Transition state searches not converging after many attempts

[Peter Walton]

Hi all,

I have been running a series of transition state searches on a variety of systems and have had problems with convergence on a number of occasions. 

An example input file is as follows:

echo

START

memory 30000 mb

charge 1

geometry

N        1.2784000000      2.2340000000      0.7657000000

C        0.0386000000      1.6875000000      0.9700000000

C       -0.1139000000      0.3090000000      0.7239000000

C        0.8954000000     -0.3942000000      0.0964000000

C        2.0991000000      0.2610000000     -0.2347000000

C        2.2729000000      1.5715000000      0.1355000000

C        0.7720000000     -1.8641000000     -0.2404000000

O        1.6653000000     -2.4630000000     -0.7806000000

O       -0.4055000000     -2.3530000000      0.1254000000

C       -0.6566000000     -3.7737000000     -0.1668000000

C       -2.0214000000     -4.1134000000      0.3814000000

C       -1.1500717721      2.5750972948     -0.5863924494

F       -2.3844232408      2.1244935510     -0.5768453184

F       -1.1007652500      3.8895572517     -0.4616266341

H       -0.5164832712      2.1701916866     -1.3732619684

H       -0.6035000000      2.2162000000      1.6628000000

H       -1.0362000000     -0.1790000000      1.0057000000

H        3.1909000000      2.1193000000     -0.0261000000

H        1.4336000000      3.1961000000      1.0439000000

H        2.8895000000     -0.2774000000     -0.7390000000

H        0.1424000000     -4.3522000000      0.2962000000

H       -0.5918000000     -3.8970000000     -1.2479000000

H       -2.0682000000     -3.9409000000      1.4576000000

H       -2.2332000000     -5.1692000000      0.2018000000

H       -2.8018000000     -3.5271000000     -0.1059000000

end

basis

 * library 6-311+G(2d,p)

end

dft

 xc b3lyp

 odft

 mult 2

 iterations 300

 direct

 grid huge

end

driver

 xyz geo-opt

end

stepper

 TS

 maxiter 1000

 track nmode 1

end

task dft saddle

I have tried increasing the size of the integration grid and adjusting the input geometry to ensure the system is not in a local minima yet when the TS search calculation is finished the converged geometry is not the exact transition state as there is more than one imaginary frequency. This has been the case after up to 14 transition state searches in some cases and so i was wondering if there was anything you could suggest to help find the final transition state geometry?

Kind Regards

Peter

[Edoardo Aprà]

These are the changes that should get you on a TS (after removing the stepper input section)

basis spherical
 * library 6-311+G(2d,p)
end

...

task dft freq

driver
 inhess 2
 maxiter 80

end

task dft saddle

[Peter Walton]

Thank you very much for your suggestion. I ran this calculation over the weekend and I'm just checking the final geometry with a frequency calculation now.

[Peter Walton]

Hi there,

Many thanks for your help already with this issue. After multiple attempts with this altered input file it still fails to converge and I am still seeing multiple imaginary frequencies despite three TS searches. Do you have any other ideas that may help?

Many thanks 

Peter

[Edoardo Aprà]

Increasing the accuracy of the integrals with the line "tolerances acccoul 14" makes the calculation more numerically stable.

https://nwchemgit.github.io/Density-Functional-Theory-for-Molecules.html#tolerances-screening-tolerances

The attached input works using NWChem 7.0.2. If you still cannot converge to the transition state, please post the output file.

...

 grid huge
 tolerances acccoul 14
end

[Peter Walton]

Hi Edoardo,

Unfortunately after attempting this alteration to the input file I am still struggling to converge to the transition state. Below is a link to download the output file as requested. 

Kind Regards

Peter

https://drive.google.com/file/d/1CoihJivMpQz7QmB5pnp_3eqjo0ELPpgD/view?usp=sharing

[Edoardo Aprà]

Could you please describe the results of your analysis of this output file?

You seem to have started from a TS (1 negative frequency) and the TS search did not change the geometry too much.

[Peter Walton]

Hi Edoardo,

It seems that my frequency calculation on both the input and output geometries show that there is more than one imaginary frequency. I have attached a link of the frequency calculation done on the input geometry and you can see the multiple imaginary frequencies. Perhaps the freq calculations are not being done correctly?

Kind Regards

Peter

https://drive.google.com/file/d/1rUs6im7Rn7RW7_1u4O51wNsD_qiXSYUh/view?usp=sharing

[Edoardo Aprà]

Here are two input files that gets a TS with one negative frequencies.

The first one used the cheaper def2-svp basis set.

The second uses as starting geometry the TS found with the def2-svp basis, but uses (as in your original input) the 6-311+G(2d,p) basis

[Sibo Lin]

I found the following code to work in NWChem 6.8 and give one imaginary frequency without having to switch to huge grids or smaller basis sets. The key is first doing a constrained optimization in zmatrix coordinates, then doing an unconstrained cartesian saddle optimization. Kudos to Edo for helping me figure out how to set all of that up with one input deck. This boilerplate procedure has found a few TSs for me now, with the exception of some beta-hydride elimination TSs, which seem to require a very small sadstp (0.01 works, but 0.03 is too big) and many iterations during the cartesian saddle optimization. Hope it works for you! - Sibo

echo

START sadts

memory stack 200 mb heap 1000 mb global 2800 mb

basis spherical

* library 6-311+G(2d,p)

end

dft

xc b3lyp

odft

mult 2

iterations 300

direct

grid fine

end

geometry adjust

  zcoord

    bond 2 12 r constant

  end

end

driver

  maxiter 200

end

task dft optimize

driver

  maxiter 0

  xyz optConst

end

task dft optimize ignore

geometry noautoz

  load optConst-000.xyz

end

driver

  clear

  maxiter 200

  #sadstp 0.01

  xyz optSad

end

task dft saddle

task dft freq numerical