[molpro-user] negative eigenvalues in basis construction in CCSD-F12

[Mariusz Radon]

Dear Molpro Experts:

>From time to time I experience the following problem with negative
eigenvalues in my CCSD-F12 calculations (actually, in the preceding
MP2-F12); sample output is appended below. This problem occurs when I
use the diagonal Ansatz with unfixed amplitudes [ansatz=3C(D)]; it is
not present with the default one [ansatz=3C(FIX)]. Nevertheless, I have
some reasons to try other Ansatze as well.

As you may see from the sample output, everything seems to converge
fine. The final (relative) energies are also close to the benchmarks
obtained from basis set extrapolations. My question is whether this
warning should be considered as fatal (therefore, should I treat these
calculations as "failed") or maybe the program is able to recover from
this problem (meaning that I may still trust the results).

Thank you in advance for your replies.


With kind regards,
Mariusz




 CPU time for transformed integrals            1929.09 sec
 Multipassing in F12-Matrix calculation: NPASS =  2    NLHS =   339
 CPU time for F12 matrices                      124.80 sec


 ITER.     SQ.NORM      CORR.ENERGY   TOTAL ENERGY  ENERGY CHANGE
VAR         CPU   MICRO    DIIS
 ? Warning
 ? Encountered negative eigenvalues in basis construction.
S[<mn|X|op>[AA]] is not positive semi-definite!
 ? The problem occurs in SvdConstructOrthoBasis
 The offending spectrum starts with:
      -6.01E-08  -4.95E-08  -4.80E-08  -4.75E-08  -4.18E-08  -3.33E-08
-3.30E-08  -3.03E-08  -3.02E-08
 ? Warning
 ? Encountered negative eigenvalues in basis construction.
S[<mn|X|op>[BB]] is not positive semi-definite!
 ? The problem occurs in SvdConstructOrthoBasis
 The offending spectrum starts with:
      -6.00E-08  -4.94E-08  -4.77E-08  -4.74E-08  -4.18E-08  -3.32E-08
-3.30E-08  -3.02E-08  -3.00E-08
 ? Warning
 ? Encountered negative eigenvalues in basis construction.
S[<mn|X|op>[AB]] is not positive semi-definite!
 ? The problem occurs in SvdConstructOrthoBasis
 The offending spectrum starts with:
      -6.00E-08  -5.93E-08  -4.94E-08  -4.84E-08  -4.79E-08  -4.74E-08
-4.71E-08  -4.68E-08  -4.18E-08
   1      1.46367881    -1.82122672 -1617.59623048    -1.9055E+00
4.63E-01     58.58 58 50 45   0  0
   2      1.46476339    -1.82236905 -1617.59737281    -1.1423E-03
1.57E-04     63.05  0  0  0   1  1
   3      1.46509730    -1.82272189 -1617.59772565    -3.5284E-04
1.42E-06     67.80  0  0  0   2  2
   4      1.46510409    -1.82272472 -1617.59772848    -2.8312E-06
1.09E-08     72.71  0  0  0   3  3
   5      1.46510501    -1.82272474 -1617.59772850    -2.1830E-08
1.42E-10     77.79  0  0  0   4  4

 - - Continuing with F12/conv. amplitude coupling turned on.

   6      1.46511632    -1.82503709 -1617.60004085    -2.3124E-03
3.83E-04    113.82 27 27 31   1  1
   7      1.46511559    -1.82499860 -1617.60000235     3.8497E-05
5.57E-08    123.68  4  4  4   2  2
   8      1.46511573    -1.82499975 -1617.60000351    -1.1569E-06
8.57E-11    131.60  2  2  2   3  3

 CPU time for iterative RMP2-F12                131.60 sec

 DF-RMP2-F12 doubles corrections:
 - - - - - - - - - - - - - - - -
  Approx.                             TOTAL           A-B            A-A
           B-B
  RMP2-F12/3C(D)                  -0.258048804   -0.223258684
-0.020206705   -0.014583415
  RMP2-F12/3*C(D)                 -0.255773796   -0.222151917
-0.019459662   -0.014162217
  RMP2-F12/3*C(DX)                -0.255767327   -0.222141827
-0.019462726   -0.014162775
  RMP2-F12/3*C(FIX,DX)            -0.278206090   -0.244610605
-0.019790395   -0.013805090


(...)


-- 
Dr Mariusz Radon, Ph.D.
Coordination Chemistry Group
Faculty of Chemistry
Jagiellonian University
ul. Ingardena 3, 30-060 Krakow, Poland
http://www2.chemia.uj.edu.pl/~mradon