[molpro-user] lmp2 geometry optimisation convergence problem
Falk Richter
falk.richter at univ-pau.fr
Fri Oct 7 16:47:16 BST 2011
Hi,
I'm trying df-mp2 optg for some molecules having all some valence
angles close to 180
and optg is jumping arround close to a presumed equilibrium geometry,
but does not converge.
The "step size" (what is given in the "convergence line" under "next" in
the optimization point resumee)
is lower than 0.05 but actually never falls below 0.01. If I guess
right, it this value which must fall below 0.01
in order to get the domains frozen for all the optimization points.
(i.e. for all optimization points to have the same domains)
The reason why this value is always higher than 0.01 are probably the
large changes of some torsional angles connected
to some angles close to 180 deg (the potential in torsional direction
is then very shallow)
and consequently the reason for nonconvergence is the changing domain
definition at each optimization point.
If this is the case it would make sense to adjust this 0.01 trashold.
So, is there any way to adjust this treshold ?
Is this a possible interpretation of something that can happen
or is there some error from my part (see also the input) ?
Thanks
Falk Richter
***,
MEMORY,50,M
file,2,141.wfu
Angstrom
noorient
basis=avdz
geometry={
C;
C,1,R2;
C,2,R3,1,A3;
C,1,R4,3,A4,2,D4;
H,1,R5,3,A5,2,D5;
C,2,R6,3,A6,1,D6;
C,2,R7,3,A7,1,D7;
C,3,R8,2,A8,7,D8;
N,7,R9,6,A9,2,180.0;
N,6,R10,7,A10,9,0.0;
N,4,R11,5,A11,1,180.0;
N,8,R12,2,A12,3,180.0;
}
R2 = 1.54457775
R3 = 1.43601809
A3 = 57.5707
R4 = 1.44771117
A4 = 121.4228
D4 = 251.9241
R5 = 1.10987590
A5 = 121.3899
D5 = 106.9102
R6 = 1.45523163
A6 = 120.9002
D6 = 254.8190
R7 = 1.45286408
A7 = 121.3321
D7 = 109.2881
R8 = 1.35230000
A8 = 147.1963
D8 = -78.0080
R9 = 1.15978019
A9 = 145.1733
D9 = 178.3186
R10= 1.15998441
A10= 146.0559
D10= -1.3742
R11= 1.16060342
A11= 148.3197
D11= 181.8337
R12= 1.17484329
A12= 163.3735
D12= 180.5848
{df-hf;
wf,72,1,0;
accu,16;
}
{df-lmp2;
PIPEK,METHOD=3,DELETE=1}
optg;
More information about the Molpro-user
mailing list