Question about Lusol_obj.repcol

[zhe hu]

I met a problem in testing lusol_obj.repcol().

I replace one column of matrix A with itself, then recalculate sum of elements in the matrix. Result1 and result2 are always same. But sometimes result2 becomes NaN. When scale of A is bigger as 1000, result will never be correct.

Here is my code in Matlab:

n = 20;

A = sprand(n,n,0.2)/10+speye(n);

lu_options = lusol_obj.luset('nzinit',200000,'pivot','TCP');

one_vector = ones(n,1);
mylu = lusol_obj(A,lu_options);
result1 = one_vector'*mylu.mulA(one_vector)

mylu.repcol(A(:,10),10);   %replace col 10 with himself
result2 = one_vector'*mylu.mulA(one_vector)

result:

test2

result1 =

54.5849

result2 =

54.5849

test2

result1 =

53.1647

result2 =

8.4224e+252

Version of MATLAB, and make matlab_test:

                       < M A T L A B (R) >
              Copyright 1984-2012 The MathWorks, Inc.
                 R2012b (8.0.0.783) 64-bit (maci64)
                          August 22, 2012

test_addcol_lhr02: passed
test_addrow_lhr02: passed
test_delcol_lhr02: passed
test_delrow_lhr02: passed
test_factorize_lhr02: passed
test_mulA_lhr02: passed
test_mulAt_lhr02: passed
test_r1mod_lhr02: passed
test_repcol_lhr02: passed
test_reprow_lhr02: passed
test_solveA_lhr02: passed
test_solveAt_lhr02: passed

and I also tested it in ubuntu 13.04, matlab R2012b. Same problem exist.

$ gfortran --version
GNU Fortran (GCC) 4.8.1
Copyright (C) 2013 Free Software Foundation, Inc.

[Nick Henderson]

Thanks for reporting this. It turned out to be a problem passing a sparse vector into the repcol method. I've updated lusol_obj.m to densify all input vectors. Please give it a try. If things seem to work for you, I will post another release.

[zhe hu]

It works! thank you!