Are their any clear indications that the SCF will not converge for a SPE calculation in Gaussian?

08 August 2015 11 2K Report

Of 54 atoms in my structure (a solute and a water molecule) in an implicit solvent (water), using Gaussian 09, I performed a high level DFT geometry optimisation: wb97xd/6-311++g(2df,2p). Using the optimised structure I am trying to perform an MP2/aug-cc-pVTZ single point energy calculation. It is taking a very long time to complete (16 cores on 1 node and I'm still waiting after 3 days). I have included scf=(qc,maxcycle=1024).

My question; given the output below does it look as though the calculation will ever complete? Or, is it 'circling the drain'?

I would also like to know what "QCLLim is confused" means.

Many thanks

Anthony

Iteration 1 A*A^-1 deviation from unit magnitude is 2.78D-15 for 295.

Iteration 1 A*A^-1 deviation from orthogonality is 3.09D-15 for 2663 2446.

Iteration 1 A^-1*A deviation from unit magnitude is 2.66D-15 for 1526.

Iteration 1 A^-1*A deviation from orthogonality is 5.12D-08 for 2296 2294.

Iteration 2 A*A^-1 deviation from unit magnitude is 2.55D-15 for 280.

Iteration 2 A*A^-1 deviation from orthogonality is 2.16D-15 for 3291 32.

Iteration 2 A^-1*A deviation from unit magnitude is 6.66D-16 for 236.

Iteration 2 A^-1*A deviation from orthogonality is 2.68D-16 for 2588 970.

Gradient too large for Newton-Raphson -- use scaled steepest descent instead.

Minimum is close to point 3 DX= 3.60D-02 DF= -1.61D-04 DXR= 4.30D-02 DFR= 1.85D-03 which will be used.

Gradient too large for Newton-Raphson -- use scaled steepest descent instead.

Accept linear search using points 1 and 2.

Gradient too large for Newton-Raphson -- use scaled steepest descent instead.

Accept linear search using points 1 and 2.

LinEq1: Iter= 0 NonCon= 1 RMS=1.79D-05 Max=1.27D-03 NDo= 1

AX will form 1 AO Fock derivatives at one time.

LinEq1: Iter= 1 NonCon= 1 RMS=5.25D-06 Max=3.44D-04 NDo= 1

LinEq1: Iter= 2 NonCon= 1 RMS=1.79D-06 Max=8.20D-05 NDo= 1

LinEq1: Iter= 3 NonCon= 1 RMS=3.51D-07 Max=2.27D-05 NDo= 1

LinEq1: Iter= 4 NonCon= 1 RMS=1.63D-07 Max=9.52D-06 NDo= 1

LinEq1: Iter= 5 NonCon= 1 RMS=4.58D-08 Max=2.70D-06 NDo= 1

LinEq1: Iter= 6 NonCon= 1 RMS=2.06D-08 Max=1.19D-06 NDo= 1

LinEq1: Iter= 7 NonCon= 0 RMS=1.04D-08 Max=2.30D-07 NDo= 1

Linear equations converged to 5.524D-08 5.524D-07 after 7 iterations.

Accept linear search using points 1 and 2.

Minimum is close to point 2 DX= -1.07D-03 DF= -1.41D-10 DXR= 1.07D-03 DFR= 1.14D-06 which will be used.

LinEq1: Iter= 0 NonCon= 1 RMS=4.41D-08 Max=2.39D-06 NDo= 1

LinEq1: Iter= 1 NonCon= 1 RMS=2.65D-08 Max=2.19D-06 NDo= 1

LinEq1: Iter= 2 NonCon= 1 RMS=1.14D-08 Max=2.49D-07 NDo= 1

LinEq1: Iter= 3 NonCon= 1 RMS=6.58D-09 Max=2.22D-07 NDo= 1

LinEq1: Iter= 4 NonCon= 1 RMS=4.14D-09 Max=1.91D-07 NDo= 1

LinEq1: Iter= 5 NonCon= 1 RMS=1.50D-09 Max=5.61D-08 NDo= 1

LinEq1: Iter= 6 NonCon= 1 RMS=1.08D-09 Max=2.73D-08 NDo= 1

LinEq1: Iter= 7 NonCon= 1 RMS=1.14D-09 Max=1.38D-08 NDo= 1

LinEq1: Iter= 8 NonCon= 1 RMS=1.18D-09 Max=2.65D-08 NDo= 1

LinEq1: Iter= 9 NonCon= 1 RMS=1.26D-09 Max=3.05D-08 NDo= 1

LinEq1: Iter= 10 NonCon= 1 RMS=1.18D-09 Max=1.61D-08 NDo= 1

LinEq1: Iter= 11 NonCon= 1 RMS=6.90D-10 Max=1.29D-08 NDo= 1

LinEq1: Iter= 12 NonCon= 1 RMS=5.19D-10 Max=8.95D-09 NDo= 1

LinEq1: Iter= 13 NonCon= 1 RMS=5.84D-10 Max=8.23D-09 NDo= 1

LinEq1: Iter= 14 NonCon= 1 RMS=4.89D-10 Max=5.58D-09 NDo= 1

LinEq1: Iter= 15 NonCon= 1 RMS=5.05D-10 Max=6.42D-09 NDo= 1

LinEq1: Iter= 16 NonCon= 1 RMS=3.51D-10 Max=3.42D-09 NDo= 1

LinEq1: Iter= 17 NonCon= 0 RMS=2.44D-10 Max=2.83D-09 NDo= 1

Linear equations converged to 3.476D-10 3.476D-09 after 17 iterations.

LinEq1: Iter= 0 NonCon= 1 RMS=1.88D-08 Max=2.20D-07 NDo= 1

LinEq1: Iter= 1 NonCon= 1 RMS=5.39D-09 Max=8.12D-08 NDo= 1

LinEq1: Iter= 2 NonCon= 1 RMS=4.73D-09 Max=6.98D-08 NDo= 1

LinEq1: Iter= 3 NonCon= 1 RMS=2.82D-09 Max=3.38D-08 NDo= 1

LinEq1: Iter= 4 NonCon= 1 RMS=1.95D-09 Max=2.85D-08 NDo= 1

LinEq1: Iter= 5 NonCon= 1 RMS=1.80D-09 Max=2.01D-08 NDo= 1

LinEq1: Iter= 6 NonCon= 1 RMS=1.08D-09 Max=1.82D-08 NDo= 1

LinEq1: Iter= 7 NonCon= 1 RMS=6.16D-10 Max=7.13D-09 NDo= 1

LinEq1: Iter= 8 NonCon= 1 RMS=5.44D-10 Max=6.63D-09 NDo= 1

LinEq1: Iter= 9 NonCon= 1 RMS=6.08D-10 Max=7.66D-09 NDo= 1

LinEq1: Iter= 10 NonCon= 1 RMS=5.10D-10 Max=7.21D-09 NDo= 1

LinEq1: Iter= 11 NonCon= 1 RMS=3.96D-10 Max=4.01D-09 NDo= 1

LinEq1: Iter= 12 NonCon= 1 RMS=2.71D-10 Max=3.59D-09 NDo= 1

LinEq1: Iter= 13 NonCon= 1 RMS=2.04D-10 Max=2.27D-09 NDo= 1

LinEq1: Iter= 14 NonCon= 1 RMS=2.36D-10 Max=2.26D-09 NDo= 1

LinEq1: Iter= 15 NonCon= 1 RMS=2.44D-10 Max=2.76D-09 NDo= 1

LinEq1: Iter= 16 NonCon= 1 RMS=2.08D-10 Max=2.57D-09 NDo= 1

LinEq1: Iter= 17 NonCon= 0 RMS=1.96D-10 Max=2.10D-09 NDo= 1

Linear equations converged to 2.220D-10 2.220D-09 after 17 iterations.

QCLLim is confused: Bigger=T Turned=T

NLin= 3 IMin12= 1 2 I12= 0 2 IX= 1 XLMin= 0.000D+00 XLMax= 0.000D+00

X = 0.000D+00 1.000D+00 2.000D+00

DE= 0.000D+00 5.184D-11 8.322D-10

LinEq1: Iter= 0 NonCon= 1 RMS=1.15D-08 Max=1.38D-07 NDo= 1

LinEq1: Iter= 1 NonCon= 1 RMS=6.34D-09 Max=1.01D-07 NDo= 1

LinEq1: Iter= 2 NonCon= 1 RMS=4.57D-09 Max=7.15D-08 NDo= 1

LinEq1: Iter= 3 NonCon= 1 RMS=2.12D-09 Max=3.09D-08 NDo= 1

LinEq1: Iter= 4 NonCon= 1 RMS=2.10D-09 Max=2.33D-08 NDo= 1

LinEq1: Iter= 5 NonCon= 1 RMS=1.28D-09 Max=1.45D-08 NDo= 1

LinEq1: Iter= 6 NonCon= 1 RMS=7.91D-10 Max=1.05D-08 NDo= 1

LinEq1: Iter= 7 NonCon= 1 RMS=5.56D-10 Max=6.68D-09 NDo= 1

LinEq1: Iter= 8 NonCon= 1 RMS=4.93D-10 Max=7.23D-09 NDo= 1

LinEq1: Iter= 9 NonCon= 1 RMS=6.15D-10 Max=8.48D-09 NDo= 1

LinEq1: Iter= 10 NonCon= 1 RMS=7.11D-10 Max=8.51D-09 NDo= 1

LinEq1: Iter= 11 NonCon= 1 RMS=5.40D-10 Max=7.07D-09 NDo= 1

LinEq1: Iter= 12 NonCon= 1 RMS=4.21D-10 Max=5.07D-09 NDo= 1

LinEq1: Iter= 13 NonCon= 1 RMS=3.13D-10 Max=3.44D-09 NDo= 1

LinEq1: Iter= 14 NonCon= 1 RMS=2.42D-10 Max=2.47D-09 NDo= 1

LinEq1: Iter= 15 NonCon= 1 RMS=2.01D-10 Max=2.45D-09 NDo= 1

LinEq1: Iter= 16 NonCon= 1 RMS=1.85D-10 Max=2.36D-09 NDo= 1

LinEq1: Iter= 17 NonCon= 1 RMS=1.64D-10 Max=1.52D-09 NDo= 1

LinEq1: Iter= 18 NonCon= 0 RMS=1.09D-10 Max=1.34D-09 NDo= 1

Linear equations converged to 1.381D-10 1.381D-09 after 18 iterations.

Restarting incremental Fock formation.

LinEq1: Iter= 0 NonCon= 1 RMS=6.06D-08 Max=6.90D-07 NDo= 1

LinEq1: Iter= 1 NonCon= 1 RMS=1.27D-08 Max=1.37D-07 NDo= 1

LinEq1: Iter= 2 NonCon= 1 RMS=4.40D-09 Max=6.68D-08 NDo= 1

LinEq1: Iter= 3 NonCon= 1 RMS=2.64D-09 Max=3.13D-08 NDo= 1

LinEq1: Iter= 4 NonCon= 1 RMS=1.95D-09 Max=2.42D-08 NDo= 1

LinEq1: Iter= 5 NonCon= 1 RMS=1.71D-09 Max=2.54D-08 NDo= 1

LinEq1: Iter= 6 NonCon= 1 RMS=1.21D-09 Max=1.66D-08 NDo= 1

LinEq1: Iter= 7 NonCon= 1 RMS=7.63D-10 Max=1.13D-08 NDo= 1

LinEq1: Iter= 8 NonCon= 0 RMS=5.20D-10 Max=6.37D-09 NDo= 1

Linear equations converged to 7.330D-10 7.330D-09 after 8 iterations.

Search did not lower the energy significantly.

No lower point found -- try reversing direction.

Restarting incremental Fock formation.

Search did not lower the energy significantly.

No lower point found -- switch to scaled steepest descent.

Restarting incremental Fock formation.

Minimum is close to point 10 DX= 0.00D+00 DF= 0.00D+00 DXR= 0.00D+00 DFR= 0.00D+00 which will be used.

LinEq1: Iter= 0 NonCon= 1 RMS=7.08D-08 Max=7.87D-07 NDo= 1

LinEq1: Iter= 1 NonCon= 1 RMS=1.34D-08 Max=1.47D-07 NDo= 1

LinEq1: Iter= 2 NonCon= 1 RMS=4.89D-09 Max=6.60D-08 NDo= 1

LinEq1: Iter= 3 NonCon= 1 RMS=2.35D-09 Max=2.87D-08 NDo= 1

LinEq1: Iter= 4 NonCon= 1 RMS=1.74D-09 Max=2.46D-08 NDo= 1

LinEq1: Iter= 5 NonCon= 1 RMS=1.27D-09 Max=1.70D-08 NDo= 1

LinEq1: Iter= 6 NonCon= 1 RMS=8.90D-10 Max=1.12D-08 NDo= 1

LinEq1: Iter= 7 NonCon= 1 RMS=7.69D-10 Max=1.18D-08 NDo= 1

LinEq1: Iter= 8 NonCon= 1 RMS=8.62D-10 Max=1.07D-08 NDo= 1

LinEq1: Iter= 9 NonCon= 1 RMS=1.01D-09 Max=1.30D-08 NDo= 1

LinEq1: Iter= 10 NonCon= 0 RMS=7.41D-10 Max=8.17D-09 NDo= 1

Linear equations converged to 8.552D-10 8.552D-09 after 10 iterations.

Restarting incremental Fock formation.

Search did not lower the energy significantly.

No lower point found -- try reversing direction.

Daniel Smykowski

A maxcycle=1024 is pointless for such system, 100-150 should be enough to reach the convergence. Did the geometry optimization end up with success? Or it has just reached the max number of atomic steps? Please provide your input file.

Anthony Nash

The geometry optimisation worked perfectly fine.

What is the rationale behind 1024 cycles being pointless for such a system?

It seems that your system oscillates. Have you tried to change the scf convergence criteria as well as the algorithm, e.g. switching to qc/xqc? Using ultrafine grid may also be a potential solution.

Thanks for that suggestion. I'm currently trying qc, I've yet to try xqc or an ultrafine grid (it is on fine, by default). I'll try those.

Bartosz Trzaskowski

"What is the rationale behind 1024 cycles being pointless for such a system?"

The optimizing algorithm in gaussian is really efficient, so (as Daniel suggested) going above 150 (well, perhaps 200) steps rarely makes sense. If you need that many steps it means that either:

- you started very far from the stationary point, which is a waste of computational power and time, or

- your system is oscilating between two geometries, and it won't optimize even in a hige number of steps. So it's better to let it run for a smaller number of steps and then take a look at geometry and/or try different keywords/methods, then let it run for a long time and waste computational powern and time.

Hi Bartosz,

Thanks for that answer. The structure was initially optimised on an ultrafine grid using wb97xd/6-311++G(2df,2p) (Freq found a minimum). I am now trying a single point energy calculation using MP2/aug-cc-pVTZ on an ultrafine grid. Unfortunately, just using the default SCF algorithm and the QC algorithm the system is oscillating, I'm hoping that xqc will make the difference.

xqc may help. Another idea is to:

- first, change the geometry of the oscilating structure a bit, so it's further from the minimum (e.g. by making one bond longer than usual) and then:

- try a combination o different algorithms (opt=RFO, opt=EF) and different coordinate system (fopt=z-matrix, opt=cartesian - this uses a different optimization approach even if you don't change the geometry specification). From my experience the latter option is particualrly useful in diffucult cases.

A. M. Verma

Hi Anthony

Since ab initio methods take much time as you know already it deals with numerous more variable than what DFT deals with. So I think you need not to worry about it unless DFT is taking less than an hours to complete the same job. I think DFT must've also taken around more than 3 hours.

I've also done one QST2 optimization with one DFT functional and MP2. DFT took 15-16 hours whereas MP2 took 6 days for the same job.

And I don't agree with 1024 maxcycles. You should've taken 200 cycles then depending upon the optimization trajectory you could've changed your geometry.

Hi A. M. Verma,

Thanks for that information. It was very interesting to note the time difference between the QST2 optimisation using DFT and MP2.

With regards to max cycles I've actually removed that entry because after careful study it turns out the default is 512 if SCF=QC is included (which it is in my case).

Chamikara Herath

Hello I am Chamikara,

I am running a DFT calculation for a Metal-Aminoacid fragment of a protein. my calculation is still running but log file continuously print this.

and this is my input file

%NProcShared=8

%Chk=complex_charged_new_02.chk

#n b3lyp/6-311g(d,p) pop=mk Opt scf=qc

Title

2 1

is this an error or can I let the calculation run and check ?? please advice me

How to get hydrogen bond analysis using MDAnalysis working with custom atom names?

How to add periodic rendering in PyMol?

How to resolve GROMACS mdrun seg fault during minimisation?

How do derive force constants from QM into MD for drugs and small compounds?

How to increase the number of optimisation steps in ORCA?

Are these QC DFT settings appropriate?

Looking for CHI (transmembrane modelling software)?

How to change the amino acid in a Gromacs topology and force field file?

How to calculate whether a disease outcome in a population is by chance or significant?

How to determine whether two time-series trajectories are significantly different?

error in running a model in Abaqus

error in modelling model in Abaqus

MGL tools

i have some problem with chtMultiRegionSimpleFoam

Resistor definition in Piezoelectric bimorph with ANSYS workbench

Arena simulation - how do i model the following model in ARENA (see description) (new to ARENA)?

How to repeat an operation on an atom selection for multiple times in PyMol?

What is the difference between genotype and haplotype data in the ped format?

Is there an open source free software for text mining and sentiment analysis?

What software programs can directly simulate Additive Manufacturing Directed Energy Deposition (DED)?