Models created by MSMBuilder often must have atoms interpolated into their structure as a result of the compressed trajectory format that is used to build the MSM. This can lead to errors in some cases; typically these errors propagate rapidly enough to crash a simulation within 1ns, often much more quickly.
A utility needs to be added to test every generated model. The utility should work on a cell-by-cell basis (e.g. 1 thread per cell) and record the first model in that cell that does not lead to an error. Any subsequent models that do lead to errors should be replaced with the stored good model. Velocities are randomized at simulation time, so even models with the same positions will produce different trajectories, meaning that the proposed replacement scheme should not have a heavy impact AWE's convergence rate.
jeffkinnison
commented
8 years ago
Models created by MSMBuilder often must have atoms interpolated into their structure as a result of the compressed trajectory format that is used to build the MSM. This can lead to errors in some cases; typically these errors propagate rapidly enough to crash a simulation within 1ns, often much more quickly.
A utility needs to be added to test every generated model. The utility should work on a cell-by-cell basis (e.g. 1 thread per cell) and record the first model in that cell that does not lead to an error. Any subsequent models that do lead to errors should be replaced with the stored good model. Velocities are randomized at simulation time, so even models with the same positions will produce different trajectories, meaning that the proposed replacement scheme should not have a heavy impact AWE's convergence rate.