Motivation

The Biharmonic Problem

Discrete Setting
Weakly Imposed Boundary Condition
Conclusion

Some Mixed Finite Element Methods for the Biharmonic

Problem

Bishnu P. Lamichhane, [email protected]

School of Mathematical and Physical Sciences, Faculty of Science and Information Technology, University of
Newcastle, Australia

Monash University
13th April, 2017, Melbourne

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Discrete Setting
Weakly Imposed Boundary Condition
Conclusion

Table of Contents

1 Motivation

2 The Biharmonic Problem

The Standard Formulation
A Varaitonal Formulation
Finite Element Method

3 Discrete Setting
Algebraic Formulation
Numerical Results

4 Weakly Imposed Boundary Condition

5 Conclusion

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Discrete Setting
Weakly Imposed Boundary Condition
Conclusion

How Does A Finite Element Method Work?

The idea is quite simple. Decompose any continuous complex problem using discrete
simple elements. Most of the computational methods and algorithms are based on
ideas like “divide and conquer”, “divide and simplify”.
Rectification of the circle as limit of inscribed regular polygons (Felippa).

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Discrete Setting
Weakly Imposed Boundary Condition
Conclusion

A Finite Element Model of an Engineering System: Felippa

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Discrete Setting
Weakly Imposed Boundary Condition
Conclusion

A Finite Element Model of Crash Simulation

Measured (black) and computed (red) impact on the wall is plotted.

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Discrete Setting
Weakly Imposed Boundary Condition
Conclusion

A Finite Element Model of Crash Simulation: Volume and

Surface Discretization

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Discrete Setting
Weakly Imposed Boundary Condition
Conclusion

Biomechanics Modeling

Simulation of vehicle-accident (right), the rat’s brain (left down) and the deformation
field of the cortex after accident (left up).
NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Discrete Setting
Weakly Imposed Boundary Condition
Conclusion

Filtering the Mixture of Gaussian and Impulsive Noise

Use the finite element smoothing for the mixture noise. Here {zi }N
i=1 is the corrupted signal but u(x, y) is the recovered signal. We want to solve the
minimisation problem with V as the finite element space:

N Z " !2 !2 #
X 2 ∂u ∂u
min (u(xi , yi ) − zi ) +λ + dx dy
u∈V Ω ∂x ∂y
i=0
| {z } | {z }
control the error control the smoothness

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Discrete Setting
Weakly Imposed Boundary Condition
Conclusion

Heat Generation through Friction, L. & Falletta

∂ui (t)
ρ i ci − ∇ · (αi ∇ui (t)) = fi (t) in Ωi , i = 1, 2, Ω̄ := Ω̄1 ∪ Ω2
∂t
The transmission conditions on the interface are given in terms of heat fluxes:
α1 ∇u1 (t, x) · n1 = cD βvp − γ̄c p[u], and α2 ∇u2 (t, x) · n2 = cD (1 − β)vp + γ̄c p[u],

cD = 0.2, γ̄c = 1W/N degree Celsius, p = 1M P a on Γ

Here αi : heat conduction coefficients, v: sliding velocity, ni : outward normals, p:

pressure, γc : heat transfer parameter, cD : frictional constant, β = α1α+α
1
2
, and [u]:
the difference of temperature between two bodies. Combining this model with the
elasticity model, one obtains a thermo-mechanical problem. NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Discrete Setting
Weakly Imposed Boundary Condition
Conclusion

Heat Generation through Friction, L. & Falletta

∂ui (t)
ρ i ci − ∇ · (αi ∇ui (t)) = fi (t) in Ωi , i = 1, 2, Ω̄ := Ω̄1 ∪ Ω2
∂t
The transmission conditions on the interface are given in terms of heat fluxes:
α1 ∇u1 (t, x) · n1 = cD βvp − γ̄c p[u], and α2 ∇u2 (t, x) · n2 = cD (1 − β)vp + γ̄c p[u],

cD = 0.2, γ̄c = 1W/N degree Celsius, p = 1M P a on Γ

Here αi : heat conduction coefficients, v: sliding velocity, ni : outward normals, p:

pressure, γc : heat transfer parameter, cD : frictional constant, β = α1α+α
1
2
, and [u]:
the difference of temperature between two bodies. Combining this model with the
elasticity model, one obtains a thermo-mechanical problem. NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem The Standard Formulation
Discrete Setting A Varaitonal Formulation
Weakly Imposed Boundary Condition Finite Element Method
Conclusion

The Standard Formulation

Let Ω in R2 be a bounded convex domain with polygonal boundary ∂Ω and outward
pointing normal n on ∂Ω. Consider the biharmonic equation

∆2 u = f in Ω (1)

with clamped boundary condition

∂u
u= = 0 on ∂Ω. (2)
∂n
Here
∂2 ∂2 ∂2
∆= + + .
∂x2 ∂y 2 ∂z 2
The other set of boundary conditions is (simply supported boundary condition)

u = ∆u = 0 on ∂Ω.
NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem The Standard Formulation
Discrete Setting A Varaitonal Formulation
Weakly Imposed Boundary Condition Finite Element Method
Conclusion

The Standard Formulation

Let Ω in R2 be a bounded convex domain with polygonal boundary ∂Ω and outward
pointing normal n on ∂Ω. Consider the biharmonic equation

∆2 u = f in Ω (1)

with clamped boundary condition

∂u
u= = 0 on ∂Ω. (2)
∂n
Here
∂2 ∂2 ∂2
∆= + + .
∂x2 ∂y 2 ∂z 2
The other set of boundary conditions is (simply supported boundary condition)

u = ∆u = 0 on ∂Ω.
NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem The Standard Formulation
Discrete Setting A Varaitonal Formulation
Weakly Imposed Boundary Condition Finite Element Method
Conclusion

Applications

The fourth order problem occurs in many applications. Examples are

Kirchhoff plate
Stokes problem
strain gradient elasticity
Cahn–Hilliard equation of phase separation
thin plate splines
This problem is studied extensively by many authors. See, e.g., Ciarlet[78], Ciarlet
and Raviart [74], Babuksa, Osborn and Pitkaranta [80], Ciarlet and Glowinski [75],
Falk [78], Falk and Osborn [80], Monk [87], Brenner and Sung [05] etc.

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem The Standard Formulation
Discrete Setting A Varaitonal Formulation
Weakly Imposed Boundary Condition Finite Element Method
Conclusion

A Variational Formulation
The variational form of the biharmonic problem is found by multiplying both sides of
the equation by a test function v ∈ H02 (Ω) and then integrating by parts. Note that
H 2 (Ω) is a Hilbert space (so-called Sobolev space) defined by

H 2 (Ω) = {v ∈ L2 (Ω), ∇v ∈ L2 (Ω), ∇∇v ∈ L2 (Ω)},

and
∂v
H02 (Ω) = {v ∈ H 2 (Ω), v = = 0 on ∂Ω}.
∂n
Thus integrating by parts the left hand side of
Z Z
2
∆ u v dx = f v dx,
Ω Ω

and using the boundary condition, we get

Z Z
∆u ∆v dx = f v dx.
Ω Ω
NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem The Standard Formulation
Discrete Setting A Varaitonal Formulation
Weakly Imposed Boundary Condition Finite Element Method
Conclusion

A Variational Formulation
The variational form of the biharmonic problem is found by multiplying both sides of
the equation by a test function v ∈ H02 (Ω) and then integrating by parts. Note that
H 2 (Ω) is a Hilbert space (so-called Sobolev space) defined by

H 2 (Ω) = {v ∈ L2 (Ω), ∇v ∈ L2 (Ω), ∇∇v ∈ L2 (Ω)},

and
∂v
H02 (Ω) = {v ∈ H 2 (Ω), v = = 0 on ∂Ω}.
∂n
Thus integrating by parts the left hand side of
Z Z
2
∆ u v dx = f v dx,
Ω Ω

and using the boundary condition, we get

Z Z
∆u ∆v dx = f v dx.
Ω Ω
NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem The Standard Formulation
Discrete Setting A Varaitonal Formulation
Weakly Imposed Boundary Condition Finite Element Method
Conclusion

A Variational Formulation

Defining two bilinear forms

Z Z
a(u, v) = ∆u ∆v dx, and `(v) = f v dx,
Ω Ω

we arrive at the variational problem for the biharmonic equation with the clamped
boundary condition: Find u ∈ H02 (Ω) such that

a(u, v) = `(v), v ∈ H02 (Ω).

The variational problem can also be recast as a minimisation problem:

Z Z
1
J(u) = inf2 J(v) with J(v) = (∆v)2 dx − f v dx. (3)
v∈H0 (Ω) 2 Ω Ω

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem The Standard Formulation
Discrete Setting A Varaitonal Formulation
Weakly Imposed Boundary Condition Finite Element Method
Conclusion

A Variational Formulation
Abstract setting
Many partial differential equations can be written as this kind of variational equation
involving a Hilbert space. Generically, we consider a problem of finding u ∈ V such
that
a(u, v) = `(v), v ∈ V,
where V is a Hilbert space, and a(·, ·) and `(v) are bilinear forms.

We have seen the bilinear forms a(·, ·) and `(·) for biharmonic equation. For the
Poisson equation
−∆u = f in Ω
with the boundary condition u = 0 on ∂Ω, we have `(·) as defined above but
Z
a(u, v) = ∇u · ∇v dx.
Ω

and u, v ∈ V = H01 (Ω). NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem The Standard Formulation
Discrete Setting A Varaitonal Formulation
Weakly Imposed Boundary Condition Finite Element Method
Conclusion

Existence, Uniqueness and Stability

The existence, uniqueness and stability: from Lax-Milgram theorem.
Theorem
Let ` ∈ V ∗ , V ∗ is the dual space of V . If the bilinear form a(·, ·) is continuous, i.e.,

|a(u, v)| ≤ βkukH 1 (Ω) kvkH 1 (Ω) , β > 0 on V ×V

and coercive on V , i.e.,

|a(u, u)| ≥ αkuk2H 1 (Ω) , α > 0 on V,

the continuous problem has a unique solution, which depends continuously on the
right-hand side (well-posed).

For our Poisson problem, continuity follows from Cauchy-Schwartz inequality and
coercivity follows from Poincaré inequality:
k∇vkL2 (Ω) ≥ CkvkH 1 (Ω) , v ∈ V = H01 (Ω).
NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem The Standard Formulation
Discrete Setting A Varaitonal Formulation
Weakly Imposed Boundary Condition Finite Element Method
Conclusion

Finite Element Method

The main idea: replace the continuous space V by a discrete one Vh . Here the
subscript h refers to the fact that the discrete space Vh is based on some finite
element mesh Th having mesh-size h. The discrete problem is: given ` ∈ V ∗ , find
uh ∈ Vh such that
a(uh , vh ) = `(vh ), vh ∈ Vh ,
which yields the algebraic system

A~u = f~, A ∈ Rn×n , ~u, f~ ∈ Rn , n := dim Vh .

If Vh ⊂ V and the continuous problem is well-posed, the discrete problem is also

well-posed due to Lax-Milgram theorem.

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem The Standard Formulation
Discrete Setting A Varaitonal Formulation
Weakly Imposed Boundary Condition Finite Element Method
Conclusion

Finite Element Method

Some conditions for the discrete space Vh :

Easy and efficient to compute a(uh , vh ) and `(vh ).
Should approximate the continuous space in some sense.
Matrix A is sparse.
Matrix A inherits the properties of the original continuous problem (e.g.
symmetricity and positive-definiteness).

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem The Standard Formulation
Discrete Setting A Varaitonal Formulation
Weakly Imposed Boundary Condition Finite Element Method
Conclusion

Finite Element Method

For example, for Ω ⊂ Rd , d ∈ {1, 2, 3}, the continuous space

H 1 (Ω) = {u ∈ L2 (Ω), ∇u ∈ [L2 (Ω)]d }

can be replaced by a discrete space

Vh = span{φ1 , φ2 , · · · , φn } ⊂ H 1 (Ω).
Finite element basis functions in 1D
φ

A finite element basis function in 2D

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem The Standard Formulation
Discrete Setting A Varaitonal Formulation
Weakly Imposed Boundary Condition Finite Element Method
Conclusion

Finite Element Method

Ceá Lemma: Let Vh ⊂ V . Let u and uh be the solutions of the continuous and the
discrete problem, respectively. Then, the following a priori estimate holds
β
ku − uh kH 1 (Ω) ≤ C inf ku − vh kH 1 (Ω) , C = ,
vh ∈Vh α
where β is the continuity constant, and α is the coercivity constant.

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem The Standard Formulation
Discrete Setting A Varaitonal Formulation
Weakly Imposed Boundary Condition Finite Element Method
Conclusion

A Finite Element Method for Biharmonic Problem

Existence and uniqueness
For the continuous formulation continuity follows from Cauchy-Schwarz inequality
and coercivity follows from Poincaré-Friedrichs inequality. Use the standard norm
k · kH 2 (Ω) .

Conforming finite element method

A conforming finite element method is obtained if we replace H02 (Ω) by a discrete
space (finite dimensional) Vh ⊂ H02 (Ω). The discrete variational formulation of the
problem is to find uh ∈ Vh such that

a(uh , vh ) = `(vh ), vh ∈ Vh ,

where Z Z
a(uh , vh ) = ∆uh ∆vh dx, `(vh ) = f vh dx.
Ω Ω
NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem The Standard Formulation
Discrete Setting A Varaitonal Formulation
Weakly Imposed Boundary Condition Finite Element Method
Conclusion

A Variational Formulation
11
00
11
00 (Inner circle) Gradient
(Outer circle) Second derivative
Normal derivative
11
00
A finite element space for a trian- 11 Function value
00
gle is constructed by using Argyris
1
0
element. Locally an Argyris trian- 0
1
11
00
00
11 11
00
00
11
gle has 21 degrees of freedom.

Figure: Degrees of freedom for an Argyris triangle

Disadvantages of a conforming approach are:
Construction of such a space is difficult (H 2 means the Hessian is L2 )
We need more degrees of freedom to construct such a space
The system matrix is thus large
The discretization is based on a larger stencil
NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem The Standard Formulation
Discrete Setting A Varaitonal Formulation
Weakly Imposed Boundary Condition Finite Element Method
Conclusion

A Mixed Formulation
A constrained minimization problem is obtained by introducing φ = ∆u[Ciarlet ’78]
[based on H 1 -space]. The variational equation of this constraint is obtained by
multiplying both sides by a function q ∈ H 1 (Ω) and integrating by parts:
Z
∇u · ∇q + φ q dx = 0, q ∈ H 1 (Ω).
Ω

Let V = H01 (Ω) 2

× L (Ω) equipped with the inner product (·, ·)V defined as
((u, φ), (v, ψ))V = (∇u, ∇v)0,Ω + (φ, ψ)0,Ω ,
and Z
W = {(v, ψ) ∈ V : ∇v · ∇q + ψ q dx = 0, q ∈ H 1 (Ω)}.
Ω
New formulation
Replacing ∆v by ψ in the original functional J(v) our minimization problem is
Z Z
1
J (u, φ) = inf J (v, ψ), J (v, ψ) = |ψ|2 dx − f v dx. (4)
(v,ψ)∈W 2 Ω Ω NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem The Standard Formulation
Discrete Setting A Varaitonal Formulation
Weakly Imposed Boundary Condition Finite Element Method
Conclusion

A Mixed Formulation
A constrained minimization problem is obtained by introducing φ = ∆u[Ciarlet ’78]
[based on H 1 -space]. The variational equation of this constraint is obtained by
multiplying both sides by a function q ∈ H 1 (Ω) and integrating by parts:
Z
∇u · ∇q + φ q dx = 0, q ∈ H 1 (Ω).
Ω

Let V = H01 (Ω) 2

× L (Ω) equipped with the inner product (·, ·)V defined as
((u, φ), (v, ψ))V = (∇u, ∇v)0,Ω + (φ, ψ)0,Ω ,
and Z
W = {(v, ψ) ∈ V : ∇v · ∇q + ψ q dx = 0, q ∈ H 1 (Ω)}.
Ω
New formulation
Replacing ∆v by ψ in the original functional J(v) our minimization problem is
Z Z
1
J (u, φ) = inf J (v, ψ), J (v, ψ) = |ψ|2 dx − f v dx. (4)
(v,ψ)∈W 2 Ω Ω NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem The Standard Formulation
Discrete Setting A Varaitonal Formulation
Weakly Imposed Boundary Condition Finite Element Method
Conclusion

A Saddle Point Formulation

The problem (4) can also be recast as a saddle point formulation: given ` ∈ H −1 (Ω)
find ((u, φ), p) ∈ V × H 1 (Ω) [(u, φ, p) ∈ H01 (Ω) × L2 (Ω) × H 1 (Ω)] so that
R R R
RΩ φ ψ dx+ Ω
(∇v · ∇p + ψ p) dx, = Ω f v dx., (v, ψ) ∈ V ,
Ω
(∇u · ∇q + φ q) dx = 0, q ∈ H 1 (Ω).
(5)
This is called Ciarlet-Raviart mixed formulation (Ciarlet and Raviart [74] or Ciarlet
[78]).

Saddle point formulation here is minimising some energy functional under

some constraint.

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Algebraic Formulation
Discrete Setting
Numerical Results
Weakly Imposed Boundary Condition
Conclusion

Discretization
The discrete biharmonic problem is: given ` ∈ H −1 (Ω), find
((uh , φh ), ph ) ∈ V h × Sh [(uh , φh , ph ) ∈ Sh0 × Mh × Sh ] such that
Z
R
φh ψh dx + Ω
∇vh · ∇ph + ψh ph dx = `(vh ), (vh , ψh ) ∈ V h ,
Ω (6)
R
Ω
∇uh · ∇qh + ψh qh dx = 0, qh ∈ Sh ,

where V h = Sh0 × Mh , and Mh ⊂ L2 (Ω) is another FE space. Standard choice is

Mh = Sh .
A clever choice
But if we choose Mh and Sh so that the basis functions of Sh and Mh can be chosen
to satisfy a biorthogonality relation, the Gram matrices corresponding to the form
Z Z
φh ψh dx and ψh qh dx will be diagonal.
Ω Ω
NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Algebraic Formulation
Discrete Setting
Numerical Results
Weakly Imposed Boundary Condition
Conclusion

Discretization
The discrete biharmonic problem is: given ` ∈ H −1 (Ω), find
((uh , φh ), ph ) ∈ V h × Sh [(uh , φh , ph ) ∈ Sh0 × Mh × Sh ] such that
Z
R
φh ψh dx + Ω
∇vh · ∇ph + ψh ph dx = `(vh ), (vh , ψh ) ∈ V h ,
Ω (6)
R
Ω
∇uh · ∇qh + ψh qh dx = 0, qh ∈ Sh ,

where V h = Sh0 × Mh , and Mh ⊂ L2 (Ω) is another FE space. Standard choice is

Mh = Sh .
A clever choice
But if we choose Mh and Sh so that the basis functions of Sh and Mh can be chosen
to satisfy a biorthogonality relation, the Gram matrices corresponding to the form
Z Z
φh ψh dx and ψh qh dx will be diagonal.
Ω Ω
NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Algebraic Formulation
Discrete Setting
Numerical Results
Weakly Imposed Boundary Condition
Conclusion

Algebraic Formulation
The algebraic formulation of the discrete saddle point problem is given by
0 AT
    
0 uh fh
 0 M D   φh  =  0  , (7)
A D 0 ph 0
where fh is the discretized vector associated with the linear form `(vh ).
Static Condensation
The matrix D is diagonal or quasi-diagonal → easy static condensation of
unknowns φh and ph . A quasi-diagonal matrix:
" #
D 1 (diagonal) 0
D= . (8)
R (rectangular) D 2 (diagonal)

After static condensation

 
AT D −1 M D −1 A uh = fh . NUlogo
(9)
Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Algebraic Formulation
Discrete Setting
Numerical Results
Weakly Imposed Boundary Condition
Conclusion

Construction of Mh
Let Sh = span{ϕ1 , · · · , ϕn }, and Mh = span{µ1 , · · · , µn } with
supp µi = supp ϕi , 1 ≤ i ≤ n. These two bases form a biorthogonal system if
Z
µi ϕj dx = cj δij , cj 6= 0, 1 ≤ i, j ≤ n. (10)
Ω
R
Biorthogonality ⇒ the mass matrix D ( Ω
µh qh dx) is diagonal.

Rectangular or hexahedral meshes

Finite element basis functions based on Gauss-Lobatto quadrature nodes ⇒ the
construction of Mh satisfying all assumptions [L ’06].

Simplicial meshes
Construction of Mh only for the linear case. Higher order case is pretty complicated.
Quadratic case: one should be satisfied with a quasi-diagonal matrix.
NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Algebraic Formulation
Discrete Setting
Numerical Results
Weakly Imposed Boundary Condition
Conclusion

Construction of Mh
Let Sh = span{ϕ1 , · · · , ϕn }, and Mh = span{µ1 , · · · , µn } with
supp µi = supp ϕi , 1 ≤ i ≤ n. These two bases form a biorthogonal system if
Z
µi ϕj dx = cj δij , cj 6= 0, 1 ≤ i, j ≤ n. (10)
Ω
R
Biorthogonality ⇒ the mass matrix D ( Ω
µh qh dx) is diagonal.

Rectangular or hexahedral meshes

Finite element basis functions based on Gauss-Lobatto quadrature nodes ⇒ the
construction of Mh satisfying all assumptions [L ’06].

Simplicial meshes
Construction of Mh only for the linear case. Higher order case is pretty complicated.
Quadratic case: one should be satisfied with a quasi-diagonal matrix.
NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Algebraic Formulation
Discrete Setting
Numerical Results
Weakly Imposed Boundary Condition
Conclusion

Error Estimates: Clamped Boundary Condition

Theorem

Let ((u, φ), p) ∈ V × Q be the solution of the continuous problem and

((uh , φh ), ph ) ∈ V h × Sh be that of discrete one. Let u ∈ W k+1,∞ (Ω) ∩ H02 (Ω),
φ ∈ H k (Ω). Then under some reasonable assumptions, there exists a constant C > 0
independent of the mesh-size h so that
 1

k(u − uh , φ − φh )kV ≤ C hk |u|k+1,Ω + hk |φ|k,Ω + hk− 2 kukk+1,∞,Ω . (11)

Rectangular grid gives better approximation if the solution is smooth enough.

k(u − uh , φ − φh )kV ≤ Chk (|u|k+1,Ω + kukk+3,Ω )
[Girault and Raviart, ’86, Page 108] and [Li ’99]. We get the same estimate even if
Mh = Sh . However, the static condensation is not so easy, and if it is done the
system matrix is dense. NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Algebraic Formulation
Discrete Setting
Numerical Results
Weakly Imposed Boundary Condition
Conclusion

Error Estimates: Clamped Boundary Condition

Theorem

Let ((u, φ), p) ∈ V × Q be the solution of the continuous problem and

((uh , φh ), ph ) ∈ V h × Sh be that of discrete one. Let u ∈ W k+1,∞ (Ω) ∩ H02 (Ω),
φ ∈ H k (Ω). Then under some reasonable assumptions, there exists a constant C > 0
independent of the mesh-size h so that
 1

k(u − uh , φ − φh )kV ≤ C hk |u|k+1,Ω + hk |φ|k,Ω + hk− 2 kukk+1,∞,Ω . (11)

Rectangular grid gives better approximation if the solution is smooth enough.

k(u − uh , φ − φh )kV ≤ Chk (|u|k+1,Ω + kukk+3,Ω )
[Girault and Raviart, ’86, Page 108] and [Li ’99]. We get the same estimate even if
Mh = Sh . However, the static condensation is not so easy, and if it is done the
system matrix is dense. NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Algebraic Formulation
Discrete Setting
Numerical Results
Weakly Imposed Boundary Condition
Conclusion

Numerical Results
∆2 u = f with the clamped boundary condition in Ω := (0, 1)2 . u(x, y) = ex + (x + 1) ey x2 y 2 (1 − x)2 (1 − y)2 .

Table: Discretization errors in different norms for the clamped boundary condition, triangles
and linear
ku−uh k0,Ω |u−uh |1,Ω kφ−φh k0,Ω
level # elem. kuk0,Ω |u|1,Ω kφk0,Ω
0 32 5.34290e-01 6.32693e-01 6.32041e-01
1 128 3.26972e-01 0.71 4.01635e-01 0.66 5.16879e-01 0.29
2 512 1.30302e-01 1.33 1.89139e-01 1.09 3.34937e-01 0.63
3 2048 3.99107e-02 1.71 8.32646e-02 1.18 1.88319e-01 0.83
4 8192 1.08809e-02 1.87 3.88438e-02 1.10 9.92016e-02 0.93
5 32768 2.82773e-03 1.94 1.89646e-02 1.03 5.08074e-02 0.97
6 131072 7.19891e-04 1.97 9.41839e-03 1.01 2.56967e-02 0.98
7 524288 1.81559e-04 1.99 4.70081e-03 1.00 1.29204e-02 0.99

The convergences for ku − uh k0,Ω is O(h2 ) for ku − uh k1,Ω is O(h) and for
kφ − φh k0,Ω is O(h). Better than predicted by the theory. NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Algebraic Formulation
Discrete Setting
Numerical Results
Weakly Imposed Boundary Condition
Conclusion

Table: Discretization errors in different norms for the clamped boundary condition,
rectangles and bilinear
ku−uh k0,Ω |u−uh |1,Ω kφ−φh k0,Ω
level # elem. kuk0,Ω |u|1,Ω kφk0,Ω
0 16 3.63341e-01 0 5.05331e-01 0 5.77392e-01 0
1 64 2.18880e-01 0.73 3.06596e-01 0.72 4.61843e-01 0.32
2 256 8.02639e-02 1.45 1.40782e-01 1.12 2.84812e-01 0.70
3 1024 2.36341e-02 1.76 6.40446e-02 1.14 1.56135e-01 0.87
4 4096 6.36319e-03 1.89 3.07178e-02 1.06 8.14364e-02 0.94
5 16384 1.64784e-03 1.95 1.51580e-02 1.02 4.15487e-02 0.97
6 65536 4.19102e-04 1.98 7.55147e-03 1.01 2.09804e-02 0.99
7 262144 1.05669e-04 1.99 3.77214e-03 1.00 1.05415e-02 0.99

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Algebraic Formulation
Discrete Setting
Numerical Results
Weakly Imposed Boundary Condition
Conclusion

Table: Discretization errors in different norms for the clamped boundary condition, triangles
and quadratic
ku−uh k0,Ω |u−uh |1,Ω kφ−φh k0,Ω
level # elem. kuk0,Ω |u|1,Ω kφk0,Ω
0 32 7.21130e-02 1.73764e-01 2.24893e-01
1 128 7.67699e-03 3.23 4.61574e-02 1.91 7.39127e-02 1.61
2 512 8.36813e-04 3.20 1.18158e-02 1.97 2.48864e-02 1.57
3 2048 1.00174e-04 3.06 2.97468e-03 1.99 8.77516e-03 1.50
4 8192 1.24057e-05 3.01 7.44945e-04 2 3.12711e-03 1.49
5 32768 1.54909e-06 3 1.86302e-04 2 1.11290e-03 1.49
6 131072 1.93720e-07 3 4.65777e-05 2 3.95040e-04 1.49

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Algebraic Formulation
Discrete Setting
Numerical Results
Weakly Imposed Boundary Condition
Conclusion

Table: Discretization errors in different norms for the clamped boundary condition,
rectangles and quadratic serendipity
ku−uh k0,Ω |u−uh |1,Ω kφ−φh k0,Ω
level # elem. kuk0,Ω |u|1,Ω kφk0,Ω
0 16 3.51301e-02 1.36042e-01 1.57615e-01
1 64 3.71421e-03 3.24 3.38321e-02 2.01 4.40283e-02 1.84
2 256 4.20670e-04 3.14 8.46059e-03 2 1.09815e-02 2.00
3 1024 5.08994e-05 3.05 2.11683e-03 2 2.57099e-03 2.09
4 4096 6.32166e-06 3.01 5.29376e-04 2 6.05036e-04 2.09
5 16384 7.89256e-07 3 1.32357e-04 2 1.45262e-04 2.06
6 65536 9.86327e-08 3 3.30902e-05 2 3.54722e-05 2.03

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Discrete Setting
Weakly Imposed Boundary Condition
Conclusion

Table: Discretization errors in different norms for the clamped boundary condition,
rectangles and biquadratic
ku−uh k0,Ω |u−uh |1,Ω kφ−φh k0,Ω
level # elem. kuk0,Ω |u|1,Ω kφk0,Ω
0 16 2.35528e-02 1.27617e-01 2.46599e-02
1 64 3.15699e-03 2.90 3.33867e-02 1.93 3.23521e-03 2.93
2 256 4.01609e-04 2.97 8.43997e-03 1.98 4.14979e-04 2.96
3 1024 5.04224e-05 2.99 2.11584e-03 2 5.28399e-05 2.97
4 4096 6.30974e-06 3 5.29325e-04 2 6.68037e-06 2.98
5 16384 7.88934e-07 3 1.32354e-04 2 8.40308e-07 2.99

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Discrete Setting
Weakly Imposed Boundary Condition
Conclusion

Nitsche Formulation

1 The main difficulty of getting the optimal error estimate is the strong imposition
of Dirichlet boundary condition.
2 Use Nitsche technique (as in Discontinuous Galerkin) to weaken the imposition
of the Dirichlet boundary condition.

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Discrete Setting
Weakly Imposed Boundary Condition
Conclusion

Nitsche Formulation

Consider the non-homogeneous clamped boundary condition with

1 1
gD ∈ H 2 (Γ), gN ∈ H − 2 (Γ):

∂u
u = gD and = gN on Γ, (12)
∂n
we have the minimisation problem (4) with
Z Z
1 1
J (v, ψ) = |ψ|2 dx + kv − gD k21 ,Γ − f v dx,
2 Ω 2 2
Ω

with
Z Z Z
∂q
V = {(v, ψ) ∈ V : ψ q dx − hv, ∆qi = gN q dσ − gD dσ, q ∈ Q}.
Ω Γ Γ ∂n

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Discrete Setting
Weakly Imposed Boundary Condition
Conclusion

Discrete Nitsche Formulation

Let Ch be the collection of boundary edges of the triangulation of Ω. Given
` ∈ H −1 (Ω), find ((uh , φh ), ph ) ∈ V h × Qh such that

a((uh , φh ), (vh , ψh ))+ b((vh , ψh ), ph ) = `(vh ), (vh , ψh ) ∈ V h ,

(13)
b((uh , φh ), qh ) = g(qh ), qh ∈ Qh ,

where with γ > 0

Z
a((uh , φh ), (vh , ψh )) = φh ψh dx + γhuh vh i 12 ,h ,
ZΩ
b((vh , ψh ), qh ) = ψh qh dx − huh , ∆qh i
Ω
Z
`(vh ) = f vh dx + γhgD , vh i 21 ,h and
ZΩ Z
∂qh
g(qh ) = gN qh dσ − gD dσ.
Γ Γ ∂n
NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Discrete Setting
Weakly Imposed Boundary Condition
Conclusion

Discrete Nitsche Formulation: Mesh-dependent norms

Here we use the the mesh-dependent norm on the boundary Γ induced by the inner
product
X 1 Z
kuh k21 ,h = u2 dσ.
2 he e h
e∈Ch

X 1 Z
huh vh i 12 ,h = uh vh dσ,
he e
e∈Ch
Z Z
∂qh
huh , ∆qh i = ∇uh · ∇qh dx − uh dσ.
Ω Γ ∂n

We will use the mesh-dependent norm for uh ∈ Sh ,

kuh k21,h = k∇uh k20,Ω + kuh k21 ,h .

2

NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Discrete Setting
Weakly Imposed Boundary Condition
Conclusion

Discrete Nitsche Formulation: Mesh-dependent norms

1 We also need another norm on the boundary Γ of Ω defined as

X Z
2
kuh k− 1 ,h = he u2h dσ.
2
e∈Ch e

2 Use the following mesh-dependent norm for the Lagrange multiplier qh ∈ Qh as

kqh k2Q = kqh k20,Ω + k∆qh k2∗,h ,

where
h∆qh , vh i
k∆qh k∗,h = sup .
vh ∈Sh kvh k1,h
3 With these norms the discrete saddle point problem is well-posed.
4 I also hope that an optimal priori error estimates can be proved, which I will be
doing in a few weeks time.
NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem
 Motivation
The Biharmonic Problem
Discrete Setting
Weakly Imposed Boundary Condition
Conclusion

Conclusion and Remarks

Conclusion
Mixed finite element methods for the biharmonic equation with a clamped
boundary condition.
The new idea is to use different basis functions satisfying a biorthogonality
relation. This gives an efficient numerical scheme.
Improved error estimate with the Nitsche technique.

Future work
Complete the numerical simulation for the new formulation.
Apply the formulation to more applied problems.
Efficient iterative solver (preconditioner).
Extend the idea to the plate theory.

Thank You NUlogo

Bishnu P. Lamichhane, [email protected] Some Mixed Finite Element Methods for the Biharmonic Problem