The multiple state points code is significantly slower. Fortunately, the Golden search optimization converges very quickly, e.g. within 10 to 15 iterations. Still, I should talk with Michael to figure out how I can make the MBAR routine faster. For example, most of the state points are unnecessary at distant temperatures, but the code currently solves all of them simultaneously. Also, I rebuild my u_kn matrix for every iteration of mu, and then I create a new instance of the MBAR class. Even when I provide a good initial guess for the reduced free energies, I think it is slow to build the matrix that is extremely large.
The multiple state points code is significantly slower. Fortunately, the Golden search optimization converges very quickly, e.g. within 10 to 15 iterations. Still, I should talk with Michael to figure out how I can make the MBAR routine faster. For example, most of the state points are unnecessary at distant temperatures, but the code currently solves all of them simultaneously. Also, I rebuild my u_kn matrix for every iteration of mu, and then I create a new instance of the MBAR class. Even when I provide a good initial guess for the reduced free energies, I think it is slow to build the matrix that is extremely large.