Dear Julian,
Following your previous guidance，I manually adjusted the Buckingham parameter A and was able to run the relax fitting. Thank you very much!
But the error cannot be further reduced, even if I adjusted the weights multiple times. I guess it stuck in a local minimum, I want to use genetic fitting to determine a series of local minima. After reading the manual, I found some parameters to control the genetic fitting, but there are problems with my input format. Could you please further explain the format of these parameters, especially “discrete”.
I want to acquire the optimal 10 combinations after every 20 iterations, can “best” achieve the function? And the genetic algorithm’s maximum iteration also controlled by “maxcyc”?
The input file is attached at the end, which is my test file, so the parameters may not be very reasonable.
Kind regards,
Xiaohan
input.gin (1.4 KB)
Dear Xiaohan,
In my experience of fitting during the last 30 years I don’t think I’ve found a case of a local minimum in fitting parameter space for a standard ionic model force field. Therefore I don’t think genetic algorithms will solve your problem, especially as you can’t run a relax fit with genetic algorithms. If the error (sum of squares) won’t go any lower you may have reached the limit of your model for your system. My advice would be to think about the underlying physics of the system and whether your model captures everything that is important & to improve the model.
For genetic algorithms, the “discrete” option generates a grid of points that control the resolution of the parameter over the range specified. The parameter is specified as a binary number, which is why the input is 2^N where N is the value input for each parameter. “maxcyc fit” controls the number of fitting cycles (as opposed to “maxcyc opt” which is optimisation of structures).
Hope that’s some help.
Regards,
Julian
Dear Julian,
Thank you for your advice!
I will further consider and improve my model.
Kind regards,
Xiaohan