I am about to compute reaction steps for a series of compounds using DFT. A key point in our discussion is the dependency of the experimental rate of reaction w.r.t. solvent used, so it is very important that we incorporate solvation effects in our computational models.

I am aware of the best practices in choosing functionals, dispersion corrections, basis sets, etc.. By reading a little I decided on using the SMD model for implicit solvation. I may use either ORCA or NWChem.

My questions are:

  1. a. Should geometries be optimised in solvent, in the gas phase or both? b. In case I use only gas phase geometries, should I calculate frequencies using implicit solvation (I doubt) or can I use the calculated free energy of solvation (done in a single point calculation) to correct the gas phase free energy?
  2. a. In case I decide on using (one or two) explicit solvent molecules to better model the system, are there best practices on using them together with an implicit solvation model (in terms of calculating free energy of solvation and in deciding when it has converged/is adequate)? b. What about the points addressed in 1. (gas phase versus condensed optimisation and frequency calculations) for this case?

References from literature are greatly appreciated. I am looking forward to hear from you!

  • 2
    $\begingroup$ These questions also can depend on the system size. If you are pushing the limits of a quantum calculation you might be a bit more limited. If the system is somewhat smaller you can explore all these options, or perhaps using you could use a smaller model system to get a feel for how the system is behaving. This can help build your intuition as well for determining what is appropriate. $\endgroup$
    – brose
    Jan 31, 2017 at 0:07
  • 4
    $\begingroup$ I have written about it here: dx.doi.org/10.1039/C5CP00628G $\endgroup$
    – Jan Jensen
    Jan 31, 2017 at 8:57
  • 1
    $\begingroup$ @JanJensen thanks for sharing, this has truly clarified my doubts. Even though all answers where very useful, I wish I could accept your comment as the definite one. $\endgroup$ Feb 1, 2017 at 16:08

2 Answers 2


1) Yes, geometries must be calculated both in the gas phase and in the solvent. To analize your results in the solvent you need something to compare that's why the gas phase calculation is always needed and often is considered "the only source of truth".

Frequencies must be calculated both in the solvent and in the gas phase as well, on the respective converged geometries.

2) I never worked with explicit solvent but I feel that the use cases of two methods are not overlapping. When performing calculations on a multi step reaction usually you don't have enough computational resources to calculate all the solvent molecules needed. This is why people use implicit solvent calculations. If you want to benchmark your solvent calculations, compare it with the gas phase ones and evaluate wether there is some improvement in the replication of experimental data.

  • 1
    $\begingroup$ Thanks for your answer! A few points: a. how bad are going to be the results if I optimise geometries only in the gas phase and do single point calculations using implicit solvation? b. could you provide a reference to an article using the strategy you described? $\endgroup$ Jan 30, 2017 at 18:56
  • 4
    $\begingroup$ I'll be honest, very little will change in the final reaction profile. However it is advised to reoptimize the geometry in the solvent starting from the already converged structure in the gas phase, since it will not be a long computation. I don't have any specific reference for this, but you can search a supporting information of a relevant article and check their procedure. $\endgroup$
    – user288431
    Jan 30, 2017 at 19:06

For explicit solvent interactions I would recommend for example FDE (Frozen-Density-Embedding) or embedding techniques in general.

In FDE the subsystems (solvent/solute molecules) "see" only the electron density of the other subsystems.

One problem you should consider when calculation with explicit solvent interactions is that molecules move all the time.

So to really get accurate measurements for that case you should consider doing an MD simulation to get a couple of snapshots for your QM-Level energy calculations.

Since this will get extremely expensive for your case, I wouldn't recommend using explicit interactions.

If you omit explicit interactions, consider using a Continuum Solvation model like COSMO/PCM or force field methods.

Optimize your geometries in gas phase (so you have a converged geometry already), but for calculations, optimize them again with solvent environments.

The change might not be big, but can be significant nonetheless.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.