Tips for (TS) optimization

Go back to Main Page, Group Pages, Feliu Maseras and Group, Computational Resources, Scripts, The tips only work in Gaussian 03/09

Internal Coordinates

Keywords "nosymm" and "cartesian" may help you to avoid the horrible "error in internal coordinates"

Also, you can go on cesca and try version b, c, d, which apparently defines internal coordinates slightly differently.

If you use keyword "cartesian", set "maxcycles=100" because at 101 it starts going crazy for no reason. Then, wait and restart the job.

If you need "cartesian" but also a constrained optimization, a useful tip is to add a -1 in front of frozen atoms:

H -1 0.000 0.000 0.000

Standard Optimization

Remebmber to use GaussView's Force Field optimization before going further if your structure is sketchy

If it is still "ugly", you may use your favourite DFT method with a reduced basis set such as STO-3G

For most SCF convergence problems, scf=qc is what you need. G09.d1 on cesca has a new scf=yqc.

Now, on g09.d1 version, the option XQC is available. It runs a normal SCF and if it fails, it starts to do a QC type of SCF. It is useful if in some steps of the optimization the calculation fails but in other goes well because it reduces the computational time considerably.

TS search

Useful Keywords

You will have notice that this is the hardest. Some keywords that helped me:

maxstep=X to see where the algorithm "lost" the TS, and restart with calcfc. Very slow method. Alternatively, ask me (cgoehry@iciq.es) my script to see in molden which atoms are concerned by the Eigenvalue-following TS search

vshift=X in case the optimization goes up and down very quickly (with metal atoms).

Useful tricks

The calculation of force constants (calcfc) is extremely demanding. So you can do it, or use one of the following ways around (use checkpoint files):

• Compute the force constant at a lower level, and use it at your favourite level. Ex:

     #p "method" freq=noraman  STO-3G
     --Link1--
     #p opt=(TS,readfc,noeigentest)  "method" 6-31G*

• Use the QST3 algorithm. Remember in Gaussian you have to put in order reactant, product and THEN TS.

• A possibility if you want to "renew" your force constant set:

when you find what a good guess for your ts, export it (ts.xyz), optimise normally with maxstep=10,maxcycles=10 or so, you'll go down (product.xyz), then "mirror" the change in coordinates. Ex:

     paste product.xyz ts.xyz > temp.xyz
     awk '{print $1,2*$2-$6,2*$3-$7,2*$4-$8}' temp.xyz > reactant.xyz

Then, use those structure within a QST3 scheme. This trick also works if the TS search "looses" your TS.

• One common problem in big molecules is the slowly convergence at final points of TS (when you know that it is your TS but the geometry oscillates around the saddle point). I usually restart

the calculation using IOP(1/8=x) where x can go from 1 to 10 normally. It reduces the stepsize in the optimization (the default is 0.1 Bohr or Radian, and the modified is stepsize=0.01·x Bohr or Radian). I usually select 4 or 5.