Weak Formulations and Discretisation: Finite Element Models For Nonlinear Analysis of Solids and Structures

Weak Formulations and discretisation
Finite Element Models for Nonlinear Analysis of Solids and

Structures
José M.a Goicolea
Computational Mechanics Group,

Escuela de Ingenieros de Caminos, Universidad Politécnica de Madrid
8 march 2012
Contents
1 Preliminary concepts and notation

Kinematics, Stress measures and derivatives
2 Principle of virtual work and variational principles

Virtual displacements and variations
Strong form in Eulerian and Lagrangian description
PVW in Eulerian and Lagrangian descriptions
Principle of stationary potential energy
Other variational principles
3 Discretisation
Shape functions
Discretised equilibrium equations
More on shape functions
Matrix notation
Outline


3 Discretisation
Shape functions
Matrix notation
Directional derivative
The directional (or Gâteaux) derivative is a concept which is widely

used for linearisation of the equilibrium equations in Finite Elements.
Assume a general function F(x) of vector unknowns x. The value of
F may be scalar, vector, or tensor of any order. Considering a point
x0 and an increment u,

d
DF(x0 )[u] = F(x0 + u)
d =0
If a gradient operator exists, then
DF(x0 )[u] = ∇F · u
The directional derivative is linear and satisfies the usual product rule
and chain rule of derivatives.
JM Goicolea (UPM) Weak Formulations and discretisation 08/03/2012 12 / 54

Linearisation of strain measures
For increment ∆u from current configuration
x=X +u 7→ x + ∆u = X + u + ∆u
Deformation Gradient F
∂∆u
DF [∆u] = = ∇0 ∆u
∂X
∂∆u ∂x
= = (∇∆u)F
∂x ∂X
Notation for gradient operators
∂(•)
def ∂(•i )
∇0 ∆u = ⇒
∂X ∂XJ
def ∂(•) ∂(•i )
∇∆u = ⇒
∂x ∂xj

Linearisation of strain measures (2)
Green strain E = 12 (C − 1) and Cauchy-Green tensor C = F T F
1
DE[∆u] = (F T DF [∆u] + DF T [∆u]F )
2
1
= F T (∇0 ∆u) + (∇0 ∆u)T F
2
1
= F T (∇∆u + (∇∆u)T F = F T ε(∆u)F
2
1
= DC[∆u]
2
Determinant J = det F
DJ[∆u] = (DJ) · ∇0 ∆u = JF −T :(∇0 ∆u) = J(∇·∆u)

∂XQ
=J ∆up,Q = J∆up,p
∂xp

Outline


3 Discretisation
Shape functions
Matrix notation
Virtual displacements
∂u B ∂t B
(δu = 0)
B δu
definition and properties

Arbitrary Field δu(x), for x ∈ B, from the current position.
Time invariant
Compatible with essential boundary conditions: δu = 0 for x ∈ ∂u B
Variation of magnitudes through directional derivative:

d
δF(u) = DF(u)[δu] = F(u + δu)
d =0

Variation of kinematic variables
Displacement gradient (spatial):
δ∇u = ∇δu = ∇0 δuF −1
1
δε = (∇δu + (∇δu)T ) = ε(δu)
2
Deformation gradient:
∂δui ∂δui ∂xj
δF = ∇0 δu = (∇δu)F , or δFiJ = =
∂XJ ∂xj ∂XJ
−1 −1 ∂δuk
δF −1 = −F −1 ∇δu, or δFIj = −FIk
∂xj
Strain tensor
1
δE = (F T δF + δF T F )
2
1 T
= F (∇δu + (∇δu)T )F = F T δεF
2
Updated Lagrangian description: governing
equations
strong formulation in 3D
Field Equations
Elastic material behaviour: σ = g(F )
u∗ ∂t B
Compatibility (strain):
e = 21 ∇u + (∇u)T − (∇u)T ∇u
tdS n t∗
B b
Equilibrium (stress): dS
∇·σ T + ρb = 0; σ = σT ∂u B
σpi,p + ρbi = 0; σij = σji
Unknowns
Displacements
Boundary Conditions u(x) : B → R3
in ∂t B: σ T ·n = t
in ∂u B: u = u
Lagrangian description: governing equations
strong formulation in 3D
Field Equations
Elastic material behaviour: P = h(F )
Compatibility (strain):
u∗ ∂t B0
E = 12 ∇0 u + (∇0 u)T + (∇0 u)T ∇0 u T dS0 N T∗
Equilibrium (stress): ρ0 b
dS0 B0
∇0 ·P + ρ0 b = 0; F P T = P F T (1) ∂u B0
PiQ,Q + ρ0 bi = 0; FiQ PjQ = PiQ FjQ
(2) Unknowns
Displacements
Boundary Conditions u(X) : B0 → R3
in ∂t B0 : P ·N = T
in ∂u B0 : u = u
PVW in updated Lagrangian description (I)
Consider equation of equilibrium (strong form):
∇·σ T + ρb = 0 ∀x ∈ B
Dot product with an arbitrary (but compatible) virtual displacement field

δu(x) and integrate in B:
Z Z
δu·(∇·σ T ) dV + δu·ρb dV = 0
B B
The integrand in the first term may be transformed as

δu·(∇·σ T ) = ∇·(δu·σ T ) − σ T :∇δu
δuq σpq,p = (δuq σpq ),p − σpq δuq,p
Substituting into the integral one can apply the divergence theorem,
Z Z Z
T T
δu·(∇·σ ) dV = (δu·σ )·n dS − σ T :∇δu dV
B ∂B B

PVW in updated Lagrangian description (II)
Considering the boundary conditions (δu = 0 in ∂u B, σ T ·n = t in ∂t B),
Z Z Z
T
− σ :∇δu dV + δu·t dS + δu·ρb dV = 0 ∀δucomp
B ∂t B B
| {z } | {z }
δW int δW ext
which may be condensed as
−δW int + δW ext = 0 ∀δu
It has been shown that the strong form leads to the weak form (PVW).
The converse may also be proved, hence the PVW is a necessary and
sufficient condition for equilibrium.
Due to the symmetry of σ, the internal virtual work may be expressed as
Z Z
int T 1 T
δW = σ : ∇δu + (∇δu) dV = σ:δε dV
B 2 B

PVW in Lagrangian description (I)
Consider equation of equilibrium (strong form):
∇0 ·P + ρ0 b = 0 ∀X ∈ B0
Dot product with an arbitrary (but compatible) virtual displacement field
δu(X) and integrate in B0 :
Z Z
δu·(∇0 ·P ) dV0 + δu·ρ0 b dV0 = 0
B0 B0
The integrand in the first term may be transformed as

δu·(∇0 ·P ) = ∇0 ·(δu·P ) − P :∇0 δu
δup PpQ,Q = (δup PpQ ),Q − PpQ δup,Q
Substituting into the integral one can apply the divergence theorem,
Z Z Z
δu·(∇0 ·P ) dV0 = (δu·P )·N dS0 − P :∇0 δu dV0
B0 ∂B0 B0

PVW in Lagrangian description (II)
Considering the boundary conditions (δu = 0 in ∂u B0 , P ·N = T in ∂t B0 ),

and taking into account the notation ∇0 δu = δF ,
Z Z Z
− P :δF dV0 + δu·T dS0 + δu·ρ0 b dV0 = 0 ∀δucomp (3)
B0 ∂t B0 B0
| {z } | {z }
δW int δW ext
which may be condensed as
−δW int + δW ext = 0 ∀δu
It has been shown that the strong form leads to the weak form (PVW).
The converse may also be proved, hences it is a necessary and sufficient
condition for equilibrium.

PVW in Lagrangian description (III)
Considering the symmetry of S = F −1 P , whereby P = F S, and that
δF = ∇0 δu = (∇δu)F , the density of internal virtual work may be
expressed as
δw0int = P :∇0 δu = (F S):(∇δuF ) = S:(F T ∇δuF )

T1 T
= S: F ∇δu + (∇δu) F = S:δE
2
and the internal virtual work
Z
int
δW = S:δE dV0
B0
This expression is more useful, as it is generally simpler to obtain S from

constitutive equations, which due to objectivity are often expressed in
terms of E or equivalently C = 2E + 1. As an example, for a hyperelastic
material,
∂Ψ ∂Ψ
Ψ(C) ⇒ S = =2
∂E ∂C
The total potential energy functional is expressed as:

Z Z Z
Πp (u) = Ψ(X, F ) dV0 − ρ0 b · u dV0 − t · u dS0 (4)
B0 B0 ∂t B0
Equation (3) is equivalent to establishing the stationarity condition of

the functional (4):
Πp stationary ⇒ δΠp (u) = 0 (5)
We say that (3) is the variational equation for the problem (5), and
that equation (1) is the Euler-Lagrange equation associated to the
variational problem (5).

Stationary condition for functional Π(u):
δΠ(u) = DΠ(u)[δu] = δΠint + δΠext = 0 ∀δu
Considering the domain B0 is fixed we take variations of the integrand:

Z Z
∂Ψ
δΠint = :δF dV = P :δF dV (6)
B0 ∂F B0
Z Z
δΠext = − ρ0 b·δu dV − T ·δu dS (7)
B0 ∂t B0
Example: St Venant-Kirchhoff elasticity

1 ∂Ψ
Ψ(X, E) = E:D:E ⇒ P = = F (D:E) (8)
2 ∂F
(Also: S = ∂Ψ/∂E = D:E, and P = F S.)

We may formulate other variational principles different to the

stationary total potential energy (5)
Such functionals allow as variable fields other quantities apart from
the displacements u(X), and are the basis of the so-called mixed or
hybrid elements
As an example of such multifield functionals, we consider that due to
Hu-Washizu (formulated here for small strains ε):
ΠW (u, ε, σ) =
Z Z
[Ψ(x, ε) + σ: (∇s u − ε) − ρb · u] dV − t · u dS
B ∂t B

Mixed Finite Elements u-p-θ
Nearly incompressible nonlinear elasticity

(Simó & Taylor, CMAME 1991).
u (displacements); p (pressure); θ (volume ratio).
1
u = φ(X, t) − φ(X, 0); θ = J = det(F ); p= tr(σ) .
3
Strong formulation
0 = ∇·σ T + ρb ;
J1
p= tr(σ) ;
θ3
θ=J.

Mixed Finite Elements u-p-θ
Modified variables:
1/3
e = θ eT · F
e =F e;
F F; C
J
e = 2 ∂W (C);
S e e ·S
e = J −1 F
σ e ·F
e ; T
e = p1 + σ
σ e dev
∂C
Weak formulation (updated lagrangian): ∀{δu(x), θ(x), p(x)}
Z Z Z
0= ∇δu:(e σ dev + p1) dV − δu·ρb dV − δu·t dS
B B ∂t B
Z
e
tr σ p
0= δθ − dV
B 3θ J
Z
θ
0= δp 1 − dV
B J
δθ, δp are condensed at element level
Outline


3 Discretisation
Shape functions
Matrix notation
Discretisation – Shape functions
Let us consider one element, with nodes a = 1 . . . n. The shape functions
interpolate the geometry. As the mesh is Lagrangian,
n
X n
X
X= Na (X)X a ⇒ x= Na (X)xa
a=1 a=1
Considering u = x − X, it follows that

n
X
u= Na (X)ua
a=1
Galerkin: identical interpolation to virtual displacements:

n
X
δu = Na (X)δua
a=1

Discretisation – Interpolation of deformations
Consider the deformation gradient,

n
! n
X X
F = ∇0 Na (X)xa = xa ⊗ ∇0 Na ;
a=1 a=1
!
X X
FiJ = xai Na = xai Na,J
a ,J a
In a similar fashion,
n
! n
X X
δF = ∇0 δu = ∇0 Na (X)δua = δua ⊗ ∇0 Na ;
a=1 a=1
!
X X
(δF )iJ = δuai Na = δuai Na,J
a ,J a

Discretised equilibrium equations I
With these results we can discretise the virtual work of internal forces. It is
computed as a sum over all nodes a = 1, . . . n.
Z Z n
!
X
δW int = P :δF dV0 = PpQ δuap Na,Q dV0
B0 B0 a=1
The nodal virtual displacements are constant values and may be taken out
of the integral:
Xn Z X n Z
int
δW = δuap PpQ Na,Q dV0 = δua · P ·∇0 Na dV0
a=1 B0 a=1 B0
For ease of expression in what follows we may focus on the contribution of

a given node a,
Z Z
int(a)
δW = δua · P ·∇0 Na dV0 = δua · ∇0 Na ·P T dV0
B0 B0
| {z }
f int
a
Discretised equilibrium equations II
We deduce the expression for the internal nodal force,
Z
f int
a = P ·∇0 Na dV0 (9)
B0
We shall now introduce the standard notation in the Finite Element jargon
of B operators for interpolation of gradients, which at the basic continuum
level are the gradients of the shape functions:
∂Na
B 0a = ∇0 Na , 0
BQa = Na,Q =
∂XQ
In the above expression, B a is a vector, the gradient interpolation operator
for node a. The internal virtual work is then
Z Z
int(a)
δW0 = δua · P ·B a dV0 = δuar PrQ BQa dV0
B0 B0
Z Z
int
fa = P ·B a dV0 ≡ PjQ BQa dV0
B0 B0
Discretised equilibrium equations III
The other integrals in the PVW expression (3), representing the virtual
work of
Pexternal forces, are discretised straightforwardly considering
δu = na=1 Na (X)δua :
Z Z
δW0ext = δu·T , dS0 + δu·ρ0 b dV0
∂t B0 B0
n
X Z Z
= δua · Na T dS0 + Na ρ0 b dV0
a=1 ∂t B0 B0
The term in parentheses is the external nodal force,

Z Z
f ext
a = N a T dS 0 + Na ρ0 b dV0 (10)
∂t B0 B0
Taking the above into account, the discretised expression for the PVW is

δW0 = −δW0int + δW0ext = δua · −f int a + fa
ext
= 0 ∀δua
Discretised equilibrium equations IV
which leads to the following discrete system of algebraic equations:
f int ext
a (ub ) = f a , a = 1 . . . n .
In the previous derivation for simplicity it has been considered that the
external loads (volume forces ρ0 b and surface loads T ) are constant, i.e.
they do not depend on the configuration or motion of the body. Hence the
RHS term f ext
a is constant, representing a conservative loading.
On the contrary, the LHS term f int a (ub ) does depend on the configuration
of the body, which may be represented for the discrete problem by the
nodal displacements {ub , b = 1, . . . n}. This dependency is nonlinear for a
general case.
If the system is linear (e.g. linear elasticity and small displacements) it can
be shown easily that the dependence of f int a (ub ) is linear, and the resulting
system is a linear system of algebraic equations:
f int
a (ub ) = K ab ·ub ; K ab ·ub = f ext
a .
Discretised equilibrium equations V
In the expressions (9) and (10) in principle the integrals cover the complete
reference domain B0 or its traction boundary ∂t B0 . A key aspect of the
finite element method is to define the shape functions Na (X) with
(e)
compact support within subdomains {B0 , e = 1, . . . nelem } called
elements. These element subdomains are non-overlapping and cover the
complete domain,
n[
elem
(e) (ei ) (ej )
B0 = B0 ; µ(B0 ∩ B0 )=0
e=1
It makes sense to perform the integral over each element subdomain, and
eventually add up all the contributions:
Z
f int(e)
a = P ·B a dV0
(e)
B0
Z Z
f ext(e)
a = Na T dS0 + Na ρ0 b dV0
(e) (e)
∂t B0 B0
Discretised equilibrium equations VI
The advantage of this subdivision is that, as we shall see, the algorithmic

computations to be performed within each element in order to evaluate the
integrals can be exactly the same, and therefore programmed within the
so-called element subroutines and carried out in a very efficient and
versatile manner.
For calculating the sum of all the element contributions we must consider
that a given node may in general be shared by several adjoining elements.
Therefore the process must take into account the global node numbers and
it is called assembly. This process considers the connectivity of elements
which is read in with the mesh at the beginning of computations and is
generally fixed for the rest of the deformation.

Discretisation: finite-dimensional solution spaces
The (generally polynomic) shape functions form a basis for the

finite-dimensional spaces U h and V h used for finding the approximate
solution: NA , A = 1 . . . nnod :
( n
)
X nod
U h = uh ∈ U | uh = u ∀x ∈ ∂u B; uh = uA NA (ξ) ;
A=1
( n
)
X nod
Vh = δuh ∈ V | δuh = 0 ∀x ∈ ∂u B; δu = h
δuA NA (ξ) .
A=1
Sufficient conditions for convergence:

1 Variational index m: order of derivatives of unknowns in weak form
2 Shape functions Na (X) should be C m−1 in the edges between
elements and C m piecewise within each element subdomain.
3 Completeness: shape functions must be capable of representing exactly
polynomic terms of order ≤ m in cartesian coordinates (patch test)

Parent element coordinates
Each element B (e) is mapped into a parent element, as a “bi-unit

cube”.
= [−1, 1] × · · · × [−1, 1]
| {z }
ndim
Within this parent element a parent coordinate system {ξ} is also

defined:
nenod
X
φ : ξ ∈ 7→ x ∈ B (e) ; x = ϕ(ξ) = xa Na (ξ) (11)
a=1
where xa (a = 1, . . . nenod ) are the nodal coordinates of element e

1D Elements
Linear interpolation functions (global level)


 x − xA−1
 xA−1 < x < xA
NA (x) = x hA−1 −x

 A+1 xA < x < xA+1
hA
N1 N2
1 2 3 4 1 2 3 4
N3 N4
1 2 3 4 1 2 3 4

1D Elements
Global and parent references

x − xA x − xA
NB (x) = = (12)
xB − xA he
1 + ξb ξ
Nb (ξ) = (13)
2
A B
a=1 b=2
ξa = −1 ξ ξb = 1

1D Elements
Interpolation of displacement field
uh (x) = uA NA (x) + uB NB (x) (14)

uh (ξ) = u1 N1 (ξ) + u2 N2 (ξ) (15)
Changing from parent element to global coordinates

2x − xA − xB
ξ(x) = (16)
he
e
h ξ + xA + xB
xe (ξ) = (17)
2
X2
1
= Na (ξ)xea = ((1 − ξ)xa + (1 + ξ)xb )
2
a=1

1D Elements
Integration in parent element domain

Z +1
e 1
Kab = Na,ξ (ξ)ENb,ξ (ξ) ∂x dξ (18)
−1 ∂ξ
substituting:

∂ 1 1
Na,ξ = (1 + ξa ξ) = ξa (19)
∂ξ 2 2
e
∂x ∂ h ξ + xA + xB he
= = (20)
∂ξ ∂ξ 2 2
in (18), we obtain:

E 1 −1
Ke = e (21)
h −1 1

Isoparametric elements
Q1 element (four node linear quad):
η
6
3 4 3
,L
y , L 6
, L ξ
4 , - 2
6,, L
L
B L
B 2 ?
B 1 2
1 -
x 2 -
Interpolation:
P4 P4
x(ξ, η) = a=1 xa Na y(ξ, η) = a=1 yI Na
P4 P4
u(ξ, η) = a=1 uI Na v(ξ, η) = a=1 vI Na
with: N (ξ, η) = 14 (1 + ξa ξ)(1 + ηa η).

Jacobian of the isoparametric mapping and derivatives of shape

functions
∂x x,ξ x,η
J (ξ, η) = =
∂ξ y,ξ y,η
where:
4
X 4
X
x,ξ = xea Na,ξ x,η = xea Na,η
a=1 a=1
X4 X4
y,ξ = yae Na,ξ y,η = yae Na,η
a=1 a=1

Computation of derivatives of shape functions:

−1 ξ,x ξ,y 1 y,η −x,η
J (ξ, η) = = ; j = det [J(ξ, η)]
η,x η,y j −y,ξ x,ξ
Na,x = Na,ξ ξ,x + Na,η η,x
Na,y = Na,ξ ξ,y + Na,η η,y
Numerical integration: Several quadrature types are available:

1 Trapezoidal rule
2 Simpson’s formula
3 Gauss quadrature
4 Lobatto Quadrature

CST element (Constant Strain Triangle)

ξ2 x(ξ) x2
2 (0, 1)
3 2
ξ1 1 x1
3 (0, 0) 1 (1, 0)
I Shape functions are planes:
N1e = ξ1
N2e = ξ2
N3e = ξ3 = 1 − ξ1 − ξ2

CST element
I Integration of functions
Z nint
X
∂x(ξ b )
f (x)dV = wl f (x(ξ b )) det
B(e) ∂ξ
b=1
I Jacobian matrix
!
∂x1 ∂x1 ∂x1 ∂x1
∂x ∂ξ1 − ∂ξ3 ∂ξ2 − ∂ξ3
= ∂x2 ∂x2 ∂x2 ∂x2
∂ξ ∂ξ1 − ∂ξ3 ∂ξ2 − ∂ξ3

LST element (Linear Strain Triangle)

ξ2 x(ξ) x2
2 (0, 1) 5
3 2
5 4 (1/2, 1/2)
6 4
(0, 1/2)
ξ1 1
3 6 1 (1, 0) x1
(1/2, 0)
I Shape functions
N1e = 2ξ1 N4e = 4ξ1 ξ2

N2e = 2ξ2 N5e = 4ξ2 ξ3
N3e = 2ξ3 N6e = 4ξ3 ξ1

Numerical Integration
Gauss quadrature:
Z 1 Z 1 nint
X
g(ξ, η)dξdη = Wb g(ξ˜b , η̃b )
−1 −1 b=1
lint = 1 lint = 4 lint = 9

η η η
6 -6
- 6-
-
c c c c c
q
√1 3
6 5
c c c c ?
3
6
- ?
- -
c c
6ξ 6ξ
c c c ?
ξ ?

3D elasticity
Interpolation of strain field (matrix B)

 
Na,x 0 0
 0 Na,y 0 
 
 0 0 Na,z 

B=  a = 1 . . . nnenod
Na,y Na,x 0  
Na,z 0 Na,x 
0 Na,z Na,y

Isoparametric elements in 3D
Trilinear brick
5 ξ = ξ(ξ, η, ζ)
5 8
8
6 6 7
7 1 4
1
4
2 Ωe
2 3
3
x = x(x, y, z)
1
Na (ξ, η, ζ) = (1 + ξa ξ)(1 + ηa η)(1 + ζa ζ)
8

Further reading
Peter Wriggers
Nonlinear finite element methods .
Springer, 2008
Javier Bonet and Richard D Wood
Nonlinear continuum mechanics for finite element analysis .
Cambridge University Press, 1997
T Belytschko, Wing K Liu and Brian Moran
Nonlinear finite elements for continua and structures .
Wiley & Sons, New York, 2000
Mike Crisfield
Nonlinear Finite Element Analysis of Solids and Structures, Vols I & II
Wiley, 1991, 1997
Klaus J. Bathe
Finite Element Procedures,
Prentice-Hall, 1996

Weak Formulations and Discretisation: Finite Element Models For Nonlinear Analysis of Solids and Structures

Uploaded by

Copyright:

Available Formats

Weak Formulations and Discretisation: Finite Element Models For Nonlinear Analysis of Solids and Structures

Uploaded by

Document Information

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

Weak Formulations and Discretisation: Finite Element Models For Nonlinear Analysis of Solids and Structures

Uploaded by

Copyright:

Available Formats

Weak Formulations and discretisation

Finite Element Models for Nonlinear Analysis of Solids and

José M.a Goicolea

Computational Mechanics Group,

1 Preliminary concepts and notation

2 Principle of virtual work and variational principles

1 Preliminary concepts and notation

2 Principle of virtual work and variational principles

The directional (or Gâteaux) derivative is a concept which is widely

If a gradient operator exists, then

JM Goicolea (UPM) Weak Formulations and discretisation 08/03/2012 12 / 54

For increment ∆u from current configuration

JM Goicolea (UPM) Weak Formulations and discretisation 08/03/2012 13 / 54

Green strain E = 12 (C − 1) and Cauchy-Green tensor C = F T F

DJ[∆u] = (DJ) · ∇0 ∆u = JF −T :(∇0 ∆u) = J(∇·∆u)

JM Goicolea (UPM) Weak Formulations and discretisation 08/03/2012 14 / 54

1 Preliminary concepts and notation

2 Principle of virtual work and variational principles

definition and properties

JM Goicolea (UPM) Weak Formulations and discretisation 08/03/2012 16 / 54

Dot product with an arbitrary (but compatible) virtual displacement field

The integrand in the first term may be transformed as

JM Goicolea (UPM) Weak Formulations and discretisation 08/03/2012 20 / 54

which may be condensed as

−δW int + δW ext = 0 ∀δu

JM Goicolea (UPM) Weak Formulations and discretisation 08/03/2012 21 / 54

The integrand in the first term may be transformed as

JM Goicolea (UPM) Weak Formulations and discretisation 08/03/2012 22 / 54

Considering the boundary conditions (δu = 0 in ∂u B0 , P ·N = T in ∂t B0 ),

which may be condensed as

−δW int + δW ext = 0 ∀δu

JM Goicolea (UPM) Weak Formulations and discretisation 08/03/2012 23 / 54

This expression is more useful, as it is generally simpler to obtain S from

The total potential energy functional is expressed as:

Equation (3) is equivalent to establishing the stationarity condition of

Πp stationary ⇒ δΠp (u) = 0 (5)

JM Goicolea (UPM) Weak Formulations and discretisation 08/03/2012 25 / 54

Stationary condition for functional Π(u):

δΠ(u) = DΠ(u)[δu] = δΠint + δΠext = 0 ∀δu

Considering the domain B0 is fixed we take variations of the integrand:

Example: St Venant-Kirchhoff elasticity

JM Goicolea (UPM) Weak Formulations and discretisation 08/03/2012 26 / 54

We may formulate other variational principles different to the

JM Goicolea (UPM) Weak Formulations and discretisation 08/03/2012 27 / 54

Nearly incompressible nonlinear elasticity

JM Goicolea (UPM) Weak Formulations and discretisation 08/03/2012 28 / 54

1 Preliminary concepts and notation

2 Principle of virtual work and variational principles

Considering u = x − X, it follows that

Galerkin: identical interpolation to virtual displacements:

JM Goicolea (UPM) Weak Formulations and discretisation 08/03/2012 31 / 54

Consider the deformation gradient,

JM Goicolea (UPM) Weak Formulations and discretisation 08/03/2012 32 / 54

For ease of expression in what follows we may focus on the contribution of

The term in parentheses is the external nodal force,

The advantage of this subdivision is that, as we shall see, the algorithmic

JM Goicolea (UPM) Weak Formulations and discretisation 08/03/2012 38 / 54

The (generally polynomic) shape functions form a basis for the

Sufficient conditions for convergence: