Example: COSMO: HCl¶
Computing solvent effects, with the COSMO model, is illustrated in the HCl example.
After a non-solvent (reference) calculation, which is omitted here, two solvent runs are presented, with somewhat different settings for a few input parameters. The block key Solvation controls all solvent-related input.
All subkeys in the SOLVATION block are discussed in the User’s Guide. Most of them are rather technical and should not severely affect the outcome. Physically relevant is the specification of the solute properties, by the SOLVENT subkey: the dielectric constant and the effective radius of the solvent molecule.
Note that a non-electrostatic terms as a function of surface area is included in the COSMO calculation, by setting the values for CAV0 and CAV1 in the subkey SOLVENT of the key SOLVATION. In ADF2010 one should explicitly include such values for CAV0 and CAV1, otherwise this non-electrostatic term will be taken to be zero, since the defaults have changed in ADF2010.
A rather strong impact on the computation times has the method of treating the ‘C-matrix’. There are 3 options (see the User’s Guide): EXACT is the most expensive, but presumably most accurate. POTENTIAL is the cheapest alternative and is usually quite adequate. EXACT uses the exact charge density for the Coulomb interaction between the molecular charge distribution and the point charges (on the Van der Waals type molecular surface) which model the effects of the solvent. The alternatives, notably ‘POTENTIAL’, use the fitted charge density instead. Assuming that the fit is a fairly accurate approximation to the exact charge density, the difference in outcome should be marginal.
$ADFBIN/adf << eor TITLE HCl(1) Solv-excl surfac; Gauss-Seidel (old std options) SYMMETRY NOSYM ATOMS Cartesian H 0.000000 0.000000 0.000000 R=1.18 Cl 1.304188 0.000000 0.000000 R=1.75 END Fragments H t21.H Cl t21.Cl End SOLVATION Solvent epsilon=78.8 radius=1.4 cav0=1.321 cav1=0.0067639 SurfaceType esurf DivisionLevel ND=4 min=0.5 Ofac=0.8 ChargeUpdate Method=Gauss-Seidel DiscAttributes SCale=0.01 LEGendre=10 TOLerance=1.0d-2 SCF Variational C-Matrix Exact END NOPRINT Bas EigSFO EKin SFO, frag, functions EPRINT SCF NoEigvec END END INPUT eor rm TAPE21 logfile
In the second solvent run, another (technical) method is used for determining the charge distribution on the cavity surface (conjugate-gradient versus Gauss-Seidel in the previous calculation), and the POTENTIAL variety is used for the C-matrix handling. The results show that it makes little difference in outcome, but quite a bit in computation times.
$ADFBIN/adf << eor TITLE HCl(9) NoDisk and Cmatrix potential FRAGMENTS H t21.H Cl t21.Cl END ATOMS Cartesian H 0.000000 0.000000 0.000000 R=1.18 Cl 1.304188 0.000000 0.000000 R=1.75 END SOLVATION Solvent epsilon=78.8 radius=1.4 cav0=1.321 cav1=0.0067639 SurfaceType esurf DivisionLevel ND=4 min=0.5 Ofac=0.8 ChargeUpdate Method=conjugate-gradient SCF Variational C-Matrix POTENTIAL END NOPRINT Bas EigSFO EKin SFO, frag, functions EPRINT SCF NoEigvec END END INPUT eor