Hi,
This is a general question for all relax developers and contributors, as well as relax users. This important question is - how do you perform a
grid search in a relaxation dispersion analysis for large spin clusters? Is there a different
technique for determining the starting parameter values? How is the
starting point of optimisation chosen?
When there is a cluster of many
spins, the grid search is prohibitively expensive. My idea would be to let the user specify a directory of relax results from a previous dispersion analysis. For example an analysis whereby no clustering has been used. Then the optimised parameters from the earlier analysis is used as the starting point of the clustered analysis.
This then raises an important question. What do you do for the shared cluster parameters? This is only for population and kex parameters. Is an average value used? Or do you hold all non-cluster specific parameters (R20 and all parameters with delta_omega) constant and perform a grid search on the others? Do you know what is normally done in the field?
Cheers,
Edward
P. S. Nothing is implemented yet, so large spin clusters will currently choke the auto-analysis.
Relaxation dispersion and grid searches with large spin clusters.