CASINO manual - Theory of Condensed Matter

• By default, gaussian uses spin configurations (combinations of Slater determinants) in a
CASSCF calculation. It is best to converge the CASSCF state that you want using this option.
However, for input to the QMC code, the wave function must be in terms of Slater determinants.
In principle, the ‘SlaterDet’ option to CASSCF will do this, but I never succeeded in getting it
to work. Instead, specifying IOp(4/21=10) does the trick, as does IOp(4/46=3).
• For large CASSCF calculations on the alpha cluster Columbus, the diagonalization method must
be changed by specifying IOp(5/51=1).
• In large CASSCF calculations where very many determinants are involved, gaussian currently
only prints the first fifty determinants in the expansion (those with the largest coefficients).
The significance of the truncation may be judged by looking at the sum of the squares of the
coefficients that gaussiantoqmc outputs when it reads the wave function. Unfortunately, this
truncation prevents a full ‘HF test’ (i.e., running the wave function in VMC without a Jastrow
factor and checking that the result agrees with that of gaussian), but the energy returned by
such a test should be above that reported by gaussian.
• As well as this truncation to fifty determinants, gaussian has a formatting error which means
that if the index number of a determinant is greater than 99,999 then it is replaced by stars and
is thus useless. gaussiantoqmc deals with this by simply throwing away such configurations
which further truncates the expansion.
• gaussian switches to direct mode for large CASSCF calculations and in doing so automatically
stops printing the definitions of the Slater determinants used in the calculation. In order to reconstruct
the wave function we do of course need to know what the Slater determinants are. gaussian
may be persuaded to print them by using IOp(4/46=3) IOp(4/21=10) IOp(4/21=100).
The first two of these both tell gaussian to use Slater determinants (I specified both to be on
the safe side) and the last one in theory tells it to ‘just print the configurations’ although in fact
it still proceeds to do the calculation as well.
• Restarting a CASSCF calculation from a previously converged run fails when symmetry is
switched on. Use ‘Nosymm’ to avoid this problem.
TD-HF and TD-DFT
As far as converting the resulting wave function for use in casino is concerned, these two methods
are no different to CIS apart from the issue of normalization. In CIS, the default output is normalized
so that the sum of the squares of the coefficients is equal to unity. In a TD-HF or TD-DFT calculation
(which involves solving a non-Hermitian eigenvalue problem) a different scheme is used which
essentially means that the sum of the squares of the coefficients is arbitrary.
It should also be noted that gaussian cannot do gradients within TD-DFT yet and so cannot relax
excited states.
109