ERROR: non-positive definite matrix in Cholesky decomposition (routine: usad)

[JakubV]

Dear Turbomole Experts,

I wanted to compute NUHFI with TM default suggested ECP and basis sets def2-QZVPP for N,H,F and I atoms.
For uranium, I downloaded basis set "HGBSP3-9" from https://www.basissetexchange.org/

(there is no basis set for uranium above triple zeta in the standard library, sadly. HGBSP3-9 is possibly overkill - much above def2-QZVPP level and used for all-electron calculations..., but I could not find def-QZVPP for ECP calculation)

Somewhere after I decided EHT initial guess it gave me an error message about negative eigenvalues in Cholesky decomposition.
What should be a possible workaround, please?

0

 NUMBER OF ELECTRONS IN YOUR MOLECULE IS   74


 AUTOMATIC OCCUPATION NUMBER ASSIGNMENT ESTABLISHED !
 FOUND CLOSED SHELL SYSTEM !
 HOMO/LUMO-SEPARATION :  0.000171
 ORBITAL  SYMMETRY      ENERGY     DEFAULT
 (SHELL)    TYPE                 OCCUPATION
    34    34a          -0.39807      2
    35    35a          -0.36079      2
    36    36a          -0.33264      2
    37    37a          -0.24197      2
    38    38a          -0.24179      0
    39    39a          -0.24024      0
    40    40a          -0.23511      0

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!! WARNING ! HOMO/LUMO-SEPARATION LOWER THAN 0.05 AU !!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

 DO YOU ACCEPT THIS OCCUPATION ?  DEFAULT=y

y
 ERROR: non-positive definite matrix in Cholesky decomposition (routine: usad)
    of overlap matrix in project.f

MODTRACE: no modules on stack

 negative root in usad, see output
 ehtinp ended abnormally
 ehtinp ended abnormally
program stopped.

Best regards,
JakubV

[antti_karttunen]

Hi,

For uranium, the def-TZVPP basis set has ECP of 60 electrons, leaving 32 electrons in the valence space. This would mean that your valence space is in terms of orbitals:

s: 5s+6s+7s
p: 5p+6p+7p
d: 5d+6d
f:  5f

In the def-TZVPP basis set, you have 10 s functions, 9 p functions, 5 d functions, 4 f functions, and 3 g polarization functions. So I would say that at least on paper the basis set sounds more like a QZVPP basis set. Of course some higher angular momentum polarization functions could be added for example from cc-pVQZ-PP. cc-pVQZ-PP could also be an alternative basis set to consider (although as a generally contracted basis set it will be computationally more costly). It is available here in TM format: http://www.grant-hill.group.shef.ac.uk/ccrepo/

Best,
Antti

[uwe]

Hello,

if you use define to prepare the input, it will most likely assign ECPs to the default basis set def-SV(P) automatically.

If you then switch to an all-electron basis set, please make sure to remove the ECPs before continuing. Not sure if this is related to the error you get, but I  ran into this myself some time ago...