NEB Gmax convergence question


Just Got Here
I've been running a few NEB calculations to try and find the MEP and I often get the message

@neb  NEB calculation converged
@neb  However NEB Gmax not converged...Try increasing the number of beads.


I then use the python script available in the following link to take the path and add interpolated frames, and then resubmitted the calculation (doubling the number of beads)

http://verahill.blogspot.com.au/2014/06/580-python-script-interpolate-between.html

So far I've not been able to achieve convergence even when using hundreds of beads within a very small region of the pathway (located around the TS)

I'm wondering what else I can try to achieve it.

Clicked A Few Times
Hi,

Have you ever managed to reach convergence ?

I am experiencing the same issue.

Thank you very much

Forum Vet
I have tried the example h2o inversion in the manual and found there is no problem to run this NEB calculation using NWCHEM6.8.
e.g.,
 Step    Intrinsic E   ...     Minimum E      ...      Walltime
---- -------------- -------------- -------------- --------------
@neb 18 -75.961531 -75.973966 164.5
                                     ok       ok       ok

The last step gets for oKs
     ok       ok       ok       ok  

it,converged=                   ... T
@neb NEB calculation converged

According to the manual, NEB in NWCHEM can be used to optimize MEP at all levels of theory, e.g., SCF, HF, DFT, PSPW, BAND, MP2, RIMP2, CCSD, TCE, etc., and there is the other way for this in NWCHEM, i.e., zero string temperature method, also at all levels of theory, both of which, unlike IRC, do not need the knowledge of the structure of a transition state. There are two examples to use the above two methods together to find an MEP in QA.

Forum Vet
I have tried QA test neb-isobutene, an example to use NEB and zero string temperature method together to optimize the MEP of the rearrangement of isobutene, but got different result from those of the test. This test does not get the four oKs at all for me with the original setting, but the test got one.
I get
...
it,converged= 10 F
@neb NEB calculation not converged
@neb Try increasing number of steps or...Try increasing the number of beads.

Task  times  cpu:      613.8s     wall:      852.1s

The test got
@neb NEB calculation not converged

Task  times  cpu:      299.5s     wall:      433.8s
but the walltime already reached 14323.1s.

This test cannot converge only through the increase of NEB steps, perhaps the initial geometry or other setting in the input is unsuitable.

When the NEB input like H2O is used, it does not converge at the 30th step, but gets 3 oks, perhaps still requiring a lot of steps to achieve 4 because of the fluctuating optimization process.

Forum Vet
Now I am trying QA test neb-fch3cl and find NEB cannot converge with the original input. After I increased the NEB steps to 30, it got two oKs. This test contains solvation model.
This test cannot converge only through the increase of NEB steps, perhaps the initial geometry or other setting in the input is unsuitable.
I will put the results on Github, because the log files are too big.

Forum Vet
Thanks Dr. Ebylaska a lot for his explanaition of the reason of the nonconvergence of NEB and the Python code possibly to overcome this.
Zero string temperature method has a bug for the example HCN-HNC owing to the recent modification.
Thanks very much for the immediate fixation of this from Dr. Edoapra and the verification from Dr. Ebylaska.

I will try these.

You can find these on Github.

Clicked A Few Times
Hi,

Thank you very much for your time and very detailed replies !
Unfortunately, I am afraid my simulation is a QM/MM with NEB, which make my problem very different. Do you Dr. Edoapra and Dr. Ebylaska might be able to help me ?

Thanks


Forum >> NWChem's corner >> Running NWChem