I want to optimize a transition state in Q-chem and i was wondering if i can do this in Gaussian first and then read the *.fchk file to speed up my calculation in Q-chem? I found the “SCF_GUESS = READ” is used for a similar way but mostly for commanding two tasks in the same input file. So i found the wrong keyword or i missed the true examples?
Actually, I use the same original structure to optimize the TS in both Q-chem (JOB_TYPE=TS) and Gaussian (opt=(TS, CalcFC, NoEigenTest)), but it takes much more time for Q-chem and even now is calculating. Maybe it is the “CalcFC, NoEigenTest” in Gaussian accelerate its calculation. So is there any keyword in Q-chem can do this?
(1) SCF_GUESS = READ is not going to read anything other than Q-Chem scratch files, and that is what’s used for restart (rather than FChk file).
(2) To get better/faster convergence with TS jobs, try starting with an exact Hessian calculation (JOBTYPE=FREQ) and then reading that into the subsequent TS search by adding GEOM_OPT_HESSIAN = READ to the $rem section.
You can use the utility fch2qchem in the open source package MOKIT to transfer MOs(Molecular Orbitals) from Gaussian → Q-Chem. For example,
fch2qchem h2o.fch
A Q-Chem input file h2o.in will be generated. And a directory containing scratch files will be generated and put into $QCSCRATCH/. Then one can use the following command
qchem h2o.in h2o.out h2o
to submit the Q-Chem job. If the same computational level (theoretical method, basis set, etc) are used in Gaussian/Q-Chem, for Q-Chem the SCF in a HF(Hartree-Fock) calculation will be converged in 1 cycle, and the SCF in a DFT calculation will be converged in 5 cycles. So this utility can be applied to a single-point calculation. Unfortunately, this utility cannot be used to transfer Hessian matrix to help geometry optimization convergence.
The slower convergence for DFT may be the result of different grids. I believe that the default grid in recent versions of Gaussian (e.g., g16) is close to Q-Chem’s EML (99,590), which can be requested in Q-Chem using XC_GRID = 000099000590.
Actually, I have used “GEOM_OPT_HESSIAN = READ” as you suggested and decreased the setting of “SCF_CONVERGENCE” and " THRESH" to optimize my TS structure. The result is when I only use “GEOM_OPT_HESSIAN = READ” , my task hasn’t end and if I further reduce the convergence criteria, the time shrinks a lot (about 46 hours) although still longer than using Gaussian09 (15 hours using same amount of cpu). I should have noticed this sooner.
Maybe if I change the grids, it will be better? I will try this. And I think I also should do the test in which I only reduce the convergence criteria but not read Hessian. But I am not that sure if this is in need.
And also I found the “Displacement” is still “NO” although the task ends normally, should I do something to solve this? Displacement 0.059624 0.001200 NO
Regarding grid, I don’t recommend XC_GRID = 000099000590 unless you are specifically trying to make contact with Gaussian (i.e., reproduce absolute energies), or if you think you that somehow the default grid is problematic. Q-Chem default grids are appropriate for most applications and much faster than unpruned (99,590).
Okay! I may need to compare my results with those in published paper. So I will test and if it only add a bit of time, I think it will be better to do so. Thanks for reminding me.