Dear Dr. Klamt,
Actually, this was just pnitroaniline (C6H6N2O2).
But this was CPCM + SMD calculation. As Dr. Zou said, SMD is
constructed over IEFPCM (if I use the right words), at least in
Gaussian.
I actually got to check it without SMD option (same molecule),
and in this case
1) the "error on total polarization charges" is
0.00492 for CPCM and 0.00493 for IEFPCM, so, seems to me, it's
all right with CPCM in Gaussian, for all that.
I also checked some archived files, in which a molecule of 71
atom has this number 0.00015 for CPCM; another one is 88 atoms
and the error value is 0.01296 (but here it was a 'generic'
solvent; similar case has 0.01106 for 92 atoms); and I also
found a rather big molecule with 258 atoms, for which CPCM "error
on total polarization charges" is 0.01800 for one
conformer and 0.02019 for another one.
2) yes, escaped charge is not reported now also for CPCM, so it
is probably due to a threshold.
So, in fact, Dr. Zou told all I needed at the moment.
It's just that SMD is a model which was not build with CPCM. And
CPCM alone is as good (from that "error" side) as IEFPCM.
3) there is an IOp option (3/70=20000)
to "Do COSMO style CPCM: Klamt radii, iterative (implies
g03defaults)", and it does not return any "error
on total polarization charges" (maybe this is why You were
confused by this message). The corresponding part of output is
(same molecule/solvent):
Gaussian default CPCM:
No special actions if energy rises.
Inv3: Mode=1 IEnd=
6203532.
Iteration 1 A*A^1 deviation from unit
magnitude is
3.55D15 for 771.
Iteration 1 A*A^1 deviation from
orthogonality is
2.58D15 for 583 203.
Iteration 1 A^1*A deviation from unit
magnitude is
3.55D15 for 771.
Iteration 1 A^1*A deviation from
orthogonality is
1.56D15 for 1219 49.
Error on total polarization charges =
0.00492
SCF Done: E(RB3LYP) =
492.131647316 A.U. after
13 cycles
NFock= 13 Conv=0.26D08 V/T= 2.0086
Gaussian "COSMOstyle CPCM":
No special actions if energy rises.
Integral accuracy reduced to 1.0D05 until final
iterations.
Initial convergence to 1.0D05 achieved. Increase
integral accuracy.
SCF Done: E(RB3LYP) =
492.120662663 A.U. after
15 cycles
NFock= 15 Conv=0.48D08 V/T= 2.0086
Seems to be two quite different algorithms – and
there actually are two different algorithms, as IOp(3/140) possible
values suggest:
Override PCM solution method.
0 Leave
unchanged.
1 Force
inversion.
2 Force
iterative.
3 Force simultaneous in
L502. (Gaussian 09 IOp Reference, 2nd ed., p.
72).
Looks like that "error on total
polarization charges" appears only for the inversion method
of solving PCM (the first case, as I can guess).
I am starting to feel uneasy if this discussion does not fall
under 'reverse engineering' or something else which is not
permitted under Gaussian license...
Yours sincerely,
Igors
Mihailovs
ISSP UL
On 10/09/17 14:25, Andreas Klamt
klamt__cosmologic.de wrote:
Dear Igors,
Can you tell me how big the molecule with escaped charge of
0.46 e is?
This either means, that too small radii have been used
somewhere, or that the molecule is very big, so that even with
COSMO (CPCM) the error is that large. By the way, this error
does not depend on epsilon.
And what does the
"Error on total polarization charges = 0.30616 "mean?
It is claimed to be only 0.01 for IEFPCM.
Andreas
Am 09.09.2017 um 21:24 schrieb Igors Mihailovs
igorsm###cfi.lu.lv:
Dear Dr. Zou,
Thank You for the further clarification!
One another thing I am a little curious about  only in
CPCM+SMD calculation the escaped charge is reported
(in my case, enormous 0.46984). Is that because of some
threshold for things to appear in output, or due to a
different algorithm in use (for IEFPCM, even with #P there
is no sign of escaped charge)?
Dear Dr. Klamt,
That's why I was surprised by the results I had got,
because I have read before, well, at least the abstract of
the article You cited (due to the lack of access) and
Your article of 2011 (DOI: 10.1002/wcms.56)
where You argued that, although it was proven that the
best value of x in the equation for COSMO is 0.5,
"[u]nfortunately, several reimplementations of COSMO,
including CPCM, set the value of x to zero either
by default or fixed." Can this be the reason? Or is it
simply because SMD model in Gaussian is optimized for
IEFPCM, not CPCM? (Maybe Dr. Zou can comment on
that.)
You also mention that "in the literature the COSMO
approximation is often claimed to be applicable only to
polar solvents, it needs to be emphasized that for
neutral compounds the COSMO approximation with x = 0.5
provides very accurate results down to ε = 2". Maybe
this is why COSMO is not that popular? On the other
hand, if, as Dr. Zou said, the differences in
computational cost arise only in really large molecules,
maybe software developers just feel it is not that
urgent to implement COSMO if they already have another
methods implemented…
By "the equation" I mean the equation of COSMO scaling
factor for unscreened charge density,
f (ε) = (ε – 1) /
(ε + x)

To mention some other details, the differing part in the
output is the following:
***** CPCM_SMD.out
No special actions if energy rises.
Using charges instead of weights in
PCMQM.
Inv3: Mode=1 IEnd=
5838075.
Iteration 1 A*A^1 deviation from
unit
magnitude is 2.78D15 for 865.
Iteration 1 A*A^1 deviation from
orthogonality is 1.65D15 for
711 187.
Iteration 1 A^1*A deviation from
unit
magnitude is 3.11D15 for 813.
Iteration 1 A^1*A deviation from
orthogonality is 1.65D15 for
1142 64.
Escaped charge = 0.46984
Error on total polarization charges =
0.30616
SCF Done: E(RB3LYP) =
492.141763479 A.U.
after 13 cycles
NFock= 13 Conv=0.34D08 V/T=
2.0086
SMDCDS (nonelectrostatic)
energy
(kcal/mol) = 1.55
***** IEFPCM_SMD.OUT
No special actions if energy rises.
Inv3: Mode=1 IEnd=
5838075.
Iteration 1 A*A^1 deviation from
unit
magnitude is 3.55D15 for 317.
Iteration 1 A*A^1 deviation from
orthogonality is 2.67D15 for
629 474.
Iteration 1 A^1*A deviation from
unit
magnitude is 3.33D15 for 317.
Iteration 1 A^1*A deviation from
orthogonality is 2.19D15 for
545 224.
Error on total polarization charges =
0.01601
SCF Done: E(RB3LYP) =
492.138961313 A.U.
after 13 cycles
NFock= 13 Conv=0.35D08 V/T=
2.0086
SMDCDS (nonelectrostatic)
energy
(kcal/mol) = 1.55
Yours
sincerely,
Igors
Mihailovs
ISSP UL
On 08/09/17 22:59, Lufeng Zou
g09gv5_+_gmail.com wrote:
Hello Igors,
One thing to add is that SMD is a solvation model
(IEFPCM using SMD radii for electrostatic part, plus
nonelectrostatic terms), so please do not combine SMD
with other solvation models. The program should have
recognized this error and abort, thank you for
catching the bug.
As for the cost, please rest assured that
difference between IEFPCM and CPCM will be minor for
most calculations. The only exception is that for very
large molecules with hundreds of atoms, more MEMORY
(RAM) is required for IEFPCM than CPCM.
On 09/09/17 04:16, Andreas
Klamt klamt#cosmologic.de wrote:


Prof. Dr. Andreas Klamt
CEO / Geschäftsführer
COSMOlogic GmbH & Co. KG
Imbacher Weg 46
D51379 Leverkusen, Germany
phone +492171731681
fax +492171731689
email klamt+/cosmologic.de
web www.cosmologic.de
[University address: Inst. of Physical and
Theoretical Chemistry, University of Regensburg]
HRA 20653 Amtsgericht Koeln, GF: Prof. Dr. Andreas Klamt
Komplementaer: COSMOlogic Verwaltungs GmbH
HRB 49501 Amtsgericht Koeln, GF: Prof. Dr. Andreas Klamt
