
b
u = b u.
In the following we show that B
h
is H
1,b
-coercive.
Lemma 3.4 Let b and c be constant and hc
1
2
. Then there is a C > 0 such that for all
v H
1,b
() we have
B
h
(v, v)
_
hb v
2
L
2
()
+v
2
L
2
()
+
_
|b n|v
2
ds
_
. (87)
Proof: Using b = 0 and c = const we have
_
cv(b v) dx =
1
2
c
_
_
bv
2
_
dx =
1
2
c
_
b nv
2
ds.
20
Then, using (85), (73) and hc
1
2
we obtain
B
h
(v, v) = B(v, v) +hb v
2
L
2
()
+
1
2
hc
_
b nv
2
ds
0
v
2
L
2
()
+
1
2
_
|b n|v
2
ds +hb v
2
L
2
()
+
1
2
hc
_
b nv
2
ds
hb v
2
L
2
()
+
0
v
2
L
2
()
+
1
4
_
|b n|v
2
ds
_
hb v
2
L
2
()
+v
2
L
2
()
+
_
|b n|v
2
ds
_
,
(88)
where = min(,
0
, 1/4).
Motivated by (87) we dene following norm on H
1,b
():
v
H
1,b
()
=
_
hb v
2
L
2
()
+v
2
L
2
()
+
_
|b n|v
2
ds
_1
2
. (89)
Hence, with (87) we have shown H
1,b
-coercivity of B
h
(, ),
B
h
(v, v) v
2
H
1,b
()
.
Additionally, by using Cauchy-Schwarzs inequality we nd that F
h
() is continuous,
|F
h
(v)| |
_
fv dx| +h|
_
fb v dx| +|
_
b ngv ds|
f
L
2
()
v
L
2
()
+hf
L
2
()
b v
L
2
()
+Cg
L
2
()
__
|b n|v
2
ds
_
1
2
C
F
v
H
1,b
()
.
We can then deduce stability of the method as follows
u
h
2
H
1,b
()
B
h
(u
h
, u
h
) = F
h
(u
h
) C
F
u
h
H
1,b
()
. (90)
Hence, we have
hb u
h
2
L
2
()
+u
h
2
L
2
()
+
_
|b n|u
2
h
ds = u
h
2
H
1,b
()
_
C
F
_
2
, (91)
i.e. we have control of u
h
and b u
h
. In fact, the streamline diusion method is stable.
Remark 3.5 We recall, see Section 3.2, that the standard Galerkin method for the linear advection
equation oers no control of b u
h
which might lead to highly oscillating numerical solutions,
hence an unstable scheme. Only by adding diusion in streamline direction we gain control of
b u
h
and a stable scheme. We will show later, that in contrast to that, the discontinuous
Galerkin discretizations of the linear advection equation is stable without streamline diusion.
Theorem 3.6 Let u H
p+1
() be the solution to (59) with constant b and c. Furthermore, let
u
h
V
c
h,p
the solution to (83). Then,
u u
h
H
1,b
()
Ch
p+1/2
|u|
H
p+1
()
. (92)
21
Proof: Let e = u u
h
= with = u I
h
u and = u
h
I
h
u where I
h
: H
2
() V
c
h,p
is the
interpolation operator as dened in Denition 2.10. Then
e
2
H
1,b
()
B
h
(e, e) = B
h
(e, ) = B
h
(e, ), (93)
where we used Galerkin orthogonality property B
h
(e, ) = 0 for V
c
h,p
. Using the denition of
B
h
(, ), b = 0 and the inequality ab
4
a
2
+
1
b
2
we obtain
e
2
H
1,b
()
B
h
(e, ) =
_
b ne ds
=
_
b ne ds
h
4
b e
2
+
1
h
2
+
h
4
b e
2
+
2
h
b
2
+
4
e
2
+
c
2
2
+
4
e
2
+
c
2
h
2
b
2
+
4
_
|b n| e
2
ds +
1
|b n|
2
ds
2
e
2
H
1,b
()
+C
_
h
1
2
+hb
2
+
_
|b n|
2
ds
_
.
Subtracting
2
e
2
H
1,b
()
on both sides and multiplying with
2
we obtain
e
2
H
1,b
()
C
_
h
1
2
+hb
2
+
_
|b n|
2
ds
_
. (94)
Using the interpolation estimates (81) we obtain (92).
We see, that the streamline diusion discretization is of order O(h
p+1/2
) which is half an order
higher than the O(h
p
) obtained for the standard Galerkin discretization, see (77). Due to the
streamline diusion term (86), the streamline diusion discretization is coercive with respect to
the H
1,b
-norm which includes the hb v
2
L
2
()
term, see (87) and (89). Thereby, in the proof
of Theorem 3.6 we could subtract/hide the term
h
4
b e
2
from/in the left hand side term
e
2
H
1,b
()
. The remaining term h
1/2
b is O(h
p+1/2
).
In contrast to that the standard Galerkin discretization is coercive with respect to a weaker
norm which does not include b v
2
, see (73). Thereby, in the proof of Theorem 3.3 the b
2
term could not be hidden in the left hand side and is nally estimated as O(h
p
).
Remark 3.7 Note, that there are many possibilities known as how to stabilize the standard Galerkin
method of the linear advection equation: e.g. by using residual-free bubbles [12, 14], or subgrid mod-
eling [21], among many others. However, for the purpose of this lecture it is sucient to have error
estimation results available for the classic SDFEM, only.
22
v
+
wrt. element .
4 Higher order DG discretizations of the linear advection equation
4.1 Mesh related function spaces
We begin by introducing some notation. As before, we assume that the domain can be subdivided
into shape regular meshes T
h
= {} consisting of elements . Here, h denotes the piecewise constant
mesh function dened by h|
H
m
(), T
h
}. (95)
Denition 4.2 (Interior faces:
I
) Let and
. We dene
I
to be the union of all interior faces of T
h
.
Denition 4.3 (Traces v
+
, v
H
1
() for each T
h
. By v
H
1
() the trace v
+
belongs to L
2
(),
and traces of v H
1
(T
h
) belong to T(T
h
) :=
T
h
L
2
().
Finally, we dene mesh related (or broken) gradient, divergence and Laplace operators.
Denition 4.4 (
h
,
h
and
h
) We dene broken operators by restriction to each element
T
h
as follows:
The broken gradient operator
h
: H
1
(T
h
) [L
2
(T
h
)]
d
is dened by
(
h
v)|
:= (v|
), T
h
, (96)
for v H
1
(T
h
), where (v)
i
=
x
i
v, i = 1, . . . , d.
The broken divergence operator
h
: [H
1
(T
h
)]
d
L
2
(T
h
) is dened by
(
h
)|
= (|
), T
h
, (97)
for [H
1
(T
h
)]
d
, where =
1id
x
i
i
.
Finally, the broken Laplace operator
h
: H
2
(T
h
) L
2
(T
h
) is dened by
(
h
u)|
:= (u|
), T
h
, (98)
for u H
2
(T
h
), where u = u =
1id
2
x
i
u.
23
4.2 A variational formulation of the linear advection equation
Like in Section 3.1 here we consider the linear advection equation
Lu := (bu) +cu = f in , u = g on
, (99)
where f L
2
(), b [C
1
()]
d
, c L
() and g L
2
(
), where
() and a number
0
> 0 such that
c(x) +
1
2
b(x) = c
2
0
(x)
0
> 0. (101)
In the following we derive a variational formulation for u H
1,b
(T
h
) where
H
1,b
(T
h
) = {u L
2
() : Lu = (bu) +cu|
L
2
(), T
h
}, (102)
is a broken space which is the mesh related (broken) counterpart of the function space H
1,b
()
dened in (64). Note, that H
1,b
() H
1,b
(T
h
). Given an element T
h
, we multiply (99) by a
test function v H
1,b
(T
h
), integrate over
_
( (bu) +cu) v dx =
_
fv dx,
and integrate by parts
(bu) v dx +
_
cuv dx +
_
b nuv ds =
_
fv dx. (103)
We sum over all elements T
h
and replace u on
(bu)
h
v dx +
_
cuv dx +
T
h
_
\
b nuv ds +
_
+
b nuv ds
=
_
fv dx
_
b ngv ds v H
1,b
(T
h
).
As functions u H
1,b
(T
h
) may be discontinuous across edges e =
between neighboring
elements and
, n), where u
+
:= u
+
and u
:= u
, respectively, are the interior and exterior traces of u on . Thereby, the variational
formulation of (99) is given by: nd u H
1,b
(T
h
) such that
B
h
(u, v) = F(v) v H
1,b
(T
h
), (104)
where
B
h
(u, v) =
_
(bu)
h
v dx +
_
cuv dx +
T
h
_
\
H(u
+
, u
, n)v ds +
_
+
b nuv ds,
F(v) =
_
fv dx
_
, n) is said to be consistent if
H(u, u, n) = b nu. (106)
Furthermore, H(u
+
, u
, n) is said to be conservative if
H(u
+
, u
, n) = H(u
, u
+
, n). (107)
Lemma 4.6 (Consistency) Let problem (99) be discretized based on (104) and (105). Then, the
discretization is consistent, i.e. the exact solution u H
1,b
() H
1,b
(T
h
) to (99) satises
B
h
(u, v) = F(v) v H
1,b
(T
h
), (108)
if and only if the numerical ux function H is consistent, i.e.
H(u, u, n) = b nu. (109)
Proof: After integrating by parts (104), and rearranging terms we see that (104) is equivalent to
_
(f
h
(bu) cu) v dx +
T
h
_
\
_
b nu
+
H(u
+
, u
, n)
_
v ds
b n
_
g u
+
_
v ds = 0 v H
1,b
(T
h
). (110)
For the exact (and smooth) solution u H
1,b
() to problem (99), the rst and third term in (110)
vanishes. Thereby, we obtain
T
h
_
\
(b nu H(u, u, n)) v ds = 0 v H
1,b
(T
h
), (111)
hence we have consistency if and only if H(u, u, n) = b nu.
Lemma 4.7 (Global conservation property) Let problem (99) with c 0 be discretized based
on (104) and (105). Then, the discretization is conservative, i.e.
_
b ng ds +
_
+
b nu ds =
_
f dx, (112)
if and only if the numerical ux function H is conservative, i.e.
H(u
+
, u
, n) = H(u
+
, u
, n).
Proof: By setting v 1 in (104) with c 0, we obtain
T
h
_
\
H(u
+
, u
, n) ds +
_
+
b nu ds =
_
f dx
_
b ng ds.
Then we note that in the sum
between neighboring
elements and
. Thereby writing in
terms of interior faces e
I
we obtain
e
I
_
e
H(u
+
, u
, n) + H(u
, u
+
, n) ds +
_
b ng ds +
_
+
b nuds =
_
f dx.
Hence we see that (112) holds if and only if the numerical ux function is conservative.
In the following we introduce two dierent numerical ux functions which are both consistent
and conservative.
25
The mean value ux First we dene the mean value ux (also named central ux in nite
volume schemes) as follows
H
mv
(u
+
, u
, n) = b n{u}, (113)
where
{u} =
1
2
_
u
+
+u
_
(114)
denotes the mean value of u
+
and u
, n) =
_
b nu
,
b nu
+
, for (b n)(x) 0, i.e. x
+
,
, (115)
where
and
+
are the inow and outow boundaries of element dened as follows
+
= {x , b(x) n(x) 0} = \
.
(116)
This ux always takes the value from upstream (upwind) direction. This numerical ux is consistent
and conservative. Additionally, as we will show later, a discretization based on this ux is stable.
Generic ux The mean value ux and the upwind ux can be written as follows
H
b
0
(u
+
, u
, n) = b n{u} +b
0
[u], (117)
where
[u] = u
+
u
(118)
denotes the (simple) jump of u. By setting b
0
= 0 the generic ux (117) reduces to the mean value
ux (113) and by setting b
0
=
1
2
|b n| we obtain the upwind ux (115).
Theorem 4.8 (Coercivity) Let B
h
(, ) be given by
B
h
(u, v) =
_
(bu)
h
v dx +
_
cuv dx
+
T
h
_
\
H
b
0
(u
+
, u
, n)v ds +
_
+
b nuv ds,
(119)
where H
b
0
as dened in (117) represents the mean value ux or the upwind ux depending on b
0
= 0
or b
0
=
1
2
|b n| respectively. Then for all v H
1,b
(T
h
) we have
B
h
(v, v) = c
0
v
2
+
e
I
_
e
b
0
[v]
2
ds +
1
2
_
|b n| v
2
ds.
(120)
Proof: First we rewrite
(bv) v dx =
1
2
_
b v
2
dx
=
1
2
_
_
bv
2
_
dx +
1
2
_
bv
2
dx
=
1
2
_
b n(v
+
)
2
ds +
1
2
_
bv
2
dx.
26
Furthermore, we have
1
2
T
h
_
b n(v
+
)
2
ds +
T
h
_
\
b n{v} v ds
=
1
2
e
I
_
e
b n
+
_
(v
+
)
2
(v
)
2
_
ds +
e
I
b n
+
1
2
(v
+
+v
)(v
+
v
) ds = 0.
Thereby,
B
h
(v, v) =
_
(bv)
h
v dx +
_
cv
2
dx +
T
h
_
\
(b n{v} +b
0
[v]) v ds +
_
+
b nv
2
ds
=
_
_
c +
1
2
b
_
v
2
dx +
T
h
_
\
b
0
[v]v ds +
_
+
b nv
2
ds
1
2
_
b nv
2
ds
=
_
c
2
0
v
2
dx +
e
I
b
0
[v][v] ds +
1
2
_
+
b nv
2
ds
1
2
_
b nv
2
ds
= c
0
v
2
+
e
I
_
e
b
0
[v]
2
ds +
1
2
_
|b n| v
2
ds,
where we used hypothesis (101). Hence, we have shown (120).
Denition 4.9 Motivated by the coercivity (120) we dene the DG norm | |
b
0
by
|v|
2
b
0
= c
0
v
2
+
e
I
_
e
b
0
[v]
2
ds +
1
2
_
|b n| v
2
ds (121)
From the coercivity of B
h
, (120), we immediately obtain the stability of the discontinuous Galerkin
discretization in the DG norm | |
b
0
as follows:
|v|
2
b
0
= B
h
(v, v) = F(v) =
_
fv dx +
_
|b n| gv ds
f
L
2
()
v
L
2
()
+
__
|b n| g
2
ds
_
1/2
__
|b n| v
2
ds
_
1/2
C|v|
b
0
,
for f L
2
() and g L
2
(
e
I
_
e
b
0
[v]
2
ds +
1
2
_
|b n| v
2
ds C
2
. (122)
Remark 4.10 We note, that the discontinuous Galerkin discretization based on the upwind ux,
i.e. b
0
=
1
2
|b n|, has an improved stability as compared to the discretization based on the mean
value ux where b
0
= 0. In fact, from (122) we can see that we have control of
e
I
_
e
[v]
2
ds for
b
0
= 0 which we do not have for b
0
= 0. As we will see later, this translates to a reduced order of
convergence for b
0
= 0 as compared to the case b
0
=
1
2
|b n|. In fact, the DG discretization based
on the upwind ux turns out to be stable whereas the DG discretization based on the mean value
ux is unstable.
This behavior corresponds to the dierence in terms of stability and order of convergence of
the standard (continuous) Galerkin method in comparison to the streamline diusion method as
discussed in Sections 3.2 and 3.3, respectively.
27
4.4 The discontinuous Galerkin discretization
Denition 4.11 For p 0 we dene the space of discontinuous piecewise polynomials of degree p:
V
d
h,p
= {v
h
L
2
() :v
h
|
Q
p
( ) if is the unit hypercube, and
v
h
|
P
p
( ) if is the unit simplex, T
h
},
(123)
where P
p
and Q
p
are the spaces of polynomials and tensor product polynomials of degree p.
Remark 4.12 Note that V
h
:= V
d
h,p
H
m
(T
h
) H
1
(T
h
) H
1,b
(T
h
), m > 1, and u H
1,b
()
H
1,b
(T
h
) but V
h
H
1,b
(), i.e. DG methods are non-conforming nite element methods.
For obtaining the DG discretization of the linear advection equation (99) we now replace the
functions u, v H
1,b
(T
h
) in (104) by discrete functions u
h
, v
h
V
h
: nd u
h
V
h
such that
B
h
(u
h
, v
h
) = F(v
h
) v
h
V
h
, (124)
where
B
h
(u, v) =
_
(bu)
h
v dx +
_
cuv dx +
T
h
_
\
H(u
+
, u
, n)v ds +
_
+
b nuv ds,
F(v) =
_
fv dx
_
, n) = b n{u} +b
0
[u], (126)
which reduces to the mean value ux H
mv
for b
0
= 0 and the upwind ux H
uw
for b
0
=
1
2
|b n|.
Existence and uniqueness of a discrete solution Writing the discrete function u
h
(x) =
0j<N
h
u
j
j
(x), in terms of the basis functions
j
V
h
, 0 j < N
h
, where N
h
= #V
h
, and
varying v
h
over all basis functions, v
h
=
i
, 0 i < N
h
, problem (124) can be rewritten as a linear
system
Au = b, (127)
where A R
N
h
N
h
with A
ij
= B
h
(
j
,
i
), 0 i, j < N
h
, b R
N
h
with b
i
= F(
i
), 0 i < N
h
,
and u R
N
h
. First we note, that from the coercivity (120) of B
h
we have
i
A
ij
j
=
i
B
h
(
j
,
i
)
j
= B
h
(
j
j
,
i
i
) = B
h
(, ) c
0
2
0
2
,
with a constant
0
> 0, i.e. A is positive denite. Given two solution vectors u
1
and u
2
, with
Au
1
= b and Au
2
= b, respectively, we obtain Au
1
Au
2
= 0 and 0 = (u
1
u
2
)
i
A
ij
(u
1
u
2
)
j
,
thus u
1
= u
2
. Hence the linear mapping u Au is injective. We then use following
Lemma 4.13 For a linear mapping A : U W of nite dimensional spaces with dimU = dimW
following properties are equivalent: A is injective, A is surjective, and A is bijective.
Proof: Any introductory book to linear algebra
We conclude that u Au is bijective. Hence there is a unique discrete solution u
h
to (124).
28
Consistency and Galerkin orthogonality: According to Lemma 4.6 and due to the consis-
tency of the numerical ux H we conclude that this discretization is consistent, i.e. the exact
solution u H
1,b
() to (99) satises
B
h
(u, v) = F(v) v H
1,b
(T
h
). (128)
Again we obtain the Galerkin orthogonality
B
h
(u u
h
, v
h
) = 0 v V
h
, (129)
Before deriving a priori error estimates for the discontinuous Galerkin discretization (124), we
give the denition of the L
2
-projector onto V
d
h,p
together with some approximation estimates.
4.5 The local L
2
-projection and approximation estimates
Denition 4.14 (Local L
2
-projection) Let p 0 and V
d
h,p
be the discontinuous nite element
space dened in (123). Then, by P
d
h,p
we denote the L
2
-projection onto V
d
h,p
, i.e. given a u L
2
()
we dene P
d
h,p
u V
d
h,p
by
_
_
u P
d
h,p
u
_
v
h
dx = 0 v
h
V
d
h,p
. (130)
We use the short notation P
h
u instead of P
d
h,p
u when it is clear which projection is meant.
Given a T
h
, we set v
h
0 on
T
h
with
(u P
h
u) v
h
dx = 0 v
h
V
d
h,p
. (131)
P
h
restricted to T
h
is an L
2
()-projection which is why P
h
is also called local L
2
-projection.
Remark 4.15 We note that P
d
h,p
u may in fact be discontinuous from element to element. Fur-
thermore, the approximation error estimates for the L
2
-projection are similar to the approximation
estimates for the interpolation I
c
h,p
dened in Section 2.2.4. However, the L
2
-projection additionally
oers the local projection property given in (131). This property will be used when deriving a priori
error estimates for the discontinuous Galerkin discretization. We note, that in continuous nite el-
ement methods we used the interpolation operator I
c
h,p
for ensuring global continuity of I
c
h,p
u V
c
h,p
.
There, we could have also used a global L
2
-projection P
c
h,p
onto the continuous discrete functions
with P
c
h,p
u V
c
h,p
. However, this operator does not have a local projection property.
Analogous to the interpolation estimates in Section 2.2.4 we have following approximation
estimates for the L
2
-projection:
Corollary 4.16 (Local approximation estimates for the L
2
-projection) Let p 0 and P
h
u :=
P
d
h,p
u be the L
2
-projection dened in Denition 4.14. Suppose u|
in H
s+1
(), s
0, for T
h
.
Then
u P
h
u
H
m
()
Ch
t+1m
|u|
H
t+1
()
, (132)
where t
= min(s
, p), T
h
.
Again, for suciently smooth functions u H
p+1
(), i.e. s
|u|
H
p+1
()
, (133)
29
while for functions with a lower smoothness, i.e. u H
s+1
() with s
< p, we have
u P
h
u
H
m
()
Ch
s+1m
|u|
H
s+1
()
, (134)
for m s
-norm:
u P
h
u
L
()
Ch
p+1
|u|
H
p+1,
()
, (135)
Furthermore, analogous to Theorem 2.14 we have
u P
h
u
L
2
()
Ch
p+1/2
|u|
H
p+1
()
. (136)
4.6 A priori error estimates
Theorem 4.17 (A priori error estimate, [13]) Let u H
p+1
() be the exact solution to the
linear advection equation (99). Furthermore, let u
h
V
d
h,p
be the solution to
B
h
(u
h
, v
h
) = F(v
h
), v
h
V
d
h,p
,
where
B
h
(u, v) =
_
(bu)
h
v dx +
_
cuv dx
+
T
h
_
\
(b n{u} +b
0
[u]) v ds +
_
+
b nuv ds,
F(v) =
_
fv dx
_
b ngv ds.
Then, for b
0
=
1
2
|b n|, i.e. when using the upwind ux, we have
|u u
h
|
b
0
Ch
p+1/2
|u|
H
p+1
()
, (137)
and for b
0
= 0, i.e. when using the mean value ux, we have
|u u
h
|
b
0
Ch
p
|u|
H
p+1
()
, (138)
where
|v|
2
b
0
= c
0
v
2
+
e
I
_
e
b
0
[v]
2
ds +
1
2
_
|b n| v
2
ds (139)
is the DG norm as dened in Denition 4.9.
Proof: Let e = u u
h
= with = u P
h
u and = u
h
P
h
u where P
h
:= P
d
h,p
is the
L
2
-projection onto V
h
:= V
d
h,p
as dened in Denition (4.14). Then, by triangle inequality,
|e|
b
0
||
b
0
+||
b
0
. (140)
For the rst term we use approximation results for P
h
u:
c
0
L
2
()
Cu P
h
u
L
2
()
Ch
p+1
|u|
H
p+1
()
,
_
_
e
I
_
e
b
0
[]
2
ds
_
_
1/2
C
e
I
u P
h
u
L
2
(e)
Ch
p+1/2
|u|
H
p+1
()
,
_
1
2
_
|b n|
2
ds
_
1/2
Cu P
h
u
L
2
()
Ch
p+1/2
|u|
H
p+1
()
,
(141)
30
see (133) for m = 0 and (136), and hence
||
b
0
h
p+1/2
|u|
H
p+1
()
. (142)
The second term in (140) we rewrite as follows
||
2
b
0
= B
h
(, ) = B
h
( e, ) = B
h
(, ), (143)
where we used coercivity (120) of B
h
and the Galerkin orthogonality property B
h
(e, ) = 0 for
V
h
. Using the denition of B
h
(, ) we obtain
||
2
b
0
=
_
(b
h
+c) dx +
T
h
_
\
(b n{} +b
0
[]) ds +
_
+
b n ds.
Next observe that
h
V
h
, so that, by the denition of the projector P
d
h,0
,
_
_
P
d
h,0
b
_
dx = 0.
Using this, together with the approximation estimate (135) for p = 0, the Cauchy-Schwarz inequal-
ity, the inverse inequality
H
1
()
Ch
m1
H
m
()
for V
h
, and the approximation estimate
(133) for m = 0, we deduce that
_
(b
h
+c) dx =
T
h
_
(b +c) dx
=
T
h
_
__
P
d
h,0
b b
_
+c
_
dx
C
L
2
()
_
_
T
h
_
P
0
h
b b
L
()
||
H
1
()
+
L
2
()
_
2
_
_
1
2
C
L
2
()
_
_
T
h
_
h
|b|
H
1,
()
h
1
L
2
()
+
L
2
()
_
2
_
_
1
2
C
L
2
()
L
2
()
Ch
p+1
|u|
H
p+1
()
||
b
0
. (144)
Furthermore, using Cauchy-Schwarz inequality and approximation estimate (136) we nd
_
+
b n ds
__
+
|b n|
2
ds
_
1
2
__
+
|b n|
2
ds
_
1
2
Ch
p+1/2
|u|
H
p+1
()
||
b
0
.
(145)
Finally, we have
T
h
_
\
(b n{} +b
0
[]) ds =
e
I
_
e
(b n{} +b
0
[]) [] ds.
31
If b
0
=
1
2
|b n| we have b n |b n| = 2b
0
and obtain
_
e
(b n{} +b
0
[]) [] ds
_
e
b
1/2
0
(2{} + []) b
1/2
0
|[]| ds
__
e
b
0
(2{} + [])
2
ds
_
1
2
__
e
b
0
[]
2
ds
_
1
2
Ch
p+1/2
|u|
H
p+1
()
__
e
b
0
[]
2
ds
_
1
2
,
and hence
e
I
_
e
(b n{} +b
0
[]) [] ds Ch
p+1/2
|u|
H
p+1
()
_
_
e
I
_
e
b
0
[]
2
ds
_
_
1
2
Ch
p+1/2
|u|
H
p+1
()
||
b
0
.
(146)
However, if b
0
= 0, the norm ||
b
0
does not include
_
e
[]
2
ds which is why we cannot bound
_
e
b n{}[] ds in terms of ||
b
0
. Thereby, we are forced to use the inverse inequality
[]
L
2
(e)
Ch
1
2
L
2
()
,
to bound the L
2
(e)-norm by the L
2
()-norm. Hence, instead of (146) we obtain
_
e
b n{}[] ds C[]
L
2
(e)
[]
L
2
(e)
Ch
p+1/2
|u|
H
p+1
()
h
1
2
L
2
()
,
and hence,
e
I
_
e
b n{}[] ds Ch
p
|u|
H
p+1
()
L
2
()
Ch
p
|u|
H
p+1
()
||
b
0
.
(147)
Combining (144), (145) and (146) we obtain (137), while (144), (145) and (147) gives (138).
From the proof above we see that it is essential whether the term
e
I
_
e
b
0
[v]
2
ds is included in
the DG norm with b
0
= 0 or with b
0
= 0. In the former case we can bound the interior face terms
e
I
_
e
(b n{} +b
0
[]) [] ds (148)
in terms of ||
b
0
whereas in the latter case we are forced to use the inverse inequality due to which
we loose half an order of h. Recalling the discussion in Remark 4.10 we see that the stability of the
discretization, in particular of the interior face terms, is connected to the order of convergence. For
suciently smooth solutions, u H
p+1
(), the discretization based on the upwind ux is of order
O(h
p+1/2
) and the discretization based on the mean value ux is of the order O(h
p
). In contrast
to that the order of convergence will be reduced for solutions with a lower smoothness. In fact, for
u H
s+1
() with s < p the estimates (137) and (138) are replaced by the estimates
|u u
h
|
b
0
Ch
s+1/2
|u|
H
s+1
()
, (149)
for the upwind ux, and by
|u u
h
|
b
0
Ch
s
|u|
H
s+1
()
, (150)
32
for the mean value ux, respectively. In particular, for u H
1
() we see that the discontinuous
Galerkin solution based on the mean value ux does not converge under h 0. In fact, for
u H
1
(), the discretization based on the mean value ux is unstable, whereas the upwind ux
yields a stable discretization.
We recall that the dierence in stability and the order of convergence of the discontinuous
Galerkin discretization based on the mean value as compared to the upwind ux corresponds to
the dierence in the standard (continuous) Galerkin method as compared to the streamline diusion
method discussed in Sections 3.2 and 3.3, respectively.
Finally, we note that estimate (137) which is suboptimal by h
1/2
as compared to the O(h
p+1
)
approximation order of V
d
h,p
, Peterson [37] conrmed by considering socalled Peterson meshes,
that O(h
p+1/2
) is actually a sharp estimate.
4.7 The discontinuous Galerkin discretization based on upwind
As shown in the previous section the discontinuous Galerkin discretization for the linear advection
equation based on the upwind ux is stable whereas that based on the mean value ux is unstable.
Therefore, in this subsection we concentrate on the discretization based on the upwind ux while
ignoring the discretization based on the mean value ux.
First, we recall the discontinuous Galerkin discretization (124) for the linear advection: nd
u
h
V
d
h,p
such that
(bu
h
)
h
v
h
dx +
_
cu
h
v
h
dx +
T
h
_
\
H(u
+
h
, u
h
, n)v
+
h
ds +
_
+
b nu
h
v
h
ds
=
_
fv
h
dx
_
b ngv
h
ds v
h
V
d
h,p
.
As shown in the previous sections the numerical ux function must be consistent and conservative,
see Denition 4.5. The upwind ux
H
uw
(u
+
, u
, n) =
_
b nu
,
b nu
+
, for (b n)(x) 0, i.e. x
+
,
is consistent and conservative and we have shown in previous sections that it yields a stable scheme.
Using this ux the discontinuous Galerkin discretization of the linear advection equation is given
by: nd u
h
V
d
h,p
such that
(bu
h
)
h
v
h
dx +
_
cu
h
v
h
dx +
T
h
_
\
b nu
h
v
+
h
ds +
T
h
_
+
b nu
+
h
v
+
h
ds
=
_
fv
h
dx
_
b ngv
h
ds v
h
V
d
h,p
. (151)
Integrating back by parts on each element we obtain following equivalent form of the discretiza-
tion: nd u
h
V
d
h,p
such that
_
(
h
(bu
h
) +cu
h
) v
h
dx
T
h
_
\
b n(u
+
h
u
h
)v
+
h
ds
_
b nu
h
v
h
ds
=
_
fv
h
dx
_
b ngv
h
ds v
h
V
d
h,p
. (152)
In contrast to the discontinuous Galerkin discretization of viscous terms where several dierent
discretization schemes based on dierent choices of numerical uxes can be derived, see Section 5,
the discretization in (152) is known as the (one and only reasonable) discontinuous Galerkin
discretization of the linear advection equation.
33
4.7.1 The importance of the inter-element jump terms
We note that the discrete functions u
h
V
d
h,p
may be discontinuous on interior faces between
neighboring elements. Originating from the upwind ux, the interior face terms
T
h
_
\
b n(u
+
h
u
h
)v
+
h
ds (153)
include the jump [u
h
] = u
+
h
u
h
of the discrete solution and impose continuity of the discrete solu-
tion in a weak sense. If these inter-element jump terms were neglected the discrete problem would
due to the discontinuity of the ansatz and test functions decouple into one local problem per
element with no data coupling to neighboring elements and hence to an inconsistent discretization.
For getting further insight into the importance of these jump terms, let us, for a moment,
evaluate the bilinear form in (152) for functions u
h
, v
h
V
c
h,p
C
0
(), i.e. the discrete space V
c
h,p
of continuous piecewise polynomial functions given in Denition 2.6. Then the jump terms vanish
and we obtain: nd u
h
V
c
h,p
such that
_
(
h
(bu
h
) +cu
h
) v
h
dx
_
b nu
h
v
h
ds =
_
fv
h
dx
_
b ngv
h
ds, (154)
for all v
h
V
c
h,p
. This is the standard Galerkin discretization with weakly imposed boundary con-
ditions of the linear advection equation based on continuous nite element methods as introduced
in Section 3.1.2. We recall that this discretization is unstable and of order O(h
p
). In contrast to
that we have seen that the discontinuous Galerkin discretization (152) based on the upwind ux
is stable and of order O(h
p+1/2
). From this we see, that allowing inter-element discontinuities and
imposing continuity in a weak sense has a stabilizing eect on the discretization and yields an
improved order of convergence as compared to the standard (continuous) Galerkin discretization.
Finally, given a T
h
and setting v
h
0 on all
= ,
T
h
, in (152) we obtain
_
( (bu
h
) +cu
h
) v
h
dx
_
b nu
+
h
v
+
h
ds =
_
fv
h
dx
_
b nu
h
v
+
h
ds, (155)
where u
h
is replaced by g on
h
instead of the boundary function g as in (154). In fact, u
h
, i.e. the value of u
h
on a neighboring element, can be considered as boundary function to the linear advection problem
(99) localized on . In summary, using the upwind ux in the discontinuous Galerkin discretization
of the linear advection equation corresponds to a weak imposition of boundary conditions on each
element T
h
.
4.7.2 The global and local conservation property
We recall from Lemma 4.7 that a discretization based on any conservative ux H has a global
conservation property. The upwind ux is conservative, hence the discretization based on the
upwind ux, see (151) or (152), has a global conservation property.
Ignoring Lemma 4.7 for a moment we want to (re)show the global conservation property of
the discretization based on the upwind ux. To this end, we rewrite (151) with c 0 in terms of
interior faces e
I
. For any two neighboring elements and
with e :=
+
= we
34
rewrite b n
v
+
on
as b (n
) u
+
on
+
and obtain
(bu
h
)
h
v
h
dx +
e
I
_
e
b nu
+
h
_
v
+
h
v
h
_
ds +
_
+
b nu
+
h
v
+
h
ds
=
_
fv
h
dx
_
b ngv
h
ds v
h
V
d
h,p
. (156)
Setting v
h
1 on we obtain the global conservation property:
_
b ng ds +
_
+
b nu
+
h
ds =
_
f dx. (157)
Furthermore, given a T
h
and setting v
h
1 on and v
h
0 on all
= ,
T
h
, in (151)
with c 0 or in (156) we obtain the local conservation property:
_
b nu
h
ds +
_
+
b nu
+
h
ds =
_
f dx, (158)
where u
h
is replaced by g on
.
4.7.3 Consistency
We recall from Lemma 4.7 that a discretization based on any consistent ux H is consistent. The
upwind ux is consistent, hence the discretization based on the upwind ux, see (151) or (152),
is consistent. In the following we present another way to show that the discretization based on
the upwind ux is consistent. From (152) we see that the discrete solution u
h
V
d
h,p
satises the
primal residual form:
T
h
_
R(u
h
)v dx +
T
h
_
\
r(u
h
)v ds +
_
(u
h
)v ds = 0 v V
d
h,p
, (159)
where R(u
h
), r(u
h
) and r
(u
h
) denote the element, interior face and boundary residuals, respec-
tively, given by
R(u
h
) = f
h
(bu
h
) cu
h
in , T
h
,
r(u
h
) = b n(u
+
h
u
h
) on
\ , T
h
,
r
(u
h
) = b n(u
h
g) on
,
and r
(u
h
) 0 on
+
. We see that the exact solution u H
1,b
() to (99) satises
T
h
_
R(u)v dx +
T
h
_
\
r(u)v ds +
_
(u)v ds = 0 v H
1,b
(T
h
), (160)
with R(u) = 0, r(u) = 0 and r
N
are disjoint subsets with union , that is
D
N
= and
D
N
= . Furthermore, we
assume that
D
= . Problem (162) represents the general Dirichlet-Neumann problem of Poissons
equation. In case of
D
= and g
D
= 0 this represents the Dirichlet problem with homogeneous
boundary conditions given by
u = f in , u = 0 on . (163)
Let H
m
(T
h
) be the broken Sobolev space consisting of functions v L
2
() whose restriction to
each element T
h
belongs to the Sobolev space H
m
(). Analogously, by [H
m
(T
h
)]
d
we denote
the d-vector valued broken Sobolov space.
5.1 The system and primal ux formulation
In this and the following sections we recall the common framework for deriving DG discretizations of
the homogeneous Dirichlet problem (163) from [2] and apply it to the Dirichlet-Neumann problem
(162). We begin by rewriting (162) as a rst-order system as follows
= u, = f in , u = g
D
on
D
, n u = g
N
on
N
. (164)
Assuming u H
2
(T
h
) and [H
1
(T
h
)]
d
we multiply the rst and second equation by test functions
[H
1
(T
h
)]
d
and v H
1
(T
h
), respectively, integrate over an element T
h
, and integrate by
parts. Thus
_
dx =
_
u dx +
_
u nds,
_
v dx =
_
fv dx +
_
nv ds,
(165)
where n is the unit outward normal vector to . Then, we sum (165) over all elements T
h
. As
u and may be discontinuous across inter-element faces \ , T
h
, they must be replaced by
numerical ux functions u and which are approximations to u and = u, respectively. Thus
we obtain following system ux formulation: nd u H
2
(T
h
) and [H
1
(T
h
)]
d
such that
_
dx =
_
u
h
dx +
T
h
_
u nds [H
1
(T
h
)]
d
, (166)
_
h
v dx =
_
fv dx +
T
h
_
nv ds v H
1
(T
h
). (167)
Here, u : H
1
(T
h
) T(T
h
) is a scalar numerical ux function, and : H
2
(T
h
)[H
1
(T
h
)]
d
[T(T
h
)]
d
is a vector-valued numerical ux function. Depending on the particular choice of the numerical ux
functions u and several dierent DG discretizations can be derived, each with specic properties
with respect to stability and accuracy.
Denition 5.1 We say that the numerical uxes u and are consistent if
u(v) = v, (v, v) = v, on
I
, (168)
whenever v is a smooth function satisfying the Dirichlet boundary conditions. Furthermore, we say
that u and are conservative if they are single-valued on
I
.
36
The term conservative comes from the following useful property, which holds whenever the vector
ux is single-valued: If S is the union of any collection of elements, we obtain
_
S
h
v dx =
_
S
fv dx +
S
_
\S
nv ds +
_
S
nv ds
=
_
S
fv dx +
e
I
S
n
+
(v
+
v
) ds +
_
S
nv ds,
(169)
for v 0 on \ S, where
I
S
are the faces interior to S. Taking v 1 on S we then have
_
S
fv dx +
_
S
nv ds = 0, (170)
i.e. a global and local conservation property similar to conservation properties of the DG discretiza-
tion of the linear advection in Section 4.7.
Equations (166) and (167) represent a rst order system in u and with (d + 1) as many
unknowns as the original (scalar) problem in u. In order to reduce the problem size, the auxiliary
variable is usually eliminated to gain a so-called primal formulation involving only the primal
variable u. To this end, we perform a second integration by parts on each element in (166) and
set =
h
v which gives us
_
h
v dx =
_
h
u
h
v dx +
T
h
_
( u u)n
h
v ds. (171)
Substituting (171) into (167) we obtain the primal ux formulation: nd u H
2
(T
h
) such that
B
h
(u, v) =
_
fv dx v H
2
(T
h
),
where the bilinear form
B
h
(, ) : H
2
(T
h
) H
2
(T
h
) R is dened by
B
h
(u, v) =
_
h
u
h
v dx
T
h
_
nv ds +
T
h
_
( u u)n
h
v ds. (172)
This bilinear form is denoted by
B
h
(and not B
h
) as it includes the (still unspecied) numerical
uxes u and . Furthermore,
B
h
includes through the specication of u and on the boundary
all boundary data terms.
Finally, we note that
B
h
in (172) is an element-based form, i.e. it is given in terms of
.
This means that each interior face e =
I
occurs twice in the sum over all elements (once in
_
and once in
_
for
= and e =
I
where each interior face occurs only
once. However, before doing so, we introduce some more notation.
Denition 5.2 Let e
I
be an interior edge between two adjacent elements
+
and
with unit
outward normal vectors, n
+
, n
R
d
, respectively. Let q T(T
h
) and [T(T
h
)]
d
be the traces
of a scalar and a vector valued function, respectively. Then, we dene the mean value and jump
operators, {{}} and [[]], as follows
{{q}} =
1
2
(q
+
+q
), [[q]] = q
+
n
+
+q
,
{{}} =
1
2
(
+
+
), [[]] =
+
n
+
+
.
37
Denition 5.3 On boundary edges e the mean value and jump operators are dened by
{{q}} = q
+
, [[q]] = q
+
n
+
,
{{}} =
+
, [[]] =
+
n
+
.
Based on these notations, we can show following result which will frequently be used to transfer
between element-based and face-based forms.
Lemma 5.4 Again, let q T(T
h
) and [T(T
h
)]
d
, then
_
\
+
n
+
q
+
ds =
_
I
{{}} [[q]] ds +
_
I
[[]] {{q}} ds, (173)
+
n
+
q
+
ds =
_
{{}} [[q]] ds +
_
I
[[]] {{q}} ds. (174)
Proof: On
I
we have
{{}} [[q]] + [[]] {{q}} =
1
2
(
+
+
) (q
+
n
+
+q
)+
1
2
(
+
n
+
+
)(q
+
+q
)
=
1
2
(
+
n
+
q
+
+
n
+
q
+
+
+
n
)+
1
2
(
+
n
+
q
+
+
q
+
+
+
n
+
q
)
=
+
n
+
q
+
+
using n
= n
+
in the last identity. On we have {{}}[[q]] =
+
n
+
q
+
.
Using (174) and the Gauss integral formula we obtain following result.
Corollary 5.5 Let v H
1
(T
h
) and [H
1
(T
h
)]
d
, then
_
h
v dx =
_
h
v dx +
_
{{}} [[v]] ds +
_
I
[[]] {{v}} ds. (175)
Proof: Using the Gauss integral formula on each T
h
,
_
dx =
_
nds, (176)
for := v [H
1
(T
h
)]
d
, and summing over all T
h
we obtain
_
h
v dx +
_
h
v dx =
nv ds =
_
{{}} [[v]] ds +
_
I
[[]] {{v}} ds,
(177)
which shows (175).
We now proceed in transferring the element-based form (172) into a face-based form. To this
end, we use equation (174) twice (once for = and q = v, and once for =
h
v and q = uu),
and rewrite (172) as follows
B
h
(u, v) =
_
h
u
h
v dx
_
{{ }} [[v]] ds
_
I
[[ ]] {{v}} ds
+
_
{{
h
v}} [[ u u]] ds +
_
I
[[
h
v]] {{ u u}} ds,
which results in following face-based primal ux form,
B
h
(u, v) =
_
h
u
h
v dx +
_
([[ u u]] {{
h
v}} {{ }} [[v]]) ds
+
_
I
({{ u u}} [[
h
v]] [[ ]] {{v}}) ds.
(178)
38
5.2 The DG discretization: Consistency and adjoint consistency
Let V
h
:= V
d
h,p
H
2
(T
h
) be the space of discontinuous piecewise polynomials of degree p as dened
in (123). Then the discontinuous Galerkin discretization in face-based ux formulation is given by:
nd u
h
V
h
such that
B
h
(u
h
, v
h
) =
_
fv
h
dx v
h
V
h
. (179)
where
B
h
is as dened in (178) and the uxes u and are still unspecied.
Denition 5.6 (Consistency) Let u H
2
() be the exact solution to problem (162). Then, the
discretization (179) of (162) is consistent if and only if
B
h
(u, v) =
_
fv dx v H
2
(T
h
). (180)
Theorem 5.7 Let
B
h
(, ) be given by (178). Then the discretization (179) of the homogeneous
Dirichlet problem (163) is consistent if and only if the numerical uxes u and are consistent,
u(v) = v, (v, v) = v on
I
, (181)
Proof: Let u be the solution to problem (162). Setting = u in (175) we obtain
_
u
h
v dx =
_
uv dx +
_
{{u}} [[v]] ds +
_
I
[[u]] {{v}} ds, (182)
for v H
2
(T
h
). Substituting this into
B
h
(u, v), cf. (178), gives
B
h
(u, v) =
_
h
uv dx +
_
{{
h
u}} [[v]] ds +
_
I
[[
h
u]] {{v}} ds
+
_
([[ u u]] {{
h
v}} {{ }} [[v]]) ds +
_
I
({{ u u}} [[
h
v]] [[ ]] {{v}}) ds,
where u = u(u) and = (u, u). Using {{u}} = u, [[u]] = 0, {{
h
u}} = u, [[
h
u]] = 0, and
u = f, we obtain
B
h
(u, v) =
_
fv dx +
_
([[ u]] {{
h
v}} + (u {{ }}) [[v]]) ds
+
_
I
(({{ u}} u) [[
h
v]] [[ ]] {{v}}) ds.
(183)
If the numerical ux u is consistent, i.e. u(u) = u on
I
, then [[ u]] = 0 and {{ u}} = u on
I
.
If the numerical ux is also consistent, i.e. (u) = u, then [[ ]] = 0 and {{ }} = u on
I
.
Inserting these relations in (183) we obtain
B
h
(u, v) =
_
fv dx. (184)
Hence, the primal formulation is consistent. This argument can easily be reversed. Assuming that
the primal formulation is consistent, i.e. Equation (184) holds, then, in view of (183), we recognize
that consistency (181) of the numerical uxes is necessary.
Remark 5.8 Theorem 5.7 can be generalized to the Dirichlet-Neumann problem (162): The dis-
cretization (179) of (162) is consistent if and only if
u(v) = v, (v, v) = v on
I
D
, n (v, v) = n v on
N
. (185)
39
Corollary 5.9 (Galerkin orthogonality) Let (179) be a consistent discretization. Then, the
error u u
h
is orthogonal (with respect to
B
h
(, )) to the discrete function space V
h
, i.e.
B
h
(u u
h
, v
h
) = 0 v
h
V
h
. (186)
Proof: Subtract (179) from (180) for v := v
h
V
h
H
2
(T
h
).
As we will see later, in addition to the consistency of a discretization also the so-called adjoint
consistency of the discretization is of interest. Whereas optimal estimates in the energy norm
(H
1
-norm) can be derived for consistent discretizations, the derivation of optimal error estimates
in the L
2
-norm requires the application of a duality argument which requires the discretization to
be adjoint consistent. In that sense, adjoint consistency represents an additional measure of quality
of a discretization.
Denition 5.10 (Adjoint consistency) Given a function j
L
2
(), let z H
2
() be the
exact solution to the dual or adjoint problem
z = j
in , z = 0 on . (187)
Then, the discretization (179) of the homogeneous Dirichlet problem (163) is adjoint consistent if
and only if
B
h
(v, z) =
_
v dx v H
2
(T
h
). (188)
Remark 5.11 Related to the adjoint problem (187) we note that
The dierential operator in (187) is the adjoint operator L
= e = u u
h
which
gives e
2
=
B
h
(e, z). An adjoint problem like in (187) will also be required when deriving
error estimates with respect to target quantities of the form J(v) =
_
v dx which gives
J(e) =
B
h
(e, z).
Theorem 5.12 Let
B
h
(, ) be given by (178). Then the discretization (179) of the homogeneous
Dirichlet problem (163) is
adjoint consistent
if and only if the numerical uxes u and are conservative, i.e.
[[ u]] = 0, and [[ ]] = 0. (189)
Proof: Let z H
2
() be the solution to (187), we set = z and v = w in (175) and obtain
_
z
h
wdx =
_
wdx +
_
{{z}} [[w]] ds +
_
I
[[z]] {{w}} ds. (190)
Substituting u by w and v by z in (178) and using (190) results in
B
h
(w, z) =
_
wdx +
_
{{z}} [[w]] ds +
_
I
[[z]] {{w}} ds
+
_
I
({{ u(w) w}} [[z]] [[ (w)]] {{z}}) ds,
40
which reduces to following face-based dual form
B
h
(w, z) =
_
wdx +
_
I
({{ u(w)}} [[z]] [[ (w)]] {{z}}) ds.
As z H
2
() we have {{z}} = z, [[z]] = 0, {{z}} = z and [[z]] = 0. Therefore,
B
h
(w, z) =
_
wdx +
_
[[ u(w)]] z ds
_
I
[[ (w)]] z ds, (191)
which reduces to
B
h
(w, z) =
_
h
u
h
h
v
h
dx +
_
([[ u
h
u
h
]] {{
h
v
h
}} {{
h
}} [[v
h
]]) ds
+
_
I
({{ u
h
u
h
}} [[
h
v
h
]] [[
h
]]{{v
h
}}) ds =
_
fv
h
dx v
h
V
h
, (192)
where u
h
:= u(u
h
) and
h
:= (u
h
, u
h
). Here the numerical ux functions u and are still
unspecied. (192) results in a consistent discretization of Problem (162) provided the uxes u and
are consistent, i.e. u(u) = u and (u, u) = u. Depending on the specic choice of u and
several dierent DG methods can be derived, each with specic properties with respect to stability
and accuracy. Before continuing with the derivation of specic DG discretizations we rst collect
some elementary relations of the mean value and jump operators:
Lemma 5.14 Let {{}} and [[]] be the mean value and jump operators dened in Denition 5.2.
Furthermore, let q T(T
h
) and [T(T
h
)]
d
, then
{{{{q}}}} = {{q}}, {{[[q]]}} = [[q]], [[[[q]]]] = 0, [[{{q}}]] = 0, (193)
{{{{}}}} = {{}}, {{[[]]}} = [[]], [[[[]]]] = 0, [[{{}}]] = 0. (194)
5.3.1 The SIPG and NIPG methods and the method of Baumann-Oden
The symmetric interior penalty method (SIPG): Let the uxes u
h
and
h
be given by
u
h
= {{u
h
}},
h
= {{
h
u
h
}}
ip
(u
h
) on
I
, (195)
u
h
= g
D
,
h
=
h
u
h
ip
(u
h
) on
D
, (196)
u
h
= u
h
,
h
= g
N
n on
N
, (197)
where
ip
(u
h
) = [[u
h
]] = C
IP
p
2
h
[[u
h
]] on
I
, (198)
ip
(u
h
) = (u
h
g
D
)n = C
IP
p
2
h
(u
h
g
D
)n on
D
. (199)
41
Using Lemma 5.14 we obtain [[ u
h
]] = [[{{u
h
}}]] = 0, {{ u
h
}} = {{{{u
h
}}}} = {{u
h
}}, {{
h
}} = {{{{
h
u
h
}}}}
{{
ip
(u
h
)}} = {{
h
u
h
}}
ip
(u
h
), and [[
h
]] = [[{{
h
u
h
}}]] [[
ip
(u
h
)]] = 0 on
I
, and (192) reduces
to the symmetric interior penalty method given by: nd u
h
V
h
such that
_
h
u
h
h
v
h
dx +
_
D
([[u
h
]] {{
h
v
h
}} {{
h
u
h
}} [[v
h
]]) ds +
_
D
[[u
h
]] [[v
h
]] ds
=
_
fv
h
dx
_
D
g
D
n
h
v
h
ds +
_
D
g
D
v
h
ds +
_
N
g
N
v
h
ds (200)
for all v
h
V
h
.
The non-symmetric interior penalty method (NIPG): Let the uxes u
h
,
h
be given by
u
h
= {{u
h
}} +n
+
[[u
h
]],
h
= {{
h
u
h
}}
ip
(u
h
) on
I
, (201)
u
h
= 2u
h
g
D
,
h
=
h
u
h
ip
(u
h
) on
D
, (202)
and by (197). We use n
+
[[u
h
]] = n
+
(u
+
h
n
+
+u
h
n
) = u
+
h
u
h
and [[{{u
h
}}]] = 0, and obtain
[[ u
h
]] = [[u
+
h
u
h
]] = (u
+
h
u
h
)n
+
+ (u
h
u
+
h
)n
= 2(u
+
h
n
+
+u
h
n
) = 2[[u
h
]],
{{ u
h
}} = {{{{u
h
}}}} +{{n [[u
h
]]}} = {{u
h
}} +
1
2
(u
+
h
u
h
+u
h
u
+
h
)) = {{u
h
}}.
Then, (192) reduces to the non-symmetric interior penalty method: nd u
h
V
h
such that
_
h
u
h
h
v
h
dx +
_
D
([[u
h
]] {{
h
v
h
}} {{
h
u
h
}} [[v
h
]]) ds +
_
D
[[u
h
]] [[v
h
]] ds
=
_
fv
h
dx +
_
D
g
D
n
h
v
h
ds +
_
D
g
D
v
h
ds +
_
N
g
N
v
h
ds (203)
for all v
h
V
h
. We note, that the only dierence of this discretization to the SIPG discretization
in (200) is the sign of the
_
D
[[u
h
]] {{
h
v
h
}} ds and
_
D
g
D
n
h
v
h
ds term.
The method of Baumann-Oden (BO): Let the uxes u
h
and
h
be given by
u
h
= {{u
h
}} +n
+
[[u
h
]],
h
= {{
h
u
h
}} on
I
, (204)
u
h
= 2u
h
g
D
,
h
=
h
u
h
on
D
, (205)
and by (197). Then we obtain the method by Baumann and Oden: nd u
h
V
h
such that
_
h
u
h
h
v
h
dx +
_
D
([[u
h
]] {{
h
v
h
}} {{
h
u
h
}} [[v
h
]]) ds
=
_
fv
h
dx +
_
D
g
D
n v
h
ds +
_
N
g
N
v
h
ds (206)
for all v
h
V
h
. We note, that this discretization can be obtained from the NIPG discretization in
(203) simply by ignoring the interior penalty term
ip
(u
h
) = [[u
h
]]. However, we will show later,
that the method of Baumann-Oden is unstable whereas the NIPG discretization is stable due to
the stabilizing eect of the interior penalty term.
42
Unied description for SIPG, NIPG and Baumann-Oden: The discretizations derived
above can be written in unied form as follows: nd u
h
V
h
such that
B
h
(u
h
, v
h
) = F
h
(v
h
) v
h
V
h
, (207)
where
B
h
(u, v) =
_
h
u
h
v dx
+
_
D
([[u]] {{
h
v}} {{
h
u}} [[v]]) ds +
_
D
[[u]] [[v]] ds,
F
h
(v) =
_
fv dx +
_
D
g
D
n v ds +
_
D
g
D
v ds +
_
N
g
N
v ds, (208)
and the constants and are given by
SIPG : = 1, > 0,
NIPG : = 1, > 0, (209)
Baumann-Oden : = 1, = 0.
We note that in the primal form
B
h
as dened in (172) or (178) the numerical uxes u and were
still unspecied. Furthermore,
B
h
(implicitly) included all boundary data functions. In contrast to
that B
h
as given in (208) includes no boundary data. Instead, all boundary data terms have been
moved to the right hand side and are now included in F
h
(v
h
).
5.3.2 The original DG discretization of Bassi and Rebay (BR1)
Let us choose the uxes u
h
and
h
to be given by
u
h
= {{u
h
}},
h
= {{
h
u
h
}}
br1
(u
h
) on
I
, (210)
u
h
= g
D
,
h
=
h
u
h
br1
(u
h
) on
D
, (211)
u
h
= u
h
,
h
= g
N
n on
N
, (212)
with
br1
(u
h
) =
br1
(u
h
) = {{L
g
D
(u
h
)}}, (213)
where the so-called global lifting operator including Dirichlet boundary values,
L
g
D
: T(T
h
)
d
h,p
:= [V
d
h,p
]
d
[H
1
(T
h
)]
d
,
is a vector-valued ane operator dened by: Let L
g
D
(w)
d
h,p
be the solution to
_
L
g
D
(w) dx =
_
I
[[w]] {{}} ds
_
D
(w g
D
)n ds
d
h,p
. (214)
Here, T(T
h
) := L
2
(
I
) denotes the space of traces of functions v H
1
(T
h
). Furthermore, we
consider the global lifting operator L
0
1
with homogeneous Dirichlet boundary values, which is the
vector-valued linear operator given by: Let L
0
(w)
d
h,p
be the solution to
_
L
0
(w) dx =
_
[[w]] {{}} ds
d
h,p
. (215)
1
We note that in some publications, the global lifting operator is dened as l0 : [T(T
h
)]
d
d
h,p
with
Z
l0() dx =
Z
{{}} ds
d
h,p
,
for which we then have L0(w) = l0([[w]]).
43
In view of (214) and (215), we have
_
L
g
D
(w) dx =
_
L
0
(w) dx +
_
D
g
D
n ds. (216)
Using the numerical uxes u
h
and
h
as given in (210)-(212) the DG discretization (192) reduces
to: nd u
h
V
h
such that
_
h
u
h
h
v
h
dx +
_
D
([[u
h
]] {{
h
v
h
}} {{
h
u
h
}} [[v
h
]]) ds
_
D
{{L
g
D
(u
h
)}} [[v
h
]] ds
=
_
fv
h
dx
_
D
g
D
n
h
v
h
ds +
_
N
g
N
v
h
ds v
h
V
h
. (217)
Using the denition (214) for L
g
D
we can rewrite
_
L
g
D
(u
h
)
h
v
h
dx =
_
[[u
h
]] {{
h
v
h
}} ds +
_
D
g
D
n
h
v
h
ds. (218)
Furthermore, using the relation (215) of L
0
we can rewrite
_
h
u
h
L
0
(v
h
) dx =
_
{{
h
u
h
}} [[v
h
]] ds, (219)
_
L
g
D
(u
h
) L
0
(v
h
) dx =
_
{{L
g
D
(u
h
)}} [[v
h
]] ds. (220)
Substituting these relations into (217) we obtain the discretization: nd u
h
V
h
such that
_
(
h
u
h
+L
g
D
(u
h
)) (
h
v
h
+L
0
(v
h
)) dx =
_
fv
h
dx +
_
N
g
N
v
h
ds v
h
V
h
, (221)
or equivalently, using (216): nd u
h
V
h
such that
B
h
(u
h
, v
h
) = F
h
(v
h
) v
h
V
h
, (222)
where
B
h
(u, v) =
_
(
h
u +L
0
(u)) (
h
v +L
0
(v)) dx,
F
h
(v) =
_
fv dx +
_
N
g
N
v ds
_
D
g
D
n (
h
v +L
0
(v)) ds.
(223)
This is the original method of Bassi and Rebay introduced in [5] for which, however, several problems
have been observed: In contrast to most other DG discretizations where an element communicates
with its direct neighboring elements only, the stencil of the BR1 discretization is considerably larger
as it includes also neighbors of neighbors. Furthermore, this discretization is unstable. In fact, we
obtain
B
h
(v, v) =
h
v +L
0
(v)
2
L
2
()
,
which vanishes on the set
Z := {v V
h
:
h
v +L
0
(v) = 0},
where Z \ {0} can, in general, be nonempty. This discretization is called the BR1 discretization in
order to distinguish it from the modication of Bassi and Rebay, the so-called BR2 discretization,
which we will introduce in the following subsection.
44
5.3.3 The modied DG discretization of Bassi and Rebay (BR2)
Let us choose the uxes u
h
and
h
to be given by
u
h
= {{u
h
}},
h
= {{
h
u
h
}}
br2
(u
h
) on
I
, (224)
u
h
= g
D
,
h
=
h
u
h
br2
(u
h
) on
D
, (225)
u
h
= u
h
,
h
= g
N
n on
N
, (226)
with
br2
(u
h
) =
br2
(u
h
) = C
BR2
{{L
e
g
D
(u
h
)}} for e
I
D
, (227)
where the so-called local lifting operator including Dirichlet boundary conditions, L
e
g
D
: L
2
(e)
d
h,p
, is a vector-valued ane operator dened by: L
e
g
D
(w)
d
h,p
is the solution to
_
L
e
g
D
(w) dx =
_
e
(w g
D
) n ds
d
h,p
, for e
D
(228) _
L
e
g
D
(w) dx =
_
e
[[w]] {{}} ds
d
h,p
, on e
I
,
and L
e
g
D
(w) is dened to be zero for e
N
. The local lifting operator with homogeneous Dirichlet
boundary conditions L
e
0
is dened accordingly. In particular, for e
D
we have
_
L
e
g
D
(w) dx =
_
L
e
0
(w) dx +
_
e
g
D
n ds
d
h,p
. (229)
We note, that L
e
g
D
(w) has support (i.e. is non-equal zero) only on the (one or two) elements
sharing the edge e. Furthermore, L
e
g
D
(w) does not depend on g
D
on interior edges e
I
. Using
the numerical uxes u
h
and
h
as given in (224)-(226) the DG discretization (192) reduces to: nd
u
h
V
h
such that
_
h
u
h
h
v
h
dx+
_
D
([[u
h
]] {{
h
v
h
}} {{
h
u
h
}} [[v
h
]]) ds
_
D
C
BR2
{{L
e
g
D
(u
h
)}} [[v
h
]] ds
=
_
fv
h
dx
_
D
g
D
n
h
v
h
ds +
_
N
g
N
v
h
ds v
h
V
h
. (230)
Using the denition (228) of L
e
g
D
we can rewrite
e
I
D
C
BR2
_
L
e
g
D
(u
h
) L
e
0
(v
h
) dx =
_
D
C
BR2
{{L
e
g
D
(u
h
)}} [[v
h
]] ds. (231)
Substituting this, (218) and (219) into (230) yields: nd u
h
V
h
such that
_
(
h
u
h
h
v
h
+L
g
D
(u
h
)
h
v
h
+
h
u
h
L
0
(v
h
)) dx
+
e
I
D
C
BR2
_
L
e
g
D
(u
h
) L
e
0
(v
h
) dx =
_
fv
h
dx +
_
N
g
N
v
h
ds v
h
V
h
, (232)
or equivalently, using (216) and (229): nd u
h
V
h
such that
B
h
(u
h
, v
h
) = F
h
(v
h
) v
h
V
h
, (233)
45
where
B
h
(u, v) =
_
(
h
u
h
v +L
0
(u)
h
v +
h
u L
0
(v)) dx +
e
I
D
C
BR2
_
L
e
0
(u) L
e
0
(v) dx,
F
h
(v) =
_
fv dx +
_
N
g
N
v ds
_
D
g
D
n (
h
v +C
BR2
L
e
0
(v)) ds. (234)
We note, that (232) can be obtained also by replacing
_
L
g
D
(u
h
)L
0
(v
h
) dx in (221) by (231). The
BR2 discretization has several advantages over the BR1 scheme. The stencil of BR2 scheme includes
only rst neighbors instead of additional second neighbors as does the BR1 scheme. Furthermore,
as we will show later, the BR2 discretization is stable, provided C
BR2
is larger than the number of
neighboring elements, i.e. C
BR2
> 3 for triangular elements and C
BR2
> 4 for quadrilateral elements.
Finally, the BR2 discretization diers from the SIPG discretization in Subsection 5.3.1 only in the
denition of the stabilization/penalization term
br2
(u
h
) versus
ip
(u
h
). We will show later, that
the BR2 scheme has the advantage that a lower bound of the constant C
BR2
is known (number of
neighboring elements), whereas the constant C
IP
occurring in the symmetric interior penalty term
must be larger than a constant C
0
IP
> 0 which in general is not known. However, due to the use
of lifting operators the BR2 discretization is signicantly more complicated and more computing
time expensive than the SIPG discretization.
5.4 Consistency, adjoint consistency, continuity and coercivity
Corollary 5.15 (Consistency) SIPG, NIPG, Baumann-Oden, BR1 and BR2 are consistent dis-
cretizations, i.e. the exact solution u H
2
() to (162) satises
B
h
(u, v) = F
h
(v) v H
2
(T
h
). (235)
Proof: We have u(u) = u. Furthermore, we have
ip
(u) =
br1
(u) =
br2
(u) = 0 for u H
2
()
which is continuous. Thereby (u, u) = u. Finally, we have n (u, u) = g
N
= n u on
N
and hence consistency by Theorem 5.7 and Remark 5.8.
Corollary 5.16 (Galerkin orthogonality) We have the Galerkin orthogonality
B
h
(u u
h
, v
h
) = 0 v
h
V
h
. (236)
Proof: Subtract (207) from (235) for v
h
V
h
H
2
(T
h
).
Corollary 5.17 (Adjoint consistency) SIPG, BR1 and BR2 for the homogeneous Dirichlet prob-
lem (163) are adjoint consistent. NIPG and Baumann-Oden are adjoint inconsistent.
Proof: As shown in Subsection 5.3.1 we have [[
h
]] = 0 and [[ u
h
]] = 0 for the SIPG method
and similarly for BR1 and BR2. However, for NIPG and Baumann-Oden we have [[
h
]] = 0 and
[[ u
h
]] = 2[[u
h
]]. Adjoint (in)consistency now follows from Theorem 5.12.
In the following we show that the bilinear form corresponding to the method by Baumann-Oden,
and the bilinear forms of the SIPG, the NIPG and the BR2 methods are continuous.
Lemma 5.18 (Continuity of Baumann-Oden) Let B
h
be given by
B
h
(u, v) =
_
h
u
h
v dx +
_
D
([[u]] {{
h
v}} {{
h
u}} [[v]]) ds, (237)
with = 1. Then,
|B
h
(u, v)| |u|
|v|
, u, v H
2
(T
h
), (238)
46
for any > 0, where
|v|
2
=
h
v
2
L
2
()
+
_
1
(n {{v}})
2
ds +
_
D
[v]
2
ds. (239)
Furthermore, (238) holds also for = 1.
Proof: We have
|
_
h
u
h
v dx|
h
u
L
2
()
h
v
L
2
()
.
Furthermore, we use
|
_
D
[[u]] {{
h
v}} ds| = |
_
1/2
[u]
1/2
n {{
h
v}} ds|
__
D
[u]
2
ds
_
1/2
__
1
(n {{
h
v}})
2
ds
_
1/2
,
(240)
and (240) with u and v exchanged, and obtain
|B
h
(u, v)| |
_
h
u
h
v dx| +|
_
D
[[u]] {{
h
v}} ds| +|
_
D
{{
h
u}} [[v]] ds|
h
u
L
2
()
h
v
L
2
()
+
__
D
[u]
2
ds
_
1/2
__
1
(n {{
h
v}})
2
ds
_
1/2
+
__
1
(n {{
h
u}})
2
ds
_
1/2
__
D
[v]
2
ds
_
1/2
h
u
2
L
2
()
+
_
1
(n {{
h
u}})
2
ds +
_
D
[u]
2
ds
_
1/2
_
h
v
2
L
2
()
+
_
1
(n {{
h
v}})
2
ds +
_
D
[v]
2
ds
_
1/2
|u|
|v|
,
where we used the Cauchy-Schwarz inequality
i
a
i
b
i
_
i
a
2
i
_
1/2
_
i
b
2
i
_
1/2
.
Lemma 5.19 (Continuity of NIPG and SIPG) Let B
h
be given as in (208) with > 0 and
= 1 for NIPG and = 1 for SIPG. Then there is a constant 1 < C 2 such that
|B
h
(u, v)| C|u|
|v|
, u, v H
2
(T
h
), (241)
where the norm | |
is as dened (239).
Proof: Using the Cauchy-Schwarz inequality
_
ab
__
a
2
_
1/2
__
b
2
_
1/2
for a =
1/2
[u] and b =
1/2
[v] we obtain
|
_
D
[u][v] ds|
__
D
[u]
2
ds
_
1/2
__
D
[v]
2
ds
_
1/2
.
Thereby, using (238) for = 1 and = 1 we obtain
|B
h
(u, v)| |u|
|v|
+|
_
D
[u][v] ds|
_
|u|
2
+
_
D
[u]
2
ds
_
1/2
_
|v|
2
+
_
D
[v]
2
ds
_
1/2
2|u|
|v|
.
47
and hence (241).
We note that > 0 in (238) may be any positive constant. In contrast to that > 0 in (241)
is the constant of the interior penalty term (198).
Lemma 5.20 (Continuity of BR2, [11]) Let B
h
be given as in (234). Then there is a constant
C > 1 such that
|B
h
(u, v)| C|u|
L
e
0
|v|
L
e
0
u, v H
2
(T
h
), (242)
where the | |
L
e
0
-norm is given by
|v|
2
L
e
0
=
h
v
2
L
2
()
+
e
I
L
e
0
(v)
2
L
2
()
. (243)
Proof: First, we note that L
0
(v) =
e
L
e
0
(v) on . Furthermore, since the support of each L
e
0
is the union of (one or two) elements sharing the edge e, we have
e
I
L
e
0
(v)
2
L
2
()
=
T
h
e
L
e
0
(v)
2
L
2
()
. (244)
Thereby, using Cauchy-Schwarz inequality,
_
N
i=1
a
i
_
2
N
i=1
1
N
i=1
a
2
i
= N
N
i=1
a
2
i
,
L
0
(v)
2
L
2
()
=
_
e
L
e
0
(v)
_
2
dx N
_
e
(L
e
0
(v))
2
dx = N
e
L
e
0
(v)
2
L
2
()
, (245)
where N is the number of faces e of an element . And, due to (244)
L
0
(v)
2
L
2
()
N
T
h
e
L
e
0
(v)
2
L
2
()
= N
e
I
L
e
0
(v)
2
L
2
()
. (246)
Given B
h
as in (234), using Cauchy-Schwarz, (246) and again Cauchy-Schwarz we obtain
B
h
(u, v) =
_
(
h
u
h
v +L
0
(u)
h
v +
h
u L
0
(v)) dx +
e
I
D
C
BR2
_
L
e
0
(u) L
e
0
(v) dx
h
u
L
2
()
h
v
L
2
()
+L
0
(u)
L
2
()
h
v
L
2
()
+
h
u
L
2
()
L
0
(v)
L
2
()
+
e
I
D
C
BR2
L
e
0
(u)
L
2
()
L
e
0
(v)
L
2
()
_
_
2
h
u
2
L
2
()
+L
0
(u)
2
L
2
()
+
e
I
D
C
BR2
L
e
0
(u)
2
L
2
()
_
_
_
_
2
h
v
2
L
2
()
+L
0
(v)
2
L
2
()
+
e
I
D
C
BR2
L
e
0
(u)
2
L
2
()
_
_
C|u|
L
e
0
|v|
L
e
0
with C = (N +C
BR2
)
2
.
Based on the relations shown in the proof of Lemma 5.20 coercivity of BR2 is easily obtained.
Theorem 5.21 (Coercivity of BR2, [11]) Let B
h
be given as in (234). Then there is a constant
C
0
BR2
> 0 (C
0
BR2
= 3 on triangles, C
0
BR2
= 4 on quadrilaterals) such that for all C
BR2
> C
0
BR2
we have
following coercivity property: There is a constant > 0 such that
B
h
(v, v) |v|
2
L
e
0
v H
2
(T
h
), (247)
where the | |
L
e
0
-norm is as dened in (243).
48
Proof: Using 2ab a
2
+
1
b
2
and (245) we obtain
2
_
h
v L
0
(v) dx
h
v
2
L
2
()
+
1
L
0
(v)
2
L
2
()
h
v
2
L
2
()
+
N
e
I
D
L
e
0
(v)
2
L
2
()
,
(248)
Then, using the denition (234) of B
h
we have
B
h
(v, v) =
h
v
2
L
2
()
+ 2
_
L
0
(v)
h
v dx +
e
I
D
C
BR2
L
e
0
(v)
2
L
2
(e)
(1 )
h
v
2
L
2
()
+
e
I
D
_
C
BR2
_
L
e
0
(v)
2
L
2
(e)
,
(249)
and hence (247) with = min(1 , C
BR2
N
h
v
h
v dx +
_
D
( 1){{
h
v}} [[v]] ds =
h
v
2
L
2
()
(251)
for = 1.
We see, that the bilinear form B
h
for Baumann-Oden is coercive only with respect to the
H
1
(T
h
)-seminorm. In particular, for any v
h
V
d
h,0
we have B
h
(v
h
, v
h
) = 0, i.e. the method
of Baumann-Oden is unstable. However, considering the discretization of u + cu = f with
c c
0
> 0 we obtain B
h
(v, v) c
0
v
2
L
2
()
+
h
v
2
L
2
()
, hence coercivity in the H
1
(T
h
)-norm.
Finally, in order to show coercivity of the NIPG and SIPG discretization we rst recall the
following standard inverse estimate: There is a constant C > 0 such that
v
h
L
2
()
Ch
1
v
h
L
2
()
v
h
V
h
, (252)
In the following we quote from [40], p. 208, a generalization of this estimate to v
h
V
d
h,p
.
Lemma 5.23 (Inverse estimate on V
d
h,p
) Let T
h
be a shape regular mesh. Then, there is a con-
stant C 0 such that for any T
h
we have
v
h
L
2
()
C
p
2
v
h
L
2
()
, v
h
V
d
h,p
. (253)
Furthermore, let us recall following trace inequality:
Lemma 5.24 (Multiplicative trace inequality, [38]) Let T
h
, with diameter h
and radius
r
< r
< h
, c > 0, then
v
2
L
2
()
C
_
h
1
v
2
L
2
()
+v
L
2
()
v
L
2
()
_
v H
1
(). (254)
49
Theorem 5.25 (Coercivity of NIPG and SIPG, [38]) Let B
h
be given as in (208) with =
C
IP
p
2
h
. Then there is a constant C
0
IP
0 (C
0
IP
= 0 for NIPG, i.e. = 1, and C
0
IP
> 0 for SIPG,
i.e. = 1), such that for all C
IP
> C
0
IP
we have following coercivity property: There is a constant
> 0 such that
B
h
(v
h
, v
h
) |v
h
|
2
v
h
V
d
h,p
, (255)
where
|v|
2
=
h
v
2
L
2
()
+
_
1
(n {{v}})
2
ds +
_
D
[v]
2
ds. (256)
Proof: This proofs follows the proof given in [38]. We begin by considering the term
_
e
(n v)
2
ds
on e
I
D
. By employing the trace inequality (254) and the inverse estimate (253) we obtain
for all v
h
V
d
h,p
,
_
e
(n v
h
)
2
ds C
_
h
1
v
h
2
L
2
()
+v
h
L
2
()
2
v
h
L
2
()
_
C
_
1
h
+
p
2
_
v
h
2
L
2
()
C
p
2
v
h
2
L
2
()
C
C
IP
v
h
2
L
2
()
,
(257)
and hence
1
(n {{v
h
}})
2
ds
C
C
IP
h
v
h
2
L
2
()
, (258)
where we used (a +b)
2
2(a
2
+b
2
) for a = n v
+
h
and b = n v
h
. From (208) we have
B
h
(v, v) =
_
h
v
h
v dx +
_
D
( 1){{
h
v}} [[v]] ds +
_
D
[v]
2
ds. (259)
For = 1 the second term vanishes and we obtain using (258)
B
h
(v
h
, v
h
) |v
h
|
2
_
1
C
C
IP
_
v
h
2
L
2
()
+ (1 )
_
D
[v]
2
ds.
Thereby, for any C
IP
> C
0
IP
= 0 we nd a 0 < 1/(1 +C/C
IP
) such that (255) holds.
For = 1 the second term in (259) is bounded using ab
4
a
2
+
1
b
2
,
2
_
e
{{
h
v}} [[v]] ds 2
_
e
n {{
h
v}}[v] ds 2
__
e
1
(n {{
h
v}})
2
ds
_
1/2
__
e
[v]
2
ds
_
1/2
4
4
_
e
1
(n {{
h
v}})
2
ds +
1
_
e
[v]
2
ds,
and hence
2
_
D
{{
h
v}} [[v]] ds
_
1
(n {{
h
v}})
2
ds
1
D
[v]
2
ds. (260)
We then obtain
B
h
(v
h
, v
h
) |v
h
|
2
_
1 ( +)
C
C
IP
_
v
h
2
L
2
()
+ (1
1
)
_
D
[v]
2
ds.
Hence, we require
1 ( +)
C
C
IP
0 and 1
1
0.
50
The second inequality is fullled if 0 < 11/ and > 1. On the other hand the rst inequality
requires that 1 C/C
IP
(1 +C/C
IP
) and hence
0 <
1 C/C
IP
1 +C/C
IP
<
1 C/C
IP
1 +C/C
IP
=
C
IP
C
C
IP
+C
,
for C
IP
> C
0
IP
> 0, e.g. C
0
IP
= C where C is the constant in (258).
We emphasize that for the NIPG method any choice of the interior penalty constant C
IP
> 0
gives a stable scheme. In contrast to that the SIPG method requires C
IP
> C
0
IP
> 0 for stability
with a constant C
0
IP
which is in general not known. However, numerical experiments showed that
C
IP
= 10 20 is a good choice a for large variety of problems.
We note that whereas continuity of B
h
could be shown on H
2
(T
h
) coercivity of B
h
on H
2
(T
h
)
does not hold, see Prop. 4.4 in [42]. However, coercivity of B
h
on the discrete function space V
d
h,p
as shown in Theorem 5.25 is sucient for proving existence and uniqueness of the discrete solution
u
h
V
d
h,p
. This has been discussed in more detail in Section 4.4 for the DG discretization of the
linear advection equation.
Remark 5.26 The estimates (257) and (258) motivate the particular choice of = C
IP
p
2
h
.
5.5 A priori error estimates
In this section we give a a priori error estimates for the NIPG and SIPG discretization.
Lemma 5.27 (A priori error estimate for NIPG and SIPG) Let u H
p+1
() be the exact
solution to Poissons equation (162). Furthermore, let u
h
V
d
h,p
be the solution to
B
h
(u
h
, v
h
) = F
h
(v
h
) v
h
V
d
h,p
,
where B
h
is as given in (208) with = 1 (NIPG) or = 1 (SIPG) and = C
IP
p
2
h
, C
IP
> C
0
IP
, cf.
Theorem 5.25. Then
|u u
h
|
Ch
p
|u|
H
p+1
()
, (261)
where | |
B
h
(, ) = B
h
( e, ) = B
h
(, ),
where we used Galerkin orthogonality (236). Using continuity of B
h
, (241), we obtain
||
2
B
h
(, ) C||
||
.
In summary, we obtain
|u u
h
|
||
+||
C||
.
Thereby, employing (239) and the approximation estimates in Section 4.5 we obtain
|u u
h
|
2
C
_
2
L
2
()
+
_
1
(n {{}})
2
ds +
_
D
[]
2
ds
_
C
_
Ch
2p
+
1
C
IP
h
p
2
Ch
2(p1/2)
+Ch
2(p+1/2)
_
|u|
2
H
p+1
()
,
and thus (261).
51
Remark 5.28 We note that estimate (261) corresponds to the H
1
-error estimate
u u
h
H
1
()
Ch
p
|u|
H
p+1
()
. (262)
shown in Section 2 for the continuous Galerkin discretization of Poissons equation. Estimate
(261) as well as (262) is of optimal order p which corresponds to the order of approximation of
polynomials of degree p in the H
1
-norm, cf. estimate (133).
Furthermore, we recall that for the continuous Galerkin discretization of Poissons equation an a
priori error estimate in the L
2
-norm has been obtained via a duality argument (Aubin-Nitsche)
which is based on the denition of an appropriate dual (or adjoint) problem. We will use this
technique also for the interior penalty discontinuous Galerkin discretization. However, as we will
see in the following, application of an duality argument requires an adjoint consistent discretization.
Lemma 5.29 (L
2
-error estimates for NIPG and SIPG) Let u H
p+1
() be the exact solu-
tion to Poissons equation (162). Furthermore, let u
h
V
d
h,p
be the solution to
B
h
(u
h
, v
h
) = F
h
(v
h
) v
h
V
d
h,p
,
where B
h
is as given in (208) with = 1 (NIPG) or = 1 (SIPG) and = C
IP
p
2
h
, C
IP
> C
0
IP
, cf.
Theorem 5.25. Then, for NIPG:
u u
h
L
2
()
Ch
p
|u|
H
p+1
()
, (263)
and for SIPG:
u u
h
L
2
()
Ch
p+1
|u|
H
p+1
()
, (264)
Proof: For simplicity we assume homogeneous Dirichlet boundary conditions. We recall from the
proof of Theorem 5.12 that given a function j
L
2
(), and the solution z H
2
() to the adjoint
problem
z = j
in , z = 0 on , (265)
the bilinear form
B
h
, see (178) with discretization specic numerical uxes u
h
and
h
, can be
written as
B
h
(w, z) =
_
wdx +
_
[[ u(w)]] z ds
_
I
[[ (w)]]z ds, (266)
see Equation (191). In particular, for the SIPG method and Dirichlet boundary conditions we have
[[
h
]] = 0 and [[ u
h
]] = 0, cf. proof to Corollary 5.17, and nd
B
s
h
(w, z) =
_
wdx, (267)
where B
s
h
(, ) denotes the bilinear form of the symmetric interior penalty DG discretizations given
in (208) with = 1. We dene z
s
to be the solution to (265) for j
:= e. We assume that
z
s
H
2
() and z
s
H
2
()
Ce
L
2
()
which is satised if is a convex polygon, for example. In
view of (267) we have
B
s
h
(w, z
s
) =
_
ewdx. (268)
Now choosing w = e we obtain
e
2
L
2
()
=
_
e
2
dx = B
s
h
(e, z
s
) = B
s
h
(e, z
s
z
h
) |e|
|z
s
z
h
|
, (269)
52
where we used Galerkin orthogonality for z
h
= P
d
h,p
z
s
V
d
h,p
and continuity (241) of B
h
. Thus
using (261) and approximation estimates for z
s
z
h
we obtain
e
2
L
2
()
|e|
|z
s
z
h
|
Ch
p
|u|
H
p+1
()
Ch|z|
H
2
()
Ch
p+1
|u|
H
p+1
()
e
L
2
()
,
and hence (264).
For the NIPG method the above argument fails because the method does not satisfy the adjoint
consistency condition (188). In fact, for the NIPG method and homogeneous Dirichlet boundary
conditions we have [[
h
]] = 0 and [[ u
h
]] = 2[[u
h
]], cf. proof to Corollary 5.17, and
B
n
h
(w, z) =
_
wdx + 2
_
ewdx, (271)
is mesh-dependent. Furthermore, z
n
is not regular which is why we do not obtain an additional
order of h from z
n
z
h
as we do in the case of the SIPG method.
5.6 Numerical results
In the following we investigate the experimental order of convergence in the H
1
- and the L
2
-norm of
the SIPG ( = 1) and the NIPG ( = 1) discretizations, see Section 5.3.1. According to Theorem
5.25 the penalization parameter is given by = C
IP
p
2
h
. In this example we choose C
IP
= 4.
Let us consider the following model problem: Let = (0, 1)
2
and consider Poissons equation
(162) with forcing function f which is chosen so that the analytical solution to (162) is given by
u(x) = sin(
1
2
x
1
) sin(
1
2
x
2
). (272)
We impose Dirichlet boundary conditions where the boundary value function g
D
on
D
= =
is prescribed based on the analytical solution u.
Figure 2 plots the error in the H
1
()-seminorm, |uu
h
|
H
1
()
, against the number of elements.
We see that for a given polynomial degree p the discretization errors of the SIPG and the NIPG
discretization almost coincide. Furthermore, we see that for the discretizations with polynomial
degree p = 1, . . . , 5, the discretization error in the H
1
-seminorm is of order O(h
p
) which is in
agreement with the theoretical order of convergence, see Lemma 5.27.
Figure 3a) shows that the error in the L
2
()-norm of the SIPG discretization for the polynomial
degrees p = 1, . . . , 5, is of order O(h
p+1
) which again is in perfect agreement with the theoretical
result, see Lemma 5.29. In comparison to that, Figure 3b) shows the L
2
()-error of the NIPG
discretization. Here, we see that the discretization behaves like O(h
p+1
) for odd p and like O(h
p
)
for even p. This sub-optimal convergence of the NIPG method is attributed to the lack of adjoint
consistency and the resulting lack of smoothness of the adjoint solution, see Lemma 5.29. We note
that similar results for a dierent test case have been obtained in [23].
Figures 2 and 3 show that there is a signicant advantage of using higher order discretizations
over using low order discretization methods. In fact, in Figure 2 we see that the discretization error
in the H
1
()-seminorm for p = 3 on the coarsest mesh is of similar size as the error for p = 1 on
the nest mesh. Similarly, in Figure 3a) the discretization error in the L
2
()-norm for p = 4 on the
coarsest mesh is comparable to the error for p = 1 on the nest mesh. We emphasize that here the
solutions are of similar accuracy although the nest mesh has by a factor of 16384 more elements
53
1e-10
1e-08
1e-06
0.0001
0.01
1
10 100 1000 10000 100000
1
1
1
2
1
3
1
4 1
5
SIPG,p=1
SIPG,p=2
SIPG,p=3
SIPG,p=4
SIPG,p=5
NIPG,p=1
NIPG,p=2
NIPG,p=3
NIPG,p=4
NIPG,p=5
|
u
u
h
|
H
1
(
)
elements
Figure 2: Model problem: The discretization error |u u
h
|
H
1
()
of the SIPG and NIPG methods
with p = 1, . . . , 5 is of order O(h
p
).
1e-12
1e-10
1e-08
1e-06
0.0001
0.01
1
10 100 1000 10000 100000
1
2
1
3
1
4
1
5
1
6
SIPG,p=1
SIPG,p=2
SIPG,p=3
SIPG,p=4
SIPG,p=5
u
h
L
2
(
)
elements
1e-12
1e-10
1e-08
1e-06
0.0001
0.01
1
10 100 1000 10000 100000
1
2
1
2
1
4
1
4 1
6
NIPG,p=1
NIPG,p=2
NIPG,p=3
NIPG,p=4
NIPG,p=5
u
h
L
2
(
)
elements
(a) (b)
Figure 3: Model problem: Convergence of the discretization error u u
h
L
2
()
for a) the SIPG
and b) the NIPG methods with global mesh renement.
1e-12
1e-10
1e-08
1e-06
0.0001
0.01
10 100 1000 10000 100000
SIPG,p=1
SIPG,p=2
SIPG,p=3
SIPG,p=4
SIPG,p=5
u
h
L
2
(
)
DoFs
1e-13
1e-12
1e-11
1e-10
1e-09
1e-08
1e-07
1e-06
1e-05
0.0001
0.001
0.01
0.01 0.1 1 10 100 1000
SIPG,p=1
SIPG,p=2
SIPG,p=3
SIPG,p=4
SIPG,p=5
u
h
L
2
(
)
time (s)
(a) (b)
Figure 4: Model problem: The discretization error u u
h
L
2
()
of the SIPG method plotted a)
against the number of degrees of freedom (DoFs) and b) against the computing time in seconds.
54
than the coarsest mesh. Clearly, a discretization method of higher order requires more degrees
of freedom (DoFs) per element, 25 DoFs/element for p = 4 in comparison to 4 DoFs/element for
p = 1 in this case, but still the p = 1 discretization requires a factor of more than 2600 as many
DoFs for the same accuracy as the p = 4 discretization. In more detail this is seen in Figure 4a)
which plots the L
2
-error against the number of DoFs. The large factor in the number of DoFs
for the specic accuracy translates into a large factor in the computing time required. In fact, in
Figure 4b) we see that the discretization with p = 1 on the nest mesh requires a by a factor of
several thousands larger computing time for the same accuracy as the discretization with p = 4 on
the coarsest mesh.
Admittedly, the model problem considered here is ideal in the sense that the geometry (unit
square) and the governing equations (Poissons equation) are particularly simple, also the solution
is perfectly smooth. However, also for more complicated problems like aerodynamic ows, see e.g.
Section 9.4, a signicant gain of higher order methods over low order methods can be expected.
6 Consistency and adjoint consistency for linear problems
We recall that one of the most important properties of a discretization is its consistency with the
dierential equations to be discretized. In fact, consistency ensures that the right equations are
solved. In nite element methods consistency directly implies the well-known Galerkin orthogonal-
ity. Provided the discretization is stable and using standard interpolation/approximation estimates
this gives optimal a priori order estimates in the so-called energy norm, like e.g. the
H
1-norm
for Poissons equation.
Furthermore, we recall that in continuous as well as in discontinuous Galerkin nite element
methods a duality argument has been applied for deriving error estimates in the L
2
-norm. This
approach introduces an appropriate adjoint (dual) problem, which is then used to represent the
L
2
-norm of the discretization error e = uu
h
in terms of the discretization and the adjoint solution
z. Again by Galerkin orthogonality and by using smoothness properties of the adjoint solution the
L
2
-error estimates are derived, see Theorem 2.17 for the continuous Galerkin discretization and
Lemma 5.29 for the discontinuous Galerkin discretization of Poissons equation.
Optimal order L
2
-error estimates depend on the applicability of the duality argument as well
as on the smoothness of the adjoint solution. Both, however, are connected to the so-called adjoint
consistency of the discretization. As we have seen in Lemma 5.29 adjoint consistency of the SIPG
results in optimal error estimates in the L
2
-norm whereas the lack of adjoint consistency in the
case of the NIPG discretization results in a suboptimal order of convergence in the L
2
-norm.
In the following we introduce the adjoint consistency analysis following [27]. In particular, we
revisit the consistency property and extend the denition of adjoint consistency for the homoge-
neous Dirichlet problem of Poissons equation, see Denition 5.10, to the general case of linear
problems with inhomogeneous boundary conditions. We then give a general framework for ana-
lyzing consistency and adjoint consistency and apply it to the interior penalty DG discretization
of Poissons equation and to the upwind DG discretization of the linear advection equation. The
generalization of this analysis to nonlinear problems will be given in Section 8.5.
6.1 Denition of consistency and adjoint consistency
Let be a bounded open domain in R
d
with boundary . Given the linear problem
Lu = f in , Bu = g on , (273)
55
where f L
2
(), g L
2
(), L denotes a linear dierential operators on , and B denotes a linear
boundary operator on . Let J be a linear target functional given by
J(u) = (j
, u)
+ (j
, Cu)
udx +
_
Cuds, (274)
where j
L
2
(), j
L
2
(), C is an operator on which may be dierential, and (, )
and (, )
denote the L
2
() and L
2
() scalar products, respectively. We assume that the target functional
(274) is compatible with the primal problem (273), i.e. we assume that there are linear operators
L
, B
and C
+ (Bu, C
z)
= (u, L
z)
+ (Cu, B
z)
. (275)
Then, L
, B
and C
z = j
in , B
z = j
on . (276)
In an adjoint-based optimization framework, see e.g. [20], this ensures that
J(u) = (u, j
+ (Cu, j
= (u, L
z)
+ (Cu, B
z)
= (Lu, z)
+ (Bu, C
z)
= (f, z)
+ (g, C
z)
.
(277)
Let be subdivided into shape-regular meshes T
h
= {} consisting of elements and let V
h
be
a discrete function space on T
h
. Furthermore, let problem (273) be discretized as follows: nd
u
h
V
h
such that
B
h
(u
h
, v
h
) = F
h
(v
h
) v
h
V
h
, (278)
where B
h
(, ) is a bilinear form and F
h
() a linear form including the prescribed primal force and
boundary data functions f and g. Then the discretization (278) is said to be consistent if the exact
solution u V to the primal problem (273) satises:
B
h
(u, v) = F
h
(v) v V, (279)
where V is a suitably chosen function space such that u V and V
h
V . Furthermore, the
discretization (278) is said to be adjoint consistent if the exact solution z V to the adjoint
problem (276) satises:
B
h
(w, z) = J(w) w V. (280)
In other words, a discretization is adjoint consistent if the discrete adjoint problem is a consistent
discretization of the continuous adjoint problem.
6.2 The consistency and adjoint consistency analysis
Based on the denition of consistency and adjoint consistency in the previous subsection we now
follow [27] and outline a framework for analyzing consistency and adjoint consistency of discontin-
uous Galerkin discretizations. This framework can also be used to nd specic terms due to which
some DG discretizations may not be adjoint consistent. In these cases the analysis gives some
insight into how an adjoint inconsistent DG discretization together with a specic target functional
could be modied to recover an adjoint consistent discretization.
Given a primal problem, a discontinuous Galerkin discretization of the problem and a target
functional, the adjoint consistency analysis consists of the following steps:
56
Derivation of the continuous adjoint problem: Let the primal problem be given by
(273). Furthermore, assume that J() is a linear (or linearized) target functional as in (274)
which is compatible with (273). Then, we derive the continuous adjoint problem with con-
tinuous adjoint boundary conditions as given in (276).
Consistency analysis of the discrete primal problem: We rewrite the discontinuous
Galerkin discretization (278) of problem (273) in following element-based primal residual
form: nd u
h
V
h
such that
_
R(u
h
)v
h
dx +
T
h
_
\
r(u
h
)v
h
+(u
h
)
h
v
h
ds
+
_
(u
h
)v
h
+
(u
h
)
h
v
h
ds = 0 v
h
V
h
, (281)
where R(u
h
) denotes the element residual, r(u
h
) and (u
h
) denote the interior face residuals,
and r
(u
h
) and
(u
h
) denote the boundary residuals. We note, that this is a generalization
of (159) to include also
h
v
h
terms. According to (279) the discretization (278) is consistent
if the exact solution u to (273) satises
_
R(u)v dx +
T
h
_
\
r(u)v +(u)
h
v ds
+
_
(u)v +
(u)
h
v ds = 0 v V, (282)
which holds provided u satises
R(u) = 0 in , T
h
,
r(u) = 0, (u) = 0 on \ , T
h
, (283)
r
(u) = 0,
(u) = 0 on .
Adjoint consistency of element, interior face and boundary terms: Given the dis-
cretization (278) and the target functional (274), we rewrite the discrete adjoint problem:
nd z
h
V
h
such that
B
h
(w
h
, z
h
) = J(w
h
) w
h
V
h
, (284)
in following element-based adjoint residual form: nd z
h
V
h
such that
_
w
h
R
(z
h
) dx +
T
h
_
\
w
h
r
(z
h
) +w
h
(z
h
) ds
+
_
w
h
r
(z
h
) +w
h
(z
h
) ds = 0 w
h
V
h
, (285)
where R
(z
h
), r
(z
h
),
(z
h
), r
(z
h
) and
(z
h
) denote the element, interior face and bound-
ary adjoint residuals, respectively. According to (280) the discretization (278) is adjoint
consistent if the exact solution z V to (276) satises
_
wR
(z) dx +
T
h
_
\
wr
(z) +w
(z) ds
+
_
wr
(z) +w
(z) ds = 0 w V, (286)
57
which holds provided z satises
R
(z) = 0 in , T
h
,
r
(z) = 0,
(z) = 0 on \ , T
h
, (287)
r
(z) = 0,
(z) = 0 on .
We note that the adjoint problem and consequently the adjoint consistency of a discretization
depends on the specic target functional J() under consideration. Given a target functional of the
form (274), we see that R
(z) depends on j
(), and r
(z) depends on j
J(u
h
) = J(i(u
h
)) +
_
r
J
(u
h
) ds, (288)
where i() and r
J
() are functions to be specied. A modication of a target functional is called
consistent if
J(u) = J(u) holds for the exact solution u. Thereby, the modication in (288) is
consistent if the exact solution u satises i(u) = u and r
J
(u) = 0. Although the true value of the
target functional is unchanged,
J(u) = J(u), the computed value J(u
h
) of the target functional
is modied, and more importantly for nonlinear
J functionals,
J
[u
h
] diers from J
[u
h
]. This
modication can be used to recover an adjoint consistent discretization. We note, that (288) is not a
unique choice of a consistent modication of J(); other examples are
J(u
h
) = J(u
h
)+
_
R
J
(u
h
) dx,
with R
J
(u) = 0; or
J(u
h
) = m(J(u
h
), J(i(u
h
))) with i(u) = u and m(j, j) = j. However the
consistent modication as given in (288) will be sucient for our purpose.
In the following subsections we will perform the complete consistency and adjoint consistency
analysis as outlined above for the interior penalty discontinuous Galerkin discretization of the
Dirichlet-Neumann problem (162) of Poissons equation and for the upwind discontinuous Galerkin
discretization of the linear advection equation.
6.3 Adjoint consistency analysis of the IP discretization
We consider the Dirichlet-Neumann boundary value problem (162) of Poissons equation,
u = f in , u = g
D
on
D
, n u = g
N
on
N
, (289)
where f L
2
(), g
D
L
2
(
D
) and g
N
L
2
(
N
) are given functions. We assume that
D
and
N
are disjoint subsets with union . We also assume that
D
is nonempty.
6.3.1 The continuous adjoint problem to Poissons equation
In order to derive the continuous adjoint problem, we multiply the left hand side of (289) by z and
integrate twice by parts over the domain . Thereby, we obtain
(u, z)
= (u, z)
(n u, z)
= (u, z)
+ (u, n z)
(n u, z)
.
Splitting the boundary terms according to =
D
N
and shuing terms we arrive at
(u, z)
+ (u, n z)
D
+ (n u, z)
N
= (u, z)
+ (n u, z)
D
+ (u, n z)
N
.
Comparing with the compatibility condition (275), we see that for Lu = u in and
Bu = u, Cu = n u on
D
,
Bu = n u, Cu = u on
N
,
58
the adjoint operators are given by L
z = z on and
B
z = z, C
z = n z on
D
,
B
z = n z, C
z = z on
N
.
In particular, for
J(u) =
_
udx +
_
Cuds
=
_
udx +
_
D
j
D
n uds +
_
N
j
N
uds,
(290)
the continuous adjoint problem is given by
z = j
in , (291)
subject to the boundary conditions
z = j
D
on
D
, n z = j
N
on
N
. (292)
6.3.2 Primal residual form of the interior penalty DG discretization
We begin by recalling the unied form of the method by Baumann-Oden, of the symmetric and
the non-symmetric interior penalty discontinuous Galerkin discretization of (289) as given in (207)
and (208): nd u
h
V
h
such that
B
h
(u
h
, v
h
) = F
h
(v
h
) v
h
V
h
, (293)
where
B
h
(u, v) =
_
h
u
h
v dx
+
_
D
([[u]] {{
h
v}} {{
h
u}} [[v]]) ds +
_
D
[[u]] [[v]] ds,
F
h
(v) =
_
fv dx +
_
D
g
D
n v ds +
_
D
g
D
v ds +
_
N
g
N
v ds. (294)
For = 0 and = 1 this represents the symmetric, whereas for = 0 and = 1 the non-
symmetric version of the interior penalty DG method. Furthermore, for = 0 and = 1 this
scheme reduces to the method of Baumann-Oden. The discretization is given in face-based form,
i.e. in terms of
_
I
. In order to rewrite this in the element-based primal residual form as given in
(281) we rst must rewrite B
h
(, ) in (294) in element-based formulation, i.e. in terms of
.
However, before doing so we introduce some more notation: In addition to the jump operator [[]]
dened in Denition (5.2) we dene the jump operator [].
Denition 6.1 Let e
I
be an interior edge between two adjacent elements
+
and
. Let
q T(T
h
) be the traces of a scalar. Then, we dene the jump operator [] by
[q] = q
+
q
on e
I
,
[q] = q
+
on e ,
Remark 6.2 The jump [] has already been used in the denition of the generic ux in (117).
Remark 6.3 Note that [[q]] = [q]n and [[q]] n = [q] for all q T(T
h
).
59
Then we show following two lemmas which will later be used to transfer from face-based to element-
based formulations.
Lemma 6.4 Let q, T(T
h
) and H
1
(T
h
), then
_
[[q]][[]] ds =
[q]ds, (295)
_
[[q]] {{
h
}} ds =
qn {{
h
}} ds, (296)
_
[[q]] {{
h
}} ds =
1
2
_
\
[[q]]
h
ds +
_
qn
h
ds. (297)
Proof: We have
_
\
[q]ds =
_
\
[[q]] nds =
_
I
[[q]] (n
+
+
+n
) ds =
_
I
[[q]] [[]] ds,
_
\
qn {{
h
}} ds =
_
I
_
q
+
n
+
+q
_
{{
h
}} ds =
_
I
[[q]] {{
h
}} ds,
1
2
_
\
[[q]]
h
ds =
_
I
[[q]] {{
h
}} ds.
and use the denitions of [[]] and {{}} on .
Lemma 6.5 Let q T(T
h
) and H
1
(T
h
), then
q
+
n
+
h
+
= q
+
n
+
{{
h
}} +
1
2
q
+
[[
h
]]. (298)
Proof:
q
+
n
+
h
+
=
1
2
(q
+
n
+
h
+
+q
+
n
+
h
) +
1
2
(q
+
n
+
h
+
q
+
n
+
h
)
= q
+
n
+
{{
h
}} +
1
2
q
+
[[
h
]].
h
u
h
h
v
h
dx +
_
\
1
2
[[u
h
]]
h
v
h
ds (299)
_
\
N
{{
h
u
h
}} nv
h
ds +
_
\
N
[u
h
]v
h
ds +
_
D
u
h
n
h
v
h
ds.
Using integration by parts on each element and summing over all elements yields
_
h
u
h
h
v
h
dx =
_
h
u
h
v
h
dx +
h
u
h
nv
h
ds. (300)
Substituting this into (299) we obtain
B
h
(u
h
, v
h
)
_
h
u
h
v
h
dx +
h
u
h
nv
h
ds +
_
\
1
2
[[u
h
]]
h
v
h
ds
_
\
N
{{
h
u
h
}} nv
h
ds +
_
\
N
[u
h
]v
h
ds +
_
D
u
h
n
h
v
h
ds.
60
This is simplied by using (298) for q := v
h
and := u
h
on \ and
h
u
h
{{
h
u
h
}} = 0 on
D
. In summary, the discretization (293) and (294) can be rewritten as
B
h
(u
h
, v
h
)
_
h
u
h
v
h
dx +
_
N
n
h
u
h
v
h
ds +
_
\
1
2
[[u
h
]]
h
v
h
ds
+
_
\
_
1
2
[[
h
u
h
]] +[u
h
]
_
v
h
ds +
_
D
u
h
n
h
v
h
ds +
_
D
u
h
v
h
ds
=
_
fv
h
dx +
_
D
g
D
n v
h
ds +
_
D
g
D
v
h
ds +
_
N
g
N
v
h
ds. (301)
This can be expressed in element-based residual form as follows: nd u
h
V
h
such that
_
R(u
h
)v
h
dx +
T
h
_
\
r(u
h
)v
h
+(u
h
)
h
v
h
ds
+
_
(u
h
)v
h
+
(u
h
)
h
v
h
ds = 0 v
h
V
h
,
where the residuals are given by R(u
h
) = f +
h
u
h
on , and
r(u
h
) =
1
2
[[
h
u
h
]] [u
h
], (u
h
) =
1
2
[[u
h
]] on \ , T
h
,
r
(u
h
) = (g
D
u
h
),
(u
h
) = (g
D
u
h
)n on
D
, (302)
r
(u
h
) = g
N
n
h
u
h
,
(u
h
) = 0 on
N
.
In particular, we see that the exact solution u H
2
() to Poissons equation (162) satises R(u) =
0, r(u) = 0, (u) = 0, r
(u) = 0 and
R(u)v dx +
T
h
_
\
r(u)v +(u)
h
v ds +
_
(u)v +
(u)
h
v ds = 0,
for all v V which is equivalent to u satisfying
B
h
(u, v) = F
h
(v) v V,
i.e. the DG discretizations as given in (293) or (301) are consistent. We note that we have shown
this property already in Corollary 5.15 which is based on the consistency of the numerical uxes
u(u), (u, u) in Theorem 5.7. In this subsection we now analyzed consistency using an alternative
way which is based on the primal residual form of the discretization. Here, consistency can easily
be checked based on primal residuals which do (or do not) vanish for the exact solution u of the
underlying equations.
We end this section by noting that from (301) we see that the discrete solution u
h
satises
following problem in a weak sense
u = f in , (303)
subject to inter-element conditions
1
2
[[
h
u]] +[u] = 0 on \ ,
[u] = 0 on \ ,
(304)
and boundary conditions
u = g
D
on
D
,
n
h
u = g
N
on
N
.
(305)
In fact, this is the mesh-dependent counterpart of the original equations (162) to be solved.
61
6.3.3 Adjoint residual form of the interior penalty DG discretization
Given the target functional dened in (290), the discrete adjoint problem (284) to the discretization
(293) and (294) is given by: nd z
h
V
h
such that
_
h
w
h
h
z
h
dx +
_
D
([[w
h
]] {{
h
z
h
}} {{
h
w
h
}} [[z
h
]] +[[w
h
]] [[z
h
]]) ds = J(w
h
),
for all w
h
V
h
. Then, in element-based form, we have: nd z
h
V
h
such that
_
h
w
h
h
z
h
dx +
_
\
N
w
h
(n {{
h
z
h
}} +[z
h
]) ds
1
2
_
\
h
w
h
[[z
h
]] ds
_
h
w
h
nz
h
ds = J(w
h
), (306)
for all w
h
V
h
. Using (300) with u
h
and v
h
replaced by z
h
and w
h
, and using (298) with q and
replaced by w
h
and z
h
yields
_
h
z
h
h
w
h
dx =
_
h
z
h
w
h
dx +
w
h
n
h
z
h
ds
=
_
w
h
h
z
h
dx +
_
w
h
n {{
h
z
h
}} +
1
2
w
h
[[
h
z
h
]]
_
ds.
Substituting this into (306) we obtain
w
h
h
z
h
dx +
_
\
w
h
_
1
2
[[
h
z
h
]] + (1 +)n {{
h
z
h
}} +[z
h
]
_
ds
1
2
_
\
h
w
h
[[z
h
]] ds +
_
N
w
h
n
h
z
h
ds
+
_
D
w
h
((1 +)n
h
z
h
+z
h
) ds
_
h
w
h
nz
h
ds
=
_
w
h
j
dx +
_
D
w
h
nj
D
ds +
_
N
w
h
j
N
ds, (307)
and we arrive at the element-based adjoint residual form: nd z
h
V
h
such that
_
w
h
R
(z
h
) dx +
T
h
_
\
w
h
r
(z
h
) +w
h
(z
h
) ds
+
_
w
h
r
(z
h
) +w
h
(z
h
) ds = 0 w
h
V
h
, (308)
where the adjoint residuals are given by R
(z
h
) = j
+
h
z
h
on , by
r
(z
h
) =
1
2
[[
h
z
h
]] (1 +)n {{
h
z}} [z
h
],
(z
h
) =
1
2
[[z
h
]], (309)
on interior faces \ , T
h
, and by
r
(z
h
) = (1 +)n
h
z
h
z
h
,
(z
h
) = (j
D
+z
h
)n on
D
,
r
(z
h
) = j
N
n
h
z
h
,
(z
h
) = 0 on
N
. (310)
From (309) we see that the exact solution z to the adjoint problem (291) satises r
(z) = 0
provided = 1. Furthermore, we have
(z) = 0 and R
(z) = 0 and
(z) = 0 on
N
, the SIPG method is also adjoint consistent on
N
. However,
on
D
the requirements r
(z) = 0 and
[u
h
](w
h
) w
h
V
h
, (312)
where
[u
h
](w
h
) = J
[u
h
](w
h
) +
_
D
r
J
[u
h
](w
h
) ds = J(w
h
)
_
D
w
h
j
D
ds. (313)
Then, the adjoint residuals on
D
are given by
r
(z
h
) = j
D
(1 +)n
h
z
h
z
h
,
(z
h
) = (j
D
+z
h
)n, on
D
, (314)
which vanish for z = j
D
. Thereby, the SIPG method is adjoint consistent also on
D
. Finally,
we see that the discrete adjoint solution z
h
must satisfy following problem in a weak sense
z = j
in , T
h
, (315)
subject to inter-element conditions
1
2
[[
h
z]] + (1 +)n {{
h
z}} +[z] = 0 on \ , T
h
,
[z] = 0 on \ ,
(316)
and boundary conditions
z = j
D
on
D
,
(1 +)n
h
z +z = j
D
on
D
, (317)
n
h
z = j
N
on
N
.
Note, that for = 1 there is a correspondence to the primal equations (303)-(305). In fact,
the discrete adjoint equations correspond to the discrete primal equations, with u, f, g
D
and
g
N
replaced by z, j
, j
D
and j
N
respectively; I.e. the discrete adjoint equation to the SIPG
discretization is equivalent to the SIPG discretization of the continuous adjoint equation.
Finally, we note that [22] considers the target functional J(u) =
_
0
n uj
D
ds,
0
D
,
which is a special case of (290) with j
0 in , j
N
0 on
N
and j
D
0 on
D
\
0
. Numerical
experiments in [22] have shown, that the (discrete) adjoint solution associated with this target
functional is non-smooth near
0
. Furthermore, it has been demonstrated that either by setting
= 0 on
0
, or by modifying the target functional appropriately, this eect vanishes, and the
adjoint solution becomes smooth. We note, that the modication of the target functional supposed
in [22] is connected to (313). However, here, we followed [27] and derived (313) in the more general
framework of consistent modications of target functionals, see (288).
63
6.4 Adjoint consistency analysis of the upwind DG discretization
In this section we apply the consistency and adjoint consistency analysis outlined in Section 6.2 to
the upwind discontinuous Galerkin discretization of the linear advection equation.
We begin by recalling the linear advection equation:
(bu) +cu = f in , u = g on
, (318)
where f L
2
(), b [C
1
()]
d
, c L
() and g L
2
(
), where
+ (u, b nz)
= (u, b z +cz)
+ (u, b nz)
+
. (320)
Comparing with (275), we see that for Lu = (bu) +cu in and
Bu = u, Cu = 0 on
,
Bu = 0, Cu = u on
+
,
the adjoint operators are given by L
z = b z +cz in and
B
z = 0, C
z = b nz on
,
B
z = b nz, C
z = 0 on
+
.
In particular, for
J(u) =
_
udx +
_
Cuds =
_
udx +
_
+
j
uds, (321)
the continuous adjoint problem is given by
b z +cz = j
in , (322)
subject to the boundary condition
b nz = j
on
+
. (323)
6.4.2 Primal residual form of the DG discretization based on upwind
We recall the discontinuous Galerkin discretization of the linear advection equation based on the
upwind ux: nd u
h
V
d
h,p
such that
(bu
h
)
h
v
h
dx +
_
cu
h
v
h
dx +
T
h
_
\
b nu
h
v
+
h
ds +
T
h
_
+
b nu
+
h
v
+
h
ds
=
_
fv
h
dx
_
b ngv
h
ds v
h
V
d
h,p
. (324)
64
Furthermore, we recall the primal residual form as given in Section 4.7.3: nd u
h
V
d
h,p
such that
T
h
_
R(u
h
)v dx +
T
h
_
\
r(u
h
)v ds +
_
(u
h
)v ds = 0 v V
d
h,p
, (325)
where R(u
h
), r(u
h
) and r
(u
h
) denote the element, interior face and boundary residuals, respec-
tively, given by
R(u
h
) = f
h
(bu
h
) cu
h
in , T
h
,
r(u
h
) = b n(u
+
h
u
h
) on
\ , T
h
,
r
(u
h
) = b n(u
h
g) on
,
and r
(u
h
) 0 on
+
. I.e. we have (282) with
(u
h
) 0 and (u
h
) 0. Furthermore, we see
that the exact solution u H
1,b
() to (318) satises (282) with R(u) = 0, r(u) = 0 and r
(u) = 0.
Thereby, (324) is a consistent discretization of (318). Finally, we see that the discrete solution u
h
to (318) must satisfy following problem in a weak sense
(bu) +cu = f in , T
h
, (326)
subject to inter-element conditions
b n[u] = 0 on \ , T
h
, (327)
and boundary conditions
b nu = b ng on
. (328)
This is the mesh-dependent counterpart of the original equations (318) to be solved.
6.4.3 Adjoint residual form of the DG discretization based on upwind
Substituting
T
h
_
\
b nu
h
v
+
ds =
T
h
_
+
\
b nu
+
h
v
ds
in (324) we nd that the discrete adjoint problem (284) to the discretization (324) is given by: nd
z
h
V
h
such that
B
h
(w
h
, z
h
)
_
w
h
(b
h
z
h
+cz
h
) dx +
+
\
w
+
h
b n[z
h
] ds = J(w
h
),
for all w
h
V
h
. Hence, for a target functional J() as in (321), we have (285) with
R
(z
h
) = j
+b
h
z
h
cz
h
in , T
h
,
r
(z
h
) = b n[z
h
] on \ , T
h
,
r
(z
h
) = j
b nz
h
on
+
,
r
(z
h
) 0 on
(z
h
) 0 on \ , T
h
, and
(z
h
) 0 on . As (286) with (287)
holds for the exact solution z to (322) and (323), we conclude, that (324) is an adjoint consistent
discretization of (318). Furthermore, we see that the discrete adjoint solution z
h
must satisfy
following problem in a weak sense
b z +cz = j
in , T
h
, (329)
subject to inter-element conditions
b n[z] = 0 on \ , T
h
, (330)
and boundary conditions
b nz = j
on
+
. (331)
This is the mesh-dependent counterpart of the adjoint equations given in Section 6.4.1.
65
7 A priori error estimates for target functionals J()
In this section we derive a priori error estimates with respect to target functionals J() for adjoint
consistent and adjoint inconsistent discontinuous Galerkin discretizations. In particular, we will see
that analogous to the (sub-)optimal order of convergence in the L
2
-norm for adjoint (in-)consistent
discretizations also the order of convergence in J() is (sub-)optimal.
We consider following general linear problem:
Lu = f in , Bu = g on , (332)
where f L
2
(), g L
2
(), L denotes a linear dierential operator on and B a linear dierential
boundary operator on . Let (332) be discretized as follows: nd u
h
V
d
h,p
such that
B
h
(u
h
, v
h
) = F
h
(v
h
) v
h
V
d
h,p
, (333)
where the bilinear form B
h
(, ) is continuous on V with respect to a specic | |-norm, i.e.
B
h
(w, v) C
B
|w| |v| w, v V.
Here, V is suitably chosen function space such that u
V V and V
h
V but possibly V
h
V .
Remark 7.1 We recall that for the DG discretization of Poissons equation we have
V = H
2
()
and V = H
2
(T
h
), and for the DG discretization of the linear advection equation we have
V =
H
1,b
() and V = H
1,b
(T
h
).
We assume that the discretization (333) is consistent, i.e. the exact solution u
V V to
(332) satises
B
h
(u, v) = F
h
(v) v V, (334)
which implies the Galerkin orthogonality
B
h
(u u
h
, v
h
) = 0 v
h
V
d
h,p
. (335)
Furthermore, we assume that following a priori error estimate in the | |-norm holds: There are
constants C > 0 and r = r(p) > 0 such that
|u u
h
| Ch
r
|u|
H
p+1
()
u H
p+1
(). (336)
Finally, we assume that the local projection operator P
d
h,p
as dened in Section 4.5 satises following
approximation estimate in the | | norm: There are constants C > 0 and r = r(p) > 0 such that
|v P
d
h,p
v| Ch
r
|v|
H
p+1
()
v H
p+1
(). (337)
In the following we want to measure the discretization error e = uu
h
not in some global norm
like the | |-norm but with respect to target functionals J() of the form
J(u) = (j
, u)
+ (j
, Cu)
udx +
_
Cuds, (338)
where j
L
2
() and j
L
2
(), and C is an dierential boundary operator on . We assume
that J() is compatible with (332) as dened in Section 6.1. Then, there are dierential operators
L
, B
and C
which are the adjoint operators to L, B and C, respectively, and the continuous
adjoint problem is given by:
L
z = j
in , B
z = j
on . (339)
Finally, we recall that a discretization together with a target functional is called adjoint consistent,
if the exact solution z
V V to the adjoint problem (339) satises:
B
h
(w, z) = J(w) w V. (340)
66
Theorem 7.2 (A priori error estimates in J()) Let the situation be as described above. Fur-
thermore, assume that j
and j
, (354)
and recall its DG discretization based on the upwind ux: nd u
h
V
d
h,p
such that
B
h
(u
h
, v
h
)
_
(bu
h
)
h
v
h
dx +
_
cu
h
v
h
dx +
T
h
_
\
b nu
h
v
+
h
ds
+
T
h
_
+
b nu
+
h
v
+
h
ds =
_
fv
h
dx
_
b ngv
h
ds v
h
V
d
h,p
. (355)
This discretization is consistent, see Section 6.4.2. Furthermore, the bilinear form B
h
is continuous
with respect to the | |
b
0
-norm given by
|v|
2
b
0
= c
0
v
2
+
e
I
_
e
b
0
[v]
2
ds +
1
2
_
|b n| v
2
ds. (356)
where b
0
=
1
2
|b n|.
Furthermore, we have an approximation estimate for the local projection operator P
d
h,p
:
|v P
d
h,p
v|
b
0
Ch
p+1/2
|v|
H
p+1
()
v H
p+1
(), (357)
see [proof of Theorem 8.20]. Finally, we recall the a priori error estimate in Theorem 4.17: There
is a constant C > 0 such that
|u u
h
|
b
0
Ch
p+1/2
|u|
H
p+1
()
u H
p+1
(), (358)
68
which in view of (357) is optimal. In the following we give a priori error estimates for the dis-
cretization error e = u u
h
measured in terms of target quantities J() of the form
J(u) =
_
udx +
_
+
j
uds, (359)
which are compatible with the linear advection, see Section 6.4.1. The corresponding continuous
adjoint problem is
b z +cz = j
in , b nz = j
on
+
. (360)
Finally, we recall from Section 6.4.3 that the discretization (355) together with the target functional
(359) is adjoint consistent.
If we now had continuity of the bilinear form B
h
(, ) in (355) all necessary conditions would
be fullled in order to employ Theorem 7.2a) for obtaining an a priori error estimate in J().
However, continuity of B
h
(, ), i.e. |B
h
(u, v)| C|u|
b
0
|v|
b
0
for all u, v H
1,b
(T
h
), is not
available. Nevertheless, there are alternative proofs which result in following estimate:
Corollary 7.4 (A priori error estimate in J()) Let B
h
(, ) and J() be given by (355) and
(359), respectively. Let j
and j
t+1
|u|
H
t+1
()
|z|
H
t+1
()
u H
s+1
(), (362)
where t = min(s, p) and
t = min( s, p).
7.2 IP DG discretization for Poissons equation: Estimates in J()
We consider the Dirichlet-Neumann boundary value problem of Poissons equation,
u = f in , u = g
D
on
D
, n u = g
N
on
N
, (363)
and its symmetric ( = 1) and unsymmetric ( = 1) interior penalty DG discretization given by:
nd u
h
V
d
h,p
such that
_
h
u
h
h
v
h
dx +
_
D
([[u
h
]] {{
h
v
h
}} {{
h
u
h
}} [[v
h
]]) ds +
_
D
[[u
h
]] [[v
h
]] ds,
=
_
fv
h
dx +
_
D
g
D
n v
h
ds +
_
D
g
D
v
h
ds +
_
N
g
N
v
h
ds v
h
V
d
h,p
. (364)
69
This discretization is consistent, see Section (6.3.2). We recall, see Lemma 5.19, that B
h
given by
the left hand side in (364) is continuous,
B
h
(w, v) C
B
|w|
| v|
w, v V,
with respect to the | |
2
-norm dened by
|v|
2
=
h
v
2
L
2
()
+
_
1
(n {{v}})
2
ds +
_
D
[v]
2
ds.
Furthermore, we have an approximation estimate for the local projection operator P
d
h,p
:
|v P
d
h,p
v|
Ch
p
|v|
H
p+1
()
v H
p+1
(), (365)
see proof of Lemma 5.27. Finally, we recall the a priori error estimate in Lemma 5.27: There is a
constant C > 0 such that
|u u
h
|
Ch
p
|u|
H
p+1
()
u H
p+1
(), (366)
which in view of (365) is optimal. In the following we give a priori error estimates for the dis-
cretization error e = u u
h
measured in terms of target quantities J() of the form
J(u) =
_
udx +
_
D
j
D
n uds +
_
N
j
N
uds, (367)
which are compatible with Poissons equation, see Section 6.3.1. The corresponding continuous
adjoint problem is
z = j
in , z = j
D
on
D
, n z = j
N
on
N
. (368)
Finally, we recall from Section 6.4.3 that the symmetric version of the IP discretization (364)
together with following consistent modication
J(u
h
) = J(u
h
)
_
D
(u
h
g
D
)j
D
ds,
[u
h
](w
h
) = J(w
h
)
_
D
w
h
j
D
ds
(369)
of the target functional (367) is adjoint consistent whereas the unsymmetric version is adjoint incon-
sistent. Having collected these results we now can use Theorem 7.2 to obtain following estimates.
Corollary 7.6 (A priori error estimate in J()) Let B
h
(, ) be given by the left hand side of
(364). Furthermore, let
J() be given by (369) and (367). Let j
, j
D
and j
N
in (367) be smooth
functions such that the adjoint solution z to (368) is smooth, z H
p+1
(). Then, we have following
estimate for an adjoint consistent SIPG discretization:
|J(u)
J(u
h
)| Ch
2p
|u|
H
p+1
()
|z|
H
p+1
()
u H
p+1
(). (370)
and for an adjoint inconsistent IP discretization, e.g. NIPG:
|J(u) J(u
h
)| Ch
p
|u|
H
p+1
()
u H
p+1
(). (371)
Proof: Use (341) in Theorem 7.2 with r(p) = r(p) = p and (342) with r(p) = p.
70
In this proof we applied statement a) of Theorem 7.2. In the following we give an alternative proof,
see also [23], which is connected to part c) of Theorem 7.2 as well as to the proof of Lemma 5.29.
Proof: (Alternative proof of Corollary 7.6, [23]): Let B
s
h
(, ) and B
n
h
(, ), and F
s
h
() and
F
n
h
() denote the bilinear and linear forms in (364) for, respectively, the symmetric ( = 1) and
nonsymmetric ( = 1) interior penalty DG discretizations. Then the discrete solutions u
s
h
and u
n
h
satisfy the following problems: nd u
s
h
V
d
h,p
such that
B
s
h
(u
s
h
, v
h
) = F
s
h
(v
h
) v
h
V
d
h,p
; (372)
and nd u
n
h
V
d
h,p
such that
B
n
h
(u
n
h
, v
h
) = F
n
h
(v
h
) v
h
V
d
h,p
, (373)
respectively. Furthermore, let z
s
and z
n
be the analytical solutions to following adjoint problems:
nd z
s
H
2
(T
h
) such that
B
s
h
(w, z
s
) = J(w) w H
2
(T
h
); (374)
and nd z
n
H
2
(T
h
) such that
B
n
h
(w, z
n
) = J(w) w H
2
(T
h
), (375)
respectively. Then we obtain
J(u) J(u
s
h
) = B
s
h
(u u
s
h
, z
s
) = B
s
h
(u u
s
h
, z
s
z
h
), (376)
due to Galerkin orthogonality, where z
h
V
d
h,p
is any discrete function. We note that due to the
adjoint consistency of B
s
h
the solution z
s
to the adjoint problem (374) coincides with the solution z
to the continuous adjoint problem (368) and thus is smooth. By using coercivity of B
s
h
, smoothness
of the adjoint solution z
s
H
p+1
(), and estimates (366) and (365) in (376) we obtain (370) for
the SIPG discretization. Similarly, for the NIPG discretization we obtain
J(u) J(u
n
h
) = B
n
h
(u u
n
h
, z
n
) = B
n
h
(u u
n
h
, z
n
z
h
). (377)
In the following we rewrite the error representation (377) in terms of the (smooth) adjoint solution
z
s
instead of the solution z
n
(374) which is mesh-dependent and in general non-smooth. Before we
proceed, we note that
B
n
h
(w
h
, v
h
) = B
s
h
(w
h
, v
h
) + 2
_
D
[[w
h
]] {{
h
v
h
}} dx.
Thereby and by using the Galerkin orthogonality of the NIPG discretization we obtain:
J(u) J(u
n
h
) = B
n
h
(u u
n
h
, z
n
) = B
s
h
(u u
n
h
, z
s
)
= B
n
h
(u u
n
h
, z
s
) 2
_
D
[[u u
n
h
]] {{
h
z
s
}} dx
= B
n
h
(u u
n
h
, z
s
z
h
) 2
_
D
[[u u
n
h
]] {{
h
z
s
}} dx,
Ch
2p
|u|
H
p+1
()
|z|
H
p+1
()
+C|u u
h
|
z
s
H
2
(T
h
)
Ch
2p
|u|
H
p+1
()
|z|
H
p+1
()
+Ch
p
|u|
H
p+1
()
z
s
H
2
(T
h
)
Ch
p
|u|
H
p+1
()
,
where we used z
s
H
2
(T
h
)
C. Hence we have (371).
We see, that the order of convergence in J() is O(h
2p
) for SIPG and O(h
p
) for NIPG provided
both primal and adjoint solutions are smooth, u H
p+1
() and z H
p+1
(). If, however, u or z
are less regular we obtain an estimate with a correspondingly reduced order of convergence in J():
71
1e-12
1e-11
1e-10
1e-09
1e-08
1e-07
1e-06
1e-05
0.0001
0.001
0.01
10 100 1000 10000 100000
1
2
1
4
1
6
1
8
SIPG,p=1
SIPG,p=2
SIPG,p=3
SIPG,p=4
|
J
1
(
u
)
J
1
(
u
h
)
|
elements
1e-12
1e-11
1e-10
1e-09
1e-08
1e-07
1e-06
1e-05
0.0001
0.001
0.01
10 100 1000 10000 100000
1
2
1
2
1
4
1
4
1
6
NIPG,p=1
NIPG,p=2
NIPG,p=3
NIPG,p=4
NIPG,p=5
|
J
1
(
u
)
J
1
(
u
h
)
|
elements
(a) (b)
Figure 5: Example 1: Convergence of the error |J
1
(u) J
1
(u
h
)| for a) the SIPG and b) the NIPG
discretizations with global mesh renement.
Corollary 7.7 (A priori error estimate in J() with reduced regularity) Let B
h
(, ) be given
by the left hand side of (364) and let
J() be given by (369) and (367). Assume that u H
s+1
()
and z H
s+1
() hold for the exact solutions u and z to the primal and adjoint problems (363) and
(368), respectively. Then, we have following estimates for an adjoint consistent SIPG discretization:
|J(u) J(u
h
)| Ch
t+
t
|u|
H
t+1
()
|z|
H
t+1
()
u H
s+1
(), (378)
and for an adjoint inconsistent discretization, e.g. NIPG:
|J(u) J(u
h
)| Ch
t
|u|
H
t+1
()
u H
s+1
(), (379)
where t = min(s, p) and
t = min( s, p).
7.3 Numerical results
Example 1 We begin by investigating the experimental order of convergence of the SIPG and
NIPG discretizations when measuring the error in terms of specic target quantities J().
To this end we revisit the experimental model problem introduced in Section 5.6. This problem
is based on Poissons equations with an inhomogeneous Dirichlet boundary value function g
D
and
a forcing function f chosen so that the analytical solution u is given by (272).
First, we choose the target quantity to represent the (weighted) mean value of u over , i.e.
J
1
(u
h
) =
_
u
h
dx; (380)
here, we dene the weight function j
by
j
(x) = sin(x
1
) sin(x
2
).
Thereby the true value of the target quantity is given by J(u) = 0.1801265486975. We note that
the target quantity (380) is compatible with Poissons equation (289). In fact, it is a special case
of the target quantity given in (290) with
N
= and j
D
= 0 on
D
= .
Figure 5a) shows the error of the SIPG discretization measured in terms of the target quantity
J
1
() given by (380). We see that under global mesh renement the error |J
1
(u) J
1
(u
h
)| behaves
like O(h
2p
) which is in perfect agreement with the theoretical order of convergence, see estimate
(370). Figure 5b) shows the respective plot for the NIPG discretization. Here, we see that the
error |J
1
(u) J
1
(u
h
)| behaves like O(h
p+1
) for odd p and like O(h
p
) for even p. This convergence
72
1e-12
1e-10
1e-08
1e-06
0.0001
0.01
1
10 100 1000 10000
1
1
1
2
1
3
SIPG,p=1
SIPG,p=2
SIPG,p=3
|
J
2
(
u
)
J
2
(
u
h
)
|
elements
1e-12
1e-10
1e-08
1e-06
0.0001
0.01
1
10 100 1000 10000
1
4
1
6
1
8
SIPG,p=1
SIPG,p=2
SIPG,p=3
|
J
2
(
u
)
J
2
(
u
h
)
|
elements
(a) (b)
Figure 6: Example 2: Convergence of a) the error |J
2
(u) J
2
(u
h
)| and b) the error |J
2
(u)
J
2
(u
h
)|
for the SIPG discretization with global mesh renement.
behavior is similar to the convergence behavior in the L
2
()-norm encountered for the NIPG scheme
in Section 5.6. Again, due to the lack of adjoint consistency the order of convergence in J
1
() for
the NIPG scheme is lower than in the case of the adjoint consistent SIPG scheme. We note that
similar results for a dierent test case have been obtained in [23].
Example 2 We consider the same model problem as in the previous example. However, instead
of the mean value quantity (380) here we now choose the target quantity to represent the mean
value of the normal derivative of u over the boundary = , i.e.
J
2
(u
h
) =
_
j
D
n
h
u
h
ds, (381)
with j
D
1 on
D
= . Thereby the true value of the target quantity is given by J(u) = 2. We
note that this target quantity is compatible with Poissons equation (289). In fact, it is a special
case of the target quantity given in (290) with
N
= , and j
= 0 on . Furthermore, we note
that the solution z to the corresponding continuous adjoint problem
z = 0 in , z = j
D
on
D
(382)
is given by z 1 on . Figure 6a) shows that the error |J
2
(u)J
2
(u
h
)| behaves like O(h
p
) for the
SIPG discretizations with p = 1, 2, 3. Following the discussion in Section 6.3.3 we recognize that the
SIPG discretization in combination with the target functional J
2
() in (381) is adjoint inconsistent.
Thus the order of convergence O(h
p
) encountered in Figure 6a) is, in fact, the expected order of
convergence for this adjoint inconsistent discretization, see estimate (371).
However, we recall from Sections 6.3.3 and 7.2 that the following modication
J
2
(u
h
) := J
2
(u
h
)
_
(u
h
g
D
)j
D
ds (383)
of the target functional J
2
(u
h
) leads to an adjoint consistent discretization. Here, is the penaliza-
tion parameter of the IP discretization and g
D
is the boundary value function of the model problem
considered. Note, that
J
2
(u
h
) in (383) is a consistent modication of J
2
(u
h
) as the true value of
the target quantity is unchanged:
J
2
(u) = J
2
(u) = 2 holds for the exact solution u.
Figure 6b) shows the error of the SIPG discretization measured in terms of the modied target
quantity
J
2
() given in (383). We see that the consistent modication of the target functional leads
to a signicant increase in the accuracy and the order of convergence of the discretization. In
73
0
0.2
0.4
0.6
0.8
1 0
0.2
0.4
0.6
0.8
1
-1.4
-1.2
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
z
h
for J
2
, adjoint inconsistent
0
0.2
0.4
0.6
0.8
1 0
0.2
0.4
0.6
0.8
1
-1.01
-1.005
-1
-0.995
-0.99
-0.985
z
h
for
J
2
, adjoint consistent
(a) (b)
Figure 7: Example 2: Discrete adjoint solution z
h
corresponding a) to the adjoint inconsistent
SIPG discretization with J
2
(u
h
) and b) to the adjoint consistent SIPG discretization with
J
2
(u
h
).
fact, we see that under global mesh renement the error |J
2
(u)
J
2
(u
h
)| behaves like O(h
2(p+1)
)
which is even larger than the expected order O(h
2p
), see estimate (370), of an adjoint consistent
discretization.
In the following we want to highlight the connection between adjoint consistency and the
smoothness of the adjoint solution. To this end, Figure 7a) shows the discrete adjoint solution
z
h
connected to the (original) target quantity J
2
(); i.e. z
h
is the solution to the discrete adjoint
problem given at the top of Section 6.3.3 with right hand side J
2
(). In Figure 7a) we see that z
h
is
irregular in the neighborhood of the boundary. We note that this irregularity does not vanish under
mesh renement. Thereby, the discrete adjoint solution does not converge to the exact solution,
z 1, of the continuous adjoint problem (382). This behavior corresponds to the fact that the
SIPG discretization in combination with the target quantity J
2
() is adjoint inconsistent.
In comparison to that, Figure 7b) shows the discrete adjoint solution z
h
connected to the
modied target quantity
J
2
(). Here, we see that z
h
is perfectly smooth. Furthermore, we note
that z
h
converges to the exact adjoint solution z 1. In fact, we have z
h
1 from the second
coarsest mesh onwards. That is, the discrete adjoint solution is a consistent discretization of the
continuous adjoint solution. In other words: the SIPG discretization in combination with the
modied target quantity
J
2
() is adjoint consistent.
Finally, we recall that the experimental order of convergence of the error |J
2
(u)
J
2
(u
h
)| of the
adjoint consistent SIPG discretization behaves like O(h
2(p+1)
) which is two powers of h larger than
the theoretically expected order O(h
2p
), see estimate (370). A possible reason for this might be a
too simple model problem in combination with a particularly simple target quantity which results
in the constant continuous adjoint solution z 1.
Example 3 In order to demonstrate that the estimate (370) is sharp we consider the following
problem: Let = (0, 1) (0.1, 1) and consider Poissons equation (162) with forcing function f
which is chosen so that the analytical solution to (162) is given by
u(x) =
1
4
(1 +x
1
)
2
sin(2x
1
x
2
). (384)
We note that this is a modication of the problem considered in [23]. Again, we impose Dirichlet
boundary conditions where the boundary value function g
D
on
D
= is prescribed based on the
solution u. We consider the target quantity J
3
(u
h
) and its modication
J
3
(u
h
) given as follows
J
3
(u
h
) =
_
j
D
n
h
u
h
ds, (385)
J
3
(u
h
) = J
3
(u
h
)
_
(u
h
g
D
)j
D
ds. (386)
74
1e-12
1e-10
1e-08
1e-06
0.0001
0.01
1
10 100 1000 10000 100000
1
1
1
2
1
3
SIPG,p=1
SIPG,p=2
SIPG,p=3
|
J
3
(
u
)
J
3
(
u
h
)
|
elements
1e-12
1e-10
1e-08
1e-06
0.0001
0.01
1
10 100 1000 10000 100000
1
2
1
4
1
6
SIPG,p=1
SIPG,p=2
SIPG,p=3
|
J
3
(
u
)
J
3
(
u
h
)
|
elements
(a) (b)
Figure 8: Example 3: Convergence of a) the error |J
3
(u) J
3
(u
h
)| and b) the error |J
3
(u)
J
3
(u
h
)|
for the SIPG discretization with global mesh renement.
0
0.2
0.4
0.6
0.8
10.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
-1.2
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
z
h
for J
3
, adjoint inconsistent
0
0.2
0.4
0.6
0.8
10.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
-1.2
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
z
h
for
J
3
, adjoint consistent
(a) (b)
Figure 9: Example 3: Discrete adjoint solution z
h
corresponding a) to the adjoint inconsistent
SIPG discretization with J
3
(u
h
) and b) to the adjoint consistent SIPG discretization with
J
3
(u
h
).
and choose j
D
L
2
() to be given by
j
D
(x) =
_
_
exp
_
4
1
16
((x
1
1
4
)
2
1
8
)
2
_
for x (0,
1
4
) (0.1, 1),
exp
_
4
1
16
((x
1
3
4
)
2
1
8
)
2
_
for x (
3
4
, 1) (0.1, 1),
1 for x (
1
4
,
3
4
) (0.1, 1),
0 elsewhere on .
Thereby, the true value of the target quantity is J
3
(u) = 1.2825165799606.
Figure 8a) shows that the convergence behavior of the error |J
3
(u) J
3
(u
h
)| behaves like O(h
p
)
which is in perfect agreement with the estimate (371) for an adjoint inconsistent discretization.
Furthermore, Figure 8b) shows that the convergence behavior of the error |J
3
(u)
J
3
(u
h
)| behaves
like O(h
2p
) which is as expected, see the estimate (370), for an adjoint consistent discretization.
Finally, Figure 9a) shows the discrete adjoint solution z
h
connected to the (original) target
quantity J
3
(). We see that in the neighborhood of the bottom boundary [0, 1] {0.1} the
discrete adjoint solution is irregular which corresponds to the fact that the SIPG discretization
in combination with the target quantity J
3
() is adjoint inconsistent. In contrast to that the
corresponding Figure 9b) shows that the discrete adjoint solution z
h
connected to the modied
target quantity
J
3
() is entirely smooth which corresponds to the fact that the SIPG discretization
in combination with the modied target quantity
J
3
() is adjoint consistent.
75
8 Discontinuous Galerkin discretizations of
the compressible Euler equations
The compressible Euler equations are a nonlinear system of conservation equations (conservation
of mass, momentum and energy) describing inviscid compressible ows frequently used as a simple
model for gas ows. In order to introduce some new notation we rst consider a system of linear
hyperbolic equations.
8.1 Hyperbolic conservation equations
Given a nal time T > 0, we consider the following system of conservation equations,
t
u +
d
i=1
x
i
f
c
i
(u) = 0 in (0, T] ,
u(0, ) = u
0
() in ,
(387)
where is a bounded connected domain in R
d
, d 1, u = (u
1
, . . . , u
m
)
, F
c
(u) = (f
c
1
(u), . . . , f
c
d
(u))
and f
c
i
: R
m
R
m
, i = 1, . . . , d, are continuously dierentiable. In particular, we will be concerned
with the solution of the stationary system of conservation equations,
F
c
(u) = 0 in , (388)
subject to appropriate boundary conditions described below. We say that (387) is hyperbolic, if
the matrix
B(u, ) :=
d
i=1
i
A
i
(u) (389)
has m real eigenvalues and a complete set of linearly independent eigenvectors for all vectors
= (
1
, . . . ,
d
) R
d
. Here, A
i
(u) denotes the Jacobi matrix of the ux f
c
i
(u), i.e.
A
i
(u) :=
u
f
c
i
(u), i = 1, . . . , d. (390)
The system of conservation equations (387) must be supplemented by appropriate boundary
conditions; for example at inow/outow boundaries, we require that
B
(u, n) (u g) = 0, on (391)
where n denotes the unit outward normal vector to the boundary = and g is a (given) vector
function. Here, B
(u, n) = P
P
1
, (392)
where P = [r
1
, . . . , r
m
] denotes the mmmatrix of eigenvectors of B(u, n) and
= diag(min(
i
, 0))
and
+
= diag(max(
i
, 0)) the m m diagonal matrix of the negative/positive eigenvalues of
B(u, n) with Br
i
=
i
r
i
, i = 1, . . . , d.
Example 8.1 The simplest hyperbolic problem is given by the linear advection equation
(bu) = f in , u = g on
, (393)
i.e. the model problem previously considered in (99) with vanishing reaction term c = 0. In fact,
setting F
c
(u) = (b
1
u, . . . , b
d
u)
(u, n)(u g) =
(u g) = 0 on ,
where
are
dened by
u =
_
v
1
v
2
E
_
_
, f
c
1
(u) =
_
_
v
1
v
2
1
+p
v
1
v
2
Hv
1
_
_
and f
c
2
(u) =
_
_
v
2
v
1
v
2
v
2
2
+p
Hv
2
_
_
, (395)
where , v = (v
1
, v
2
)
, p and E denote the density, velocity vector, pressure and specic total
energy, respectively. Additionally, H is the total enthalpy given by
H = E +
p
= e +
1
2
v
2
+
p
, (396)
where e is the specic static internal energy, and the pressure is determined by the equation of
state of an ideal gas
p = ( 1)e, (397)
where = c
p
/c
v
is the ratio of specic heat capacities at constant pressure, c
p
, and constant volume,
c
v
; for dry air, = 1.4. The ux Jacobians A
i
(u) dened in (390) are given by
A
1
(u) =
_
_
_
_
0 1 0 0
v
2
1
+
1
2
( 1)v
2
(3 )v
1
( 1)v
2
1
v
1
v
2
v
2
v
1
0
v
1
_
1
2
( 1)v
2
H
_
H ( 1)v
2
1
( 1)v
1
v
2
v
1
_
_
_
_
,
A
2
(u) =
_
_
_
_
0 0 1 0
v
1
v
2
v
2
v
1
0
v
2
2
+
1
2
( 1)v
2
( 1)v
1
(3 )v
2
1
v
2
_
1
2
( 1)v
2
H
_
( 1)v
1
v
2
H ( 1)v
2
2
v
2
_
_
_
_
.
Finally, the eigenvalues of the matrix B(u, n) =
2
i=1
n
i
A
i
(u) are
1
= v n c,
1
=
2
= v n,
1
= v n +c (398)
where c =
_
p/ denotes the speed of sound. Considering the signs of
i
, i = 1, . . . , 4, we
distinguish four cases of boundary conditions (391):
supersonic inow:
i
< 0, i = 1, . . . , 4,
subsonic inow:
i
< 0, i = 1, 2, 3,
4
> 0,
subsonic outow:
1
< 0,
i
> 0, i = 2, 3, 4, and
supersonic outow:
i
> 0, i = 1, . . . , 4.
Each eigenvalue smaller than zero corresponds to an inow characteristic. The number of variables
to be prescribed on the boundary depend on the number of inow characteristics.
77
8.3 The DG discretization of the compressible Euler equations
We begin by introducing the vector-valued counterpart of the discrete function space V
d
h,p
dened
in (123). Let V
d
h,p
be the nite element space consisting of discontinuous vectorvalued polynomial
functions of degree p 0, dened by
V
d
h,p
= {v
h
_
L
2
()
m
: v
h
|
[Q
p
( )]
m
if is the unit hypercube, and
v
h
|
[P
p
( )]
m
if is the unit simplex, T
h
},
(399)
For deriving discontinuous Galerkin discretizations of the compressible Euler equations we pro-
ceed similarly as we did for the case of the linear advection equation in Section 4.7. In order to
introduce a weak formulation of (388) we multiply it by an arbitrary smooth (vector-)function v
and integrate by parts over an element in the mesh T
h
; thereby, we obtain
F
c
(u) v dx +
_
F
c
(u) nv ds = 0. (400)
To discretize (400), we replace the analytical solution u by the Galerkin nite element approxi-
mation u
h
and the test function v by v
h
, where u
h
and v
h
both belong to the nite element space
V
d
h,p
. In addition, since the numerical solution u
h
is discontinuous between element interfaces, we
must replace the ux F
c
(u) n by a numerical ux function H(u
+
h
, u
h
, n), which depends on both
the interior and outertrace of u
h
on , T
h
, and the unit outward normal n to . Thereby,
summing over the elements in the mesh T
h
, yields the discontinuous Galerkin discretization of
(388) as follows: nd u
h
V
d
h,p
such that
F
c
(u
h
)
h
v
h
dx +
T
h
_
H(u
+
h
, u
h
, n) v
+
h
ds = 0 v
h
V
d
h,p
. (401)
We remark that the replacement of the ux F
c
(u) n by the numerical ux function H(u
+
h
, u
h
, n)
on the boundary of element , in T
h
, corresponds to the weak imposition of the boundary data.
Like in the case of the linear advection equation in Section 4 the numerical ux H(, , ) must be
consistent and conservative. We recall that
(i) H(, , )|
= F
c
(v) n T
h
;
(ii) H(, , ) is conservative, i.e. given any two neighboring elements and
= , noting that n
= n, we have that
H(v, w, n) = H(w, v, n).
There are several numerical ux functions satisfying these conditions, such as the Godunov,
EngquistOsher, LaxFriedrichs, Roe or the Vijayasundaram ux. As examples, here we consider
three dierent numerical uxes:
The (local) LaxFriedrichs ux H
LF
(, , ) is dened by
H
LF
(u
+
, u
, n)|
=
1
2
_
F
c
(u
+
) n +F
c
(u
) n +
_
u
+
u
__
,
for T
h
, where is the maximum over u
+
and u
,
= max
v=u
+
,u
{|(B(v, n))|},
of the largest eigenvalue (in absolute value) |(B)| of the matrix B(v, n) =
d
i=0
n
i
A
i
(u)
dened in (389).
78
The Vijayasundaram ux H
V
(, , ) is dened by
H
V
(u
+
, u
, n)|
= B
+
( u, n)u
+
+B
( u, n)u
for T
h
,
where B
+
( u, n) and B
( u, n) denote the positive and negative parts, cf. (392), of the matrix
B( u, n), respectively, evaluated at an average state u between u
+
and u
.
The HLLE ux H
HLLE
(, , ) is given by
H
HLLE
(u
+
, u
, n)|
=
1
+
F
c
(u
+
) n
F
c
(u
) n
+
_
u
+
u
__
,
where
+
= max(
max
, 0) and
= min(
min
, 0).
Remark 8.2 We note that when applied to the linear advection equation (393), most numerical
uxes, in particular the numerical uxes introduced above, reduce to the upwind ux given in (115):
H
uw
(u
+
, u
, n) =
_
b nu
,
b nu
+
, for (b n)(x) 0, i.e. x
+
.
8.4 Boundary conditions
For boundary faces = we replace u
h
by an appropriate boundary function u
(u
+
h
) which
realizes the boundary conditions to be imposed.
First we dene several fareld boundary conditions:
Supersonic inow corresponds to Dirichlet boundary conditions where
u
(u) = g
D
= u
on
D,sup
.
Supersonic outow corresponds to Neumann boundary conditions where
u
(u) = u on
N
.
The subsonic inow boundary condition takes the pressure from the ow eld and imposes
all other variables based on freestream conditions u
, i.e.
u
(u) =
_
v
1,
,
v
2,
,
p(u)
1
+
_
v
2
1,
+v
2
2,
_
_
on
D,sub-in
.
Here, p p(u) denotes the pressure evaluated using the equation of state (397).
The subsonic outow boundary condition imposes an outow pressure p
out
and takes all other
variables from the ow eld, i.e.
u
(u) =
_
u
1
, u
2
, u
3
,
p
out
1
+
u
2
2
+u
2
3
2u
1
_
on
D,sub-out
.
The characteristic fareld boundary condition imposes Dirichlet boundary conditions based
on free-stream conditions on characteristic inow variables. No boundary conditions are
imposed on characteristic outow variables. This corresponds to using the Vijayasundaram
ux on the fareld boundary.
Finally, we dene following wall boundary condition:
79
For slip wall boundary conditions used at reective walls we set
u
(u) =
_
_
_
_
1 0 0 0
0 1 2n
2
1
2n
1
n
2
0
0 2n
1
n
2
1 2n
2
2
0
0 0 0 1
_
_
_
_
u on
re
, (402)
which originates from u by inverting the sign of the normal velocity component of u, i.e.
v = (v
1
, v
2
) is replaced by v
) n = 0.
Given the boundary value function u
(u
+
h
) as dened above the DG discretization of (388) including
boundary conditions is given as follows: nd u
h
V
d
h,p
such that
N
h
(u
h
, v
h
)
_
F
c
(u
h
)
h
v
h
dx +
T
h
_
\
H(u
+
h
, u
h
, n) v
+
h
ds
+
_
(u
+
h
, u
(u
+
h
), n) v
+
h
ds = 0 v
h
V
d
h,p
, (403)
where H
(u) dx +
_
[u](w) =
_
[u] wdx +
_
[Cu] C
[u]wds, (406)
where j
() and j
and j
. Here,
denotes the
(total) Frechet derivative and the square bracket [] denotes the state about which linearization is
performed. Again, we say that the target functional (405) is compatible with (404) provided the
following compatibility condition holds
(N
[u]w, z)
+ (B
[u]w, (C
[u])
z)
= (w, (N
[u])
z)
+ (C
[u]w, (B
[u])
z)
, (407)
where (N
[u])
, (B
[u])
and (C
[u])
[u], B
[u] and C
[u]. This
condition is analogous to (275), with L, B and C replaced by N
[u], B
[u] and C
[u], respectively.
Assuming that (407) holds the continuous adjoint problem associated to (404) and (406) is:
(N
[u])
z = j
[u] in , (B
[u])
z = j
[Cu] on . (408)
80
We note that in an optimization framework [20] this ensures, analogous to (277), that
J
[u](w) = (w, j
[u])
+ (C
[u]w, j
[Cu])
= (w, (N
[u])
z)
+ (C
[u]w, (B
[u])
z)
= (N
[u]w, z)
+ (B
[u]w, (C
[u])
z)
. (409)
Let N
h
: V V R be a semi-linear form, nonlinear in its rst and linear in its second argument,
such that the nonlinear problem (404) is discretized as follows: nd u
h
V
h
such that
N
h
(u
h
, v
h
) = 0 v
h
V
h
. (410)
Then, the discretization (410) is said to be consistent if the exact solution u V to the primal
problem (404) satises the following equation:
N
h
(u, v) = 0 v V. (411)
Furthermore, the discretization (410) is said to be adjoint consistent if the exact solutions u, z V
to the primal and adjoint problems (404) and (408), respectively, satisfy the following equation:
N
h
[u](w, z) = J
[u](w) w V, (412)
where N
h
[u] denotes the Frechet derivatives of N
h
(u, v) with respect to u.
In other words, a discretization is adjoint consistent if the discrete adjoint problem is a consistent
discretization of the continuous adjoint problem. Finally, we note that in case of a linear problem
and target functional the denition of adjoint consistency in (412) reduces to the denition of linear
adjoint consistency given in Section 6. The denition of adjoint consistency for nonlinear problems
as given in (412) was introduced by Lu [36]. Furthermore, we note that [36] also gives a denition
of asymptotically adjoint consistent methods.
8.5.1 The consistency and adjoint consistency analysis
Based on the denition of consistency and adjoint consistency in the previous subsection we now
follow [27] and generalize the framework for analyzing consistency and adjoint consistency of dis-
continuous Galerkin discretizations for linear problems as given in Section 6 to the case of nonlinear
problems. We recall that this framework can also be used to nd specic terms due to which some
DG discretizations may not be adjoint consistent. In these cases the analysis gives some insight
into how an adjoint inconsistent DG discretization together with a specic target functional could
be modied to recover an adjoint consistent discretization.
Given a primal problem, a discontinuous Galerkin discretization of the problem and a target
functional, the adjoint consistency analysis consists of the following steps:
Derivation of the continuous adjoint problem: Let the primal problem be given by
(404). Furthermore, assume that J() is a nonlinear functional (405) which is compatible with
the primal problem (404). Then we derive the continuous adjoint problem (408) including
adjoint boundary conditions.
We note that the derivation of the adjoint operator (N
[u])
[u])
[u])
and (C
[u])
assumed to be connect
to B, C, N and (N
[u])
[u])
and C
[u])
[u])
.
81
Consistency analysis of the discrete primal problem: We rewrite the discontinuous
Galerkin discretization (410) of problem (404) in the following element-based primal residual
form: nd u
h
V
h
such that
T
h
_
R(u
h
)v
h
dx +
T
h
_
\
r(u
h
)v
h
ds +
_
(u
h
)v
h
ds = 0 v
h
V
h
, (413)
where R(u
h
), r(u
h
) and r
(u
h
) denote the element, interior face and boundary residuals,
respectively. According to (411), the discretization (410) is consistent if the exact solution u
to (404) satises
T
h
_
R(u)v dx +
T
h
_
\
r(u)v ds +
_
(u)v ds = 0 v V, (414)
which holds provided u satises
R(u) = 0 in , T
h
, r(u) = 0 on \ , T
h
, r
(u) = 0 on . (415)
Derivation of the discrete adjoint problem Given the discretization (410), the target
functional (405) and its linearization (406), we derive the discrete adjoint problem: nd
z
h
V
h
such that
N
[u
h
](w
h
, z
h
) = J
[u
h
](w
h
) w
h
V
h
. (416)
N
[u
h
] is called the Jacobian of the numerical scheme and is required also for implicit and ad-
joint methods, e.g. Newton iteration, a posteriori error estimation, adjoint-based adaptation,
see [25], and for optimization.
Adjoint consistency of element, interior face and boundary terms We rewrite the
discrete adjoint problem (416) in element-based adjoint residual form: nd z
h
V
h
such that
T
h
_
w
h
R
[u
h
](z
h
) dx +
T
h
_
\
w
h
r
[u
h
](z
h
) ds +
_
w
h
r
[u
h
](z
h
) ds = 0, (417)
for all w
h
V
h
, where R
[u
h
](z
h
), r
[u
h
](z
h
) and r
[u
h
](z
h
) denote the element, interior face
and boundary adjoint residuals, respectively. According to (412), the discretization (410) is
adjoint consistent if the exact solutions u and z satisfy
T
h
_
wR
[u](z) dx +
T
h
_
\
wr
[u](z) ds +
_
wr
[u](z) ds = 0 w V, (418)
which holds provided u and z satisfy
R
[u](z) = 0 in , r
[u](z) = 0 on \ , T
h
, r
[u](z) = 0 on . (419)
We note that the adjoint problem and consequently the adjoint consistency of a discretization
depends on the specic target functional J() under consideration. Given a target functional of the
form (405), we see that R
[u](z) depends on j
(), and r
[u](z) depends on j
J(u
h
) = J(i(u
h
)) +
_
r
J
(u
h
) ds, (420)
where i() and r
J
() are functions to be specied. We recall from Section 6.2 that a modication
of a target functional is called consistent if
J(u) = J(u) holds for the exact solution u.
82
8.6 Adjoint consistency analysis of DG for the compressible Euler equations
8.6.1 The continuous adjoint problem to the compr. Euler equations
The most important target quantities in inviscid compressible ows are the pressure induced drag
and lift coecients, c
dp
and c
lp
, dened by
J(u) =
_
j(u) ds =
1
C
W
p n ds, (421)
where j(u) =
1
C
p n on
W
and j(u) 0 elsewhere. Here, C
=
1
2
p
M
2
l =
1
2
|v|
2
c
2
l =
1
2
|v
|
2
or
l
= (sin(), cos())
for the
drag and lift coecient, respectively. Subscripts
indicate free-stream quantities.
In order to derive the continuous adjoint problem, we multiply the left hand side of (394) by z,
integrate by parts and linearize about u to obtain
( (F
c
u
[u](w)) , z)
= (F
c
u
[u](w), z)
+ (n F
c
u
[u](w), z)
, (422)
where F
c
u
[u] := (F
c
)
_
w, (F
c
u
[u])
z
_
+
_
w, (n F
c
u
[u])
z
_
= J
[u](w) w V, (423)
and the continuous adjoint problem is given by
(F
c
u
[u])
z = 0 in , (n F
c
u
[u])
z = j
[u] on . (424)
Using F
c
(u) n = p(0, n
1
, n
2
, 0)
on
W
, and the denition of j in (421) we obtain
p
[u](0, n
1
, n
2
, 0) z =
1
C
[u]n on
W
,
which reduces to the boundary condition of the adjoint compressible Euler equations,
(B
[u])
z = n
1
z
2
+n
2
z
3
=
1
C
n on
W
. (425)
8.6.2 Primal residual form of DG for the compr. Euler equations
Using integration by parts on (403) we obtain the residual form: nd u
h
V
d
h,p
such that
_
R(u
h
) v
h
dx +
T
h
_
\
r(u
h
) v
+
h
ds +
_
(u
h
) v
+
h
ds = 0 v
h
V
d
h,p
, (426)
where the primal residuals are given by
R(u
h
) = F
c
(u
h
) in , T
h
,
r(u
h
) = n F
c
(u
+
h
) H(u
+
h
, u
h
, n
+
) on \ , T
h
, (427)
r
(u
h
) = n F
c
(u
+
h
) H
(u
+
h
, u
(u
+
h
), n
+
) on .
Given the consistency of the numerical ux, H(w, w, n) = n F
c
(w), and the consistency of the
boundary function, i.e. u
(u) = 0 on . (428)
We conclude that (403) is a consistent discretization of (394).
83
8.6.3 Adjoint residual form of DG for the compr. Euler equations
For the target functional J() dened in (421) with Frechet derivative, J
h
[u
h
](w
h
, z
h
) = J
[u
h
](w
h
) w
h
V
d
h,p
, (429)
where
N
h
[u
h
](w, z
h
)
_
(F
c
u
[u
h
]w) :
h
z
h
dx
+
T
h
_
\
_
H
u
+(u
+
h
, u
h
, n
+
)w
+
+H
u
(u
+
h
, u
h
, n
+
)w
_
z
+
h
ds
+
_
_
H
,u
+
_
u
+
h
, u
(u
+
h
), n
+
_
+H
,u
_
u
+
h
, u
(u
+
h
), n
+
_
u
[u
+
h
]
_
w
+
z
+
h
ds. (430)
Here v H
u
+
(v
+
, v
, n) and v H
(v
+
, v
(v, w, n) =
w
H(v, w, n) =
w
H(w, v, n) =
H
u
+
(w, v, n), and
_
I
H
(u
+
h
, u
h
, n
+
)w
z
+
ds =
_
I
H
u
+
(u
h
, u
+
h
, n
)w
z
+
ds
=
_
I
H
u
+
(u
+
h
, u
h
, n
+
)w
+
z
ds,
(431)
where we exchanged notations
+
and
on
I
. Then, the discrete adjoint problem (429) with (430)
is given in adjoint residual form as follows: nd z
h
V
d
h,p
such that
_
w
h
R
[u
h
](z
h
) dx +
T
h
_
\
w
+
h
r
[u
h
](z
h
) ds +
_
w
+
h
r
[u
h
](z
h
) ds = 0, (432)
for all w
h
V
d
h,p
, where the adjoint residuals are given by
R
[u
h
](z
h
) =(F
c
u
[u
h
])
z
h
in , T
h
,
r
[u
h
](z
h
) =
_
H
u
+
(u
+
h
, u
h
, n
+
)
_
[[z
h
]] n on \ , T
h
,
r
[u
h
](z
h
) =j
[u
h
]
_
H
,u
+
+H
,u
[u
h
]
_
z
+
h
on , (433)
where H
,u
+
:= H
,u
+
(u
+
h
, u
(u
+
h
), n
+
) and H
,u
:= H
,u
(u
+
h
, u
(u
+
h
), n
+
).
Comparing the discrete adjoint boundary condition
_
H
,u
+ +H
,u
u
[u
h
]
_
z
+
h
= j
[u
h
] on , (434)
and the continuous adjoint boundary condition in (424), we notice that not all choices of H
give
rise to an adjoint consistent discretization. In fact, we require H
must depend on
u
(u
+
h
), hence H
,u
= 0. Furthermore, we require H
,u
+
= 0, as otherwise the left hand side in
(433) involves two summands which is in contrast to the continuous adjoint boundary condition
in (424). Finally, we recall that H
is consistent, H
(v, v, n) = n F
c
(v), and conclude that
84
H
is given by H
(u
+
h
, u
(u
+
h
), n) = n F
c
(u
(u
+
h
)). Employing a modied target functional
J(u
h
) = J(i(u
h
)), i.e. (288) with r
j
(u
h
) 0, (434) yields
_
n
_
F
c
u
[u
(u
+
h
)]
_
u
[u
+
h
]
_
z = j
[i(u
+
h
)]i
[u
+
h
]. (435)
We nd the modication i(u
h
) = u
(u
h
) which is consistent as i(u) = u
(u
+
h
)]
_
z = j
[u
(u
+
h
)], (436)
which represents a discretization of the continuous adjoint boundary condition in (424). In order to
obtain a discretization of the adjoint boundary condition at solid wall boundaries (425), we require
Bu
(u
+
h
) = 0 on
W
. This condition is satised by
u
(u) =
_
_
_
_
1 0 0 0
0 1 n
2
1
n
1
n
2
0
0 n
1
n
2
1 n
2
2
0
0 0 0 1
_
_
_
_
u on
W
, (437)
which originates from u by subtracting the normal velocity component of u, i.e. v = (v
1
, v
2
) is
replaced by v
n = 0.
In summary, let u
and
J be dened by
H
(u
+
h
, u
(u
+
h
), n) = n F
c
(u
+
h
),
J(u
h
) = J
(u
h
), (438)
where F
c
(u
+
h
) := F
c
(u
(u
+
h
)), J
(u
h
) := J(u
(u
h
)) and j
(u
h
) := j(u
(u
h
)), then the adjoint
residuals (433) are given by:
R
[u
h
](z
h
) =(F
c
u
[u
h
])
z
h
in , T
h
,
r
[u
h
](z
h
) =
_
H
u
+
(u
+
h
, u
h
, n
+
)
_
[[z
h
]] n on \ , T
h
,
r
[u
h
](z
h
) =j
[u
+
h
]
_
n F
c
,u
[u
+
h
]
_
z
+
h
on . (439)
In particular, the discretization (403) together with (438) is adjoint consistent as the exact solutions
u and z to (394) and (424), respectively, satisfy
R
[u](z) = 0 in , T
h
, r
[u](z) = 0 on \ , T
h
, r
[u](z) = 0 on .
Note, that the adjoint residuals in (439) reduce to the adjoint residuals of the linear advection
equation with b = 0 in Section 6.4.3, when setting F
c
(u) = bu and H
u
+
= b n.
Also note, that the standard discontinuous Galerkin discretizations for the compressible Eu-
ler equations take the same numerical ux function on the boundary as in the interior of the
domain, and simply replace u
h
in H(u
+
h
, u
h
, n) by the boundary function u
(u
+
h
) resulting in
H
(u
+
h
, u
(u
+
h
), n). Furthermore, the denition of u
in (402) based on v
)n = 0. However, v
n = 0 and Bu
(u
+
h
) = 0,
as required in (436), is not satised. Thereby, the discontinuous Galerkin discretization based on
the standard choice of H
and u
u
h
L
2
degrees of freedom
Figure 11: Ringleb ow problem: The L
2
-error of the DG(p), p = 0, . . . , 4, discretizations of the
compressible Euler equations is of order O(h
p+1
), [24].
8.7 Numerical results
Ringleb ow problem For discretizations of the 2d stationary compressible Euler equations
there are virtually no a priori error estimates available. Therefore, in the following we examine
the order of convergence of the DG discretization experimentally. In particular, we consider the
solution to the 2d compressible Euler equations for the Ringleb ow problem. This is one of the few
non-trivial problems of the 2d Euler equations for which a smooth analytical solution is known. For
this problem the analytical solution may be obtained be employing the hodograph transformation,
see [15] or the appendix of [24]. This problem represents a transonic ow in a channel, see Figure
10a), with inow and outow boundaries given by the lower and upper boundaries of the domain,
and reective (slip wall) boundaries with vanishing normal velocity, v n = 0, on the left and right
boundary. The solution to this ow problem is smooth but it is transonic with a small supersonic
region near the lower right corner. The computational domain is subdivided into quadrilateral
elements. Figure 10 shows the coarsest three meshes in a sequence of globally rened meshes. In
order to suppress the discretization eects of slip wall boundaries here we impose the boundary
condition, B
(u
+
h
, u
(u
+
h
), n) =
H(u
+
h
, u
(u
+
h
), n) and an unmodied target functional J(u
h
). To this end, we consider an inviscid
Mach M = 0.5 ow at a zero angle of attack, i.e. = 0
, respectively,
where
g
_
0
11
21
1j
v
j
+KT
x
1
_
_
and f
v
2
(u, u) =
_
_
0
12
22
2j
v
j
+KT
x
2
_
_
,
respectively, where K is the thermal conductivity coecient. Finally, the viscous stress tensor is
dened by
=
_
v + (v)
2
3
( v)I
_
,
where is the dynamic viscosity coecient, and the temperature T is given by e = c
v
T; thus
KT =
Pr
_
E
1
2
v
2
_
,
where Pr = 0.72 is the Prandtl number.
For the purposes of discretization, we rewrite the compressible NavierStokes equations (440)
in the following (equivalent) form:
x
k
_
f
c
k
(u) G
kl
(u)
u
x
l
_
= 0 in .
Here, the matrices G
kl
(u) = f
v
k
(u, u)/u
x
l
, for k, l = 1, 2, are the homogeneity tensors dened
by f
v
k
(u, u) = G
kl
(u)u/x
l
, k = 1, 2, where
G
11
=
_
_
_
_
0 0 0 0
4
3
v
1
4
3
0 0
v
2
0 1 0
_
4
3
v
2
1
+v
2
2
+
Pr
_
E v
2
__ _
4
3
Pr
_
v
1
_
1
Pr
_
v
2
Pr
_
_
_
_
,
G
12
=
_
_
_
_
0 0 0 0
2
3
v
2
0
2
3
0
v
1
1 0 0
1
3
v
1
v
2
v
2
2
3
v
1
0
_
_
_
_
, G
21
=
_
_
_
_
0 0 0 0
v
2
0 1 0
2
3
v
1
2
3
0 0
1
3
v
1
v
2
2
3
v
2
v
1
0
_
_
_
_
,
G
22
=
_
_
_
_
0 0 0 0
v
1
1 0 0
4
3
v
2
0
4
3
0
_
v
2
1
+
4
3
v
2
2
+
Pr
_
E v
2
__ _
1
Pr
_
v
1
_
4
3
Pr
_
v
2
Pr
_
_
_
_
.
88
Like for the compressible Euler equations we consider supersonic and subsonic inow and outow
boundary conditions. Furthermore, we destinguish between isothermal and adiabatic wall boundary
conditions. To this end, decomposing
W
=
iso
adia
, we set
v = 0 on
W
, T = T
wall
on
iso
, n T = 0 on
adia
, (441)
where T
wall
is a given wall temperature.
9.2 DG discretizations of the compressible Navier-Stokes equations
The derivation of discontinuous Galerkin discretizations of the compressible Navier-Stokes equations
is similar to the derivation for Poissons equation. Starting point is the compressible Navier-Stokes
equations written in terms of the homogeneity tensors G(u). In the previous section we have
concentrated on the discretization of the convective ux F
c
(u) =
x
k
f
c
k
(u) representing the
inviscid Euler part of the equations. Therefore, in the following, we can ignore the convective part
and concentrate on the remaining viscous part, i.e. we consider the discretization of
(G(u)u) =
x
k
_
G
kl
(u)
u
x
l
_
= 0 in , (442)
subject to boundary conditions given above. In index notation this writes,
x
k
_
(G(u)
kl
)
ij
x
l
u
j
_
= 0 in .
Like for Poissons equation we rewrite this problem as a rst-order system:
= G(u)u, = 0 in ,
i.e.
ik
= (G(u)
kl
)
ij
x
l
u
j
. Multiplying the rst and second equations by test functions and v,
respectively, integrating on an element T
h
, and integrating by parts, we obtain
_
: dx =
_
u
_
G
(u)
_
dx +
_
u
_
G
(u)
_
nds,
_
: vdx =
_
nv ds,
(443)
where n is the unit outward normal vector to . Here, we used
_
ik
ik
dx =
_
(G(u)
kl
)
ij
x
l
u
j
ik
dx =
_
x
l
u
j
(G(u)
kl
)
ij
ik
dx =
_
u :
_
G
(u)
_
dx.
In addition to the vector-valued discrete function space V
d
h,p
dened in (399) we now introduce
the tensor-valued discrete function space
d
h,p
consisting tensor-valued polynomial functions of
degree p 0, dened by
d
h,p
= {
_
L
2
()
42
:|
[Q
p
( )]
42
if is the unit hypercube, and
|
[P
p
( )]
42
if is the unit simplex, T
h
}.
If we sum (443) over all elements T
h
and replace u, v, and by discrete functions
u
h
, v
h
V
d
h,p
and
h
,
h
d
h,p
we obtain following discretization in the so-called ux formulation:
nd u
h
V
d
h,p
and
h
d
h,p
such that
_
h
:
h
dx =
_
u
h
_
G
(u
h
)
h
_
dx +
T
h
_
u
h
_
G
(u
h
)
h
_
nds, (444)
_
h
:
h
v
h
dx =
T
h
_
h
nv
h
ds, (445)
89
for all
h
d
h,p
and v
h
V
d
h,p
. Here, the numerical uxes u
h
and
h
are approximations to u
and = u, respectively. Depending on the particular choice of u
h
and
h
several dierent DG
methods can be derived, each with specic properties with respect to stability and accuracy.
The ux formulation (444) represents the discretization of a rst order system with unknowns
u
h
V
d
h,p
and
h
d
h,p
. However, this is (d +1) times the size of a problem involving u
h
V
d
h,p
,
only. In order to reduce the problem size, the auxiliary variable
h
in (444) and (445) is usually
eliminated to gain a primal formulation involving only the primal variable u
h
. To this end, we
perform a second integration by parts on each element in (444) and set
h
=
h
v
h
which gives
us
_
h
:
h
v
h
dx =
_
G(u
h
)
h
u
h
:
h
v
h
dx +
T
h
_
( u
h
u
h
)
_
G
(u
h
)v
h
_
nds. (446)
Substituting (446) into (445) we obtain following problem: nd u
h
V
d
h,p
such that
N
v
h
(u
h
, v
h
) = 0 v
h
V
d
h,p
, (447)
where the semilinear form
N
v
h
(, ) : [H
1
(T
h
)]
m
[H
1
(T
h
)]
m
R which is nonlinear in its rst and
linear in its second argument, is dened by
N
v
h
(u
h
, v
h
) =
_
G(u
h
)
h
u
h
:
h
v
h
dx
T
h
_
h
nv
h
ds+
T
h
_
( u
h
u
h
)
_
G
(u
h
)v
h
_
nds. (448)
Here, we use the notation
N
v
h
instead of
N
h
to underline that
N
v
h
includes the discretization of the
viscous part of the compressible Navier-Stokes equations only. Later in Section 9.2 we consider
the discretization of the complete compressible Navier-Stokes equations, including convective and
viscous parts, which will then be denoted by
N
h
.
We call (447) the primal formulation of the method and call
N
h
(, ) the primal form. This
bilinear form is denoted by
N
h
(and not N
h
) as it includes the (still unspecied) numerical uxes
u
h
and
h
. Furthermore,
B includes through the specication of u
h
and
h
on the boundary
all boundary data terms. We note, that (448) is a generalization of (172) for Poissons equation to
the case of an elliptic system of equations with diusion tensor G(u).
Finally, we note that
N
h
in (448) is the cell-based primal form, i.e. it is given in terms of
.
This means that each interior face e =
I
occurs twice in the sum over all elements (once in
_
and once in
_
for e =
= 0).
In the following, we transfer the cell-based primal form into a face-based primal form, i.e. we
rewrite
N
h
in terms of
_
I
where each interior face is treated only once. However, before doing so,
we introduce some more notation.
Mean value and jump operators First we recall the denition of mean values and jumps
operating on scalar and vector-valued functions given in Section 5.1.
Denition 9.1 Let e
I
be an interior edge between two adjacent elements
+
and
with unit
outward normal vectors, n
+
, n
R
d
, respectively. Let q T(T
h
) and [T(T
h
)]
d
be the traces
of a scalar and a vector valued function, respectively. Then, we dene the mean value and jump
operators, {{}} and [[]], as follows
{{q}} =
1
2
(q
+
+q
), [[q]] = q
+
n
+
+q
,
{{}} =
1
2
(
+
+
), [[]] =
+
n
+
+
.
90
Denition 9.2 On boundary edges e the mean value and jump operators are dened by
{{q}} = q
+
, [[q]] = q
+
n
+
,
{{}} =
+
, [[]] =
+
n
+
.
Furthermore, we recall Lemma 5.4 which has been used to transfer between cell-based and face-
based integrals:
Lemma 9.3 Again, let q T(T
h
) and [T(T
h
)]
d
, then
_
\
+
n
+
q
+
ds =
_
I
{{}} [[q]] ds +
_
I
[[]] {{q}} ds, (449)
+
n
+
q
+
ds =
_
{{}} [[q]] ds +
_
I
[[]] {{q}} ds. (450)
Proof: See Lemma 5.4.
Next we give the denition of tensor mean value and jump operators.
Denition 9.4 Let e
I
be an interior edge between two adjacent elements
+
and
with unit
outward normal vectors, n
+
, n
R
d
, respectively. Let v [T(T
h
)]
m
and [T(T
h
)]
md
be the
traces of a vector-valued and tensor-valued function, respectively. Then, we dene the mean value
and jump operators, {{}} and [[]], as follows
{{}} =
1
2
(
+
+
) on
I
, {{}} =
+
on ,
[[]] =
+
n
+
+
on
I
, [[]] =
+
n
+
on ,
[[v]] = v
+
n
+
+v
on
I
, [[v]] = v
+
n
+
on .
Now we can generalize Lemma 9.3 to systems of equations:
Lemma 9.5 Let v [T(T
h
)]
m
and [T(T
h
)]
md
, then
_
\
_
+
n
+
_
v
+
ds =
_
I
{{}} : [[v]] ds +
_
I
[[]] {{v}} ds, (451)
+
n
+
_
v
+
ds =
_
{{}} : [[v]] ds +
_
I
[[]] {{v}} ds. (452)
Proof: Employing Lemma 9.3 for any i = 1, . . . , m we obtain
_
\
+
ik
n
+
k
v
+
i
ds =
_
I
{{
ik
}}
_
v
+
i
n
+
k
+v
i
n
k
_
ds +
_
I
_
+
ik
n
+
k
+
ik
n
k
_
{{v
i
}} ds
=
_
I
{{}} : [[v]] ds +
_
I
[[]] {{v}} ds,
thus (451). Use Denition 9.4 of mean value and jump operators on for (452).
Face-based form of DG discretizations We now proceed in transferring the cell-based form
(448) into a face-based form. To this end, we use Equation (452) twice (once for =
h
and
v = v
h
, and once for = G
(u
h
)v
h
, i.e.
jl
= (G(u)
kl
)
ij
x
k
v
i
, and v = u
h
u
h
) and rewrite
(448) as follows
N
v
h
(u
h
, v
h
) =
_
G(u
h
)
h
u
h
:
h
v
h
dx
_
{{
h
}} : [[v
h
]] ds
_
I
[[
h
]] {{v
h
}} ds
+
_
[[ u
h
u
h
]] : {{G
(u
h
)v
h
}} ds +
_
I
{{ u
h
u
h
}} [[G
(u
h
)v
h
]] ds,
91
which results in following face-based primal form:
N
v
h
(u
h
, v
h
) =
_
G(u
h
)
h
u
h
:
h
v
h
dx +
_
[[ u
h
u
h
]] : {{G
(u
h
)v
h
}} {{
h
}} : [[v
h
]] ds
+
_
I
{{ u
h
u
h
}} [[G
(u
h
)v
h
]] [[
h
]] {{v
h
}} ds. (453)
Derivation of various DG discretization methods Depending on the specic choice of nu-
merical uxes u
h
and
h
several dierent discontinuous Galerkin discretizations of the compressible
Navier-Stokes equations can be derived. First we note, that for the discretizations to be consis-
tent and adjoint consistent there are requirements on the numerical uxes u
h
and
h
analog to
Theorems 5.7 and 5.12 for Poissons equations. In fact, we have
Theorem 9.6 Let
N
v
h
(, ) be given by (453). Then the discretization: nd u
h
V
d
h,p
such that
N
v
h
(u
h
, v
h
) = 0 v
h
V
d
h,p
, (454)
of a homogeneous Dirichlet problem is consistent if and only if the numerical uxes u and are
consistent, i.e.
u(v) = v, (v, v) = G(v)v on
I
, (455)
holds for all functions v [H
2
()]
m
. Furthermore, the discretization (454) is adjoint consistent if
and only if the numerical uxes u and are conservative, i.e.
[[ u(v)]] = 0, [[ (v, v)]] = 0 on
I
, (456)
holds for all functions v [H
2
()]
m
.
Proof: Analog to the proofs of Theorems 5.7 and 5.12.
We recall from the discretization of Poissons equation that the interior face terms of the sym-
metric interior penalty method, SIPG, and of the modied DG discretization of Bassi and Rebay,
BR2, are adjoint consistent. In contrast to that the method of Baumann-Oden, BO, and the non-
symmetric interior penalty method, NIPG, are adjoint inconsistent and will thus not be considered
in the following.
For SIPG and BR2 let us choose the uxes u
h
and
h
to be given by
u
h
= {{u
h
}},
h
= {{G(u
h
)
h
u
h
}} (u
h
) on
I
,
where the penalization term (u
h
) is given by
(u
h
) =
ips
(u
h
) = C
IP
p
2
he
[[u
h
]] for IP [31],
(u
h
) =
ip
(u
h
) = C
IP
p
2
he
{{G(u
h
)}}[[u
h
]] for IP [33], (457)
(u
h
) =
br2
(u
h
) = C
BR2
{{L
e
0
(u
h
)}} for BR2 [7, 8],
where the local lifting operator L
e
0
(u
h
)
d
h,p
is dened by:
_
e
L
e
0
(u
h
) : dx =
_
e
[[u
h
]] : {{G
(u
h
)}} ds
d
h,p
,
where
e
=
+
e
e
with e =
+
e
e
. Then,
[[ u
h
]] = [[{{u
h
}}]] = 0,
{{ u
h
}} = {{{{u
h
}}}} = {{u
h
}},
{{
h
}} = {{{{G(u
h
)
h
u
h
}}}} {{(u
h
)}} = {{G(u
h
)
h
u
h
}} (u
h
),
[[
h
]] = [[{{G(u
h
)
h
u
h
}}]] [[(u
h
)]] = 0,
92
the last term in (453) vanishes and thus (453) reduces to
N
v
h
(u
h
, v
h
) =
_
G(u
h
)
h
u
h
:
h
v
h
dx
_
I
[[u
h
]] : {{G
(u
h
)v
h
}} ds
I
{{G(u
h
)
h
u
h
}} : [[v
h
]] +
_
I
(u
h
) : [[v
h
]] ds +
N
,h
(u
h
, v
h
),
where the boundary term N
v
,h
(u
h
, v
h
) will be specied in the following section.
Discretization of viscous boundary terms On boundary edges we choose
u
h
= u
(u
+
h
),
h
= F
v
(u
+
h
, u
+
h
)
(u
+
h
),
where
F
v
(u
h
, u
h
) = F
v
(u
(u
h
), u
h
) = G
(u
h
)u
h
= G(u
(u
h
))u
h
on ,
and on
W,adia
, the viscous ux F
v
(u
h
, u
h
) = (0,
1j
n
x
j
,
2j
n
x
j
,
ij
v
j
n
x
i
)
.
The penalization term
(u
h
) on is given by
(u
h
) =
ips
(u
h
) = C
IP
p
2
he
(u
h
u
(u
h
)) n for IP [31],
(u
h
) =
ip
(u
h
) = C
IP
p
2
he
G
(u
h
) (u
h
u
(u
h
)) n for IP [33], (458)
(u
h
) =
br2
(u
h
) = C
BR2
L
e
(u
h
) for BR2 [7, 8],
where the local lifting operator L
e
(u
h
)
d
h,p
on is dened by:
_
L
e
(u
h
) : dx =
_
e
(u
h
u
(u
h
)) n :
_
G
(u
h
)
_
ds
d
h,p
for such that = e. Thereby the boundary term N
v
,h
(u
h
, v
h
) is given by
N
v
,h
(u
h
, v) =
_
n F
v
(u
+
h
, u
+
h
) v
+
ds
_
G
(u
+
h
)v
+
h
_
:
_
u
+
h
u
(u
+
h
)
_
nds +
_
(u
+
h
) v
+
ds.
(459)
Here, the boundary value function u
(u
+
h
) on supersonic and subsonic inow and outow boundary
conditions is given like for inviscid ows in Section 8.4. Additionally, for no-slip wall boundaries
with v = 0 we distinguish adiabatic boundary conditions, n T = 0, where
u
(u) = (u
1
, 0, 0, u
4
)
on
W,adia
, (460)
and isothermal boundary conditions, T = T
wall
, where
u
(u) = (u
1
, 0, 0, u
1
c
v
T
wall
)
on
W,iso
. (461)
Having derived DG discretizations of the compressible Euler equations and various DG dis-
cretizations of the viscous part of the compressible Navier-Stokes equations we now can combine
the discretizations of convective and viscous parts to form the DG discretization of the compressible
Navier-Stokes equations: nd u
h
V
d
h,p
such that
N
h
(u
h
, v
h
) = 0 v
h
V
d
h,p
, (462)
93
where the semilinear form N
h
(, ) : [H
1
(T
h
)]
m
[H
1
(T
h
)]
m
R is given by
N
h
(u
h
, v
h
) =
_
F
c
(u
h
)
h
v
h
dx +
T
h
_
\
H(u
+
h
, u
h
, n) v
+
h
ds
_
G(u
h
)
h
u
h
:
h
v
h
dx
_
I
[[u
h
]] : {{G
(u
h
)v
h
}} ds
I
{{G(u
h
)
h
u
h
}} : [[v
h
]] ds +
_
I
(u
h
) : [[v
h
]] ds +N
,h
(u
h
, v
h
),
and the boundary term N
,h
(u
h
, v
h
) is given by
N
,h
(u
h
, v) =
_
(u
+
h
, u
(u
+
h
), n) v
+
h
ds
_
n F
v
(u
+
h
, u
+
h
) v
+
ds
_
G
(u
+
h
)v
+
h
_
:
_
u
+
h
u
(u
+
h
)
_
nds +
_
(u
+
h
) v
+
ds.
(463)
Furthermore, on we have
H
(u
+
h
, u
(u
+
h
), n) = n F
c
(u
+
h
) = n F
c
(u
(u
+
h
)),
F
v
(u
h
, u
h
) = F
v
(u
(u
h
), u
h
) = G
(u
h
)u
h
,
G
(u
h
)u
h
= G(u
(u
h
))u
h
,
and (u
h
) and
(u
+
h
) are as given in (457) and (458), respectively. Finally, u
(u
+
h
) is given like
in Section 8.4 and in Equations (460) and (461).
9.3 Adjoint consistency analysis of DG for the compressible Navier-Stokes
equations
In this section we analyze the consistency and adjoint consistency property of the compressible
Navier-Stokes equations. In particular, we derive target functional modications which result in
an adjoint consistent discretization.
9.3.1 The continuous adjoint problem to the compressible Navier-Stokes equations
The most important target quantities in viscous compressible ows are the total (i.e. the pressure
induced plus viscous) drag and lift coecients, c
d
and c
l
, dened by
J(u) =
_
j(u) ds =
1
C
W
(p n n) ds =
1
C
W
(p n
i
ij
n
j
)
i
ds, (464)
where C
and are as in (421). In order to derive the adjoint problem, we multiply the left hand
side of (440) by z, integrate by parts and linearize about u to obtain
( (F
c
u
wF
v
u
wF
v
u
w) , z)
= ((F
c
u
F
v
u
) wF
v
u
w, z)
+ (n (F
c
u
wF
v
u
wF
v
u
w) , z)
,
where F
v
u
:=
u
F
v
(u, u) = G
[u]u and F
v
u
:=
u
F
v
(u, u) = G(u) denote the derivatives
of F
v
with respect to u and u, respectively. Using integration by parts once more, we obtain the
following variational formulation of the continuous adjoint problem: nd z such that
_
w, (F
c
u
F
v
u
)
z
_
_
w,
_
(F
v
u
)
z
__
+
_
w, n
_
(F
v
u
)
z
__
+
_
w, (n (F
c
u
F
v
u
))
z
_
_
w, (n F
v
u
)
z
_
= J
[u](w) w V.
94
Given that
J
[u](w) =
1
C
W
(p
u
[u] n
u
[u] n) w(
u
[u] n) wds
=
_
w,
1
C
(p
u
n
u
n)
_
_
w,
1
C
(
u
n)
_
W
,
(465)
we see, that the adjoint solution z satises following equation
(F
c
u
F
v
u
)
z
_
(F
v
u
)
z
_
= 0, (466)
subject to the boundary conditions on
W
=
iso
adia
,
(n (F
c
u
F
v
u
))
z +n
_
(F
v
u
)
z
_
=
1
C
(p
u
n
u
n) , (467)
(n F
v
u
)
z =
1
C
(
u
n) . (468)
At wall boundaries
W
, where v = (v
1
, v
2
)
_
_
_
_
0
(
u
n)
1
1
(
u
n)
2
2
0
_
_
_
_
, (469)
which reduces to the conditions z
2
=
1
C
1
on
W
, z
3
=
1
C
2
on
W
, and z
4
= 0 on
iso
. At
adiabatic boundaries we have n T = 0 and the last condition in (469) vanishes. Substituted into
(467) we obtain n((F
v
u
)
z) = 0 on
W
which at adiabatic boundaries reduces to nz
4
= 0. On
isothermal boundaries no additional boundary condition is obtained. In summary, the boundary
conditions of the adjoint problem (466) to the compressible Navier-Stokes equations are given by
z
2
=
1
C
1
, z
3
=
1
C
2
on
W
, z
4
= 0 on
iso
, n z
4
= 0 on
adia
. (470)
9.3.2 Primal residual form of DG for the compressible Navier-Stokes equations
Using integration by parts in (462) we obtain
N
h
(u
h
, v
h
)
_
h
F
c
(u
h
)v
h
dx +
T
h
_
\
_
H(u
+
h
, u
h
, n
+
) n F
c
(u
+
h
)
_
v
+
h
ds
(
h
F
v
(u
h
,
h
u
h
)) v
h
dx +
1
2
T
h
_
\
[[F
v
(u
h
,
h
u
h
)]] v
+
h
ds
1
2
T
h
_
\
G
(u
h
)
h
v
h
: [[u
h
]] ds +
T
h
_
\
(u
h
) : v
+
h
n
+
ds
_
F
c
(u
+
h
) F
v
(u
+
h
, u
+
h
)
_
: v
+
h
nds +N
,h
(u
h
, v
h
) = 0,
which can be expressed in the primal residual form as follows: nd u
h
V
d
h,p
such that
_
R(u
h
) v
h
dx +
T
h
_
\
r(u
h
) v
+
h
+(u
h
) : v
+
h
ds
+
_
(u
h
) v
+
h
+
(u
h
) : v
+
h
ds = 0 v
h
V
d
h,p
,
95
where the primal residuals are given by
R(u
h
) = F
c
(u
h
) + F
v
(u
h
,
h
u
h
) in , T
h
,
r(u
h
) =n F
c
(u
+
h
) H(u
+
h
, u
h
, n
+
)
1
2
[[F
v
(u
h
,
h
u
h
)]] (u
h
)n,
(u
h
) =
1
2
_
G(u
h
)[[u
h
]]
_
on \ , T
h
,
r
(u
h
) =n
_
F
c
(u
+
h
) F
c
(u
+
h
) F
v
(u
+
h
, u
+
h
) +F
v
(u
+
h
, u
+
h
)
_
(u
+
h
)n,
(u
h
) =
_
G
(u
+
h
) :
_
u
+
h
u
(u
+
h
)
_
n
_
on .
We see that the exact solution u to (440) satises
R(u) = 0, r(u) = 0, (u) = 0, r
(u) = 0,
(u) = 0,
where we used consistency of the numerical ux, H(w, w, n) = n F
c
(w), continuity of u, and
the consistency of the boundary function, i.e. u satises u
J(u
h
) = J(i(u
h
)) +
_
r
J
(u
h
) ds = J
(u
h
) +
_
r
J
(u
h
) ds. (471)
As in Section 8 for the compressible Euler equations, here we set i(u
h
) = u
(u
h
) and J
(u
h
) =
J(u
(u
h
)); r
J
(u
h
) will be specied later. Noting that u
h
[u
h
](w
h
, z
h
) =
J
[u
h
](w
h
) w
h
V
d
h,p
, (472)
where N
h
[u](w, z) is given by
N
h
[u](w, z) =
_
(F
c
u
[u]w) :
h
z dx +
T
h
_
\
H
u
+
(u
+
, u
, n
+
)w
+
[[z]] nds
+
_
_
G
[u]w
h
u
_
:
h
z dx +
_
(G(u)
h
w) :
h
z dx
I
{{G
[u]w
h
u}} : [[z]] ds
_
I
{{G(u)
h
w}} : [[z]] ds
I
{{
_
G
[u]w
h
z}} : [[u]] ds
_
I
{{G
(u)
h
z}} : [[w]] ds
+
_
[u](w) : [[z]] ds +N
,h
[u](w, z).
96
Using integration by parts this can be rewritten as follows
w(F
c
u
[u])
h
z dx +
T
h
_
\
w
+
_
H
u
+
(u
+
, u
, n
+
)
_
[[z]] nds
+
_
w
_
G
[u]
h
u
_
h
z dx
_
w
h
_
G
(u)
h
z
_
dx
1
2
T
h
_
\
_
G
[u]w
h
u
_
: [[z]] ds
1
2
T
h
_
\
(G(u)
h
w) : [[z]] ds
1
2
T
h
_
\
_
_
G
[u]w
h
z
_
: [[u]] ds +
1
2
T
h
_
\
w[[G
(u)
h
z]] ds
+
T
h
_
\
[u](w)[[z]] ds +
_
(wn) :
_
G
(u)
h
z
_
ds +N
,h
[u](w, z).
Hence, the discrete adjoint problem (472) in adjoint residual form is given as follows: nd z
h
V
d
h,p
such that
_
w R
[u
h
](z
h
) dx +
T
h
_
\
w r
[u
h
](z
h
) +w :
[u
h
](z
h
) ds
+
_
w r
[u
h
](z
h
) +w :
[u
h
](z
h
) ds = 0 w V
d
h,p
, (473)
where the adjoint residuals are given by
R
[u
h
](z
h
) =
_
F
c
u
(u
h
) G
[u
h
]u
h
_
h
z
h
+
h
_
G
(u
h
)
h
z
h
_
in , T
h
,
r
[u
h
](z
h
) =
_
H
u
+
(u
+
h
, u
h
, n
+
)
_
[[z
h
]] n
1
2
[[G
(u
h
)z
h
]]
_
[u
h
]
_
[[z
h
]] (474)
+
1
2
_
G
[u
h
]u
h
_
[[z
h
]] +
1
2
_
G
[u
h
][[u
h
]]
_
h
z
h
on \ , T
h
,
[u
h
](z
h
) =
1
2
G
[u
h
][[z
h
]] on \ , T
h
.
The adjoint boundary residuals r
and
[u]u
and F
v
u
= G(u) we see that the exact solution z to the continuous adjoint problem (466) satises
R
[u](z) = 0. In the two lines in (474) representing the face residual term r
[u
h
](z
h
) we recognize
the jump (H
u
+
)
[[z
h
]] n due to the convective part of the equations, cf. (439), furthermore the
second term in the rst line corresponding to the adjoint face residuals of the Poissons equation,
cf. (309), and nally the two terms in the second line due to the nonlinearity of the compressible
Navier-Stokes equations. Whereas the last term in the second line vanishes for a smooth exact
primal solution u, all other terms vanish for the exact solution z to the adjoint problem (466).
Thereby, the adjoint solution z satises r
[u](z) = 0. In
summary, we see that, as for the Poissons equation, the element and interior face terms of the
SIPG discretization are adjoint consistent.
97
The boundary terms of the discrete adjoint problem are given by
N
,h
[u
h
](w, z
h
)+
_
(wn) :
_
G
(u
h
)
h
z
h
_
ds
+
_
n
_
F
c
,u
[u
h
](w)
_
z
h
ds +
_
[u
h
](w) z ds,
n
_
F
v
,u
[u
h
,
h
u
h
](w) +F
v
,u
[u
h
,
h
u
h
](
h
w)
_
z
h
ds
__
_
G
[u
h
]w
_
h
z
h
_
: (u
h
u
(u
h
))nds
_
G
(u
h
)
h
z
h
_
:
_
wu
[u
h
]w
_
nds
+
_
(wn) :
_
G
(u
h
)
h
z
h
_
ds =
J
[u
h
](w).
Thus the adjoint boundary residuals in (473) on
W
are given by
r
[u
h
](z
h
) =
1
C
(p
u
n
u
n)
_
n
_
F
c
,u
F
v
,u
__
z
h
n
_
G
z
h
_
+r
J
[u
h
]
_
[u
h
]
_
z
h
+
_
G
[u
h
] : (u
h
u
(u
h
)) n
_
h
z
h
+
_
G
(u
h
) :
_
I u
[u
h
]
_
n
_
h
z
h
, (475)
[u
h
](z
h
) =
1
C
(
u
n) +
_
n F
v
,u
_
z
h
. (476)
We recall (468), F
v
,u
= G
(u), and see that the exact solutions u and z to the primal problem
(440) and the continuous adjoint problem (466)-(470) satisfy
[u](z) = 0.
We now choose the modication r
J
(u
h
) of the target functional in (471) as follows
r
J
(u
h
) =
(u
+
h
) z
_
G
(u
+
h
)
h
z
_
:
_
u
+
h
u
(u
+
h
)
_
n, (477)
with Frechet derivative
r
J
[u
h
](w) =
[u
h
](w) z
_
G
[u
h
] : (u u
(u
h
)) n
_
h
z
_
G
(u
h
) :
_
I u
[u
h
]
_
n
_
h
z
.
As the exact solution u to the primal problem satises u
(u) = u, we have r
J
(u) = 0. Hence,
(477) is a consistent modication of the target functional. Recalling (467), we see that the exact
solutions u and z satisfy
r
[u](z) =
_
[u]
_
(z
z)
_
G
[u] : (u u
(u)) n
_
(z
z)
_
G(u) :
_
I u
[u]
_
n
_
(z
z) .
Furthermore, setting z
= z on
W
we obtain r
(u
+
h
)z
_
= n ((F
v
u
)
z) = 0 on
W
the second term in (477) vanishes.
Furthermore, on adiabatic boundaries
adia
we have (u
+
h
u
(u
+
h
))
i
= 0, i = 1, 4, and on isothermal
boundaries
iso
we have (u
+
h
u
(u
+
h
))
1
= 0. Together with (470), the consistent modication (477)
reduces to
r
J
(u
h
) =
(u
+
h
) z
=
_
(u
+
h
)
_
2
1
C
1
+
_
(u
+
h
)
_
3
1
C
2
,
(478)
98
which completes the adjoint consistency analysis of the interior penalty discontinuous Galerkin
discretization of the compressible Navier-Stokes equations. Finally, we note that the consistent
modication r
J
(u
h
) given in (478) for SIPG reduces to r
J
(u
h
) =
_
u
+
h
u
(u
+
h
)
_
z
which
corresponds to the IP modication of target functionals for the Poissons equation, where r
J
(u
h
) =
(u
h
g
D
)z
, with z
= j
D
, see (311).
In summary, we have shown that the adjoint element and interior residuals R
[u
h
](z
h
), r
[u
h
](z
h
)
and
[u
h
](z
h
), see (474), vanish for the exact solutions u and z to (440) and (466), respectively.
Additionally, using an adjoint consistent treatment of convective and diusive boundary uxes,
n F
c
(u
+
h
) = n F
c
(u
(u
+
h
)), n F
v
(u
+
h
) = n F
v
(u
(u
+
h
),
h
u
+
h
), (479)
and using the following consistent modication of the target functional,
J(u
h
) = J(u
(u
h
)) +
_
(u
h
) z
ds, (480)
with z
=
1
C
(0,
1
,
2
, 0)
[u
h
](z
h
) and
[u
h
](z
h
), see (475) and (476), vanish for the exact
solutions u and z. Thereby, using the modications given in (479) and (480), we recover an adjoint
consistent discontinuous Galerkin discretizations of the compressible Navier-Stokes equations in
conjunction with total force coecients.
We note that arguments given in [36] en route to obtaining an adjoint consistent discretization
based on the BR2 scheme [7] can also be covered within the presented framework and lead to
analogous modications. Furthermore, we note that numerical experiments in [27] have conrmed
that in contrast to the original formulation in [31] the discrete adjoint solution to the adjoint
consistent discretization is entirely smooth. Furthermore, numerical tests on globally rened meshes
have shown that the adjoint consistent discretization is by a factor of 2-400 more accurate measured
in terms of viscous force coecients than the original formulation. Also a signicantly improved
order of convergence has been observed. Finally, we note that in [33], see also Section 9.4, the
interior penalty DG discretization with adjoint consistent discretization of boundary conditions
and an improved penalty term (457) has shown to be of optimal order. In fact, the accuracy of the
IP discretization in [33] is comparable to the accuracy of the BR2 discretization in [7, 8] while the
residual computation of IP is signicantly cheaper than that of BR2 which requires the additional
evaluation of local lifting operators.
9.4 Numerical results
Flow over a at plate We begin by investigating the accuracy of higher order DG discretizations
in resolving laminar boundary layers. To this end, we consider a Mach 0.01 ow with Reynolds
number 10000 horizontally passing over a at plate of length l = 2. The boundary layer solution
to this problem can be approximated using Blasius solution, see [39], for example. In Figure 14,
by [31], we compare the numerical solution computed with the DG(p) method for 1 p 3, at
x =
l
2
= 1 and a local Reynolds number Re
x
= 5000, with the Blasius solution ( = y
_
u
/(x) =
y
x
Re
x
versus u/u
u/u
0
1
2
3
4
5
6
7
8
9
0 0.2 0.4 0.6 0.8 1
DG(1)
DG(2)
DG(3)
Blasius
u/u
0
1
2
3
4
5
6
7
8
9
0 0.2 0.4 0.6 0.8 1
DG(1)
DG(2)
DG(3)
Blasius
u/u
0
1
2
3
4
5
6
7
8
9
0 0.2 0.4 0.6 0.8 1
DG(1)
DG(2)
DG(3)
Blasius
u/u
/(x) =
y
x
Re
x
versus u/u
u/u
3
4
5
6
7
8
9
0.96 0.97 0.98 0.99 1 1.01
DG(1)
DG(2)
DG(3)
Blasius
u/u
Figure 15: Zoom of the DG(p), 1 p 3, solutions on the two nest grids, [31].
latter two ner meshes is shown in the zoom depicted in Figure 15. Here, we see that there is
still a signicant dierence between the Blasius solution and the computed discontinuous Galerkin
solution with p = 1. Indeed, these gures clearly highlight the substantial gains in accuracy
attained when higherorder polynomial degrees are employed with the DG method. This is further
highlighted in Table 1, where we summarize the number of elements and the number of degrees of
freedom, orthogonal to the wall, which are required by the DG method for each polynomial degree
in order to resolve the boundary layer to a sucient accuracy that the error in computed viscous
stress forces exerted on the wall are within 5% of that computed with the Blasius solution.
DG(1) DG(2) DG(3)
elements 36 5 3
DoFs 72 15 12
Table 1: Number of elements and degrees of freedom in the boundary layer required by DG(p),
1 p 3, discretizations for approximating the viscous force up to 5%, [31].
100
10
1
10
2
10
8
10
6
10
4
10
2
10
0
10
2
1
2
1
3
1
4
1
5
p=1
p=2
p=3
p=4
SIPG
BR2
u
h
L
2
(
elements
10
1
10
2
10
10
10
8
10
6
10
4
10
2
10
0
1
2
1
4
1
6
1
8
p=1
p=2
p=3
p=4
SIPG
BR2
|
J
(
u
)
J
(
u
h
)
|
elements
Figure 16: Flow in a square domain. Comparison of the SIPG and BR2 methods when the error is
measured in terms of: (left) L
2
()-norm; (right) Weighted mean-value functional J(), [33].
Flow in a square domain In the following we examine the experimental order of convergence
of the interior penalty DG discretization, [33]. To this end we consider following model problem:
let = (0, )
2
, and supplement the compressible NavierStokes equations (440) with an inho-
mogeneous forcing function f , which is chosen so that the analytical solution to (440) is given
by
u(x) = (sin(2(x
1
+x
2
)) + 4, sin(2(x
1
+x
2
))/5 + 4, sin(2(x
1
+x
2
))/5 + 4, (sin(2(x
1
+x
2
)) + 4)
2
)
,
where the dynamic viscosity coecient has been set to 1/10. This represents a modication of
the (unsteady) test problem employed in the article [19]. In this section we shall be interested in
measuring the discretization error in terms of both the L
2
()-norm as well as in terms of a given
target functional J(). In the latter case, we consider the weighted mean-value of the density, i.e.,
J(u) J
(u) =
_
u
1
dx,
where = sin(x) sin(y); thereby, the true value of the functional is given by J(u) = 1.1685876486.
In Figure 16(a) we present a comparison of the error in the L
2
()-norm with the (square root
of the) number of elements for p = 1, 2, 3, 4, employing both the SIPG method with C
IP
= 10 and
the BassiRebay method (BR2) with C
BR2
= 4. In both cases, we observe that u u
h
L
2
()
converges to zero at the expected optimal rate O(h
p+1
) as the mesh is rened for each xed
p. Moreover, from Figure 16(b) we observe that the error in the computed target functional
J() behaves (approximately) like O(h
2p
), for each xed p, as the mesh is uniformly rened for
both of the discretization schemes considered. These rates of convergence for both the L
2
()-
norm of the error and the error in the computed target functional J() are in complete agreement
with the corresponding convergence behavior we would expect for the SIPG and BR2 methods
when applied to a linear convectiondiusion problem; see [23], for example, for the analysis of
general interior penalty DGFEMs for secondorder partial dierential equations with non-negative
characteristic form. We remark that in terms of accuracy, for a given number of elements, or
equivalently, for a xed number of degrees of freedom, both the SIPG scheme and the BR2 method
perform in a comparable manner, with the latter scheme being, in general, slightly more accurate.
However, in terms of computational resources, the time required to assemble the residual vector
of the BR2 method, which is the most computationally intensive part of the ow solver, when
explicit time-stepping schemes are employed, is signicantly more expensive than the computation
101
10
1
10
2
10
8
10
6
10
4
10
2
10
0
10
2
1
2
1
4
p=1
p=2
p=3
p=4
NIPG
STSIPG
SIPG
u
h
L
2
(
elements
Figure 17: Flow in a square domain. Comparison of the SIPG, NIPG, and STSIPG methods when
the error is measured in terms of the L
2
()-norm, [33].
10
1
10
2
10
8
10
6
10
4
10
2
10
0
1
2
1
4
p=1
p=2
p=3
p=4
|
J
(
u
)
J
(
u
h
)
|
elements
Figure 18: Flow in a square domain. Convergence of the NIPG scheme with respect to J() with
hrenement, [33].
of the corresponding quantity when the SIPG scheme is employed. More precisely, for (bi)-linear,
elements, i.e., p = 1, the BR2 method is around 38% more expensive than the SIPG scheme; this
overhead increases as the underlying polynomial degree is enriched. Indeed, for p = 2, the BR2
method is approximately 47% more expensive, and for p = 3 and p = 4 the additional work rises
to around 55%. This increase in the cpu times when the BR2 method is employed is attributed to
the computation of the lifting operator on each face of the computational mesh.
Finally, in this section we compare the performance of the SIPG method with both the corre-
sponding NIPG formulation of the underlying scheme, together with the interior penalty method
outlined in the article [31]; we shall refer to this latter scheme as the standard SIPG (STSIPG)
method. To this end, in Figure 17 we plot the L
2
()-norm of the error against the (square root of
the) number of elements for p = 1, 2, 3, 4 using each of the above schemes. In contrast to the SIPG
and BR2 methods, we now observe that uu
h
L
2
()
behaves like O(h
p+1
) for odd p and like O(h
p
)
for even p when either the NIPG method or the STSIPG scheme are employed. The sub-optimal
convergence observed when employing these two schemes is attributed to the lack of smoothness
102
10
2
10
8
10
7
10
6
10
5
10
4
10
3
10
2
1
2
1
4
1
6
p=1
p=2
p=3
SIPG
BR2
|
J
(
u
)
J
(
u
h
)
|
elements
10
2
10
4
10
3
10
2
p=1
p=2
p=3
SIPG
BR2
|
J
(
u
)
J
(
u
(
u
h
)
)
|
elements
Figure 19: Viscous ow around NACA0012 airfoil. Comparison of the SIPG and BR2 methods
employing: (left) Adjoint consistent reformulation of the drag functional; (right) Adjoint consistent
reformulation of the drag functional excluding the penalty terms, [33].
in the resulting adjoint problems, cf. [23, 28]. Moreover, the same behavior is also observed in the
functional setting; indeed, for the NIPG scheme, from Figure 18 we see that |J(u) J(u
h
)| tends
to zero at (approximately) the rate O(h
p+1
) for odd p and O(h
p
) for even p, as the mesh is uniform
rened. Analogous behavior is also observed when the error in the computed target functional J()
is evaluated using the STSIPG method; for brevity, these numerics have been omitted.
Viscous ow around a NACA0012 airfoil In this example, we consider a subsonic viscous
ow around a NACA0012 airfoil. At the fareld (inow) boundary we specify a Mach 0.5 ow at
a zero angle of attack, i.e. = 0
j(u) ds =
1
C
W
(p n n)
d
ds,
where j(u) =
1
C
p n
d
on
W
and j(u) 0 elsewhere, cf. (464) and (421). We remark that the
adjoint consistency of the SIPG scheme is based on the consistent reformulation of J() dened in
(480). With this in mind, in Figure 19(a) we present a comparison of the error in the computed
target functional with the (square root of the) number of elements for p = 1, 2, 3, employing both
the SIPG method with C
IP
= 10 and the BassiRebay method (BR2) with C
BR2
= 4. In both cases,
we observe that, asymptotically, at least, |J(u)
J(u
h
)| converges to zero at the expected optimal
rate O(h
2p
) as the mesh is rened for each xed p. Moreover, as before, we note that in terms of
accuracy, for a given number of elements, or equivalently, for a xed number of degrees of freedom,
both the SIPG scheme and the BR2 method perform in a comparable manner, though as already
noted, the SIPG scheme requires less computational eort to attain the computed solution. To
highlight the necessity of the consistent reformulation of the original target functional J() through
the additional of the term involving the penalty function
(u
h
))| with the (square root of the) number
of elements for p = 1, 2, 3 employing both the SIPG and BR2 schemes. In this case, we now
observe that there is a signicant deterioration of the error for a given mesh size and polynomial
103
Figure 20: Laminar ow around a delta wing. Geometry of the delta wing and the Mach number
isolines on several slices of the ow eld computed based on a DG(p) discretization with p = 1 (2nd
order) on the left wing and with p = 4 (5th order) on the right wing.
order when compared to the corresponding results when the penalty function modication of the
target functional has been included. Indeed, comparing Figures 19(a) and 19(b), we see that the
inclusion of the penalty function modication in the denition of
J() leads to around 23 orders
of magnitude improvement in the computed error in the drag.
Laminar ow around a delta wing As a nal example we consider a laminar ow around a
delta wing. At the fareld (inow) boundary we specify a Mach 0.3 ow at an angle = 12.5
of
attack with Reynolds number Re = 4000. On the walls of the delta wing we impose an isothermal
wall boundary condition. This is the BTC3 test case of the EU-project ADIGMA [1]. Figure 20
shows the Mach number isolines on several slices of the ow eld. The ow is computed on a coarse
mesh of 3264 elements. The corresponding surface mesh is depicted on the wing geometry. The
ow eld on the left part of the delta wing is based on a (2nd order) DG discretization with p = 1
and the right part is based on a (5th order) DG discretization with p = 4. We see that the 5th
order ow solution provides a good resolution of the primary and secondary vortices. Furthermore,
the vortices are tracked over some distance behind the wing. In contrast to that, the primary and
secondary vortices are almost indistinguishable in the 2nd order ow solution. Here, the vortices
merge and are damped out far too early. Already after a short distance behind the wing the
original vortex system is lost due to numerical viscosity. Figure 21 compares the DG(p) solutions
for p = 1, 2, 3. Here, the error is given in terms of the drag, lift and moment coecients, c
l
, c
d
and
c
m
, respectively, and is plotted against the number of degrees of freedoms (DoFs) per equation.
The horizontal line in each error plot in Figure 21 represents the error tolerances,
|J
c
l
(u) J
c
l
(u)| 10
2
,
|J
c
d
(u) J
c
d
(u)| 10
3
,
|J
cm
(u) J
cm
(u)| 10
3
,
104
1e-04
0.001
0.01
0.1
10000 100000 1e+06
dg1
dg2
dg3
0.01
e
r
r
o
r
i
n
c
l
DoFs per equ.
1e-04
0.001
0.01
0.1
10000 100000 1e+06
dg1
dg2
dg3
0.001
e
r
r
o
r
i
n
c
d
DoFs per equ.
1e-04
0.001
0.01
0.1
10000 100000 1e+06
dg1
dg2
dg3
0.001
e
r
r
o
r
i
n
c
m
DoFs per equ.
Figure 21: Laminar ow around a delta wing. Comparison of the DG(p), p = 1, 2, 3, solutions. The
error is given in terms of the drag, lift and moment coecients, c
l
, c
d
and c
m
, respectively.
as dened in ADIGMA project. Here, we see a clear advantage of using higher order DG approxi-
mations over 2nd order approximations. In fact, in terms of DoFs the DG(3) discretization is about
a factor of 10 more ecient than DG(1). This advantage further increases when stronger accuracy
requirements, i.e. smaller error tolerances, are imposed.
Acknowledgments
The author acknowledges the nancial support by the Presidents Initiative and Networking Fund of
the Helmholtz Association of German Research Centres, and the nancial support of the European
Union, under the ADIGMA project, [1]. If not indicated dierently, computations have been performed
using the DG ow solver PADGE [29] based on the deal.II library [3, 4].
References
[1] ADIGMA. Adaptive higher-order variational methods for aerodynamic applications in industry. A
specic targeted research project of the sixth European community framework programme. See
http://www.dlr.de/as/desktopdefault.aspx/tabid-2035.
[2] D. Arnold, F. Brezzi, B. Cockburn, and L. Marini. Unied analysis of discontinuous Galerkin methods
for elliptic problems. SIAM J. Numer. Anal., 39(5):17491779, 2002.
[3] W. Bangerth, R. Hartmann, and G. Kanschat. deal.II A general purpose object oriented nite element
library. ACM Transactions on Mathematical Software, 33(4), Aug. 2007.
[4] W. Bangerth, R. Hartmann, and G. Kanschat. deal.II Dierential Equations Analysis Library, Tech-
nical Reference. http://www.dealii.org/, 6.1 edition, May 2008. First edition 1999.
[5] F. Bassi and S. Rebay. A high-order accurate discontinuous nite element method for the numerical
solution of the compressible Navier-Stokes equations. J. Comp. Phys., 131:267279, 1997.
[6] F. Bassi and S. Rebay. High-order accurate discontinuous nite element solution of the 2d Euler
equations. J. Comp. Phys., 138:251285, 1997.
[7] F. Bassi and S. Rebay. GMRES discontinuous Galerkin solution of the compressible Navier-Stokes
equations. In B. Cockburn, G. Karniadakis, and C.-W. Shu, editors, Discontinuous Galerkin Methods,
volume 11, pages 197208. Springer, 1999.
[8] F. Bassi and S. Rebay. Numerical evaluation of two discontinuous Galerkin methods for the compressible
Navier-Stokes equations. Int. J. Numer. Meth. Fluids, 40:197207, 2002.
[9] C. Baumann and J. Oden. A discontinuous hp nite element method for the Euler and Navier-Stokes
equations. International Journal for Numerical Methods in Fluids, 31:7995, 1999.
105
[10] C. Baumann and J. Oden. An adaptive-order discontinuous Galerkin method for the solution of the
Euler equations of gas dynamics. International Journal for Numerical Methods in Engineering, 47:6173,
2000.
[11] F. Brezzi, G. Manzini, D. Marini, P. Pietra, and A. Russo. Discontinuous Galerkin approximations for
elliptic problems. Num. Meth. Part. Di. Eq., 16(4):365378, 2000.
[12] F. Brezzi, D. Marini, and E. S uli. Residual-free bubbles for advection-diusion problems: the general
error analysis. Numerische Mathematik, 85(1):3147, 2000.
[13] F. Brezzi, L. D. Marini, and E. S uli. Discontinuous Galerkin methods for rst-order hyperbolic problems.
Math. Models and Methods in Appl. Sci., 14(12):18931903, 2004.
[14] A. Cangiani and E. S uli. Enhanced residual-free bubble method for convection-diusion problems. Int.
J. Numer. Meth. Fluids, 47(1011):13071313, 2005.
[15] G. Chiocchia. Exact solutions to transonic and supersonic ows. Technical Report AR-211, AGARD,
1985.
[16] V. Dolejsi. On the discontinuous Galerkin method for the numerical solution of the Navier-Stokes
equations. Int. J. Numer. Meth. Fluids, 45:10831106, 2004.
[17] K. J. Fidkowski and D. L. Darmofal. A triangular cut-cell adaptive method for high-order discretizations
of the compressible Navier-Stokes equations. J. Comput. Physics, 225:16531672, 2007.
[18] K. J. Fidkowski, T. A. Oliver, J. Lu, and D. L. Darmofal. p-multigrid solution of high-order discontinuous
Galerkin discretizations of the compressible Navier-Stokes equations. J. Comp. Phys., 207(1):92113,
July 2005.
[19] G. Gassner, F. L orcher, and C.-D. Munz. A discontinuous Galerkin scheme based on a space-time
expansion II. Viscous ow equations in multi-dimensions. J. Sci. Comput., 34(3):260286, 2008.
[20] M. Giles and N. Pierce. Adjoint equations in CFD: duality, boundary conditions and solution behaviour.
AIAA, 97-1850, 1997.
[21] J.-L. Guermond. Stabilization of Galerkin approximations of transport equations by subgrid modeling.
M2AN, 33:12931316, 1999.
[22] K. Harriman, D. Gavaghan, and E. S uli. The importance of adjoint consistency in the approximation
of linear functionals using the discontinuous Galerkin nite element method. Technical report, Oxford
University Computing Laboratory, 2004.
[23] K. Harriman, P. Houston, B. Senior, and E. S uli. hpVersion discontinuous Galerkin methods with inte-
rior penalty for partial dierential equations with nonnegative characteristic form. In Recent Advances
in Scientic Computing and Partial Dierential Equations, volume 330 of Contemporary Mathematics,
pages 89119. AMS, 2003.
[24] R. Hartmann. Adaptive Finite Element Methods for the Compressible Euler Equations. PhD thesis,
University of Heidelberg, 2002.
[25] R. Hartmann. The role of the Jacobian in the adaptive Discontinuous Galerkin method for the com-
pressible Euler equations. In G. Warnecke, editor, Analysis and Numerics for Conservation Laws, pages
301316. Springer, 2005.
[26] R. Hartmann. Derivation of an adjoint consistent discontinuous Galerkin discretization of the com-
pressible Euler equations. In G. Lube and G. Rapin, editors, Proceedings of the BAIL 2006 conference,
2006.
[27] R. Hartmann. Adjoint consistency analysis of discontinuous Galerkin discretizations. SIAM J. Numer.
Anal., 45(6):26712696, 2007.
[28] R. Hartmann. Error estimation and adjoint based renement for an adjoint consistent DG discretization
of the compressible Euler equations. Int. J. Computing Science and Mathematics, 1(24):207220, 2007.
[29] R. Hartmann, J. Held, T. Leicht, and F. Prill. PADGE, Parallel Adaptive Discontinuous Galerkin
Environment, Technical reference. DLR, Braunschweig, 2008. In preparation.
106
[30] R. Hartmann and P. Houston. Adaptive discontinuous Galerkin nite element methods for the com-
pressible Euler equations. J. Comput. Phys., 183(2):508532, 2002.
[31] R. Hartmann and P. Houston. Symmetric interior penalty DG methods for the compressible Navier
Stokes equations I: Method formulation. Int. J. Num. Anal. Model., 3(1):120, 2006.
[32] R. Hartmann and P. Houston. Symmetric interior penalty DG methods for the compressible Navier
Stokes equations II: Goaloriented a posteriori error estimation. Int. J. Num. Anal. Model., 3(2):141
162, 2006.
[33] R. Hartmann and P. Houston. An optimal order interior penalty discontinuous Galerkin discretization
of the compressible NavierStokes equations. J. Comput. Phys., 227(22):96709685, 2008.
[34] P. Houston, J. Mackenzie, E. S uli, and G. Warnecke. A posteriori error analysis for numerical approxi-
mations of Friedrichs systems. Numerische Mathematik, 82:433470, 1999.
[35] P. Houston and E. S uli. hpadaptive discontinuous Galerkin nite element methods for rstorder
hyperbolic problems. SIAM J. Sci. Comp., 23(4):12261252, 2002.
[36] J. Lu. An a posteriori Error Control Framework for Adaptive Precision Optimization using Discontin-
uous Galerkin Finite Element Method. PhD thesis, M.I.T., 2005.
[37] T. E. Peterson. A note on the convergence of the discontinuous Galerkin method for a scalar hyperbolic
equation. SIAM J. Numer. Anal., 28(1):133140, 1991.
[38] S. Prudhomme, F. Pascal, J. Oden, and A. Romkes. Review of a priori error estimation for discontinuous
Galerkin methods. TICAM Report 00-27, University of Texas, 2000.
[39] H. Schlichting and K. Gersten. Boundary-Layer Theory. Springer, 2003.
[40] C. Schwab. p- and hp-Finite Element methods. Theory and Applications to Solid and Fluid Mechanics.
Oxford University Press, 1998.
[41] J. van der Vegt and H. van der Ven. Space-time discontinuous Galerkin nite element method with
dynamic grid motion for inviscid compressible ows, I. General formulation. J. Comp. Phys., 182:546
585, 2002.
[42] K. G. van der Zee. An H
1
(P
h
)-coercive discontinuous Galerkin formulation for the Poisson problem:
1-d analysis. Masters thesis, TU Delft, 2004.
107