Matrix too big?

[turbozedd]

I'm using map< size_t, map< size_t, long double > > as a sparse matrix data structure (see for implementation details: http://www.cplusplus.com/forum/general/8352/ ).

At the moment I'm importing the discrete Laplacian into my matrix (n x n, square matrix). It has no more than 5*n non-zero elements. When n > 20000, I get a segmentation fault, i.e. at about 100,000 elements.

Any ideas on getting my matrix to hold more?

[helios]

Buying more RAM? Although that doesn't really sound like it should cause a problem.
Just before the segfault, what's the size of the biggest inner map?

[kbw]

Why don't you use a single map with something like

std::map<std::pair<size_t, size_t>, long double>

That may be more space and time efficient.

[turbozedd]

Most rows have 5 elements, the rest of them have less than 5.

@kbw: it has to be implemented this way for fast multiplication, memory is slightly less important.