Coarse-Grained Force Fields Built on Atomistic …
155
Fig. 8 Comparisons of distributions of bonded terms of PEO and PDMS between AA and CG
models
The brackets p indicates ensemble-average over trajectories generated using
parameter set p = ( p j : j = 1, 2, 3, . . .). The derivatives inside of the brackets,
∂ A i
∂ p j
and
∂ A i
∂ p j
, can be calculated analytically or numerically. We choose to calculate these
terms numerically with perturbation δp j . The perturbations were set to 0.05 A for
length parameters and 0.001 kcal/mol for energy parameters. Smaller values of 0.01 A
and 0.0005 kcal/mol were tested, the derivative are almost the same.
A workflow of the automated fit is shown in Fig. 9. Because the VDW parameters
are usually coupled with the torsion parameters, the torsion parameters need to be
optimized again after the VDW parameters are adjusted in each iteration. The process
is repeated until all parameters are converged, which usually takes only a few cycles.
155
Fig. 8 Comparisons of distributions of bonded terms of PEO and PDMS between AA and CG
models
The brackets p indicates ensemble-average over trajectories generated using
parameter set p = ( p j : j = 1, 2, 3, . . .). The derivatives inside of the brackets,
∂ A i
∂ p j
and
∂ A i
∂ p j
, can be calculated analytically or numerically. We choose to calculate these
terms numerically with perturbation δp j . The perturbations were set to 0.05 A for
length parameters and 0.001 kcal/mol for energy parameters. Smaller values of 0.01 A
and 0.0005 kcal/mol were tested, the derivative are almost the same.
A workflow of the automated fit is shown in Fig. 9. Because the VDW parameters
are usually coupled with the torsion parameters, the torsion parameters need to be
optimized again after the VDW parameters are adjusted in each iteration. The process
is repeated until all parameters are converged, which usually takes only a few cycles.