Mercurial > repos > public > sbplib
view +time/CdiffNonlin.m @ 774:66eb4a2bbb72 feature/grids
Remove default scaling of the system.
The scaling doens't seem to help actual solutions. One example that fails in the flexural code.
With large timesteps the solutions seems to blow up. One particular example is profilePresentation
on the tdb_presentation_figures branch with k = 0.0005
author | Jonatan Werpers <jonatan@werpers.com> |
---|---|
date | Wed, 18 Jul 2018 15:42:52 -0700 |
parents | d1f9dd55a2b0 |
children | b5e5b195da1e |
line wrap: on
line source
classdef CdiffNonlin < time.Timestepper properties D E S k t v v_prev n end methods function obj = CdiffNonlin(D, E, S, k, t0,n0, v, v_prev) m = size(D(v),1); default_arg('E',0); default_arg('S',0); if isnumeric(S) S = @(v,t)S; end if isnumeric(E) E = @(v)E; end % m = size(D,1); % default_arg('E',sparse(m,m)); % default_arg('S',sparse(m,1)); obj.D = D; obj.E = E; obj.S = S; obj.k = k; obj.t = t0; obj.n = n0; obj.v = v; obj.v_prev = v_prev; end function [v,t] = getV(obj) v = obj.v; t = obj.t; end function [vt,t] = getVt(obj) vt = (obj.v-obj.v_prev)/obj.k; % Could be improved using u_tt = f(u)) t = obj.t; end function obj = step(obj) D = obj.D(obj.v); E = obj.E(obj.v); S = obj.S(obj.v,obj.t); m = size(D,1); I = speye(m); %% Calculate for which indices we need to solve system of equations [rows,cols] = find(E); j = union(rows,cols); i = setdiff(1:m,j); %% Calculate matrices need for the timestep % Before optimization: A = 1/k^2 * I - 1/(2*k)*E; k = obj.k; Aj = 1/k^2 * I(j,j) - 1/(2*k)*E(j,j); B = 2/k^2 * I + D; C = -1/k^2 * I - 1/(2*k)*E; %% Take the timestep v = obj.v; v_prev = obj.v_prev; % Want to solve the seq A*v_next = b where b = (B*v + C*v_prev + S); % Before optimization: obj.v = A\b; obj.v(i) = k^2*b(i); obj.v(j) = Aj\b(j); obj.v_prev = v; %% Update state of the timestepper obj.t = obj.t + obj.k; obj.n = obj.n + 1; end end end