PETSc
(the Portable, Extensible Toolkit for Scientific Computation) has
an option for writing out sparse matrices to an ASCII text file which
can be read by Matlab. This is true as of PETSc version 2.3.3 and
probably much earlier (I checked on 12 Aug 2008). PETSc's
interpretation of "can be read by Matlab" is different than our interpretation of "Matlab ASCII format."
For the latter, we merely store an ASCII file of two, three, or four
space-delimited columns of data. In contrast, PETSc actually
generates a Matlab

2 1 -1.0000000000000000e+00 1 2 -1.0000000000000000e+00 2 2 4.0000000000000000e+00 3 3 4.0000000000000000e+00 4 4 4.0000000000000000e+00 3 2 -1.0000000000000000e+00 2 3 -1.0000000000000000e+00 4 3 -1.0000000000000000e+00 3 4 -1.0000000000000000e+00 1 1 4.0000000000000000e+00would be stored in PETSc's Matlab (ASCII) format as

% Size = 4 4 % Nonzeros = 10 zzz = zeros(10,3); zzz = [ 1 1 4.0000000000000000e+00 1 2 -1.0000000000000000e+00 2 1 -1.0000000000000000e+00 2 2 4.0000000000000000e+00 2 3 -1.0000000000000000e+00 3 2 -1.0000000000000000e+00 3 3 4.0000000000000000e+00 3 4 -1.0000000000000000e+00 4 3 -1.0000000000000000e+00 4 4 4.0000000000000000e+00 ]; Mat_0 = spconvert(zzz);The latter, if run in Matlab, creates two variables in the workspace: "zzz", which is a scratch variable, and "Mat_0", which holds the sparse matrix. PETSc's interface lets users replace "Mat_0" with any desired name, as long as it is a valid Matlab variable. Both "zzz" and the matrix variable name would clobber any preexisting definitions of variables with the same name in the Matlab workspace.

PETSc's documentation does not explain (as of PETSc version 2.3.3) this output format. We had to look at the following source code file:

./mat/impls/aij/seq/aij.cand a generated example contributed by an SMC user, in order to figure out the format. However, once you figure it out, it's not hard to parse or convert via a shell script to another format.