EXCESSIVE NUMBER OF ITERATIONS IN -NTNDVD- [GAMESS]

[jonatha...@colorado.edu]

Hello,

I'm encountering an error in gamess while performing my dynamics calculations, and I was hoping someone knew how to resolve it. I received the following error during a NACME calculation using gamess-us "EXCESSIVE NUMBER OF ITERATIONS IN -NTNDVD- DURING AUGMENTED HESSIAN MATRIX DIAGONALIZATION. STOP". However, after receiving this error the calculation proceeds normally, the MCSCF calculation converges, etc. I've tried searching the internet and the gamess input for a way to increase the number of iterations so I don't receive this error, but I haven't had any luck.

I looked through the run-gamess.pl file and noticed that this error kills my dynamics calculation. In the run-gamess.pl file there was a comment that whoever wrote the script was considering commenting out this error criteria. I was wondering if I commented out this error criteria I would receive reliable results. My dynamics calculations thus far proceed normally after I tried commenting this error out, but I am unsure if the data I will receive will be correct. Attached is my TRAJ folder, which contains the gamessinput.log containing the error.

Thanks,
Jon Church

[Mario Barbatti]

Dear Jon,

I asked Aaron West about this problem (He made the NX/GAMESS interface). He wrote me the message below.

I hoope it helps,

Mario

"EXCESSIVE NUMBER OF ITERATIONS IN -NTNDVD- DURING AUGMENTED HESSIAN MATRIX DIAGONALIZATION. STOP"

I have seen this error many times and know exactly what it is.

There is no input for it, and an input would probably not help anyways.

The NEWTON-X script kills off the run if it finds this message.  That is correct.

The run should not be allowed to proceed.

(GAMESS should not go on to the NACME, so I will have to talk with someone about killing off that run.)

The error refers to a failure to solve the nonlinear equations of the MCSCF run.

The linear equation contains the electronic MCSCF orbital Hessian matrix and orbital gradient vector.

It uses a Davidson converger/solver (i.e. NTNDVD is the routine name).

This converger is the best choice for GAMESS for:

--parallel runs

--runs that require a SA-MCSCF Lagrangian (which is required for the NACME calculation).

In this case, we do not have a better MCSCF converger in GAMESS.

If one encounters this error during the dynamics after some time steps, then the active orbitals might be drastically changing.  One could change the active space size to be larger (if that is computationally affordable).

I would recommend looking at the active orbitals for the given geometry where the problem/error occurs.  If the active orbitals are different from the initial active orbitals, then one can probably not do much else here.  If the active orbitals are similar to the initial active orbitals, then I would be surprised.

In the case where the active orbitals are similar to the initial active orbitals, one could try to increase the number of iterations allowed inside of the GAMESS code itself.  However, I believe that changing this will not help much.

Open the mcscf.src file.

Search on "CALL NTNDVD".

Above each occurrence/entry in the last search, there should be a "MXIT  = 200".

One could change this to be more iterations.

[Felix Plasser]

Hi, the other option is always to see if one of the other quantum chemistry programs happens to be more stable for your specific problem.

There is of course the free Columbus program, which was specifically optimized to run smoothly and efficiently with Newton-X
http://www.univie.ac.at/columbus/

And in principle also Gaussian should let you run CASSCF dynamics but there is not so much experience with that.

Felix

--
You received this message because you are subscribed to the Google Groups "Newton-X" group.
To unsubscribe from this group and stop receiving emails from it, send an email to newtonx+u...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Mario Barbatti]

Hi,

I just received another message from Aaron. It seems that there was a mistake in his reply before. He wrote:

I must apologize since I made a mistake in my reply.

I realize now that I was a bit hasty in my reply.

The NX script should NOT kill off the trajectory when it finds this message.

I realize now that there two cases for this "NTNDVD error".

Case 1:

The error occurs at some early iteration in the MCSCF.

In later iterations, the error vanishes and the MCSCF converges normally.

Case 2:

The error occurs at some early iteration in the MCSCF and still has issues.

In that case, I believe another error will appropriately kill off the trajectory (i.e. ENERGY DID NOT CONVERGE error out present in NX script).

I reviewed my source files and find that I did remove this error out.

Again, I apologize for the previous reply with the error.

[jonatha...@colorado.edu]

Hello,

Thank you for your help. I'm trying to run a job with a larger active space and it does seem to be working thus far. However, if I continue to have problems with Gamess, I think I will give Columbus a try.

Thanks,
Jon

[jonatha...@colorado.edu]

Oh, thats great! Increasing the active space really increased the computational cost.