I think I have answered my own question. After addressing the geometric issues with my system we are finally getting SP convergence without the need to "step" my calculations for memory purposes.

I will run qc conver=5 on my larger system for comparison and let you all know how that turns out. I did some test runs on a smaller SWNT (much smaller compared to my 1400 atom sheets). When convergence is set below 6 there are calculation steps eliminated. How much of an impact this has on my end results I cannot yet say but I will share.

Dear Linda,

Would you mind share your method to circumvent the problem. I have a system, where single point calculation fail and so far dont know what to do with it.

Thank you!

-Mer-

Sure..Are you running your calculation on graphite/graphene? If so, how is set up? (ie. saturated with hydrogen, terminated with hydrogen at the edges or no hydrogen at all)

Oh, thank you.

No, i just run on some small molecule, similar to [Ru(bpy)3]2+. The opt and freq calculation terminated normally. I just want to do a single point calculation to add solvent effect.

But SP calculation failed to converge.

Thank you!

The issues on failed convergence depend on the system you are working with, and how they are set up. If you aren't running a large system like I was, the way to solve the problem will be different. Do you mind posting your input job? Just to confirm, did you add your solvent for both sets of calculations or just the single point job?

Thank you. Bellow is my input script.

%nprocshared=8

%mem=8GB

%chk=Rutpy.chk

# ub3lyp/genecp scrf=(iefpcm,solvent=acetonitrile) nosymm

Rutpy

2 3

Ru 0.00000000 0.00000100 -0.00000100

C 1.77546300 1.66061400 -1.68152200

C -0.43951500 2.18584200 -2.21333800

C 2.23592800 2.58707700 -2.61961800

C -0.03973300 3.12313600 -3.16241600

H -1.48986800 1.99647800 -2.02159800

C 1.32330200 3.32720000 -3.36903900

H 3.30009500 2.73182800 -2.76617900

H -0.78657600 3.67580700 -3.72203000

H 1.67337600 4.04901300 -4.09991500

C 2.67506100 0.83475000 -0.84526000

C 2.67507200 -0.83474700 0.84522300

C 4.07290900 0.85059700 -0.86131700

C 4.07291800 -0.85059200 0.86126800

C 4.76800500 0.00000400 -0.00002800

H 4.61685600 1.50919700 -1.52820400

H 4.61687600 -1.50919200 1.52814900

C 1.77548200 -1.66062000 1.68149000

C -0.43948900 -2.18585400 2.21332800

C 2.23596000 -2.58708700 2.61957400

C -0.03969400 -3.12315500 3.16239300

H -1.48984500 -1.99648800 2.02160300

C 1.32334400 -3.32721700 3.36899900

H 3.30013000 -2.73183400 2.76612300

H -0.78653000 -3.67582900 3.72201400

H 1.67342900 -4.04903200 4.09986900

C -1.77549200 -1.68149100 -1.66061000

C 0.43947800 -2.21335500 -2.18582800

C -2.23597200 -2.61959200 -2.58706200

C 0.03968000 -3.16242500 -3.12312300

H 1.48983400 -2.02162800 -1.99646800

C -1.32335900 -3.36902800 -3.32718300

H -3.30014100 -2.76614300 -2.73180400

H 0.78651400 -3.72205300 -3.67579300

H -1.67344500 -4.09990400 -4.04899100

C -2.67507500 -0.84523700 -0.83472500

C -2.67505900 0.84525600 0.83476300

C -4.07292300 -0.86127300 -0.85057100

C -4.07290500 0.86131000 0.85062100

C -4.76800500 0.00002200 0.00002800

H -4.61688100 -1.52814900 -1.50917400

H -4.61685100 1.52819400 1.50922700

H -5.85306800 0.00003200 0.00003400

C -1.77545300 1.68150500 1.66064000

C 0.43952700 2.21333600 2.18584300

C -2.23591600 2.61960000 2.58710400

C 0.03974700 3.16240100 3.12315100

H 1.48987900 2.02159400 1.99647800

C -1.32328800 3.36902000 3.32722300

H -3.30008300 2.76615900 2.73185900

H 0.78659300 3.72201200 3.67582100

H -1.67336000 4.09989100 4.04904200

N 0.43225900 1.46690100 -1.48536300

N 2.02288700 0.00000100 -0.00001700

N 0.43227500 -1.46690800 1.48534700

N -0.43225000 1.48536100 1.46690500

N -2.02288700 0.00000600 0.00001500

N -0.43228400 -1.48535900 -1.46689400

H 5.83800500 0.00000203 -0.00003096

Ru 0

LANL2DZ

****

C H N 0

6-31G*

****

Ru 0

LANL2DZ

I optimized and do freq calculation in vacuum condition, then intend to do a SP calculation with solvent effect included.

The first thing I would suggest is to try this job without the solvent, first, and with a different basis set. I have not used the GENECP for running ONIOM jobs, or run ONIOM jobs for that matter. If you have something set up there out of the ordinary, it wouldn't jump out at me. I have found with Gauss, start with the assumption that there is something simple set up wrong, and a small basis set without the solvent would tell you that. Then I would add in the solvent with that same basis set. If convergence fails there, look to see if your calculation is stuck in a saddle point (your numbers will oscillate back and forth if it is). Then I would re-optimize my system, and make sure my frequencies made sense,

If all goes well with those two tests, then your problem will live in the added in generic basis set. Something I would be happy to play with later, but you would probably be better suited to find someone more familiar with running those specific types of jobs.

Thank you for your usefull suggestion. That is something i heard about before but never think of using that aproach.

I will try now to see if it work.

Step by step!

Thank you very much, Linda!

-Mer-

The only thing I can add to my answer, you have a fair amount of memory, but I don't see an increase in the amount of steps you are using. With the MaxCycle=N (where n is the number of cycles), you can give your job more time to run. If you outputs aren't oscillating, and your convergence looks like it is getting close, try increasing your max cycles (my favorite fall back is 400 for a large system).

My suggestion for a small basis set as a test, is only for the sake of time, not for quality. There was nothing more frustrating for me than to run a job for 4 days only to find out then there was an issue. Hope that helps. I will check back later today.

Thank you. I am currently see if scf=qc would help. I tried to increase Maxcycle to 256 last time, but failed. If scf=qc fail then i will try step by step to find the problem.

Thank you for your concern!

Sorry for the broken answers, thinking on your question while getting ready to leave for the morning.

I just saw your answer in regards to when you added your solvent. I don't know that it makes sense to add a solvent for a single point calculation and not for the geometry optimization. Someone else may certainly disagree with me, but in my view what you are saying by doing so is that the solvent only effects the energetics and not the geometry. If I wanted to compare a system before and after the addition of a solvent, I would do a full set of calculations on it with and without. The comparison of the two outputs is what I would find interesting.

Here is a link you might find helpful:

http://www.ace-net.ca/wiki/Gaussian_Error_Messages

There are a few of these types of boards out there, it was something I found helpful when trying to understand why certain jobs were failing.

Have a great day!

Thank you. I have not yet done the opt with solvent effect. But it seems to be quite popular in my field that people do opt in gas-phased, then use that geometry to calculate other properties (SP, TD, POP..) with solvent effect included. Some said the affect of solvent on geometry is neglegible(?). But Yes! You are right. I should try my self to see if solvent make opt geometries any difference.

Thank you for the useful link, as well.

Have a great day ahead!

-Mer-

Hi,

I added scf=qc, then it converged after 100 cycles but at the value of conveg=e-5 instead of normal e-8. The energy happend to meet my expectation. I try to find if i can get it done at normal e-8 converg value.

-Mer-

The only issue I found with scf=qc conver=5, up to 5 there is an approximation that runs. The rigorous qc calculations don't start until after that level of convergence. If I can fin my old files to share with I will post them. I didn't get a chance to see if the approximation is accurate by comparing output files from systems that were better known. I remember being stunned that something so rigorous would complete quicker than its counterparts.

Would you mind posting the last two or three outputs (including the final) for your convergence failure? I should have asked for those first. When you said it crashed after 256 cycles, there is the possibility it is memory related (although I would be stunned with 8GB set aside), but just in case.....

Oh, Sorry for late reply as i had some other work.

The qc job overwritten the old output file. But it was not because of memory (which i once had, it just terminate without any notice left). This case, it clearly stated that the convergence fail.

My qc job took quite the same time with normal one and get to conver=5 after 30 cycles.