view +rv/+time/rungekuttaRV.m @ 1037:2d7ba44340d0 feature/burgers1d

Pass scheme specific parameters as cell array. This will enabale constructDiffOps to be more general. In addition, allow for schemes returning function handles as diffOps, which is currently how non-linear schemes such as Burgers1d are implemented.
author Vidar Stiernström <vidar.stiernstrom@it.uu.se>
date Fri, 18 Jan 2019 09:02:02 +0100
parents 2d7c1333bd6c
children 010bb2677230
line wrap: on
line source

% Takes one time step of size dt using the rungekutta method
% starting from v and where the function F(v,t,RV) gives the
% time derivatives. coeffs is a struct holding the RK coefficients
% for the specific method. RV is the residual viscosity which is updated
% in between the stages and after the updated solution is computed.
function v = rungekuttaRV(v, t , dt, F, RV, DvDt, coeffs)
    % Move one stage outside to avoid branching for updating the
    % residual inside the loop.
    k = zeros(length(v), coeffs.s);
    k(:,1) = F(v,t,RV.evaluate(v,DvDt(v)));

    % Compute the intermediate stages k
    for i = 2:coeffs.s
        u = v;
        for j = 1:i-1
            u = u + dt*coeffs.a(i,j)*k(:,j);
        end
        %RV.update(0.5*(u+v),(u-v)/(coeffs.c(i)*dt)); % Crank-Nicholson for time discretization
        k(:,i) = F(u,t+coeffs.c(i)*dt, RV.evaluate(u,DvDt(u)));
    end

    % Compute the updated solution as a linear combination
    % of the intermediate stages.
    u = v;
    for i = 1:coeffs.s
        u = u + dt*coeffs.b(i)*k(:,i);
    end
    %RV.update(0.5*(u+v),(u-v)/dt); % Crank-Nicholson for time discretization
    v = u;
end