Poor convergence in geometry optimization

Posted by Evgeniy on May 19, 2009 at 13:16:05:

Dear Developers and Users of Molcas,

I encountered problems in geometry optimization for an excited state treated at the CASPT2 level. From the 10th optimzation step the energy started oscillating as one can see below. I cannot figure out why the performence of the optimization algorithm got worse (poor) after the 10th step. Can it be related to some error in the gradients that are claculated numerically or to the error in caspt2 which might depend on the geometry (the low Cholesky scheme was used for two-electron integrals). I will very appreciate any help on what one can do to improve the convergence! Many thanks!

Best regards,
Evgeniy

Below are given summary of the optimization and the input file:

Energy Grad Grad
Iter Energy Change Norm Max
1 -1085.46313156 0.00000000 0.086446 0.036422
2 -1085.46207269 0.00105887 0.234372 0.084495
3 -1085.46323851 -0.00116582 0.075271 0.029517
4 -1085.46353712 -0.00029861 0.022279 0.008258
5 -1085.46363204 -0.00009491 0.014235 0.006427
6 -1085.46372071 -0.00008867 0.011162 0.004369
7 -1085.46379509 -0.00007438 0.009285 0.003310
8 -1085.46383276 -0.00003767 0.013071-0.004136
9 -1085.46387014 -0.00003737 0.007861 0.005193
10 -1085.46380096 0.00006917 0.033130-0.011681
11 -1085.46357752 0.00022345 0.065100 0.027150
12 -1085.46384444 -0.00026692 0.020402-0.007459
13 -1085.46389102 -0.00004658 0.014137 0.004522
14 -1085.46389007 0.00000095 0.011430-0.006081
15 -1085.46391963 -0.00002956 0.017116-0.005757
16 -1085.46390760 0.00001203 0.017216 0.006485
17 -1085.46384798 0.00005962 0.031721 0.010988
18 -1085.46389058 -0.00004261 0.018499-0.008240
19 -1085.46381990 0.00007069 0.051010-0.017402
20 -1085.46389382 -0.00007392 0.021647-0.006093
21 -1085.46389680 -0.00000298 0.022699-0.007744
22 -1085.46393013 -0.00003333 0.008987-0.003988
23 -1085.46393185 -0.00000171 0.015950-0.005551
24 -1085.46392854 0.00000330 0.009277-0.004313
25 -1085.46387635 0.00005219 0.041345-0.015081
26 -1085.46390709 -0.00003074 0.016117 0.006504

The input file:

&GATEWAY
coord
geom.xyz
basis
cc-pVDZ

> DO WHILE

&SEWARD
Low Cholesky

&SCF
Charge
-1
IVO

&RASSCF
Spin
1
Charge
-1
Inactive
57
Ras2
8
CiRoot
2 2 1
Levshift
2.0

&CASPT2
Multistate
2 1 2

&ALASKA
ROOT
2
DELTa
0.001

&SLAPAF
CARTesian
MAXStep
0.020
NOLIne

> ENDDO

Follow Ups:

Post a Followup

Name:
E-Mail:

Subject:

if B is 1s22s22p1, what is Li?

Passfield:

Comments:
: Dear Developers and Users of Molcas, : I encountered problems in geometry optimization for an excited state treated at the CASPT2 level. From the 10th optimzation step the energy started oscillating as one can see below. I cannot figure out why the performence of the optimization algorithm got worse (poor) after the 10th step. Can it be related to some error in the gradients that are claculated numerically or to the error in caspt2 which might depend on the geometry (the low Cholesky scheme was used for two-electron integrals). I will very appreciate any help on what one can do to improve the convergence! Many thanks! : Best regards, : Evgeniy : Below are given summary of the optimization and the input file: : Energy Grad Grad : Iter Energy Change Norm Max : 1 -1085.46313156 0.00000000 0.086446 0.036422 : 2 -1085.46207269 0.00105887 0.234372 0.084495 : 3 -1085.46323851 -0.00116582 0.075271 0.029517 : 4 -1085.46353712 -0.00029861 0.022279 0.008258 : 5 -1085.46363204 -0.00009491 0.014235 0.006427 : 6 -1085.46372071 -0.00008867 0.011162 0.004369 : 7 -1085.46379509 -0.00007438 0.009285 0.003310 : 8 -1085.46383276 -0.00003767 0.013071-0.004136 : 9 -1085.46387014 -0.00003737 0.007861 0.005193 : 10 -1085.46380096 0.00006917 0.033130-0.011681 : 11 -1085.46357752 0.00022345 0.065100 0.027150 : 12 -1085.46384444 -0.00026692 0.020402-0.007459 : 13 -1085.46389102 -0.00004658 0.014137 0.004522 : 14 -1085.46389007 0.00000095 0.011430-0.006081 : 15 -1085.46391963 -0.00002956 0.017116-0.005757 : 16 -1085.46390760 0.00001203 0.017216 0.006485 : 17 -1085.46384798 0.00005962 0.031721 0.010988 : 18 -1085.46389058 -0.00004261 0.018499-0.008240 : 19 -1085.46381990 0.00007069 0.051010-0.017402 : 20 -1085.46389382 -0.00007392 0.021647-0.006093 : 21 -1085.46389680 -0.00000298 0.022699-0.007744 : 22 -1085.46393013 -0.00003333 0.008987-0.003988 : 23 -1085.46393185 -0.00000171 0.015950-0.005551 : 24 -1085.46392854 0.00000330 0.009277-0.004313 : 25 -1085.46387635 0.00005219 0.041345-0.015081 : 26 -1085.46390709 -0.00003074 0.016117 0.006504 : The input file: : &GATEWAY : coord : geom.xyz : basis : cc-pVDZ : > DO WHILE : &SEWARD : Low Cholesky : &SCF : Charge : -1 : IVO : &RASSCF : Spin : 1 : Charge : -1 : Inactive : 57 : Ras2 : 8 : CiRoot : 2 2 1 : Levshift : 2.0 : &CASPT2 : Multistate : 2 1 2 : &ALASKA : ROOT : 2 : DELTa : 0.001 : &SLAPAF : CARTesian : MAXStep : 0.020 : NOLIne : > ENDDO

[ Follow Ups ] [ Post Followup ] [ Molcas user's WWWBoard ]