CASINO manual - Theory of Condensed Matter
CASINO manual - Theory of Condensed Matter
CASINO manual - Theory of Condensed Matter
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
CIS<br />
• One cannot use basis functions containing g and higher basis functions with a pseudopotential<br />
in g94.<br />
• Obviously, gaussiantoqmc needs the molecular orbitals (MOs) produced by the calculation.<br />
It gets these from the formatted checkpoint file which is produced by putting ‘Formcheck=(Basis,MO)’<br />
in the route section <strong>of</strong> the gaussian job. Alternatively, it may be obtained<br />
from the binary checkpoint file (.chk) using the formchk utility: see the gaussian <strong>manual</strong>.<br />
• Many (but not all) <strong>of</strong> the IOps mentioned here are described on gaussian’s website at www.<br />
gaussian.com/iops.htm.<br />
• Performing a ‘HF test’ for an excited state: It is possible to get gaussian to output a breakdown<br />
<strong>of</strong> the energy <strong>of</strong> a CIS excited state which may be compared with the results <strong>of</strong> a determinantonly<br />
VMC run. The key to this is the density used to perform the population analysis and other<br />
post-SCF calculations. By default, gaussian uses the density produced by the original SCF<br />
run. To get the kinetic, nuclear-nuclear potential and electron-nuclear potential energies you<br />
must tell it to use the one-particle CI density via ‘density=RhoCI’. You must also specify the<br />
excited state that you are interested in via ‘Root=N’ in the CIS options. With all <strong>of</strong> this done<br />
properly, gaussian produces some output like:<br />
N-N= 6.9546274D+00 E-N=-2.3781323D+01 KE= 3.3771062D+00<br />
(units <strong>of</strong> a.u.), which is hidden in the density analysis right at the end <strong>of</strong> the output.<br />
• Some trouble has been encountered with the ‘Add=N’ option to CIS (which reads converged excited<br />
states from the checkpoint file and then calculates N more). The IOp alternative which does<br />
work is IOp(9/49=2) (use guess vectors from the checkpoint file) combined with IOp(9/39=N)<br />
(make N additional guesses to those present).<br />
• Using the ‘50–50’ option to CIS to calculate singlet and triplet excitations simultaneously can<br />
cause problems. It appears best to do the singlet (‘singlets’) and triplet (‘triplets’) calculations<br />
separately.<br />
• For QMC, we want the complete CIS expansion. gaussian may be persuaded to output all<br />
excited states with coefficients > 10 −N by using IOp(9/40=N). Typically N = 5 is good<br />
enough. gaussiantoqmc outputs the sum <strong>of</strong> the square <strong>of</strong> the coefficients so that the user<br />
can see how complete the wave function is. (Standard gaussian output has the coefficients<br />
normalized so that the sum <strong>of</strong> their squares for a complete expansion would be unity.)<br />
CISD<br />
Although gaussiantoqmc cannot read a CISD wave function it might be worth mentioning that<br />
IOp(9/6)=N is equivalent to MAXCYCLE=N for such a calculation.<br />
CASSCF<br />
• As described in Foresman and Frisch’s book [22], getting CASSCF to converge for a singlet state<br />
is difficult. The following procedure normally works:<br />
1. Run a ROHF calculation for the lowest triplet state <strong>of</strong> the system and save the checkpoint<br />
file.<br />
2. Run CASSCF for the second triplet state (‘Nroot=2’) taking the initial guess from the<br />
checkpoint file. (gaussian calculates the first and second triplets but converges on the<br />
second.)<br />
3. Run CASSCF for the first triplet (‘Nroot=1’) taking the initial guess from the checkpoint<br />
file.<br />
4. Run CASSCF for the singlet excited state (‘Nroot=2’) taking the initial guess from the<br />
triplet checkpoint file.<br />
5. Finally, run CASSCF for the singlet ground state (‘Nroot=1’).<br />
108