view +time/SBPInTimeScaled.m @ 1198:2924b3a9b921 feature/d2_compatible

Add OpSet for fully compatible D2Variable, created from regular D2Variable by replacing d1 by first row of D1. Formal reduction by one order of accuracy at the boundary point.
author Martin Almquist <malmquist@stanford.edu>
date Fri, 16 Aug 2019 14:30:28 -0700
parents e95a0f2f7a8d
children 47e86b5270ad
line wrap: on
line source

classdef SBPInTimeScaled < time.Timestepper
    % The SBP in time method.
    % Implemented for A*v_t = B*v + f(t), v(0) = v0
    % The resulting system of equations is
    %   M*u_next= K*u_prev_end + f
    properties
        A,B
        f

        k % total time step.

        blockSize % number of points in each block
        N % Number of components

        order
        nodes

        Mtilde,Ktilde     % System matrices
        L,U,p,q % LU factorization of M
        e_T

        scaling
        S, Sinv % Scaling matrices

        % Time state
        t
        vtilde
        n
    end

    methods
        function obj = SBPInTimeScaled(A, B, f, k, t0, v0, scaling, TYPE, order, blockSize)
            default_arg('TYPE','gauss');
            default_arg('f',[]);

            if(strcmp(TYPE,'gauss'))
                default_arg('order',4)
                default_arg('blockSize',4)
            else
                default_arg('order', 8);
                default_arg('blockSize',time.SBPInTimeImplicitFormulation.smallestBlockSize(order,TYPE));
            end

            obj.A = A;
            obj.B = B;
            obj.scaling = scaling;

            if ~isempty(f)
                obj.f = f;
            else
                obj.f = @(t)sparse(length(v0),1);
            end

            obj.k = k;
            obj.blockSize = blockSize;
            obj.N = length(v0);

            obj.n = 0;
            obj.t = t0;

            %==== Build the time discretization matrix =====%
            switch TYPE
                case 'equidistant'
                    ops = sbp.D2Standard(blockSize,{0,obj.k},order);
                case 'optimal'
                    ops = sbp.D1Nonequidistant(blockSize,{0,obj.k},order);
                case 'minimal'
                    ops = sbp.D1Nonequidistant(blockSize,{0,obj.k},order,'minimal');
                case 'gauss'
                    ops = sbp.D1Gauss(blockSize,{0,obj.k});
            end

            I = speye(size(A));
            I_t = speye(blockSize,blockSize);

            D1 = kron(ops.D1, I);
            HI = kron(ops.HI, I);
            e_0 = kron(ops.e_l, I);
            e_T = kron(ops.e_r, I);
            obj.nodes = ops.x;

            % Convert to form M*w = K*v0 + f(t)
            tau = kron(I_t, A) * e_0;
            M = kron(I_t, A)*D1 + HI*tau*e_0' - kron(I_t, B);

            K = HI*tau;

            obj.S =    kron(I_t, spdiag(scaling));
            obj.Sinv = kron(I_t, spdiag(1./scaling));

            obj.Mtilde = obj.Sinv*M*obj.S;
            obj.Ktilde = obj.Sinv*K*spdiag(scaling);
            obj.e_T = e_T;


            % LU factorization
            [obj.L,obj.U,obj.p,obj.q] = lu(obj.Mtilde, 'vector');

            obj.vtilde = (1./obj.scaling).*v0;
        end

        function [v,t] = getV(obj)
            v = obj.scaling.*obj.vtilde;
            t = obj.t;
        end

        function obj = step(obj)
            forcing = zeros(obj.blockSize*obj.N,1);

            for i = 1:obj.blockSize
                forcing((1 + (i-1)*obj.N):(i*obj.N)) = obj.f(obj.t + obj.nodes(i));
            end

            RHS = obj.Sinv*forcing + obj.Ktilde*obj.vtilde;

            y = obj.L\RHS(obj.p);
            z = obj.U\y;

            w = zeros(size(z));
            w(obj.q) = z;

            obj.vtilde = obj.e_T'*w;

            obj.t = obj.t + obj.k;
            obj.n = obj.n + 1;
        end
    end

    methods(Static)
        function N = smallestBlockSize(order,TYPE)
            default_arg('TYPE','gauss')

            switch TYPE
                case 'gauss'
                    N = 4;
            end
        end
    end
end