Why is Matlab crashing during this program?

For the attached .m files below, when I run sandpiles.m, Matlab is asked to do an intensive computation. It very reliably fails to complete the requested computation by crashing: the entire UI and program act as if I quit the program. No error message appears. What specifically is crashing my code? I am using MATLAB 2017b

[Arvind Sathyanarayanan]

Try using the debug mode to see where exactly MATLAB gets stuck.

Also, its good practice to avoid clear all as it clears all the functions are removed from the RAM and ends up decreasing your code's performance.

EDIT: I ran your code in debug mode and it looks like you're caught in an infinite loop when you call topple at line 47, the cluprit seems to be this piece of code in topple.m:

if any(sum(A1>crit))
    A1 = topple(A1,crit);
else
    return
end

You might need to rethink your program's logic.

Answer 1

[Guillaume]

Matlab actually crashing (i.e. the application terminates) should not happen. Perhaps you run out of memory?

Matlab actually stopping running your with an error message is more likely. In this case, give us the full text of the error message, everything that is in red.

As far as I can tell, your neighbors.m is simply a convolution. A few notes on your code:

num_r = length(A(:, 1));
num_c = length(A(, 1:));

is simply

num_r = size(A, 1);  %no need to extract a column to get the height of A
num_c = size(A, 2);  %no need to extract a row to get the width of A

In any case,

N = zeros(num_r,num_c);

is simply:

N = zeros(size(A));

Now, if you have a bunch of

if somecondition
    ...
elseif someothercondition
    ...
elseif yetanothercondition
    ...
elseif 
    ...
end

you should always finish on an else, not an elsif. Particularly, as your last condition is guaranteed to be true if all previous elsif have failed.

Also, if makes sense to test for the most common condition first (i.e. the centre of the matrix) rather than last. The first condition you're testing is one that will be true for just one element.

In any case, as far as I can tell, your function is just:

function N = neighbors(A, crit)
    N = zeros(size(A));
    N(A > crit) = A(A > crit);
    N = conv2(floor(N / 4), [0 1 0; 1 0 1; 0 1 0], 'same');
end

Similarly, in your topple.m function, it looks like your

A1 = A;
[row,col] = find(A>crit);
for i = 1:length(row)
    n = floor(A(row(i),col(i))/4); % # toppled grains to neighbors
    A1(row(i),col(i)) = A(row(i),col(i)) - 4*n;
end

is simply:

A1 = A;
A1(A > crit) = mod(A1(A > crit), 4);

You need to learn to work on whole matrices rather than using overcomplicated loops.

Also,

if any(sum(A1 > crit))

would be clearer as

if any(any(A1 > crit))

or even better

if nnz(A1 > crit)

Your function recurse n itself until you've toppled all your grains. There is no need for a recursion here, you could simply use a while loop. recursion is expensive since calling a function is expensive. Matlab imposes a limit on the number of recursions but you can increase that. However if you do that, it can make your system unstable. Perhaps, that's why you crash.

function A = topple(A, crit);
    while nnz(A > crit)     %much more fficient than a recursion
       N = neighbors(A, crit);
       A(A > crit) = mod(A(A > crit), 4);
       A = A + N;
    end
end