1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
|
% --------------------------------------------
\subsection{Nonlinear scalar hyperbolic problems revisited}
% --------------------------------------------
\label{sec-hyperbolic-hdg}%
\apindex{discontinuous!hybrid}%
\cindex{problem!hyperbolic nonlinear equation}
The aim of this paragraph is to study the
hybrid discontinuous Galerkin discretization
of scalar nonlinear hyperbolic equations.
This section reuses the general framework already
introduced in section~\ref{sec-hyperbolic-hdg}.
The stationary nonlinear hyperbolic problem writes:
\ \ $(P)$: \emph{find $u$, defined in $\Omega$, such that}
\begin{subequations}
\begin{eqnarray}
{\rm div}\,\boldsymbol{f}(u)
+
\sigma u
&=&
f
\ \mbox{ in } \Omega
\label{eq-hdg-hyp-eqn}
\\
\boldsymbol{f}(u).\boldsymbol{n} &=& \Phi(\boldsymbol{n};\,u,g)
\mbox{ on } \partial\Omega
\label{eq-hdg-hyp-bc}
\end{eqnarray}
\end{subequations}
with the notations of section~\ref{sec-hyperbolic-hdg}
and where $\sigma\geq 0$ is an optional source term parameter
and $f$ is a right-hand side.
For simplicity, we consider here the stationary problem:
its extension to the non-stationary case will be considered later.
It could be done either as section~\ref{sec-hyperbolic-hdg},
with Runge-Kutta schemes, or, and it will be a new feature, implicitely,
based on BDF schemes and a Newton algorithm.
Multiplying \eqref{eq-hdg-hyp-eqn} by a test-function $v$
and integrating on $K\in\mathscr{T}_h$, we get:
\begin{eqnarray}
\int_K
\left(
\sigma u v
-
\boldsymbol{f}(u).\nabla v
\right) \,{\rm d}x
+
\int_{\partial K}
\boldsymbol{f}(u).\boldsymbol{n}
\, v
\,{\rm d}s
&=&
\int_K
f \, v
\,{\rm d}x
\label{eq-hdg-hyp-integ-part}
\end{eqnarray}
By definition of the Godunov flux $\Phi$ associated to $\boldsymbol{f}$,
a boundary condition similar to \eqref{eq-hdg-hyp-bc}
applies on $\partial K$:
\begin{eqnarray*}
\boldsymbol{f}(u).\boldsymbol{n} &=& \Phi(\boldsymbol{n};\,u,u_{ext})
\mbox{ on } \partial K
\end{eqnarray*}
where $u_{ext}$ denotes the external trace of $u$ on the boundary sides of $K$.
On all internal sides, let $\lambda=\average{u}$.
The $\lambda$ notation is now considered as an indpendent variable
and the constraint $\lambda-\average{u}=0$ will be imposed
by using a Lagrange multiplier.
We extends $\lambda$ on the boundary side by using
the boundary condition \eqref{eq-hdg-hyp-bc}.
For instance, for a transport problem,
we get $\lambda=g$ on the inflow boundary domain
and, on the rest of the boundary, $\lambda$, is the inner trace of $u$.
Let $S=\partial K_-\cap K_+$ be any internal side with orientation
such that the normal on $S$ is $\boldsymbol{n}=\boldsymbol{n}_-=-\boldsymbol{n}_+$
(See Fig.~\ref{fig-dg-neighbor},
page~\pageref{fig-dg-neighbor}).
Let $u_-$ and $u_+$ denotes the two corresponding traces of $u$ on $S$.
Then
\mbox{$\lambda=(u_-+u_+)/2$}, or equivalently
\mbox{$u_+=2\lambda-u_-$},
and thus the Godunov flux
$\Phi(\boldsymbol{n};u_-,u_+)$
can be expressed equivalently in terms of $\lambda$ and $u_-$.
Going back to an element $K\in\mathscr{T}_h$ with
$u_{ext}$ as the external trace of $u$ on $\partial K$,
we have \mbox{$u_{ext}=2\lambda-u$}
and \eqref{eq-hdg-hyp-integ-part} becomes:
\begin{eqnarray*}
\int_K
\left(
\sigma u v
-
\boldsymbol{f}(u).\nabla v
\right)
\,{\rm d}x
+
\int_{\partial K}
\Phi(\boldsymbol{n};\,u,\,2\lambda\!-\!u)
\, v
\,{\rm d}s
&=&
\int_K
f \, v
\,{\rm d}x
\end{eqnarray*}
Observe that, as $\lambda$ is now an independent variable,
the previous equation dependends only upon the value of $u$
inside the element $K$ and no more upon its external trace.
So, the problem has the desired block-diagonal structure at the
element level.
The discontinuous finite element space are defined by:
\begin{eqnarray*}
X_h &=& \{ \varphi_h \in L^2(\Omega); \ \varphi_{h|K}\in P_k, \ \forall K \in \mathscr{T}_h \} \\
M_h &=& \{ \mu \in L^2(\mathscr{S}_h); \ \mu_{h,S}\in P_k, \forall S \in \mathscr{S}_h \}
\end{eqnarray*}
where $k\geq 0$ is the polynomial degree
and $\mathscr{S}_h$ denotes the set of all sides in the mesh $\mathscr{T}_h$.
Remark that on any internal side,
a function $\varphi_h\in M_h$ is bi-valued while a
function $\mu_h\in M_h$ has only one value.
After summation of the previous equation over all $K\in\mathscr{T}_h$,
the discrete variational formulation of the problem writes:
\ \ $(FV)_h$: \emph{find $u_h\in X_h$ and $\lambda_h \in M_h$ such that}
\begin{eqnarray*}
\int_\Omega
\left(
\sigma u_h v_h
-
\boldsymbol{f}(u_h).\nabla_h v_h
\right)
\,{\rm d}x
+
\sum_{K\in\mathscr{T}_h}
\int_{\partial K}
\Phi(\boldsymbol{n};\,u_h,\,2\lambda_h\!-\!u_h)
\, v_h
\,{\rm d}s
&=&
\int_\Omega
f \, v_h
\,{\rm d}x
\\
\int_{\mathscr{S}_h^{(i)}}
(\average{u_h}-\lambda_h)
\,\mu_h
\,{\rm d}s
+
\int_{\partial\Omega}
(\boldsymbol{f}(\lambda_h).\boldsymbol{n} - \Phi(\boldsymbol{n};\,\lambda_h,g))
\,\mu_h
\,{\rm d}s
&=&
0
\end{eqnarray*}
for all $v_h\in X_h$ and $\mu_h \in M_h$.
The second equation expresses that,
on all internal sides, we impose $\lambda_h=\average{u_h}$
by using the Lagrange multiplier $\mu_h$.
This multiplier extends on the boundary for imposing
the boundary condition \eqref{eq-hdg-hyp-bc} for the
Lagrange multiplier $\lambda_h$.
\red{TODO: with this boundary condition treatment, $\lambda_h$ is undefined on the outflow boundary part.}
In the next section, we turn to the concrete example
of the linear scalar transport equation.
|
