About Foster-Boys orbitals localization method on J-coupling

[marcos87]

Dear developers,

I have a problem when I try calculate J-couplings using localized orbitals (Foster-Boys method) and RPA level approach (or SOPPA level approach) on methane molecule.
I need the analysis about the molecular orbitals contribution on each J-coupling transmition mechanism.
The program doesn´t finish properly because it has problems with start vectors when I use cc-pVTZ basis set.
On the other hand, when I remove the comand '.SOSOCC' the program finish correctly but, of course, I don´t get what I need.

How can I manipulate this variable (start vectors)?


My .dal input is:

**DALTON INPUT
.RUN PROPERTIES
**WAVE FUNCTIONS
.HF
**PROPERTIES
.SPIN-S
*LOCALI
.FOSBOY
*SPIN-S
.SOSOCC
.ABUNDA
0.04
.SELECT
2
1 2
**END OF DALTON INPUT

The output fragment inluding the error message (on the beginning of ABACUS section) is:

.
.
.
.


Old LUGDT file opened in SOSJTR.

** PPST ** NUMBER OF START VECTORS 450
EXCEED THE MAXIMUM ALLOWED NUMBER 400

--- SEVERE ERROR, PROGRAM WILL BE ABORTED ---
Date and time (Linux) : Fri May 20 12:25:56 2016
Host name : nodo8

@ MPI MASTER, node no.: 0
@ Reason: PPST: TOO MANY START VECTORS

>>>> Total CPU time used in DALTON: 23.63 seconds
>>>> Total wall time used in DALTON: 1 minute 15 seconds


QTRACE dump of internal trace stack

========================
level module
========================
10 PPST
9 RSPCTL
8 RSPLEX
.
.
.
.


My english is not good enough. I'm sorry for that.



Thanks for all.

[frj]

I cannot help you with the specific error, but want to point out that spin-spin coupling requires specialized basis sets for any decent accuracy. cc-pVTZ is not one of them, I (of course) suggest the pcJ-n.

[taylor]

First, please post your output file! Do not try to excerpt from it what you think may be important, and you do not need to post the input files because they are reprinted in the output. To some of the experts it may be (I do not say it is the case here) that some problem is apparent much earlier in the calculation.

Second, it seems clear you are blowing some fixed dimension of starting vectors in some part of the SOPPA calculation. It is likely that such a fixed dimension (it is clearly used for a test in PPST) will be defined in a header file in the "include" directory and redefining it larger there and rebuilding the code would hopefully get you past this hurdle.

Third, why do you want to use localized orbitals at all? The "orbital contributions" to a property are not observables and will likely have as much physical meaning/significance as the results of a Mulliken population analysis (say). After all, you could in principle construct a complete basis, and therefore obtain the exact SOPPA result, by putting the basis functions on Mars. Would an orbital-by-orbital analysis of the results tell you anything? No. Yet I repeat, the SOPPA results would be the exact SOPPA solution!

Fourth, I strongly endorse what Frank Jensen said. cc-pVTZ is a good basis for energies, for which it was constructed. It is therefore also pretty good for quantities like molecular gradients and geometries, and harmonic force constants, that are directly related to the energy and its curvature with molecular geometry. Properties related to internal and external electric or magnetic fields and field gradients need more care and attention to the basis, and the regular cc-pVXZ sets will not do in such cases, whether you're calculating the dipole-dipole-quadrupole hyperpolarizability or nuclear spin-spin coupling constants.

Best regards
Pete
P.S. Do not worry about your English: it is fine for communicating what you want. And after all, I was born in Britain, grew up in Australia, and lived nearly twenty years in the US, so my English is terrible!!