Can I reduce memory requirements of a script by unloading parts of a matrix I am not using anymore in the loop?

1 view (last 30 days)
I know this might have been asked several times already, but I couldn't find a feasible solution for my problem. I am working on intensive lattice simulations, and since I am also using explicit dynamics, I end up with a very big matrix (~20000 x 50000) to store the calculation results. This would be fine with me, but apparently my OS doesn't think so, as it kills the Matlab process randomly when he thinks the app is using too many of the system resources (no Matlab crash report, no way to retrieve data after hours of calculations!). For this reason, I started thinking about possibile ways to reduce the amount of memory the simulation requires. The solution for the "n+1" time step only requires the values of the function at time step "n", but since I use the big matrix to store all the time steps, the memory usage is enormous. Is there any way to "unload" all the previous results and only retain the last calculated state?
MWE:
time_steps = 50000; % Time steps
dof = 22000; % Degrees of freedom
H = zeros(dof,time_steps); % Matrix to Store Solution
H(:,1) = ones(22000,1); % Just numbers
for i=2:time_steps
H(:,i) = H(:,i-1)*K; % Solve for the next step (K is a known matrix)
end
  1 Comment
Matt J
Matt J on 16 Jun 2017
Edited: Matt J on 16 Jun 2017
If K is a non-scalar matrix, then this line
H(:,i) = H(:,i-1)*K;
should be giving you an error. Possibly, you meant H(:,i) = K*H(:,i-1) ?

Sign in to comment.

Answers (1)

Matt J
Matt J on 16 Jun 2017
Edited: Matt J on 16 Jun 2017
Just make H a vector and keep over-writing it.
H = ones(22000,1); % Just numbers
for i=2:time_steps
H = K*H; % Solve for the next step (K is a known matrix)
end
  6 Comments

Sign in to comment.

Tags

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!