dihedral_denom, rmsd comparison

Owned by Pavan Behara

May 28, 2021

3 min read

acyclic	Mean	Max	Min
a1	110.47	114.53	108.69
a2	107.42	109.48	105.99

dihedral_denom non-zero

	1	2	3	4	5	6
	bad initial values	bad initial values	good initial values	bad initial values	good initial values
	With priors (same as 1.2, penalty_additive 1.0)	w/o priors (penalty_additive 0.0;penalty_multiplicative -1.0)	With priors (same as 1.2, penalty_additive 1.0), initial values same as 1.3.0	with dihedral_denom non-zero and bad initial values for parameters	with dihedral_denom non-zero and good initial values for parameters	Openff_1.3.0
b1, length, [#6X4:1]-[#6X4:2]	1.5	1.5	1.5	1.5	1.5	1.5
b1, k, [#6X4:1]-[#6X4:2]	487.7	1106.4	517.3	487.9	517.3	517.2
b83, length, [#6X4:1]-[#1:2]	1.1	1.1	1.1	1.1	1.1	1.1
b83, k, [#6X4:1]-[#1:2]	699.9	153.2	754.1	699.9	754.1	754.1
*a1, angle, [:1]-[#6X4:2]-[:3]*	104.9	88.7	114.6	104.4	114.7	113.7
*a1, k, [:1]-[#6X4:2]-[:3]*	112.9	59.4	94.8	105.3	89.1	99.2
a2, angle, [#1:1]-[#6X4:2]-[#1:3]	94.1	88.1	113.1	94.5	112.7	114.3
a2, k, [#1:1]-[#6X4:2]-[#1:3]	42.2	63.4	64.5	45.1	62.2	66.6
X2 value	25.5052	0.0696	0.1657	25.6562	0.2389

RMSD comparison b/w bad and good parameter sets ( 2 and 3 in above table)

System	Bonds		Angles		Dihedrals		Impropers		Term.
	RMSD	denom	RMSD	denom	RMSD	denom	RMSD	denom
#2
CH4	0.001	0.05	0.006	8	0.000	0	0	20	0.001
C2H6	0.001	0.05	0.117	8	0.006	0	0	20	0.005
C3H8	0.000	0.05	0.428	8	0.507	0	0	20	0.052
C4H10	0.000	0.05	0.566	8	0.585	0	0	20	0.120
C5H12	0.000	0.05	0.543	8	0.538	0	0	20	0.139
C6H14	0.001	0.05	0.530	8	0.516	0	0	20	0.160
C7H16	0.001	0.05	0.572	8	0.999	0	0	20	0.219

#3
CH4	0.003	0.05	0.006	8	0.000	0	0	20	0.013
C2H6	0.003	0.05	0.270	8	0.006	0	0	20	0.035
C3H8	0.002	0.05	0.514	8	0.431	0	0	20	0.084
C4H10	0.001	0.05	0.742	8	0.528	0	0	20	0.216
C5H12	0.001	0.05	0.823	8	0.534	0	0	20	0.331
C6H14	0.001	0.05	0.876	8	0.549	0	0	20	0.448
C7H16	0.002	0.05	0.793	8	0.989	0	0	20	0.434

Same table with values tagged by systems

	System	Bonds		Angles		Dihedrals		Impropers		Term.	diff. in obj. fn
		RMSD	denom	RMSD	denom	RMSD	denom	RMSD	denom
#2	CH4	0.001	0.05	0.006	8	0.000	0	0	20	0.001
#3	CH4	0.003	0.05	0.006	8	0.000	0	0	20	0.013	0.012
#2	C2H6	0.001	0.05	0.117	8	0.006	0	0	20	0.005
#3	C2H6	0.003	0.05	0.270	8	0.006	0	0	20	0.035	0.030
#2	C3H8	0.000	0.05	0.428	8	0.507	0	0	20	0.052
#3	C3H8	0.002	0.05	0.514	8	0.431	0	0	20	0.084	0.032
#2	C4H10	0.000	0.05	0.566	8	0.585	0	0	20	0.120
#3	C4H10	0.001	0.05	0.742	8	0.528	0	0	20	0.216	0.096
#2	C5H12	0.000	0.05	0.543	8	0.538	0	0	20	0.139
#3	C5H12	0.001	0.05	0.823	8	0.534	0	0	20	0.331	0.192
#2	C6H14	0.001	0.05	0.530	8	0.516	0	0	20	0.160
#3	C6H14	0.001	0.05	0.876	8	0.549	0	0	20	0.448	0.288
#2	C7H16	0.001	0.05	0.572	8	0.999	0	0	20	0.219
#3	C7H16	0.002	0.05	0.793	8	0.989	0	0	20	0.434	0.215

Sulfonamides (opt only targets refits)

Refits with optgeo targets from 1.2.0 release.

	Good initial values (Frosst)	Bad initial values (1.3.0)
a30, angle	108.9	99.2
a30, k	171.1	222.4
a31, angle	101.1	89.9
a31, k	153.5	191.0
X2 values	671.66	667.67

If we look at the RMSDs of an individual molecule that shows a high gradient wrt 1.3.0

	System	Bonds		Angles		Dihedrals		Impropers		Term.
		RMSD	denom	RMSD	denom	RMSD	denom	RMSD	denom
bad params	421_C6H7N2O5S2	0.040	0.05	14.141	8	49.612	0	3.106	20	107.875
good params	421_C6H7N2O5S2	0.038	0.05	5.479	8	44.429	0	0.615	20	26.535

Should we include the dihedrals' RMSD in the objective function?

Runs with dihedral_denom non-zero failed to converge with the error Step size is too small to continue .