Introduction to Simulation - Lecture 22 Integral Equation Methods Jacob White Thanks to Deepak Ramaswamy, Michal Rewienski, Xin Wang and Karen Veroy Outline Integral Equation Methods Exterior versus interior problems Start with using point sources Standard Solution Methods in 2-D Galerkin Method Collocation Method Issues in 3-D Panel Integration SMA-HPC ©2003 MIT Interior Versus Exterior Problems Interior Exterior ∇ 2T = 0 outside ∇ 2T = 0 inside Temperature known on surface “Temperature in a Tank” Temperature known on surface “Ice Cube in a Bath” What is the heat flow? Heat Flow SMA-HPC ©2003 MIT ∂T ∫ ∂n surface Thermal = conductivity Exterior Problem in Electrostatics potential + v - ∇2Ψ = 0 Outside Ψ is given on Surface What is the capacitance? Capacitance SMA-HPC ©2003 MIT Dielectric = Permitivity ∂Ψ ∫ ∂n surface Drag Force in a Microresonator Courtesy of Werner Hemmert, Ph.D. Used with permission. Resonator Computed Forces Bottom View SMA-HPC ©2003 MIT Discretized Structure Computed Forces Top View What is common about these problems. Exterior Problems Drag Force in MEMS device - fluid (air) creates drag. Coupling in a Package - Fields in exterior create coupling Capacitance of a Signal Line - Fields in exterior. Quantities of Interest are on the surface MEMS device - Just want surface traction force Package - Just want coupling between conductors Signal Line - Just want surface charge. Exterior Problem is linear and space-invariant MEMS - Exterior Stokes Flow equation (linear). Package - Maxwell’s equations in free space (linear). Signal Line - Laplace’s equation in free space (linear). But problems are geometrically very complex! Exterior Problems Surface Why not use Finite-Difference or FEM methods 2-D Heat Flow Example T = 0 at ∞ But, must truncate the mesh ∂T Only need on the surface, but T is computed everywhere ∂n Must truncate the mesh, ⇒ T (∞) = 0 becomes T ( R ) = 0 SMA-HPC ©2003 MIT Green’s Function Laplace’s Equation ( In 2-D 2 2 ( x − x0 ) + ( y − y0 ) ) If u = log ∂ 2u ∂ 2u then 2 + 2 = 0 for all ( x, y ) ≠ ( x0 , y0 ) ∂x ∂y In 3-D If u = 1 ( x − x0 ) + ( y − y0 ) + ( z − z0 ) 2 2 2 ∂ 2u ∂ 2u ∂ 2u then 2 + 2 + 2 = 0 for all ( x, y, z ) ≠ ( x0 , y0 , z0 ) ∂x ∂y ∂z Proof: Just differentiate and see! SMA-HPC ©2003 MIT Laplace’s Equation in 2-D Simple Idea u is given on surface Surface ( x0 , y0 ) Let u = log ( ∂ 2u ∂ 2u + 2 = 0 outside 2 ∂x ∂y ( x − x0 ) + ( y − y0 ) 2 ∂ 2u ∂ 2u + 2 = 0 outside 2 ∂x ∂y 2 ) Problem Solved Does not match boundary conditions! SMA-HPC ©2003 MIT Simple Idea Laplace’s Equation in 2-D “More Points” u is given on surface ∂ 2u ∂ 2u + 2 = 0 outside 2 ∂x ∂y ( x2 , y2 ) ( x1 , y1 ) n Let u = ∑ ωi log i =1 ( ( xn , yn ) ( x − xi ) + ( y − yi ) 2 2 ) = ∑ω G ( x − x , y − y ) n i =1 i i i Pick the ωi ' s to match the boundary conditions! SMA-HPC ©2003 MIT Simple Idea Laplace’s Equation in 2-D “More Points Equations” (x , y ) t1 t1 Source Strengths selected to give correct potential at test points. ( x2 , y2 ) ( x1 , y1 ) ( xn , yn ) G ( xt − x1 , yt − y1 ) L L G ( xt − xn , yt − yn ) ω Ψ ( xt , yt ) 1 1 1 1 1 1 1 M O M M M = M M O M M G x − x , y − y L L G x − x , y − y ω n Ψ x , y 1 1 tn tn tn n tn n tn tn ( SMA-HPC ©2003 MIT ) ( ) ( ) Computational results using points approach Circle with Charges r=9.5 Potentials on the Circle n=20 SMA-HPC ©2003 MIT r R=10 n=40 Laplace’s Equation in 2-D Integral Formulation Limiting Argument Want to smear point charges onto surface Results in an Integral Equation Ψ ( x ) = ∫ G ( x, x′ ) σ ( x′ ) dS ′ surface How do we solve the integral equation? SMA-HPC ©2003 MIT Laplace’s Equation in 2-D Basis Function Approach Basic Idea n Represent σ ( x ) = ∑ ωi ϕ i ( x ) { i =1 Basis Functions Example Basis Represent circle with straight lines Assume σ is constant along each line The basis functions are “on” the surface Can be used to approximate the density May also approximate the geometry SMA-HPC ©2003 MIT Laplace’s Equation in 2-D Basis Function Approach Geometric Approximation is not new. Piecewise Straight surface basis Triangles for 2-D FEM Functions approximate the circle approximate the circle too! Ψ ( x) = ∫ approx surface SMA-HPC ©2003 MIT n G ( x, x′ ) ∑ ωiϕi ( x′ ) dS ′ i =1 Laplace’s Equation in 2-D x1 xn ln l1 x2 l2 Ψ ( x) = ∫ Basis Function Approach Piecewise Constant Straight Sections Example. 1) Pick a set of n Points on the surface 2) Define a new surface by connecting points with n lines. 3) Define ϕ i ( x ) = 1 if x is on line li otherwise, ϕ i ( x ) = 0 n n i =1 i =1 G ( x, x′ ) ∑ ωiϕi ( x′ ) dS ′ = ∑ ωi approx surface SMA-HPC ©2003 MIT G ( x, x′ ) dS ′ ∫ line l i How do we determine the ω i ' s ? Basis Function Approach Laplace’s Equation in 2-D R ( x) ≡ Ψ ( x) − ∫ Residual Definition and minimization n G ( x, x′ ) ∑ ωiϕi ( x′ ) dS ′ approx surface i =1 We will pick the ω i ' s to make R ( x ) small. General Approach: Pick a set of test functions φ1 ,K , φn , and force R ( x ) to be orthogonal to the set ∫ φ ( x )R ( x ) dS = 0 i SMA-HPC ©2003 MIT for all i. Basis Function Approach Laplace’s Equation in 2-D Residual minimization using test functions ∫ φ ( x ) R ( x ) dS = ∫ φ ( x ) Ψ ( x ) dS − ∫ ∫ i i n φi ( x ) G ( x, x′ ) ∑ ω jϕ j ( x′ ) dS ′dS = 0 j =1 approx surface We will generate different methods by chosing the φ1 ,K , φn , ( ) Collocation: φi ( x ) = δ x − xti (point-matching) Galerkin Method: φi ( x ) = ϕ i ( x ) (basis = test) SMA-HPC ©2003 MIT Basis Function Approach Laplace’s Equation in 2-D Collocation Collocation: φi ( x ) = δ ( xi ) (point-matching) ∫ δ ( x − x ) R ( x ) dS = R ( x ) = Ψ ( x ) − ∫ ti ti ( ) n ⇒ Ψ xti = ∑ ω j j =1 ( G xti , x′ ti approx surface ∫ ( n ) ∑ω ϕ ( x′) dS ′ = 0 j =1 j ) G xti , x′ ϕ j ( x′ ) dS ′ approx surface 1444424444 3 Ai , j A1,1 L L A1,n ω1 Ψ ( xt1 ) M O M M = M M O M M M An ,1 L L An ,n ω n Ψ xtn ( ) SMA-HPC ©2003 MIT j Laplace’s Equation in 2-D xn l n xt1 l1 l2 x2 Basis Function Approach Centroid Collocation for Piecewise Constant Bases ( ) n Ψ xti = ∑ ω j j =1 ∫ ( ) G xti , x′ ϕ j ( x′ ) dS ′ approx surface Collocation point in line center A1,1 L L A1,n ω1 Ψ ( xt1 ) M O M M = M M O M M M An ,1 L L An ,n ω n Ψ xtn ( ) SMA-HPC ©2003 MIT ( ) n Ψ xti = ∑ ω j j =1 ∫ G ( x , x′) dS ′ ti line j 1442443 Ai , j Laplace’s Equation in 2-D ( ) n Ψ xti = ∑ ω j j =1 Basis Function Approach Centroid Collocation Generates a nonsymmetric A ∫ G ( x , x′) dS ′ ti line j 1442443 Ai , j xt1 xt2 l1 A1,2 = l2 ∫ G ( x , x′) dS ′ ≠ ∫ G ( x t1 line 2 SMA-HPC ©2003 MIT t2 line1 , x′ ) dS ′ = A2,1 Laplace’s Equation in 2-D Basis Function Approach Galerkin Galerkin: φi ( x ) = ϕ i ( x ) (test=basis) ∫ ϕ ( x ) R ( x ) dS = ∫ ϕ ( x ) Ψ ( x ) dS − ∫ ∫ i n ϕi ( x ) G ( x, x′ ) ∑ ω jϕ j ( x′ ) dS ′dS = 0 i ∫ n ϕi ( x ) Ψ ( x ) dS = ∑ ω j approx surface j =1 14442444 3 bi A1,1 L L M O M O An ,1 L L approx surface ∫ ∫ G ( x, x′ ) ϕi ( x ) ϕ j ( x′ ) dS ′dS approx approx surface surface 1444444 424444444 3 Ai , j A1,n ω1 b1 M M M = M M M An ,n ω n bn If G ( x, x′) = G ( x′, x) then Ai , j =A j ,i SMA-HPC ©2003 MIT j =1 A is symmetric Basis Function Approach Laplace’s Equation in 2-D ln l1 xn l2 Galerkin for Piecewise Constant Bases x2 n ∫ Ψ ( x ) dS = ∑ω ∫ ∫ G ( x, x′) dS ′dS linei 14243 bi j =1 j linei line j 144424443 Ai , j A1,1 L L A1,n ω1 b1 M O M M = M M O M M M An ,1 L L An ,n ω n bn SMA-HPC ©2003 MIT 3-D Laplace’s Equation Basis Function Approach Piecewise Constant Basis Integral Equation: Ψ ( x ) = ∫ surface Discretize Surface into Panels 1 σ ( x′ ) dS ′ x − x′ n Represent σ ( x ) ≈ ∑ ωi ϕ i ( x ) { i =1 Basis Functions ϕ j ( x ) = 1 if x is on panel j Panel j ϕ ( x ) = 0 otherwise j SMA-HPC ©2003 MIT 3-D Laplace’s Equation Put collocation points at panel centroids Basis Function Approach Centroid Collocation ( ) n Ψ xci = ∑ ω j xci Collocation point j =1 G(x ∫ panel j ci ) , x′ dS ′ 14442444 3 Ai , j A1,1 L L A1,n ω1 Ψ ( xc1 ) M O M M = M M O M M M An ,1 L L An ,n ω n Ψ xcn ( ) SMA-HPC ©2003 MIT Basis Function Approach 3-D Laplace’s Equation Calculating Matrix Elements xci Collocation z point y Ai , j = x ∫ panel j 1 dS ′ xci − x′ Panel j One point quadrature Approximation Four point quadrature Approximation SMA-HPC ©2003 MIT Panel Area Ai , j ≈ xci − xcentroid j 4 0.25* Area j =1 xci − x po int j Ai , j ≈ ∑ Basis Function Approach 3-D Laplace’s Equation Calculating “Self-Term” xci Collocation z point y Ai ,i = x ∫ panel i 1 dS ′ xci − x′ Panel i One point quadrature Approximation Ai ,i = ∫ panel i SMA-HPC ©2003 MIT Ai ,i Panel Area ≈ xci − xci 1 424 3 0 1 dS ′ is an integrable singularity xci − x′ Basis Function Approach 3-D Laplace’s Equation Calculating “Self-Term” Tricks of the trade z xci Collocation y point Panel i x ∫ Ai ,i = panel i Disk of radius R surrounding collocation point ∫ Integrate in two Ai ,i = disk pieces Disk Integral has singularity but has analytic formula SMA-HPC ©2003 MIT ∫ disk 1 dS ′ xci − x′ 1 1 dS ′ + dS ′ ∫ xci − x′ rest of panel xci − x′ R 2π 1 dS ′ = ∫ xci − x′ 0 ∫ 0 1 rdrdθ = 2π R r Basis Function Approach 3-D Laplace’s Equation Calculating “Self-Term” Other Tricks of the trade z xci Collocation y point x Panel i Ai ,i = ∫ panel i 1 dS ′ xci − x′ 1 424 3 Integrand is singular 1) If panel is a flat polygon, analytical formulas exist 2) Curve panels can be handled with projection SMA-HPC ©2003 MIT Basis Function Approach 3-D Laplace’s Equation Galerkin (test=basis) n ϕ ( x ) Ψ ( x ) dS = ∑ ω ∫ ∫ ϕ ( x ) G ( x, x′ )ϕ ( x′ ) dS ′dS ∫144 2443 144444244444 3 i bi j =1 j i j Ai , j For piecewise constant Basis n 1 dS ′dS Ψ ( x ) dS ′ = ∑ ω j ∫ ∫14 ∫ panel i panel j x − x′ 243 j =1 14444244443 bi Ai , j A1,1 L L A1,n ω1 b1 M O M M = M M O M M M An ,1 L L An ,n ω n bn SMA-HPC ©2003 MIT 3-D Laplace’s Equation Basis Function Approach Problem with dense matrix Integral Equation Method Generate Huge Dense Matrices A1,1 L L A1,n ω1 Ψ ( xc1 ) M O M M = M M O M M M An ,1 L L An ,n ω n Ψ xcn ( ) Gaussian Elimination Much Too Slow! SMA-HPC ©2003 MIT Summary Integral Equation Methods Exterior versus interior problems Start with using point sources Standard Solution Methods Collocation Method Galerkin Method Next Time Æ “Fast” Solvers Use a Krylov-Subspace Iterative Method Compute MV products Approximately