forum.alglib.net

Many real life applications such as solving of electrical networks requires complex values due to reactive loads. Sparse matrix and LinCG solver has double values in matrix A and vectors x and b. I'm very interested to use this lib and algorithm because performance is about 3x better than with implementation I'm currently using in my application. I have some preliminary plans to convert double values to complex, but the job looks very hard (for me) because it is not so clear what double variables should be converted to complex and which should stay in double as a magnitude or other multiplier/merit value. sparsematrix class itself is not a problem, but at least lincgstate is. Basically everything with multiple undocumented double values. How should I proceed, or is something suitable already available?

You can forget this. Speed issue is fixed with dense matrix solver with LUP decomposition (by James McCaffrey). Improvement 15-20x compared to previous implementation. Conversion from double to Complex (System.Numerics) was piece of cake.