A SIMPLE TUTORIAL ON DISCONTINUOUS GALERKIN

METHODS

Jennifer Proft
CERMICS, ENPC

J. Proft
CERMICS, ENPC
 Outline

• Definitions
• A simple example
• Issues
• Historical development
elliptic equations
hyperbolic equations
• Discontinuous vs. continuous Galerkin
• Further discussion: convergence results, a posteriori results,
current research

J. Proft
CERMICS, ENPC

2
 Definitions
• Discontinuous Galerkin methods are those finite element methods which
utilize element-wise discontinuous basis functions in the choice of
approximation spaces.
• Let P k define the space of polynomials of degree k on element Ej . A
typical finite element subspace (broken space) is defined as
Dk (Eh) ≡ {v : v|Ej ∈ P k (Ej ), ∀j = 1, .., Nh}

Ej

• Define the “broken” norm for positive integer m

 1/2
Nh
X
|||φ|||H m ≡  kφk2H m(Ej )
j=1

J. Proft
CERMICS, ENPC

3
 A Simple Example
Consider the model problem on bounded domain
−4u = f in Ω
u = uD on ∂ΩD
Ω
∇u · n = g · n on ∂ΩN
We assume that f ∈ L2(Ω) and uD ∈ H 3/2(∂ΩD )
and g ∈ H 1/2(∂ΩN ). Then, there exists a unique
solution u ∈ H 1(Ω).

Weak variational formuations use integration by

parts (Green’s equation)
Z Z Z
− 4u = ∇u · ∇v − ∇u · n v
R R ∂R

J. Proft
CERMICS, ENPC

4
 Example: continuous Galerkin

Weak formulation
Find u ∈ H01(Ω) such that for all v ∈ H01(Ω)
Z Z Z
∇u ∇v = fv+ g · nv E1 E2
Ω Ω ∂ΩN

δΩD δΩN
Discrete weak formulation
Find uh ∈ Wh ⊂ H01(Ω) such that for all v ∈ Wh
Z Z Z
∇uh ∇v = fv+ g · nv
Ω Ω ∂ΩN

Wh(Ω) ⊂ H01(Ω) ≡ {v ∈ H 1(Ω) : v = 0 on ∂Ω}

J. Proft
CERMICS, ENPC

5
 Example: discontinuous Galerkin

Weak formulation
Find u ∈ H 1(E1) ∪ H 1(E2) such that for all
v ∈ H 1(E1) ∪ H 1(E2) E1
Z Z
∇u ∇v − ∇u · n v e1 n
δΩD
E1 e1 Z Z
− ∇u · n v = fv
∂ΩD E1

Z Z
∇u ∇v − ∇u · (−n) v E2
E2 e1 Z Z n
= fv+ g · nv e1
E2 ∂ΩN
δΩN

J. Proft
CERMICS, ENPC

6
 Example: discontinuous Galerkin

Discrete weak formulation

Find uh ∈ D(Eh) such that for all v ∈ D(Eh)
Z Z
∇uh ∇v − ∇u−
h · nv
−
E1
E1 e1 Z Z
e1 n
− ∇uh · n v = fv δΩD
∂ΩD E1

Z Z
∇uh ∇v + ∇u+
h · n v +
E2 e1 Z Z
= fv+ g · nv E2
E2 ∂ΩN n
e1
δΩN
k
Dk (Eh) ≡ {v : v|Ej ∈ P (Ej ), ∀j = 1, .., Nh}

J. Proft
CERMICS, ENPC

7
 Example: discontinuous Galerkin

Discrete weak formulation

Find uh ∈ D(Eh) such that for all v ∈ D(Eh)
X 2 Z Z E1 E2
∇uh ∇v − [∇uh · n v]
j=1 Ej e1
Z 2 Z
X Z δΩD e1 n δΩN
− ∇u− −
h · nv = fv+ g · nv
∂ΩD j=1 Ej ∂ΩN

where the ’jump’ is defined as [w] = w − − w+ for

traces
v −(x) = lim γovk (x+sne), v +(x) = lim γovk (x+sne),
(s→0−
k→∞ ) (s→0+
k→∞ )

J. Proft
CERMICS, ENPC

8
 Some Issues

• How to incorporate dirichlet boundary conditions?

• What to do with jump on boundary term (i.e. choose fluxes)?
• Are these methods consistent? Stable? Convergent?
• How efficient are they to implement?
• How accurate are their solutions?
• How do these methods compare with other methods?
• Is there a unified framework for all DG methods?

J. Proft
CERMICS, ENPC

9
 Historical development

Elliptic Equations Hyperbolic Equations

• First introduced by Nitsche in • First introduced by Reed and Hill
1971, application to second-order in 1973, application to neutron
elliptic equation transport problem
• Also known as (interior) penalty • First a priori analysis was
methods, attributed to Wheeler in published by LeSaint and Raviart
1978 in 1974
• Vigorous development in past 5 • Vigorous development in past 15
years years (including Jaffré)

J. Nitsche, “Über ein Variationsprinzip zur Lösun von Dirichlet Problemen bei Verwendung von Teilöaumen, die keinen
Randbedingungen unterworfen sind”, Abh. Math. Sem. Univ. Hamburg, 36:9-15, 1971
I. Babuška, “The finite element method with interior penalty”, Math. Comp.,27:221-228, 1973
W.Reed and T.Hill, “Triangular mesh methods for the neutron transport equation”, Los Alamos Scientific Laboratory,
LA-UR-73-479, Los Alamos, NM, 1973
P. LeSaint and P. Raviart, “On a finite element method for solving the neutron transport equations”, in Mathematical Aspects
of Finite Elements in Partial Differential Equations, Academic Press, 1974

J. Proft
CERMICS, ENPC
10
 DG on elliptic: Nitsche’s method
Consider Z Z Z
−4u = f in Ω ∇u·∇v − ∇u·n v = fv
Ω ∂Ω Ω
u = 0 on ∂Ω

Since uZvanishes on the

Z boundary, Z Z Z
∇u · ∇v − ∇u · n v − ∇v · n u − ηu v = fv
Ω ∂Ω ∂Ω ∂Ω Ω

Find uh ∈ Wh ⊂ H 1(Ω) such that for all v ∈ Wh

Z Z Z Z Z
∇uh · ∇v − ∇uh · n v − ∇v · n uh − ηuh v = fv
Ω ∂Ω ∂Ω ∂Ω Ω

consistency term
symmetry term
boundary penalty term

J. Proft
CERMICS, ENPC

11
 Interior Penalties

Mary Fanett Wheeler

Graduated 1971 Rice University (Houston, TX)
Professor at University of Texas at Austin
Likes scottie dogs and George W. Bush

Idea attributed to private communication by J. Douglas and T. Dupont.

Summarized and analyzed by D. Arnold in 1979 thesis.
• Suppose approximation space is discontinuous, then
XM Z Xm Z Z M Z
X
∇u · ∇v − [∇u · ni v] − ∇u · n v = fv
j=1 Ej i=1 ei ∂Ω j=1 Ej

M. Wheeler, “An elliptic collocation-finite element method with interior penalties”, SIAM J. Numer. Anal, 15:152-161, 1978
D. Arnold, “An interior penalty finite element method with discontinuous elements”, SIAM JNA, 19(4):742-760, 1982
J. Douglas and T. Dupont, “Interior penalty procedures for elliptic and parabolic Galerkin methods”, Lecture Notes in
Physics, 58, Springer-Verlag, Berlin, 1976.

J. Proft
CERMICS, ENPC
12
 DG Formulation
• Trick: Define average w = 12 (w − + w+) and recall jump [w] = w − = w+
then
[a b] = [a]{b} + {a}[b]
• Apply to formulation
M Z
X Xm Z Xm Z
∇u · ∇v − [∇u · ni] {v} − {∇u · ni} [v]
j=1 Ej i=1 ei i=1 ei
Z M Z
X
− ∇u · n v = fv
∂Ω j=1 Ej

• However, since u ∈ H 2(Ω) and v ∈ H 1(Ω),

Xm Z
− [∇u · ni] {v} = 0,
i=1 ei
Xm Z
− {∇v · ni} [u] = 0
i=1 ei

J. Proft
CERMICS, ENPC

13
 Symmetric Interior Penalty DG

• Idea: add interior penalty term of the form

m Z
X σ
[u] [v] = 0
|e |
i=1 ei i

M Z
X m Z
X m Z
X
∇u · ∇v − {∇u · ni} [v] − {∇v · ni} [u]
j=1 Ej i=1 ei i=1 ei
m Z M Z
σ
X Z X
+ [u] [v] − ∇u · n v = fv
i=1 ei |ei | ∂Ω j=1 Ej

J. Proft
CERMICS, ENPC

14
 Discrete Symmetric Interior Penalty DG
• Can weakly incorporate dirichlet boundary data u = g on ∂Ω
Z Z
− (∇v · n) uh = − (∇v · n) g
∂Ω ∂Ω

Find uh ∈ D(Eh) such that for all v ∈ D(Eh)

XM Z Xm Z m Z
X
∇uh · ∇v − {∇uh · ni} [v] − {∇v · ni} [uh]
j=1 Ej i=1 ei i=1 ei
m Z
σ
X Z Z
+ [uh] [v] − (∇uh · n) v − (∇v · n) uh
i=1
|e
ei i | ∂Ω ∂Ω
M Z
X Z
= fv− ∇v · n g
j=1 Ej ∂Ω

J. Proft
CERMICS, ENPC

15
 Non-Symmetric DG Method

J. Tinsley Oden
Graduated from Oklahoma State University in 1962
Director of TICAM at University of Texas at Austin
Likes hunting big game and has a ranch in Texas

Ivo Babuška
Graduated from Technical University, Prague in 1951
Professor at University of Texas at Austin
Has 4 PhD degrees and an arrest warrant.
Likes studing the history of mathematicians.

Carlos Baumann
Graduated from University of Texas in 1997
Researcher at Computational Mechanics, Inc in Austin
Oden’s student at TICAM

J. Proft
CERMICS, ENPC

16
 Non-Symmetric Discontinuous Galerkin Method

Find uh ∈ D(Eh) such that for all v ∈ D(Eh)

XM Z Xm Z m Z
X
∇uh · ∇v − {∇uh · ni} [v] + {∇v · ni} [uh]
j=1 Ej i=1 ei i=1 ei
Z Z
− (∇uh · n) v + (∇v · n) uh
∂Ω ∂Ω
M Z
X Z
= fv+ ∇v · n g
j=1 Ej ∂Ω

• Form becomes positive-definite → inf-sup condition holds → unique

solution and conditions of stability
• Consistent, locally mass-conservative formulation

C. Baumann, “An h-p adaptive discontinuous finite element method for computational fluid dynamics”, PhD thesis, The
University of Texas at Austin, 1997.
J.T. Oden, I. Babuška, and C. Baumann, “A discontinuous hp finite element method for diffusion problems”, Journ. Comput.
Phys., 146:491-519, 1998.

J. Proft
CERMICS, ENPC 17
 Non-Symmetric Interior Penalty DG Method
Béatrice Rivière
Graduated 2000 from University of Texas at Austin
Assistant Professor at University of Pittsburg
From Reunion Island, likes to eat Dodo au Vin

Mary Wheeler
(see page 12)

Vivette Girault
Professor at Paris VI
Visited University of Texas at Austin
Likes math a lot!

J. Proft
CERMICS, ENPC

18
 Non-Symmetric Interior Penalty DG Method

Find uh ∈ D(Eh) such that for all v ∈ D(Eh)

XM Z Xm Z m Z
X
∇uh · ∇v − {∇uh · ni} [v] + {∇v · ni} [uh]
j=1 Ej i=1 ei i=1 ei
Z Z
− (∇uh · n) v + (∇v · n) uh
∂Ω ∂Ω
m
σ σ
X Z Z
+ [uh][v] + uh v
|e
i=1 ei i
| |e
∂Ω i |
M Z
σ
X Z Z
= fv+ gv+ ∇v · n g
j=1 Ej
|e
∂Ω i | ∂Ω

• No longer locally mass-conservative, proved convergence rates.

B. Rivière, “Discontinuous Galerkin methods for solving the miscible displacement problem in porous media”, PhD Thesis,
The University of Texas at Austin, 2002.
B. Rivière, M. Wheeler, and V. Girault, “Improved energy estimates for interior penalty, constrained and discontinuous
Galerkin methods for elliptic problems, Part I”, Comput. Geo, 8:337-360, 1999.
B. Rivière, M. Wheeler, and V. Girault, “A priori error estimates for finite element methods based on discontinuous
approximation spaces for elliptic problems”, SIAM JNA, 39(3):902-931, 2001.
19
J. Proft
 DG on hyperbolic: Reed and Hill
• Consider the (linear) neutron transport equation for real number η and
constant vector a
ηu + ∇ · (a u) = f in Ω
Multiply by test function, integrate by parts over element E
Z Z Z Z
ηu v − u a · ∇v − a · nu↑ = fv in Ω
E E ∂E E

where upwind value u↑ = lims↓0(x − sa) is the value of u upstream in the

characteristic direction a.
• LeSaint and Raviart immediately publish convergence results (1974), prove
(suboptimal) hk convergence in L2 for general triangulations.
• Further results, extensions and numerics are published.

B. Cockburn, E. Karniadakis, C.W. Shu, “The Development of discontinuous Galerkin methods”, In Discontinuous Galerkin
Methods, Lecture Notes in Computational Science and Engineering, Springer, Berlin, 2000

J. Proft
CERMICS, ENPC

20
 DG on nonlinear hyperbolic systems
Z d Z
X d Z
X
ut v − fi(u) · ∇v + fi(u) · n v = 0
E i=1 E i=1 ∂E
• Time discretization
Problem: nonlinearities in flux function on boundary prevents
element-by-element computation using characteristics
implicit time discretization very inefficient
space-time element construction
• Analysis
Problem: nonlinear analysis extremely difficult
only 3 results exist

Jerome Jaffré
Graduated from Paris VI in 1974
Researcher at INRIA Rocquencourt
Likes geosciences a lot!

J. Jaffré, C. Johnson, A. Szeppessy, “Convergence of the discontinuous Galerkin finite element method for hyperbolic
conservation laws”, Math. Models and Meth. in Appl. Sci., 5:367-386, 1995
21
 Development of LDG method
• Cockburn & Shu: RKDG methods
– discretize in space via DG, polynomials of degree k
– discretize in time via explicit TVD Runge-Kutta (k + 1)
– employ a general slope limiter to enforce maximum principle
• Bassi & Rebay: extension to Navier-Stokes
– convection-diffusion type equations
– rewrite as first order system
• Cockburn & Shu: LDG methods
– extension to convection-diffusion systems
– local solution (paralellizable)
F. Bassi and S. Rebay, “A high-order accurate discontinuous finite element method for the numerical solution of the
compressible Navier-Stokes equations”, J. Comput. Phys., 131:267-279, 1997
B. Cockburn and C.W. Shu, “The local discontinuous Galerkin method for time-dependent convection-diffusion
systems”, SIAM JNA, 35(6):2440-2463, 1998

J. Proft
CERMICS, ENPC

22
 LDG method on transport

Clint Dawson
Graduated 1988 from Rice University
Professor at University of Texas at Austin
Grew up on a farm in rural Texas

• Cockburn & Dawson

– extension to multi-dimension tranport equation
– diffusion tensor depends on (x,t) and allows for non-invertible diffusion
tensors

∂c
+ ∇ · (uc) − ∇ · (D∇c) = 0 (x, t) ∈ Ω, t > 0,
∂t
(uc − D∇c) · n = (u · n)ĝ on ∂Ωi
(−D∇c) · n = 0 on ∂Ωo
B. Cockburn and C. Dawson, “Some extensions of the local discontinuous Galerkin method for convection-diffusion equations
in multidimensions”,

J. Proft
CERMICS, ENPC

23
 LDG method on convection-diffusion

∂c
+ ∇ · (uc − D∇c) = 0 (x, t) ∈ Ω, t > 0,
∂t
(uc − D∇c) · n = (u · n)ĝ on ∂Ωi
(−D∇c) · n = 0 on ∂Ωo

Rewrite as first order system (let D=constant)

∂c
+ ∇ · (uc − ∇ · (D∇c)) = 0
∂t
z = −D∇c
(uc − z) · n = (u · n)ĝ on ∂Ωi
z·n = 0 on ∂Ωo

J. Proft
CERMICS, ENPC

24
 LDG method
Apply DG to each equation separately
∂c
Z Z Z Z
v − (uc + z) · ∇v + (uc + z) · n v = fv
E ∂t E ∂E E

Z Z Z
zw − Dc ∇ · w + Dc w · n = 0
E E ∂E

Z Z
(uc − z) · n v = (u · n)ĝ v
∂E∩∂Ω
Zi ∂E∩∂Ωi

uc · n v = 0
∂E∩∂Ωo

J. Proft
CERMICS, ENPC
 Discrete LDG method

M Z M m Z
∂ch
X X Z X
− v− (uch + zh) · ∇v + (uc↑h + {zh}) · n [v]
j=1 Ej ∂t j=1 Ej i=1 ei
Z Z M Z
X
+ (uch · n) v + ĝu · n v = fv
∂Ωo∩∂Ω ∂Ωi∩∂Ω j=1 Ej

M Z
X M Z
X m Z
X
zh · w h + Dch ∇ · w h − {Dch} [w h] · ni = 0
j=1 Ej j=1 Ej i=1 ei

J. Proft
CERMICS, ENPC

26
 Features

DISCONTINUOUS GALERKIN FEM

• easily incorporates h- and p-adaptivity
• locally conservative and accurate
• easily handles complex geometry and nonmatching grids
• suppresses non-physical oscillations
• handles advection dominated flow regimes efficiently
• degrees of freedom associated with elements

CONTINUOUS GALERKIN FEM

• handles diffusion dominated flow regimes efficiently
• degrees of freedom usually associated with nodes

J. Proft
CERMICS, ENPC

27