Finite Element Method - 2D Mesh Generator - Metfem2D

Finite Element Method. Mathematical Basis.

Integration upon Elements.

We shall consider the linear expression derived in the Appendix B^{[1, 2, 3]}

with the boundary condition φ = γ on Γ. In such a simply case of integral – differential problems with a differential operator L = -ε∂²/∂x² - ε∂²/∂y², the variable φ in the above – written equation only consists of one scalar function φ which is the sought solution, while the constant vector f is represented by the last term in that expression. To find the solution for such a problem means to determine the values of φ(x,y) in the whole domain Ω. The values of φ on its boundary Γ are already prescribed to γ. On the other hand, at the very beginning (see Eq. (2) in the previous page) we have postulated that a function φ could be approximated by an expansion φ_A given by means of some basis functions N_m(x,y), m = 1, …,n (for more details see Appendix A). Thus another possibility to deal with the Poisson problem is just to start from the functional Π and build a set of Euler equations ∂Π/∂φ_m = 0 where m = 1, …, n and φ_m approximates value of the solution φ calculated at the m-th mesh node.

and after that we calculate the derivative ∂Π/∂φ_A,m. Moreover, let's simplify our problem by neglecting the last term in the above – presented equation and imposing φ = γ on the boundary Γ instead. In that manner, one obtains the expression

or in a simplified form

It is worth mentioning that some requirements must be imposed on the shape functions N. Namely, if n-th order derivatives occur in any term ofL then the shape functions have to be such that their n-1 derivatives (pay an attention to the above – presented equation) are continuous and finite. Therefore, generally speaking C_n-1 continuity of shape functions must be preserve.
In turn, after substituting

finally one obtains a set of equations

for m, l = 1, …, n or in matrix description

It is worth noticing that the matrix K is a symmetric one because of the symmetry in exchange of subscripts l and m in the equation.
Now, we are obliged to employ the division of our domain Ω into a set of subdomains Ωⁱ. It gives that

Therefore, after the transformation to I subdomains the expression becomes

for i = 1, …, I and m = 1, …, n. Or in matrix notation

In fact, the summation written above takes into account only these elements Ωⁱ which contribute to m-th node, however, because of the consistency in notation all elements are included in the sum with the exception that those Nⁱ_m functions for which node m does not occur in i-th element are put equal zero.
From now, the whole story is to calculate integrals

where N^i_m = L^i_m, L₃ = 1 - L₁ - L₂ (see Appendix A) whereas det(Jⁱ) – the jacobian of i-th element, T matrix together with ∇ operator in new coordinates are evaluated in Appendix C.
An integration over the i-th subdomain Ωⁱ, which is a triangular element with three nodes, enforces the transformation from n-dimensional global interpolation to the local interpolations given by means of N_ik(x,y) functions where i_k = 1, 2, 3. That is why in equations new indices i_l,i_m appear which further are allowed to take three possible values 1,2 and 3 for each element i (the local subspace).
As the next step, the Gauss quadrature is employed to compute above-written integrals numerically as it is described in the Appendix D.
And finally, after incorporating boundary conditions to equations by inserting appropriate boundary values of φ, the system of equations can be solved.

---

References

Last update: February 22, 2020
© 2013-2020 taketechease.com Privacy Policy Terms of Use