0493 Daveiga

An introduction to the Virtual Element Method
L. Beirão da Veiga
Department of Mathematics
University of Milan
in collaboration with:
F. Brezzi, A. Cangiani, D. Marini, G. Manzini, A. Russo
DURHAM, July 2014
L. Beirão da Veiga (Univ. of Milan) The Virtual Element Method DURHAM - 2014 1 / 35
The Virtual Element Method
The Virtual Element Method (VEM) is a generalization of the Finite
Element Method that takes inspiration from modern Mimetic Finite
Difference schemes.
VEM allow to use very general polygonal and polyhedral meshes,

also for high polynomial degrees and guaranteeing the patch test.
The flexibility of VEM is not limited to the mesh: an example will
be shown later.
Why polygons/polyhedrons?
The interest (and use in commercial codes? ) for polygons/polyhedra is

recently growing.
Immediate combination of tets and hexahedrons

Easier/better meshing of domain (and data) features
Automatic inclusion of “hanging nodes”
Adaptivity: more efficient mesh refinement/coarsening
Generate meshes with more local rotational simmetries
Robustness to distortion
.......
? for example CD-ADAPCO and ANSYS.
Some polytopal methods
Mimetic F.D. Shashkov, Lipnikov, Brezzi, Manzini, BdV, ....
HMM: Eymard, Droniou, ...
Polygonal FEM: Sukumar, Paulino, ...
Weak Galerkin FEM: Wang, ....
HHO: Ern, di Pietro
Polygonal DG: Cangiani, Houston, Georgoulis, ...
VEM: this talk !!!
..........
The model problem
We consider the Poisson problem in two dimensions
−∆u = f in Ω, u=0 on ∂Ω,
where
Ω ⊂ R2 is a polygonal domain;
the loading f is assumed in L2 (Ω).
Variational formulation:
 1
 find u ∈ V := H0 (Ω) such that
Z
 a(u, v ) = f v dx ∀v ∈ V ,
Ω
where Z
a(v , w) = ∇v · ∇w dx , ∀v , w ∈ V .
Ω
A Virtual Element Method
We will build a discrete problem in following form

(
find uh ∈ Vh such that
ah (uh , vh ) =< fh , vh > ∀ vh ∈ Vh ,
where
Vh ⊂ V is a finite dimensional space;
ah (·, ·) : Vh × Vh → R is a discrete bilinear form approximating the
continuous form a(·, ·);
< fh , vh > is a right hand side term approximating the load
A Virtual Element Method
We will build a discrete problem in following form

(
find uh ∈ Vh such that
ah (uh , vh ) =< fh , vh > ∀ vh ∈ Vh ,
where
Vh ⊂ V is a finite dimensional space;
ah (·, ·) : Vh × Vh → R is a discrete bilinear form approximating the
continuous form a(·, ·);
< fh , vh > is a right hand side term approximating the load
Let m ≥ 1 be a fixed integer index. Such index will represent the

degree of accuracy of the method.
The local spaces Vh|E
Let Ωh be a simple polygonal mesh on Ω. This can be any
decomposition of Ω in non overlapping polygons E with straight faces.
The space Vh will be defined element-wise, by introducing
local spaces Vh|E ;
the associated local degrees of freedom.
local spaces Vh|E ;
For all E ∈ Ωh :

Vh|E = v ∈ H 1 (E) : −∆v ∈ Pm−2 (E),

v |e ∈ Pm (e) ∀e ∈ ∂E .
local spaces Vh|E ;
For all E ∈ Ωh :

Vh|E = v ∈ H 1 (E) : −∆v ∈ Pm−2 (E),

v |e ∈ Pm (e) ∀e ∈ ∂E .
the functions v ∈ Vh|E are continuous (and known) on ∂E;

the functions v ∈ Vh|E are unknown inside the element E!
it holds Pm (E) ⊆ Vh|E
Degrees of freedom for Vh|E
The dimension of the space Vh|E is clearly
dim(Vh|E ) = Ne m + m(m − 1)/2,
with Ne number of edges of E.
The dimension of the space Vh|E is clearly
dim(Vh|E ) = Ne m + m(m − 1)/2,
with Ne number of edges of E.
pointwise values vh (ν) at all corners ν of E;

(m − 1) pointwise values on each edge:
vh (xie ) , {xie }m−1

i=1 distinct points on edge e ;
volume moments:
Z
vh · pm−2 ∀pm−2 ∈ Pm−2 (E) .
E
Depiction of the degrees of freedom for Vh|E
m=1 m=2 m=3
Green dots stand for vertex pointwise values

Red squares represent edge pointwise values
Blue squares represent internal (volume) moments
The following holds

[BdV,Brezzi,Cangiani,Manzini,Marini,Russo, M3AS 2013].
Proposition
The proposed collection of operators Vh|E → R constitutes a set of
degrees of freedom for Vh|E , ∀E ∈ Vh|E .
We already know #dofs = dim(Vh|E ).

If vh ∈ Vh|E is null on all the d.o.f.s, then it clearly vanishes on the
boundary.
The function ∆vh ∈ Pm−2 and thus
Z Z
0= vh · ∆vh = (∇vh ) · (∇vh ).
E E
The space Vh
The global space Vh is built by assembling the local spaces Vh|E as

usual:
Vh = {v ∈ H01 (Ω) : v |E ∈ Vh|E ∀E ∈ Ωh }
The total d.o.f.s are one per internal vertex, m − 1 per internal edge
and m(m − 1)/2 per element.
The choice of degrees of freedom guarantees the global continuity of

the functions in Vh .
The bilinear form ah (·, ·)
The bilinear form ah (·, ·) is built element by element

X
ah (vh , wh ) = ahE (vh , wh ) ∀ vh , wh ∈ Vh ,
E∈Ωh
where
ahE (·, ·) : Vh|E × Vh|E −→ R
are symmetric bilinear forms that mimic
ahE (·, ·) ' a(·, ·)|E
by satisfying a stability and a consistency condition.
The stability property
Stability
There exist two positive constants α∗ and α∗ , independent of h and of
E, such that
α∗ aE (vh , vh ) ≤ ahE (vh , vh ) ≤ α∗ aE (vh , vh ) ∀vh ∈ Vh|E .
The stability property guarantees that ah (·, ·) is uniformly coercive

and continuous;
clearly, it is sufficient for the (uniform) well posedness of the

discrete problem, but not for convergence.
The consistency property
Consistency
For all h and for all E ∈ Ωh it holds
ahE (p, vh ) = aE (p, vh ) ∀p ∈ Pm (E), vh ∈ Vh|E .
The consistency property
Consistency
For all h and for all E ∈ Ωh it holds
ahE (p, vh ) = aE (p, vh ) ∀p ∈ Pm (E), vh ∈ Vh|E .
NOTE: an integration by parts gives

Z
aE (p, vh ) = ∇p · ∇vh dx
EZ Z
= − (∆p) vh dx + (∇p · nE ) vh ds.
E ∂E
for all p ∈ Pm (E), vh ∈ Vh|E .

Therefore the right hand side above is explicitly computable even if we
ignore vh inside E.
The discrete load term
We consider m ≥ 2 first. Let, for all E ∈ Ωh , the approximated load fh |E

be the L2 -projection of f |E on Pm−2 (E) .
Then
X Z
(fh , vh )h := fh vh dx
E∈Ωh E
that is computable due to the internal dofs of Vh|E .
The discrete load term
We consider m ≥ 2 first. Let, for all E ∈ Ωh , the approximated load fh |E

be the L2 -projection of f |E on Pm−2 (E) .
Then
X Z
(fh , vh )h := fh vh dx
E∈Ωh E
that is computable due to the internal dofs of Vh|E .

In the case m = 1 a simple integration rule based on the vertex values
of the polygon can be used, for instance
X Z 1 X
(fh , vh )h := f dx vh (ν) dx.
E NE
E∈Ωh ν∈∂E
Note: for m = 2 better choices can be made [Ahmad, Alsaedi, Brezzi,

Marini, Russo, CMA 2013], [BdV, Brezzi, Marini, SINUM 2013].
A convergence result
Let the sequence {Ωh }h satisfy the following mesh assumptions:

each element E in Ωh is star-shaped with respect to a ball of
uniform radius or suitable union of ;
for each element E in Ωh , the lenght of all edges is comparable

with its diameter hE not needed by paying |log(he )| .
Then the following holds [... volley team ..., M3AS 2013].
Theorem
Let the stability and consistency assumptions hold. Then, if f ∈ H s (Ωh )
and u ∈ H s+1 (Ωh ), we have

|u − uh |H 1 (Ω) ≤ C hs |u|H s+1 (Ωh ) + |f |H s (Ωh )
for 0 ≤ s ≤ m and with C independent of h.
k-plain
−div(K∇u) + αu = f
exact solution:
� �
ue (x, y) = y − x + log y 3 + x + 1 − x y − x y 2 + x2 y + x2 + x3 + sin(5 x) sin(7 y) − 1
diffusion:
K(x, y) = 1
zero-order term:
α(x, y) = 1
right-hand-side:
� � 1 9 y4 6y
f (x, y) = log y 3 + x + 1 − y − 5 x + + − x y − x y 2 + x2 y + x2 + x3 − 3 + 75 sin(5 x) sin(7 y) − 3
(y 3 + x + 1)2 (y 3 + x + 1)2 y +x+1
L2 norm of the exact solution:
||ue ||0,Ω = 0.6431084584
H1 seminorm of the exact solution:
|ue |1,Ω = 5.031264492
Generated by genue on 09-Nov-2013 16:08:59

h=hmean slope: ||Π0u −u || = 4.011, ||Π∇u −u || = 4.014 h=hmean slope: ||∇Π0u −∇u || = 3.010, ||∇Π∇u −∇u || = 3.013
h e 0,Ω h e 0,Ω h e 0,Ω h e 0,Ω
−2 0
10 10
0 0
||Π u −u || ||∇Π u −∇u ||
h e 0,Ω h e 0,Ω
5 ∇ 5 ∇
h ||Π u −u || h ||∇Π u −∇u ||
h e 0,Ω h e 0,Ω
−3 −1
10 10
4 4
h h
−4 −2
10 10
h
3
h3
−5 −3
10 10
h2 h2
−6 −4
10 10
h1 h1
−7 −5
10 10
−8 −6
10 10
−2 −1 0 −2 −1 0
10 10 10 10 10 10
2
L error k=3 H 1 error
triangle0.01 triangle0.005 triangle0.001 triangle0.0005 triangle0.0001

88 polygons, hmax = 0.174116, hmean = 0.149291 177 polygons, hmax = 0.133616, hmean = 0.107452 809 polygons, hmax = 0.063418, hmean = 0.049083 1627 polygons, hmax = 0.044872, hmean = 0.034516 7921 polygons, hmax = 0.019754, hmean = 0.015641
h=hmean
0
slope: ||Π u −u ||
∇
= 5.009, ||Π u −u || = 5.017 h=hmean slope: ||∇Π0u −∇u || = 4.003, ||∇Π∇u −∇u || = 4.013
h e 0,Ω h e 0,Ω h e 0,Ω h e 0,Ω
−4 −2
10 10
0 0
||Π uh−ue||0,Ω ||∇Π u −∇u ||
h e 0,Ω
∇ 5 ∇
||Π uh−ue||0,Ω h ||∇Π u −∇u ||
h e 0,Ω
h5
−5 −3
10 10
4
h
h4
−6 −4
10 10
h3
h3
−7 −5
10 10
h2
h2
−8 −6
10 10
h1
h1
−9 −7
10 10
−10 −8
10 10
−2 −1 0 −2 −1 0
10 10 10 10 10 10
2

0
slope: ||Π0u −u ||
∇
h=hmean = 5.928, ||Π∇u −u || = 5.871 h=hmean slope: ||∇Π u −∇u ||
h e 0,Ω
= 5.020, ||∇Π u −∇u ||
h e 0,Ω
= 5.010
h e 0,Ω h e 0,Ω
−5 −4
10 10
0 0
||Π u −u || ||∇Π uh−∇ue||0,Ω
h e 0,Ω
∇ ∇
||Π u −u || 5 ||∇Π uh−∇ue||0,Ω
h e 0,Ω h
−6
10 −5
10
−7 h5 h4
10
−6
10
4
−8 h
10 h3
−7
10
3
−9 h
10
h2
−8
2 10
−10 h
10
h1
−11
h1 10
−9
10
−12 −10
10 10
−2 −1 0 −2 −1 0
10 10 10 10 10 10
2

Just for fun ...
We consider a family of meshes based on the following pattern:
Courtesy of A. Russo !!
−∆u = f
h=hmean slope: ||Π0uh−ue||0,Ω = 3.989 h=hmean slope: ||∇Π0uh−∇ue||0,Ω = 3.009
−1 1
10 10
||Π0uh−ue||0,Ω ||∇Π0uh−∇ue||0,Ω
4
h
h4
0
10
−2
10
3
h h3
−1
10
−3 2
10 h
h2
−2
10
h1
−4
10 h1
−3
10
−5 −4
10 10
−2 −1 0 −2 −1 0
10 10 10 10 10 10
L2 error k=3 H 1 error

VEM−2x2 VEM−4x4 VEM−8x8 VEM−16x16
28 polygons, Ndof=627, h = 5.21e−01, h = 4.46e−01 112 polygons, Ndof=2445, h = 2.60e−01, h = 2.23e−01 448 polygons, Ndof=9657, h = 1.30e−01, h = 1.11e−01 1792 polygons, Ndof=38385, h = 6.51e−02, h = 5.57e−02
max mean max mean max mean max mean
Generated by ploterrors on 29-Jun-2014 18:40:30

−∆u = f
h=hmean slope: ||Π0uh−ue||0,Ω = 4.991 h=hmean slope: ||∇Π0uh−∇ue||0,Ω = 3.982
−1 1
10 10
||Π0uh−ue||0,Ω ||∇Π0uh−∇ue||0,Ω
h5
h5
−2 0
10 10
4
h
h4
−3 −1
10 10
3
h
h3
−4 −2
10 10
2
h
h2
−5 −3
10 10
1
h
h1
−6 −4
10 10
−7 −5
10 10
−2 −1 0 −2 −1 0
10 10 10 10 10 10
L2 error k=4 H 1 error

VEM−2x2 VEM−4x4 VEM−8x8 VEM−16x16
28 polygons, Ndof=901, h = 5.21e−01, h = 4.46e−01 112 polygons, Ndof=3521, h = 2.60e−01, h = 2.23e−01 448 polygons, Ndof=13921, h = 1.30e−01, h = 1.11e−01 1792 polygons, Ndof=55361, h = 6.51e−02, h = 5.57e−02
max mean max mean max mean max mean
Generated by ploterrors on 29-Jun-2014 18:50:05

Construction of the stiffness matrix
We introduce the following energy projector
Π : Vh|E −→ Pm (E)
defined by, for all vh ∈ Vh|E ,

(
aE (Πvh , p) = aE (vh , p) ∀p ∈ Pm (E)/R
P0 (Πvh ) = P0 (vh )
The operator P0 is a projection on constants, that for m ≥ 2 is simply

the average, and for m = 1 the vertex value average.
NOTE: due to the consistency assumption, the projection operator

above is computable (more later).
It is immediate to check that ∀ vh , wh ∈ Vh|E
aE (vh , wh ) = aE (Πvh , Πwh ) + aE ((I − Π)vh , (I − Π)wh ).
Then the bilinear form
ahE (vh , wh ) = aE (Πvh , Πwh ) + sE ((I − Π)vh , (I − Π)wh )
is consistent and stable, provided the positive bilinear form

sE : Vh|E × Vh|E → R scales like the original bilinear form a.
It is immediate to check that ∀ vh , wh ∈ Vh|E
aE (vh , wh ) = aE (Πvh , Πwh ) + aE ((I − Π)vh , (I − Π)wh ).
Then the bilinear form
ahE (vh , wh ) = aE (Πvh , Πwh ) + sE ((I − Π)vh , (I − Π)wh )
is consistent and stable, provided the positive bilinear form

sE : Vh|E × Vh|E → R scales like the original bilinear form a.
Let now the local stiffness matrix ME
MEij := ahE (φi , φj ) ∀i, j = 1, 2, ..., N
with {φi } the canonical basis associated to the degrees of freedom.
We introduce also {mα }nα=1 a basis for the polynomial space
Pm = span{mα }nα=1 .
Since Pm ⊆ Vh|E , the matrix

 
dof1 (m1 ) dof1 (m2 ) ... dof1 (mn )
 dof2 (m1 ) dof2 (m2 ) ... dof2 (mn ) 
 
D= .. .. .. .. .
 . . . . 
dofN (m1 ) dofN (m2 ) . . . dofN (mn )
expresses the {mα } in terms of the Vh|E basis.
We have also a matrix

 
P0 φ1 ... P 0 φN
aE (φ1 , m2 ) . . . E
a (φN , m2 ) 
 
B= .. .. .. 
 . . . 
E
a (φ1 , mn ) . . . aE (φN , mn )
expressing (in terms of the bases) the right hand side in the definition
(
aE (Πvh , p) = aE (vh , p) ∀p ∈ Pm (E)/R
P0 (Πvh ) = P0 (vh )
NOTE: such matrix is computable! (integration by parts and d.o.f.s

definition)
Let
 
P0 m1 P0 m2 ... P0 mn
 0 ∇m2 , ∇m2 0,P . . . ∇m2 , ∇mn 0,P 
 
G = BD =  . .. .. .. .
 .. . . . 

0 ∇mn , ∇m2 0,P . . . ∇mn , ∇mn 0,P
Let
 
P0 m1 P0 m2 ... P0 mn
 0 ∇m2 , ∇m2 0,P . . . ∇m2 , ∇mn 0,P 
 
G = BD =  . .. .. .. .
 .. . . . 

0 ∇mn , ∇m2 0,P . . . ∇mn , ∇mn 0,P
Compute the matrices corresponding to the projection operator:
Π? = (G)−1 B , Π = DΠ? .
Let
 
P0 m1 P0 m2 ... P0 mn
 0 ∇m2 , ∇m2 0,P . . . ∇m2 , ∇mn 0,P 
 
G = BD =  . .. .. .. .
 .. . . . 

0 ∇mn , ∇m2 0,P . . . ∇mn , ∇mn 0,P
Compute the matrices corresponding to the projection operator:
Π? = (G)−1 B , Π = DΠ? .
Finally compute the local stiffness matrix

e Π? + (I − Π)T (I − Π).
ME = ΠT? G
The three dimensional case
In principle the 3D case is analogous; the degrees of freedom are

one point value per vertex
(m − 1) point values per edge
M0f , M1f , .., Mm−1
f moments per face
M0E , M1E , .., Mm−2
E moments per element.
The three dimensional case
In principle the 3D case is analogous; the degrees of freedom are

one point value per vertex
(m − 1) point values per edge
M0f , M1f , .., Mm−1
f moments per face
M0E , M1E , .., Mm−2
E moments per element.
A more efficient formulation with less degrees of freedom per face

M0f , M1f , .., Mm−2
f moments per face
can be built using the L2 projector in [Ahmed, Alsaedi, Brezzi, Marini,
Russo, CMA 2013].
Virtual Elements for Hdiv : introduction
Consider the diffusion problem in mixed form



 Find F ∈ V := Hdiv (Ω), p ∈ Q := L2 (Ω) :

 Z Z


F · G + (divG)p = 0 ∀G ∈ V ,

 Z Ω Ω Z



 (divF )q = − f q ∀q ∈ Q.
Ω Ω
We introduce the VEM spaces:
Qh = {q ∈ L2 (Ω) : q|E ∈ Pk −1 (E) ∀E ∈ Ωh } ⊂ Q,

Vh = {G ∈ Hdiv (Ω) : G|E ∈ Vh|E ∀E ∈ Ωh } ⊂ V .
What follows taken from [BdV, Brezzi, Marini, Russo, submitted].

(se also [Brezzi, Falk, Marini, M2AN, 2014])
Virtual Elements for Hdiv : local spaces
Let E ∈ Ωh . We introduce the local VEM space

n
Vh|E = G ∈ Hdiv (E) ∩ Hrot (E) : divG ∈ Pk −1 (E), rotG ∈ Pk −1 (E),
o
G|e · neE ∈ Pk (e) ∀e ∈ ∂E .
Virtual Elements for Hdiv : local spaces
Let E ∈ Ωh . We introduce the local VEM space

n
Vh|E = G ∈ Hdiv (E) ∩ Hrot (E) : divG ∈ Pk −1 (E), rotG ∈ Pk −1 (E),
o
G|e · neE ∈ Pk (e) ∀e ∈ ∂E .
This space is associated to the problem

(
divG = f1 , rotG = f2 on E,
G · n = f∂ on ∂E .
R R
that is well posed if E f1 = ∂E f∂ .
Thus
dim(Vh|E ) = 2 dim(Pk −1 (E)) + Ne dim(Pk (e)) − 1 .
Virtual Elements for Hdiv : local degrees of freedom
edge moments
Z
(G · neE )pk ∀pk ∈ Pk (e), ∀e ∈ ∂E .
e
div volume moments:

Z
(divG)pk −1 ∀pk −1 ∈ Pk −1 (E)/R .
E
edge moments
Z
(G · neE )pk ∀pk ∈ Pk (e), ∀e ∈ ∂E .
e
div volume moments:

Z
(divG)pk −1 ∀pk −1 ∈ Pk −1 (E)/R .
E
additional volume moments:

Z
G · pk ∀pk ∈ Gk⊥ (E) .
E
The space
Z

Gk⊥ = p ∈ Pk : p · ∇q = 0 ∀q ∈ Pk +1 (E) .
E
has dimension equal to Pk −1 (E).

Unisolvence. Let G ∈ Vh|E , null on all dofs. Since
• rot : Gk⊥ → Pk −1 is a bijection,

• rotG ∈ Pk −1 ,
it exists ϕ ∈ Gk⊥ such that
0 = rot(G − ϕ) =⇒ G = ∇ψ + ϕ with ψ ∈ H 1 (E).
Unisolvence. Let G ∈ Vh|E , null on all dofs. Since
• rot : Gk⊥ → Pk −1 is a bijection,

• rotG ∈ Pk −1 ,
it exists ϕ ∈ Gk⊥ such that
0 = rot(G − ϕ) =⇒ G = ∇ψ + ϕ with ψ ∈ H 1 (E).
Thus
Z
||G||2L2 (E) = G (∇ψ + ϕ)
EZ Z Z
=− (divG)ψ + (G · nE )ψ + G ϕ = 0.
E ∂E E
Virtual Elements for Hdiv : computing the L2 projection
the first set of dofs determines G · n on ∂E;

since divG ∈ Pk −1 (E), the second set of dofs determines divG.
therefore we can compute
Z Z Z
G ∇ψ = − (divG)ψ + (G · nE )ψ ∀ψ polynomial;
E E ∂E
Virtual Elements for Hdiv : computing the L2 projection
the first set of dofs determines G · n on ∂E;

since divG ∈ Pk −1 (E), the second set of dofs determines divG.
therefore we can compute
Z Z Z
G ∇ψ = − (divG)ψ + (G · nE )ψ ∀ψ polynomial;
E E ∂E
any q ∈ [Pk (E)]2 can be written as
q = p + ∇ψ , p ∈ Gk⊥ , ψ ∈ Pk +1 (E).
Thus we can compute

Z Z Z
G·q= G·p+ G · ∇ψ.
E E E
Hdiv Virtual Elements : final observations
The proposed Virtual spaces (Vh , Qh ) satisfy a commuting diagram
property.
Thus are suitable for the approximation of the problem:


 Find Fh ∈ Vh , p ∈ Qh :

 Z Z


Fh · Gh + (divGh )ph = 0 ∀Gh ∈ Vh ,
 ZΩ ΩZ




 (divFh )qh = f qh ∀qh ∈ Qh .
Ω Ω
NOTE: with the choices that we made, everything above is computable

(up to the usual VEM construction and using the local L2 projections).
Hdiv Virtual Elements : final observations
The proposed Virtual spaces (Vh , Qh ) satisfy a commuting diagram
property.
Thus are suitable for the approximation of the problem:


 Find Fh ∈ Vh , p ∈ Qh :

 Z Z


Fh · Gh + (divGh )ph = 0 ∀Gh ∈ Vh ,
 ZΩ ΩZ




 (divFh )qh = f qh ∀qh ∈ Qh .
Ω Ω
NOTE: with the choices that we made, everything above is computable

(up to the usual VEM construction and using the local L2 projections).
VEM exact sequences

A full “Safari” of VEM to appear in [BdV, Brezzi, Marini, Russo].
An application: the Cahn-Hilliard equation
With standard finite elements it is very complicated to build

spaces with global regularity higher than C 0 .
With VEM, this is instead easy to achieve. We can build elements

with arbitrary C k regularity:
[Brezzi and Marini, CMAME, 2013]: C 1 VEM for Kirchhoff plates
[BdV, Manzini, IMA J. Num. An. 2013]: C k VEM for diffusion.
We will here show some “spoiler” from a paper in collaboration

with Antonietti, Scacchi, Verani for applications to the
Cahn-Hilliard equation for phase transition.
The Cahn-Hilliard equation
We search for u : Ω × [0, T ] → R such that:

 0 2

 ∂t u − ∆ φ (u) − γ ∆u(t) = 0
 in Ω × [0, T ]
u(·, 0) = u0 (·) in Ω


∂n u = ∂n φ0 (u) − γ 2 ∆u(t) = 0 on ∂Ω × [0, T ],
where the function φ(x) = (1 − x 2 )2 /4 and γ ∈ R+ “small”.
The Cahn-Hilliard equation
We search for u : Ω × [0, T ] → R such that:

 0 2

 ∂t u − ∆ φ (u) − γ ∆u(t) = 0
 in Ω × [0, T ]
u(·, 0) = u0 (·) in Ω


∂n u = ∂n φ0 (u) − γ 2 ∆u(t) = 0 on ∂Ω × [0, T ],
where the function φ(x) = (1 − x 2 )2 /4 and γ ∈ R+ “small”.
The natural variational space is H 2 (Ω), thus a C 1 regularity is needed

for a conforming method.
• Mixed DG (Kay, Styles, Suli)
• Morley element (Elliott)
• Isogeometric Analysis (Hughes and co-workers)
• ....
C 1 VEM elements (of minimal degree)

Vh|E = v ∈ H 2 (E) : −∆2 v ∈ P0 (E),

v|e ∈ P3 (e), ∂n v|e ∈ P1 (e) ∀e ∈ ∂E .
Degrees of freedom:
k=1, p=2
Some numerical result
We apply a primal VEM C 1 discretization to the problem:

it involves Π0 , Π∇ and Π∆ projections;
it grants a conforming solution and accepts general polygons;
theoretical convergence estimates hold;
initial numerical tests are encouraging.
t=0 t = 0.03 t = 0.1
0.8 0.8
0.8
0.6 0.6
0.6
0.4 0.4
0.4
0.2 0.2
0.2
0 0
0
−0.2 −0.2
−0.2
−0.4 −0.4
−0.4
−0.6 −0.6
−0.6
−0.8 −0.8
−0.8
More VEM literature
More VEM literature not mentioned in the previous slides:
Virtual Elements for linear elasticity problems
[BdV, Brezzi, and Marini, SINUM, 2013]
A stream function formulation for Stokes
[Antonietti, BdV, Mora, Verani, SINUM, 2014]
Three dimensional compressible elasticity
[A.L. Gain, C. Talischi, G.H. Paulino, CMAME, 2014]
IN PROGRESS: nonconforming elements (Ayuso, Lipnikov,
Manzini), eigenvalue problems (Mora, Rodriguez), discrete
fracture network (Berrone at al.), contact problems (Wriggers et
al.), topology optimization (Paulino et al.), etc..
More VEM literature
More VEM literature not mentioned in the previous slides:
Virtual Elements for linear elasticity problems
[BdV, Brezzi, and Marini, SINUM, 2013]
A stream function formulation for Stokes
[Antonietti, BdV, Mora, Verani, SINUM, 2014]
Three dimensional compressible elasticity
[A.L. Gain, C. Talischi, G.H. Paulino, CMAME, 2014]
IN PROGRESS: nonconforming elements (Ayuso, Lipnikov,
Manzini), eigenvalue problems (Mora, Rodriguez), discrete
fracture network (Berrone at al.), contact problems (Wriggers et
al.), topology optimization (Paulino et al.), etc..
Regarding VEM implementation:

On M3AS: The hitchhikers guide to VEM, a paper all about VEM
implementation.
Conclusions
The Virtual Element Method is a generalization of FEM that takes

inspiration from modern mimetic schemes
The freedom that is left to the local spaces allows for a large
flexibility, for instance in terms of meshes (polygons, “hanging
nodes”), global regularity of the discrete space, definition of the
local matrixes (M-optimization), etc ..
A lot of development is still to be done in VEM, and we believe it

can be a very interesting new field of research.
Morever, more complex coding and problems need to be

challenged in order to assess the impact of VEM in applications
(various nonlinear problems already under development ...).

0493 Daveiga

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

0493 Daveiga

Uploaded by

Copyright:

Available Formats

An introduction to the Virtual Element Method

DURHAM, July 2014

VEM allow to use very general polygonal and polyhedral meshes,

The interest (and use in commercial codes? ) for polygons/polyhedra is

Immediate combination of tets and hexahedrons

? for example CD-ADAPCO and ANSYS.

Mimetic F.D. Shashkov, Lipnikov, Brezzi, Manzini, BdV, ....

HMM: Eymard, Droniou, ...

Polygonal FEM: Sukumar, Paulino, ...

Weak Galerkin FEM: Wang, ....

HHO: Ern, di Pietro

Polygonal DG: Cangiani, Houston, Georgoulis, ...

VEM: this talk !!!

−∆u = f in Ω, u=0 on ∂Ω,

We will build a discrete problem in following form

We will build a discrete problem in following form

Let m ≥ 1 be a fixed integer index. Such index will represent the

the functions v ∈ Vh|E are continuous (and known) on ∂E;

dim(Vh|E ) = Ne m + m(m − 1)/2,

with Ne number of edges of E.

dim(Vh|E ) = Ne m + m(m − 1)/2,

with Ne number of edges of E.

pointwise values vh (ν) at all corners ν of E;

vh (xie ) , {xie }m−1

m=1 m=2 m=3

Green dots stand for vertex pointwise values

The following holds

We already know #dofs = dim(Vh|E ).

The global space Vh is built by assembling the local spaces Vh|E as

The choice of degrees of freedom guarantees the global continuity of

The bilinear form ah (·, ·) is built element by element

ahE (·, ·) ' a(·, ·)|E

by satisfying a stability and a consistency condition.

α∗ aE (vh , vh ) ≤ ahE (vh , vh ) ≤ α∗ aE (vh , vh ) ∀vh ∈ Vh|E .

The stability property guarantees that ah (·, ·) is uniformly coercive

clearly, it is sufficient for the (uniform) well posedness of the

ahE (p, vh ) = aE (p, vh ) ∀p ∈ Pm (E), vh ∈ Vh|E .

ahE (p, vh ) = aE (p, vh ) ∀p ∈ Pm (E), vh ∈ Vh|E .

NOTE: an integration by parts gives

for all p ∈ Pm (E), vh ∈ Vh|E .

We consider m ≥ 2 first. Let, for all E ∈ Ωh , the approximated load fh |E

that is computable due to the internal dofs of Vh|E .

We consider m ≥ 2 first. Let, for all E ∈ Ωh , the approximated load fh |E

that is computable due to the internal dofs of Vh|E .

Note: for m = 2 better choices can be made [Ahmad, Alsaedi, Brezzi,

Let the sequence {Ωh }h satisfy the following mesh assumptions:

for 0 ≤ s ≤ m and with C independent of h.

L2 norm of the exact solution:

||ue ||0,Ω = 0.6431084584

H1 seminorm of the exact solution:

|ue |1,Ω = 5.031264492

Generated by genue on 09-Nov-2013 16:08:59

triangle0.01 triangle0.005 triangle0.001 triangle0.0005 triangle0.0001

triangle0.01 triangle0.005 triangle0.001 triangle0.0005 triangle0.0001

triangle0.01 triangle0.005 triangle0.001 triangle0.0005 triangle0.0001

L2 error k=3 H 1 error

Generated by ploterrors on 29-Jun-2014 18:40:30

L2 error k=4 H 1 error

Generated by ploterrors on 29-Jun-2014 18:50:05

We introduce the following energy projector

defined by, for all vh ∈ Vh|E ,

The operator P0 is a projection on constants, that for m ≥ 2 is simply