Fixes to bott.c

[thchr]

1. Initialize 'bott' number_list to zero before summation-loop.
2. Un and Wn matrix dimension fix before sqmatrix_resize().

I am unsure why the CHECK in sqmatrix_resize didn't produce an error before.

[stevengj]

I don't understand, sqmatrix_resize already sets the .p field. So why is this needed?

[thchr]

When the matrices Un and Wn are copied at lines 98-99 their size still equals the n from the last iteration; then, when we do sqmatrix_resize(..) with preserve_data = 1 the resize/copy loop is inconsistent because it refers to an erroneous size of the original matrix (U or W).

I don't understand why the sqmatrix_copy(..) call doesn't fail though when Un.p and U.p (or, equivalently, Wn and W) must differ after the first iteration (i.e. why does the call to the below not fail at the CHECK?)

/* A = B */
void sqmatrix_copy(sqmatrix A, sqmatrix B)
{
     CHECK(A.p == B.p, "arrays not conformant");
     blasglue_copy(A.p * A.p, B.data, 1, A.data, 1);
}

[thchr]

Or, more briefly: sqmatrix_resize only sets the .p field after it has carried out the preserve_data = 1 part: that part assumes that the .p field reflects the actual size of the original .data field.

[stevengj]

(I had forgotten that sqmatrix_copy didn't reset the p value.)

I don't understand why the check isn't failing either, unless checks are disabled (CHECK_DISABLE is defined in your config.h file)....

Can you insert a printf in the sqmatrix_copy call to see what is happening?

In any case, can you change it to put a Un.p = Wn.p = n_bands; line before the sqmatrix_copy?

[thchr]

The CHECK issue seems to have originated with this commit 225be56#diff-c1263b161d86930f3050f951a96fe060 (line 63). I had been comparing to the sqmatrix_copy(..) on the master branch, rather than the bott branch; silly. I don't recall why we made that change then.

Anyway, I've reverted that change and reinstated the original CHECK and moved the n_bands-reassignment of the .p fields up so it's made before we call sqmatrix_copy(..). This seems to provide consistent fix?

[stevengj]

don't split up "matrices" across two lines in a comment.

[thchr]

Thanks, done.