Dear Jens, Am 21.01.21 um 17:15 schrieb Jens Wehner:
> good luck. I completely agree with Rasmus, it really depends on your problem, > what the best/fastest/smallest solution is. I am sorry to hear that the > Spectra thing did not work. thanks to the fast response of the spectra author, it's running fine now, there was problem with the type definitions. In my minimal working example, spectra beats eigen by a factor of 5 in speed for a matrix dimensions >> 1000 in single threaded mode. Enabling MKL binding with multi-threading, spectra gets only a few percent faster, while it is now beaten by eigen/mkl. In addition, the are cases where spectra doesn't converge for the lowest state in energy, while it converges targeting for 2 lowest states. I refrained from testing the davidson as it is misses a user defined preconditioner. > As an intermediate solution, you can also just adapt your current lapack > wrapper to work with eigen types. It is pretty straightforward to convert > from ublas. Yes, that's the way to go. It will be lots of work but I think it will be worth it. Best regards, Peter
