Finite Element Method Sauro Succi (Non-spherical cows…)

Slides:



Advertisements
Similar presentations
Finite Difference Discretization of Hyperbolic Equations: Linear Problems Lectures 8, 9 and 10.
Advertisements

Compatible Spatial Discretizations for Partial Differential Equations May 14, 2004 Compatible Reconstruction of Vectors Blair Perot Dept. of Mechanical.
Variational Methods Applied to the Even-Parity Transport Equation
The Finite Element Method Defined
Solving Linear Systems (Numerical Recipes, Chap 2)
MECH593 Introduction to Finite Element Methods
By S Ziaei-Rad Mechanical Engineering Department, IUT.
1 Chapter 4 Interpolation and Approximation Lagrange Interpolation The basic interpolation problem can be posed in one of two ways: The basic interpolation.
An Introduction to the Finite Element Analysis
1 A component mode synthesis method for 3D cell by cell calculation using the mixed dual finite element solver MINOS P. Guérin, A.M. Baudron, J.J. Lautard.
Section 4: Implementation of Finite Element Analysis – Other Elements
1cs542g-term Notes  Preemptive make-up lecture this week  Assignment 2 due by tomorrow morning  Assignment 3 coming soon.
1cs542g-term Notes  Make-up lecture tomorrow 1-2, room 204.
Total Recall Math, Part 2 Ordinary diff. equations First order ODE, one boundary/initial condition: Second order ODE.
Materials Science & Engineering University of Michigan
12/21/2001Numerical methods in continuum mechanics1 Continuum Mechanics On the scale of the object to be studied the density and other fluid properties.
MANE 4240 & CIVL 4240 Introduction to Finite Elements Numerical Integration in 1D Prof. Suvranu De.
Finite Element Method in Geotechnical Engineering
MECH300H Introduction to Finite Element Methods
Lecture 9: Finite Elements Sauro Succi. FEM: non-spherical cows Coordinate-free: Unstructured.
MCE 561 Computational Methods in Solid Mechanics
Scientific Computing Partial Differential Equations Poisson Equation Calculus of Variations.
Finite element method 1 Finite Elements  Basic formulation  Basis functions  Stiffness matrix  Poisson‘s equation  Regular grid  Boundary conditions.
1 Numerical Integration of Partial Differential Equations (PDEs)
Taylor-Galerkin Schemes
CIS V/EE894R/ME894V A Case Study in Computational Science & Engineering HW 5 Repeat the HW associated with the FD LBI except that you will now use.
Finite Element Method.
Lecture 22 - Exam 2 Review CVEN 302 July 29, 2002.
1 Variational and Weighted Residual Methods. 2 The Weighted Residual Method The governing equation for 1-D heat conduction A solution to this equation.
Discontinuous Galerkin Methods Li, Yang FerienAkademie 2008.
1 Spring 2003 Prof. Tim Warburton MA557/MA578/CS557 Lecture 8.
6. Introduction to Spectral method. Finite difference method – approximate a function locally using lower order interpolating polynomials. Spectral method.
Summer School for Integrated Computational Materials Education 2015 Computational Mechanics: Basic Concepts and Finite Element Method Katsuyo Thornton.
1 Finite Elements. 2 Finite elements Formally identical to spectral technique Dependent variables expanded as linear combinations of basis local, non.
Parallel Solution of the Poisson Problem Using MPI
1 EEE 431 Computational Methods in Electrodynamics Lecture 18 By Dr. Rasime Uyguroglu
CHAP 3 WEIGHTED RESIDUAL AND ENERGY METHOD FOR 1D PROBLEMS
Discretization Methods Chapter 2. Training Manual May 15, 2001 Inventory # Discretization Methods Topics Equations and The Goal Brief overview.
Discretization for PDEs Chunfang Chen,Danny Thorne Adam Zornes, Deng Li CS 521 Feb., 9,2006.
Finite Element Modelling of Photonic Crystals Ben Hiett J Generowicz, M Molinari, D Beckett, KS Thomas, GJ Parker and SJ Cox High Performance Computing.
1 Chapter 4 Interpolation and Approximation Lagrange Interpolation The basic interpolation problem can be posed in one of two ways: The basic interpolation.
Finite Element Method. History Application Consider the two point boundary value problem.
BENG 276 HHMI Interfaces Lab 2: Numerical Analysis for Multi-Scale Biology Introduction to Finite Element Modeling Andrew McCulloch Department of Bioengineering.
Runge Kutta schemes Taylor series method Numeric solutions of ordinary differential equations.
1 Copyright by PZ Bar-Yoseph © Finite Element Methods in Engineering Winter Semester Lecture 7.
Matrices, Vectors, Determinants.
X1X1 X2X2  Basic Kinematics Real Applications Simple Shear Trivial geometry Proscribed homogenous deformations Linear constitutive.
Model Anything. Quantity Conserved c  advect  diffuse S ConservationConstitutiveGoverning Mass, M  q -- M Momentum fluid, Mv -- F Momentum fluid.
 Matrix Operations  Inverse of a Matrix  Characteristics of Invertible Matrices …
The Finite Element Approach to Thermal Analysis Appendix A.
1 CHAP 3 WEIGHTED RESIDUAL AND ENERGY METHOD FOR 1D PROBLEMS FINITE ELEMENT ANALYSIS AND DESIGN Nam-Ho Kim.
MATHEMATICS B.A./B.Sc. (GENERAL) THIRD YEAR EXAMINATION, 2012.
Finite Elements in multi-dimensions
2/16/2017 Prof Xin (Cindy) Wang
Boundary Element Method
Finite Element Method in Geotechnical Engineering
Advanced Engineering Mathematics 6th Edition, Concise Edition
Nodal Methods for Core Neutron Diffusion Calculations
Advanced Numerical Methods (S. A. Sahu) Code: AMC 51151
Quantum One.
Materials Science & Engineering University of Michigan
finite element method node point based strong form
Numerical Analysis Lecture 16.
finite element method node point based strong form
Linear Algebra Lecture 3.
Characteristics of Sturm-Liouville Problems
Fourier Analysis Lecture-8 Additional chapters of mathematics
Comparison of CFEM and DG methods
Linear Vector Space and Matrix Mechanics
Chapter 2 A Survey of Simple Methods and Tools
Presentation transcript:

Finite Element Method Sauro Succi (Non-spherical cows…)

Finite Elements The main of FEM is to handle real-life geometries of virtually arbitrary complexity (non spherical cows) FEM achieves this task in a mathematically most elegant and systematic way.Rooted in functional analysis: very abstract but it lands down on very practical implementation strategies with the major comfort of rigorous convergence theorems.

FEM for any geometry Most powrful instance: Coordinate free (unstructured) Arbitrary shape and connectivity

FEM Theory: Functional Analysis * Hilbert space expansions * Strong vs Weak Convergence

Expansion on basis functions KNOWN set of basis functions UNKNOWN coefficients of the expansion The basis functions form a (complete) set in Hilbert space, the functional analogue of R^N vector space. Ordinary d=3 unit vectors: i=(1,0,0); j=(0,1,0), k=(0,0,1).

Examples basis functions Fourier: (periodic geometry) Note that the basis functions are spatially extended Hermite: (harmonic oscillator) Laguerre: (radial symmetry)

What’s the point? Convergence: By sending N to infinity ANY function in the proper Hilbert space should be approximated to arbitrary accuracy: Example: Hilbert Lp: Note that this is GLOBAL statement, less demanding than LOCAL ones: WEAK convergence

Pointwise (strong) formulation Local interpolation around x=x_j: Based on global polynomials Awkward on unstructured lattices

FEM: Practice How do we compute the unknown coeff’s? * Projetion methods Define the scalar (inner) product in Hilbert space as: Where w(x) is a suitable weight to secure finiteness Mass matrix

Co-evolutionary ODE’s Liouville equation: Hilbert-Expand: Hilbert Project: Take scalar product with each phi_1 … phi_N: CO-evolution matrix ODE’s problem: Mass matrix Liouville matrix

Co-evolutionary ODE’s Suppose the basis set is ortho-normal, i.e. Then the mass matrix is the identity and each coefficient evolves on its own: The “ray” f=(f1,f2 … fN) moves in time within N-dimensional Hilberts Space spanned by phi_1 … phi_N. By making Hilbert infinite-dimensional, the exact solution is recovered. This is systematic and supported by rigorous convergence theorems. All basis functions mentioned earlier on can be made ortho-normal.

Practical concern All is well in theory, but practice often tells another story: By increasing N, orthogonality depends more and more on subtle cancellations, because the basis functions are global hence their overlap integral (mass matrix) cannot be zero unless these functions oscillate in space. This is very unhealthy from the numerical standpoint: leads to ill-fated long term solutions

FEM take stage! FEM’s are COMPACT basis functions: they are non-zero only within a finite set (whence the nomer) and strictly zero elsewhere: Piecewise linear elements (“hat functions”) Usually denoted e_j(x):

Piecewise Constant (pwc) Pretty low-order, not useful unless discontinuities are physical (shocks) Characteristic function

Piecewise linear, C1 class: Continuous with discontinuous derivatives Piecewise Linear (pwl) Higher order (quadratic, cubic and so on) are also used, esp With PDE’s with space derivatives above second order

Expansion on local basis functions Basis function (known): Coeff’s: (UNknown):

Expansion on local basis function Projection on Hilbert space Note that the mass matrix is NOT diagonal, a price to pay to locality: the finite elements overlap with neighbors The Liouville matrix cannot be diagonal as it contains space derivatives derivatives

Finite-support basis function Note that the support extends over TWO lattice spacings (generally non uniform)

Useful identities Polynomials of order 2+ are not exactly representable by linear elements:

Variational (weak) formulation Hilbert space L2: global statement Operational: for a sequence g_k, find f_N such that: Exact: Orthogonality in N-dim Hilbert space

Examples of matrices: Mass, Stiffness, Advection. using linear hat functions

FEM operators

Mass matrix Uniform mesh: Algebra on the blackboard..

Advection matrix Same as uniform mesh:

Diffusion matrix Uniform mesh:

Mass matrix: mollifier Define: F_j is a coarse-grained quantity: smoother

FEM vs FD operators

1.Form the Local Matrices (Numerical Quadrature) 2. Assemble the Global Matrices (Local to Global Mapping) 3. Time Marching M*df/dt=Lf by a suitable time-marching method 4. Solve the matrix algebra problem A*f(t+dt)=B*f(t) FEM code structure

Matrix assembly Element-wise

Matrix assembly: 1d 4 matrix elements per interval; 2 intervals per node= 8 matrix elements/node

Local to Global Mapping In d=1 it is trivial: In d>1 a bit more laborious but still simple if the FEM are Cartesian: With unstructured meshes, no global coordinate system is available. Numbering is a task intself. More on the next lecture

Time marching Index free notation: f is an array of size N: Explicit Euler: Even the explicit marching implies a matrix problem!!! (not nice) Crank-Nicolson:

Matrix Problem Index free notation: f is an array of size N: 1. Direct Methods (Gauss Elimination) Two major families: Exact (within roundoff) but expensive O(N^3). Highly sensitive to optimal numbering 2. Iterative Methods Approximate, but they involve no linear algebra other Than matrix-vector multiply, hence they take full advantage of matrix sparsity. Usually much faster, unless the matrix is very “nasty” (advection dominated). Generally recommended in d>2.

FEM for nonlinear eqs Burgers (again!) Nonlinearity brings in higher rank matrices (mode coupling) Thanks to compactness, the coupling remains local: k and l cannot be more than one spacing apart from j!

Self-advection matrix: triad Uniform mesh: Three-body coupling

Self-advection: j=k=l

Self-advection: j=k,k+1

Self-advection: j,k=j+1,l=k+1

FEM for high order PDEs Kuramoto-Sivashinsky (paradigm of chaotic PDE’s) Hat functions return ZERO: cubic elements mandatory!

Boundary Conditions 1.Periodic 2.Dirichlet 3.Neumann They are implemented at the level of the matrix and/or The right hand side, see examples

Periodic BC a10 aN,N+1 a10

Dirichlet

Von-Neumann

Assignements 1.Solve the one-dimensional Poisson equation using hat FEM (Dirichlet BC) 2.Same for ADR (Periodic BC) 3.Same for Burgers

End of Lecture

Expansion on basis functions Operators to Matrices