Table of contents

The finite element method for linear elasticity

The partial differential equation for the static linear elasticity problem

σ=fin Ωσ=2με+λIuin Ωu=gon ΩDσn=ton ΩN \def\arraystretch{1.5} \begin{array}{lc} -\nabla \cdot \bm \sigma =f & \text{in }\Omega \\ \bm \sigma = 2\mu \bm \varepsilon + \lambda \bm I \nabla \cdot \bm u & \text{in }\Omega \\ \bm u = \bm g & \text{on }\partial \Omega_D \\ \bm \sigma \cdot \bm n = \bm t & \text{on }\partial \Omega_N \end{array}

We cannot directly tackle the strong form of a differential equation to create the FEM. The first step in the FEM is to derive the weak form (or variational form) of the strong form.

Deriving the weak form

Following the recipe for deriving the weak form we multiply the PDE by vv such that v=0v=0 on ΩD\partial \Omega_D and integrate over Ω\Omega

Ω(σ)vdΩ=ΩfvdΩ\int_{\Omega } \left(-\nabla \cdot \bm \sigma \right)\cdot \bm v d\Omega =\int_{\Omega } \bm f \cdot \bm v d\Omega

1: The 2D domain.

In Figure 1, ΩD \partial \Omega_D denotes the Dirichlet boundary conditions (also known as the essential boundary conditions), where the solution u\bm u is known, and ΩN \partial \Omega_N denotes the Neumann boundary conditions (also known as the natural boundary conditions), where the force is applied.

Recall that (in 2D)

σ=[σxτxyτxyσy]\bm \sigma =\left\lbrack \begin{array}{cc} \sigma_x & \tau_{xy} \\ \tau_{xy} & \sigma_y \end{array}\right\rbrack

so the divergence of the stress is

σ=[σxx+τxyyτxyx+σyy] \def\arraystretch{2.0} \nabla \cdot \bm \sigma =\left\lbrack \begin{array}{c} \dfrac{\partial \sigma_x }{\partial x}+\dfrac{\partial \tau_{xy} }{\partial y}\\ \dfrac{\partial \tau_{xy} }{\partial x}+\dfrac{\partial \sigma_y }{\partial y} \end{array}\right\rbrack

Let

q1=[σxτxy] and q2=[τxyσy]\bm q_1 =\left\lbrack \begin{array}{c} \sigma_x \\ \tau_{xy} \end{array}\right\rbrack \text{ and } \bm q_2 =\left\lbrack \begin{array}{c} \tau_{xy} \\ \sigma_y \end{array}\right\rbrack

where q1\bm q_1 is associated with the xx-components and q2\bm q_2 is associated with the yy-components.

we can write

(σ)v=i,jσijxjvj=q1vx+q2vy\left(\nabla \cdot \bm \sigma \right)\cdot \bm v=\sum_{i,j} \dfrac{\partial \sigma_{ij} }{\partial x_j }v_j =\nabla \cdot \bm q_1 v_x +\nabla \cdot \bm q_2 v_y

then we can formulate the weak form as such:

Ω(σ)vdΩ=Ω(q1vx+q2vy)dΩ\int_{\Omega } \left(-\nabla \cdot \bm \sigma \right)\cdot \bm v d\Omega = -\int_{\Omega } \left(\nabla \cdot q_1 v_x +\nabla \cdot q_2 v_y \right)d\Omega

We can expand (vq)\nabla \cdot \left(vq\right) using the product rule

(vq)=x(vqx)+y(vqy)=product ruleqxxv+qxvx+yyv+qyvy\nabla \cdot \left(vq\right)=\dfrac{\partial }{\partial x}\left(vq_x \right)+\dfrac{\partial }{\partial y}\left(vq_y \right)\overset{\text{product rule} }{=} \dfrac{\partial q_x }{\partial x}v+q_x \dfrac{\partial v}{\partial x}+\dfrac{\partial y}{\partial y}v+q_y \dfrac{\partial v}{\partial y} (vq)=qv+vq\nabla \cdot \left(vq\right)=q\cdot \nabla v+v\nabla \cdot q

So

Ω(vq)dΩ=ΩqvdΩ+ΩvqdΩ\int_{\Omega } \nabla \cdot \left(vq\right)d\Omega =\int_{\Omega } q\cdot \nabla vd\Omega +\int_{\Omega } v\nabla \cdot qd\Omega

Recall the Gauss divergence theorem

Ω(vq)dΩ=ΩNvqnds\int_{\Omega } \nabla \cdot \left(vq\right)d\Omega =\int_{\partial \Omega_N } vq\cdot nds

which can be seen as the multivariate integration by parts formula. The theorem also describes conservation of mass in classical mechanics. What ever mass is generated (or lost) inside the body needs to be transferred across its boundary.

We then have

Ω(vq)dΩ=ΩqvdΩ+ΩvqtwodifferentialsdΩ=ΩNvqnds\int_{\Omega } \nabla \cdot \left(vq\right)d\Omega =\int_{\Omega } q\cdot \nabla vd\Omega +\int_{\Omega } v\underset{\mathrm{two}\mathrm{differentials} }{\underbrace{\nabla \cdot q} } d\Omega =\int_{\partial \Omega_N } vq\cdot nds

We move the term with the higher order differential to the left hand side

ΩvqdΩ=ΩNvqndsΩqvdΩ\int_{\Omega } v\nabla \cdot qd\Omega =\int_{\partial \Omega_N } vq\cdot nds-\int_{\Omega } q\cdot \nabla vd\Omega

So we bring it all together

Ωvxq1+vyq2dΩ=Ωq1vx+q2vydΩΩNvxq1n+vyq2nds-\int_{\Omega } v_x \nabla \cdot \bm q_1 +v_y \nabla \cdot \bm q_2 d\Omega =\int_{\Omega } \bm q_1 \cdot \nabla v_x + \bm q_2 \cdot \nabla v_y d\Omega -\int_{\partial \Omega_N } v_x \bm q_1 \cdot n+ v_y \bm q_2 \cdot nds

We substitute back q1\bm q_1 and q2\bm q_2

ΩNvxq1n+vyq2nds=ΩN(σxnx+τxyny)vx+(τxynx+σxny)vyds=ΩN(σn)vds=ΩNtvds \def\arraystretch{2.5} \begin{align*} & -\int_{\partial\Omega_N}v_{x}\bm{q}_{1}\cdot\bm{n}+v_{y}\bm{q}_{2}\cdot\bm{n}ds=\int_{\partial\Omega_N}\left(\sigma_{x}n_{x}+\tau_{xy}n_{y}\right)v_{x}+\left(\tau_{xy}n_{x}+\sigma_{x}n_{y}\right)v_{y}ds\\ & =\int_{\partial\Omega_N}\left(\bm{\sigma}\cdot\bm{n}\right)\cdot\bm{v}ds=\int_{\partial\Omega_N}\bm{t}\cdot\bm{v}ds \end{align*}

Furthermore

q1vx+q2vy=σxvxx+τxyvxy+τxyvyx+σxvyy q_1 \cdot \nabla v_x +q_2 \cdot \nabla v_y =\sigma_x \dfrac{\partial v_x }{\partial x}+\tau_{xy} \dfrac{\partial v_x }{\partial y}+\tau_{xy} \dfrac{\partial v_y }{\partial x}+\sigma_x \dfrac{\partial v_y }{\partial y} σxvxx+2τxy12(vxy+vyx)+σxvyy=σ:ε(v) \sigma_x \dfrac{\partial v_x }{\partial x}+2\tau_{xy} \dfrac{1}{2}\left(\dfrac{\partial v_x }{\partial y}+\dfrac{\partial v_y }{\partial x}\right)+\sigma_x \dfrac{\partial v_y }{\partial y}= \bm \sigma : \bm \varepsilon \left(v\right) σ:ε(v)=i,jσijεij\bm \sigma : \bm \varepsilon \left(v\right)=\sum_{i,j} \sigma_{ij} \varepsilon_{ij}

here we identified the double dot product between the stress and virtual strain. A similar derivation is used as a proof of equivalence between the principle of virtual work and the equilibrium equation. See here.

Thus, from the starting expression

Ω(σ(u))vdΩ=ΩfvdΩ\int_{\Omega } \left(-\nabla \cdot \bm \sigma (\bm u) \right)\cdot \bm v d\Omega =\int_{\Omega } \bm f \cdot \bm v d\Omega

we formulate our final weak form: Find the displacement field u\bm u, with u=g\bm u = \bm g on the boundary ΩD\partial \Omega_D, such that

Ωσ(u):ε(v)dΩ=ΩfvdΩ+ΩNtvdsv that are 0 on ΩD\boxed{ \int_{\Omega } \bm \sigma (\bm u) : \bm \varepsilon \left( \bm v\right) d\Omega =\int_{\Omega } \bm f\cdot \bm vd\Omega +\int_{\partial \Omega_N } \bm{t} \cdot \bm v ds \quad \forall v \text{ that are } 0 \text{ on } \partial \Omega_D}

More information on the weak form

More info on the weak form can be found in the section on Variational Formulation - The Weak Form.