olgerdovich
olgerdovich@yandex.ru
AFAIK, partial linear dependence affects not only SCF convergence, but also may cause higher numerical errors in gradients. Taking into account that you use opttol ten times tighter than default, it looks like a plausible cause of you trouble (without additional information)
I've met similar problems while MP2/6-31++G** optimization of complexes of difluorocarbene with various substituted pyridines. I used opttol=3e-5 and met partial linear dependence in some cases due to diffuse function of ring carbons, and some of these cases were optimizing poorly (or not at all) to this opttol, but they would have been optimized if opttol had been set to 1e-4. I removed diffuse function that cause linear dependence from the basis set and problem disappeared.
You may do the same or loose your optimization criteria
On Tue Sep 15 '15 8:40pm, Pavlo Solntsev wrote
----------------------------------------------
>I have a problem with MP2 gradient. The basis set contains diffuse functions and the geometry is oscillating. I found the similar problem on the forum and tried to follow suggestions you provided earlier. I believe I already use high enough accuracy but no luck. Could you please advice me about possible actions I may take. Thanks.
>
>
> $BASIS GBASIS=N311 Ngauss=6 ndfunc=1 npfunc=1
> diffs=.t. diffsp=.t. $END
> $CONTRL SCFTYP=RHF RUNTYP=OPTIMIZE exetyp=run
> inttyp=hondo icut=13
> itol=40 d5=.t. mplevl=2 nzvar=1 icharg=0 $END
> $STATPT NSTEP=100 opttol=1d-5 REGTOL(1)=1000,1000 $END
> $scf nconv=7 dirscf=.t. $end
> $zmat dlc=.t. auto=.t. nonvdw(1)=11,13,13,16,14,16 $end
> $mp2 method=1 ncore=0 cutoff=1d-20 $end
> $p2p p2p=.t. dlb=.t. mxbuf=2048 xdlb=.t. mixed=.t. $end
> $smp httfix=.f. $end
> $mp2grd tol1=1d-20 tol2=1d-20 $end
> $DATA
> HONO MP2/6-311++G(d,p) OPT
>Cs
>....