7.3 Finite element methods

Finite element methods are powerful tools for solving partial differential equations numerically. They break down complex problems into simpler pieces, using special functions to approximate solutions across a mesh of smaller elements.

This approach allows engineers and scientists to tackle real-world problems in areas like structural analysis, fluid dynamics, and electromagnetics. By discretizing continuous problems, finite element methods make the unsolvable solvable.

Weak Formulation of PDEs

Integral Formulation and Derivation

• Weak form relaxes continuity requirements on PDE solution through integral formulation
• Derivation multiplies PDE by test function and integrates over domain
• Green's theorem or integration by parts reduces derivative order in weak form
• Boundary conditions incorporated through boundary integrals
• Establishes variational problem equivalent to original PDE
• Existence and uniqueness of solutions analyzed using functional analysis techniques

Function Spaces and Analysis

• Choice of function spaces for trial and test functions crucial in weak formulation
• Sobolev spaces ($H^1, H^2$) commonly used for second-order elliptic problems
• Lebesgue spaces ($L^2$) employed for first-order hyperbolic equations
• Weak solutions may exist even when classical solutions do not (discontinuous solutions)
• Lax-Milgram theorem provides conditions for existence and uniqueness of weak solutions
• Céa's lemma establishes connection between weak form and best approximation in finite element space

Finite Element Basis Functions and Meshes

Mesh Generation and Element Types

• Finite element meshes discretize problem domain into smaller geometric shapes (elements)
• Common 2D element types include triangles and quadrilaterals
• 3D element types encompass tetrahedra and hexahedra
• Mesh generation algorithms create high-quality meshes (Delaunay triangulation, advancing front)
• Mesh quality metrics assess element shape, size, and distribution (aspect ratio, skewness)
• Structured meshes follow regular patterns, while unstructured meshes adapt to complex geometries

Basis Function Construction

• Basis functions typically piecewise polynomial functions with local support
• Lagrange interpolation polynomials commonly used to construct basis functions
• Partition of unity property ensures basis functions sum to one at each point in domain
• Node-based shape functions (hat functions) for linear elements
• Higher-order basis functions improve accuracy but increase computational complexity
• Hierarchical basis functions allow for p-refinement without remeshing
• Hermite basis functions incorporate derivative information at nodes

Finite Element Matrix and Load Vector Assembly

Matrix Assembly Process

• Finite element matrix (stiffness matrix) represents discretized weak form of PDE
• Assembly computes local element matrices and combines into global matrix
• Local-to-global mapping connects element-level and global degrees of freedom
• Numerical integration techniques evaluate integrals (Gaussian quadrature)
• Sparse matrix storage formats efficiently store assembled matrices (CSR, CSC)
• Assembly process parallelizable to improve computational efficiency

Load Vector and Boundary Conditions

• Load vector represents right-hand side of discretized weak form equation
• Body forces and source terms contribute to load vector
• Neumann boundary conditions incorporated into load vector
• Dirichlet boundary conditions applied through matrix modification or penalty methods
• Mixed boundary conditions combine Dirichlet and Neumann conditions
• Time-dependent problems require assembly at each time step or use of mass matrices

Linear System Solution and Interpretation

Solution Methods for Linear Systems

• Assembled finite element matrix and load vector form linear system of equations
• Direct solvers suitable for small to medium-sized problems (LU decomposition, Cholesky)
• Iterative solvers preferred for large-scale problems (conjugate gradient, GMRES)
• Preconditioning techniques improve convergence of iterative solvers (ILU, algebraic multigrid)
• Parallel solvers exploit multi-core processors and distributed computing (PETSc, Trilinos)
• Domain decomposition methods divide problem into subdomains for parallel solution

Solution Interpretation and Post-processing

• Solution vector represents coefficients of finite element basis functions
• Evaluate finite element approximation at points of interest for visualization
• Gradient recovery techniques improve accuracy of derived quantities (ZZ patch recovery)
• Error estimation methods assess solution quality (residual-based, recovery-based)
• Adaptive mesh refinement uses error estimates to selectively refine mesh
• Quantities of interest extracted from solution (stress concentrations, heat fluxes)

Accuracy and Convergence of Finite Element Methods

Error Estimation and Convergence Analysis

• A priori error estimates provide theoretical bounds on approximation error
• Error bounds typically expressed in terms of mesh size (h) and polynomial degree (p)
• A posteriori error estimates use computed solution to assess local and global errors
• Convergence rates relate error reduction to mesh refinement or polynomial degree increase
• Optimal convergence achieves best possible rate for given problem and discretization
• Superconvergence phenomena occur at specific points in domain (Gaussian points)

Adaptive Techniques and Verification

• Adaptive mesh refinement selectively refines mesh in regions of high error
• h-adaptivity adjusts element size, p-adaptivity increases polynomial degree
• hp-adaptivity combines both approaches for optimal convergence
• Verification techniques validate finite element implementations (method of manufactured solutions)
• Benchmark problems assess accuracy and efficiency of finite element methods
• Trade-off between computational cost and accuracy considered in practical applications