I am optimizing a doublet species on Gaussian using ROm06L.
The optimization converges quickly, but the frequency calculation seems to perform many SCF calculations, which I never observed with unrestricted open shell or restricted closed shell systems (I only observed 1 SCF for single point electronic energy of the optimized structure.
=> see attachment with the simple example of the methyl radical, where the optimization converged and terminated at the blue arrow on the picture.
Why is it doing all these SCF calculations? and how many are needed for the freq calculation to terminate ?
Also, if it ends up not converging, or costs too much: are there any other ways of proving that opt reached an actual minimum and to extract the thermochemistry of the species?
Thanks in advance
ROHF or ROKS doesn't supports analytical Hessian in Gaussian, therefore when doing frequency analysis task, Gaussian automatically calculates Hessian by numerical difference technique based on analytical first-order derivatives. SCF calculation is needed each time of calculation of the first-order derivatives.
I strongly discourage you from using restricted open-shell form due to many reasons: result is usually worse than unrestricted open-shell, no analytical Hessian is available, SCF convergence is commonly more difficult, evidently higher cost...
I do not know exactly how the analytical Hessian is calculated; the semi-numerical Hessian analysis however distorts each atom of the optimized molecule in x-, y-, and z-direction, respectively, and from each molecular distortion a singlepoint is calculated (thus obtaining the energy).
Example: atom 1 is moved 0.01 angstroms (or so) in x- direction, thus obtaining a slightly varied geometry, of which the energy is calculated. Then atom 1 is moved -0.01 angstroms in x-direction, and so on.
For a 30-atom molecule you get 180 singlepoint geometries (for every x-,y, and z in both directions, makes factor 6). With the energies thus obtained, the Hessian matrix (d2E/dxnxj with xn, xj are the coordinates) is built up.
- Badges
- Science topic