DFT TS search "There is an error related to the specified geometry" after more than 100 iterations

[Francesca Lønstad Bleken]

Hi, 

I am having several TS searches in which the run fails after having run for many many iterations with the following error:

 ------------------------------------------------------------------------

 driver: task_gradient failed                   0

 ------------------------------------------------------------------------

 ------------------------------------------------------------------------

  current input line : 

    21: task dft saddle

 ------------------------------------------------------------------------

 ------------------------------------------------------------------------

 There is an error related to the specified geometry

 ------------------------------------------------------------------------

The TS search runs for many iterations before they suddenly stop with this error message, and I am not able to understand why they stop. Any suggestions on how to solve this problem? I am running with NWChem6.8.1rev133.

I ususally do ts search with restart from a frequency calc that runs without problem, these are the inputs for one such job that fails:

----------------calc1.nw-------------- 

start image14_cc-pVTZ_triplet

title "image14_cc-pVTZ_triplet_freq"

geometry start units angstrom

 load format xyz image14.xyz

end

basis

 * library cc-pVTZ except Mo

 Mo library cc-pVTZ-PP

end

ecp

 Mo library cc-pVTZ-PP

end

set geometry start

charge 0

driver

 tight

 maxiter 300

 xyz #output geometry of each step

end

dft

 direct

 xc b3lyp

 convergence energy 1e-8

 convergence density 1e-7

 maxiter 300

 odft #reduntant if multiplicity !=1

 mult 3

 grid huge

 tolerances tight

end

set int:txs:limxmem               150000000

freq 

 animate

end

----------------end calc1.nw---------------

-------------calc2.nw------------

start image14_cc-pVTZ_triplet

title "image14_cc-pVTZ_triplet_ts"

restart image14_cc-pVTZ_triplet

driver

 inhess 2

end

dft

 convergence energy 1e-7

 convergence density 1e-6

 maxiter 600

 odft #reduntant if multiplicity !=1

 grid huge

 tolerances tight

end

task dft saddle

------------end calc2----------

    25

 geometry

 O                     2.12951705    -0.87904260     1.82555539

 Si                    1.69626819    -0.83238885     0.24492561

 O                    -1.63230713    -0.74112358    -0.81959178

 Si                   -0.96534609    -2.20425342    -0.46420808

 O                    -1.71831946    -2.93825641     0.79875056

 O                     0.59274676    -2.02505371    -0.00905580

 O                     1.02634428     0.59547833    -0.19173089

 H                     3.06725897    -0.80604194     2.00839155

 H                    -2.65935174    -3.09739999     0.70807949

 Mo                   -0.85881328     0.96849896    -0.50464838

 O                     3.08168546    -1.10476323    -0.60315897

 H                     3.06317665    -0.91377082    -1.54221354

 O                    -1.09134178    -3.08674246    -1.84577427

 H                    -0.74914367    -3.98209269    -1.82058262

 C                    -0.80533706     3.33494372     0.68063455

 H                    -0.74030917     2.30249357     1.11987822

 H                    -0.77660455     3.95560021     1.57760936

 H                    -1.79290357     3.48980482     0.23913171

 C                     0.30621998     3.65996983    -0.31009281

 C                     0.22437035     5.09137617    -0.83609447

 H                     0.26031072     2.98237597    -1.18605918

 H                     1.27540708     3.46703675     0.14814923

 H                    -0.72838745     5.27718599    -1.33478766

 H                     1.02115588     5.29143999    -1.55215416

 H                     0.31837519     5.80926066    -0.01942108

--------------end xyz-------------

-------One of the xyz inputs this happens with is:-----------

[Edoardo Aprà]

This could be a SCF convergence problem in the last geometry point.

However, it is hard to tell without looking at the output file.

Could you attach the output file to this forum thread?

[Francesca L. Bleken]

Hi, you are completely right, the SCF failed to converge, but this message was so far up I did not catch it. However, the if restarting the calculation from the last geometry this is not a problem, but it runs for a lot of steps before failing again. Any suggestions on how to improve this when search for TS? It is never a eproblem for ground state optimisations, only for saddle points. These saddle points seem to be quite low and flat, and I often get only one negative frequency both for 
input geometry and final geometry (which fails after some steps) from the TS serach. However the TS serach does not converge.

Best, 

Francesca

[Sibo Lin]

If your output file is on a Linux computer, can you navigate to the directory, run "more [your output file.out]  | grep @" and 
"more [your output file.out]  | grep 'Hessian ei' " and paste their outputs here?

[Francesca L. Bleken]

Thank you for your suggestion, 

Indeed, printing out the eigenvalues is informative. In som of the calculations, I get outputs similar to this (last lines):

Hessian eigenvalues: positive= 158 negative=  1 zero=  0

 Hessian eigenvalues: positive= 158 negative=  1 zero=  0
 Hessian eigenvalues: positive= 158 negative=  1 zero=  0
 Hessian eigenvalues: positive= 158 negative=  1 zero=  0
 Hessian eigenvalues: positive= 158 negative=  1 zero=  0
 Hessian eigenvalues: positive= 158 negative=  1 zero=  0
 Hessian eigenvalues: positive= 158 negative=  1 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 158 negative=  1 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 156 negative=  3 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 156 negative=  3 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0
 Hessian eigenvalues: positive= 157 negative=  2 zero=  0

and (last lines)

@   42   -1295.84220668 -1.3D-05  0.00147  0.00020  0.01237  0.04286  10455.9
@   43   -1295.84222342 -1.7D-05  0.00147  0.00020  0.01603  0.05490  10682.7
@   44   -1295.84230481 -8.1D-05  0.00117  0.00017  0.02222  0.05684  10911.9
@   45   -1295.84237810 -7.3D-05  0.00096  0.00016  0.01577  0.06938  11142.4
@   46   -1295.84238086 -2.8D-06  0.00097  0.00015  0.01180  0.03941  11368.3
@   47   -1295.84242058 -4.0D-05  0.00079  0.00015  0.01617  0.06554  11574.2
@   48   -1295.84259831 -1.8D-04  0.00067  0.00013  0.01617  0.05886  11781.8
@   49   -1295.84270738 -1.1D-04  0.00055  0.00012  0.01491  0.06620  12012.2
@   50   -1295.84275893 -5.2D-05  0.00040  0.00011  0.01789  0.06309  12221.3
@   51   -1295.84277523 -1.6D-05  0.00051  0.00011  0.01927  0.05516  12449.8
@   52   -1295.84291475 -1.4D-04  0.00045  0.00012  0.02489  0.08474  12658.5
@   53   -1295.84302710 -1.1D-04  0.00047  0.00011  0.01915  0.07508  12868.3
@   54   -1295.84306656 -3.9D-05  0.00037  0.00011  0.02924  0.09289  13076.7
@   55   -1295.84310206 -3.5D-05  0.00043  0.00011  0.02281  0.08856  13339.1
@   56   -1295.84311597 -1.4D-05  0.00032  0.00010  0.01611  0.06324  13580.0
@   57   -1295.84313081 -1.5D-05  0.00041  0.00010  0.01830  0.06423  13780.0
@   58   -1295.84317055 -4.0D-05  0.00032  0.00010  0.01557  0.04920  14042.6
@   59   -1295.84319455 -2.4D-05  0.00039  0.00010  0.01285  0.04127  14217.7

So, two negative eigenvalues (although it has one at some iterations). However, I do get this from a subsequent freqency calc on the final geometry:

          -------------------------------------------------
          NORMAL MODE EIGENVECTORS IN CARTESIAN COORDINATES
          -------------------------------------------------
             (Projected Frequencies expressed in cm-1)

                    1           2           3           4           5           6
 
 P.Frequency     -648.35       -0.00       -0.00        0.00        0.00        0.00

And visualization of the first mode looks very much like the correct ts.

Any suggestions on why the TS search gives two negative frequencies while the frequency calc only gives one?

Best regards, 

Francesca

--
You received this message because you are subscribed to a topic in the Google Groups "NWChem Forum" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/nwchem-forum/IKNxl2HlCSI/unsubscribe.
To unsubscribe from this group and all its topics, send an email to nwchem-forum...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/nwchem-forum/30ca1d81-41c8-4a8c-8ac0-d060aa27f94an%40googlegroups.com.

[Sibo Lin]

The vibrational freq calculation appears to be in Cartesian coordinates. The default coordinate system for geometry optimizations in NWChem is internal coordinates (bonds, angles, dihedrals), which usually is more efficient at finding minima than Cartesian coordinate geometry optimizations. But I have seen some transition state (saddle) geometry optimizations go off the rails in internal coordinates, but succeed with Cartesian coordinates. I have also read that Cartesian coordinates are used for saddle optimizations in other DFT programs, so I recommend you try using Cartesian coordinates. Generate a .xyz file of a guess structure. Then run the following code:

----

[first all the lines specifying job name, memory, basis sets, functional, charge, multiplicity]

geometry noautoz

  load TSguess.xyz

end

driver

  clear

  maxiter 200

  #sadstp 0.01

  xyz optSad

end

task dft saddle

task dft freq numerical

----

That said, I tried visualizing the geometry you provided into the first post, and it's not obvious to my eye what bond is supposed to be breaking in the hoped for transition state. If your guess geometry is very far from the saddle point, I don't think the saddle search algorithm will get you there. If you approximately know the distance of the key bond in the TS (from literature calculations, or previous TS calculations of your own), you can try doing a geometry optimization while holding the key bond at the desired distance to get a good guess geometry for the saddle search algorithm. See the last post of https://groups.google.com/g/nwchem-forum/c/-UHkkrLuX8o/m/H5aqn3_KAQAJ for relevant code.