I have a case where the heatmap of PureCN optimal purity/ploidy solutions for a sample shows the maximum likelihood solution (marked with a 1 on the heatmap) is located at what appears to NOT be a heatmap peak.  I examined the heatmap numbers (used PureCN:::.matrixTotalPloidyToTumorPloidy() to get the heatmap matrix), and I find that the matrix number located at the #1 solution peak doesn't match either the log.likelihood value or the total.log.likelihood value of the #1 solution.  Does the matrix factor in other numbers?

Yes, the matrix is an approximation that does not try to assign integer copy numbers to segments yet. The idea is to start the Markov chain at various places that are likely close to local optima - the fitness landscape of this problem is so spiky it's hard to get a good mixing chain otherwise.

The number 1 should still be fairly close to a peak. If not, it might be a noisy, low purity or highly heterogeneous sample. Hard to diagnose from the distance.