You can also define %mem=500Mb or more to increase the speed of the calculation. But, as Mehboob noted, we do not know the type of calculations that you are doing. Please tell the atom number and why are you doing it (no need to be very explicit).

Thanks Mehboob and thanks you too Ricardo for the valuable suggestions, but i still got error by using %nprocshared=4, even i have 4 cores in my system. %mem is working.

As i already mentioned, i'm working on semiconductor based nanoclusters. Doping of Transition metal doped in silicon or germanium nanocluster and try to investigate its structural, electronic and magnetic properties.

I think you need to find maximum core, which can be used. Please try to nproc=2,3

If you can give better initial structure, it will help to reduce the computing time.

Actually, the personal PC cannot extremely reduce the computing time. I recommend

you need to find better computing resource if you can.

This is the configuration I use, you can try this:

%mem=4000mb

%nproc=12

%chk=temp.chk

But, if you use a very thorough basis set, you cannot do this at you home pc...

Yeah sunwoo, i think you are right, but can u please tell me how can i choose better structures for reducing time? means when i ll make a structure which precautions i should remember?

Thanks

Does anyone know about "TCP Linda" for increase the capacity of gaussian? Does it helps to increase the speed of calculations?

If your research model is similar with previously reported structure, you can use it to make better initial structure. Actually, the definition of better initial structure is generally acceptable (not weird)

I have one idea to reduce computing time. You should perform the energy minimization with Chem 3D or semiemprical calculation with gaussian. Then, just perform the high level calculation.

I hope it will help.

The Linda can help to increase the performance of gaussian program. Because you can use many cpu core for one job.

Switch to Jaguar, from Schrodinger.

Thanks a lot for this valuable suggestions you all.

Are you already using RI / Coulomb fitting in your calculations? With DFT this can easily speed up calcutions by an order of magnitude and more.

I guess RI is not available in gaussian. at least it wasn't. nevertheless, Bernd is right.

I found an example for Gaussian03 with Google

What is Rl / Coulomb fitting?

In short, the often huge number of two-electron inegrals with four indices is avoided by some clever approximation using three-center integrals and downscaling some summations in the code. You can find on the net several descriptions when searching for Resolution of the identity.

Check out the section on density fitting at http://www.gaussian.com/g_tech/g_ur/m_basis_sets.htm

they write it is possible only for pure DFT functionals. Anyway, I would suggest to use Turbomole for RI

Agreed, TURBOMOLE is clearly still the fastest on the market.

What method and basis set are you using? Those are obvious places to cut cost. If you are using a hybrid-DFT like B3LYP then may be you can try some pure functionals, some of them like TPSSTPSS and M06L can be almost just as good. Also, try using those def2-nVP basis sets (you'll have to put them in yourself), they are a lot faster than Dunning sets.

if you want to run DFT calculations of say ~20 atom molecule, ORCA is the best software. I have used GAMESS,NWCHEM, Gaussian, TURBOMOLE and Dalton and ORCA is better than any of the above on a PC. if you want to know how to install it on your pc the below blog will tell you how to install and run ORCA . Try it and let me know

http://quantumchemistryniser.wordpress.com/

Agreed with Leela, but sometimes ORCA does not go well (calculations didn't converge). I finish it with Gaussian (I'm lucky and I can send my calculations to a cluster).

Anyway, if you have a NVIDIA graphic card you are able to run calculations using GPU's in GAMESS (see example on http://combichem.blogspot.com.es/2011/02/compiling-gamess-with-cuda-gpu-support.html).

But, answering your question and agreed with Sunwoo Kang , you can try to make a first optimization of you system using a semiempirical method (i.e. PM3 or AM1) and then, with the optimized geometry, perform a new geometry optimization using DFT.

Good luck!

I agree with Elisabeth that some times calculations dont converge in ORCA but if you use options like PMODEL it converges very fast. There are other options like PATOM etc . but PMODEL worked for me in most of the cases

2-3 days isn't that long. I wouldn't worry about it.

Use scf=qc or iop(5/6)=1 .. Option .

You will find all in

http://www.gaussian.com/g_tech/g_iops/ov1.htm

@Ricardo Ferreira

But in my cluster i have a option about 12 or 24 proc .

:)

If you mean you want to shorten the calculation time, with the same computer resources, with the same basis set size, the tips are: (1) Define the symmetry of the atoms. If equivalent atoms are present you can reduce the calculation resources a lot. (2) If you optimze the structure, start with smaller basis set or sometimes semi-empirical calculation might be better to start. With using the optimzed structures you can calculate with better basis set to cut back the computer resources.

"i still got error by using %nprocshared=4, even i have 4 cores in my system. %mem is working." I guess your Gaussian does not support parallel computing. Please check the license.

I'm having trouble using the Gaussian system in parallel.

I tried using the system suggested by Ricardo Ferreira

% mem = 2000mb

% nproc = 8

% chk = temp.chk

However, the calculation takes more than just using one processor.

Anyone have any suggestions of what might be happening and how to fix?

Using the program Gaussian in parallel, sometimes displayed is the message "Normal termination" repeated. Could someone tell me why?

Example:

There are more things in heaven and earth, Horatio,

than are dreamt of in your philosophy.

-- Hamlet, Act I, Scene 5

The chemist is a guest at the physicist's table and

frequently dines well.

-- Richard Bersohn

Job cpu time: 0 days 7 hours 57 minutes 48.6 seconds.

File lengths (MBytes): RWF= 909 Int= 0 D2E= 0 Chk= 16 Scr= 1

Normal termination of Gaussian 09 at Tue May 7 15:48:13 2013.

7 hours 42 minutes 34.0 seconds.

File lengths (MBytes): RWF= 909 Int= 0 D2E= 0 Chk= 16 Scr= 1

Normal termination of Gaussian 09 at Tue May 7 15:48:13 2013.

at Tue May 7 15:47:34 2013

Fabricio, I may not get what you want to ask, "Normal termination" is normal. If the calculation ends normally you must have the answer :) As for the efficiency of parallel computing some links of the gaussian can not run in parallel way. If your calculation uses such links efficiency of parallel computing might be low. You can check, from the OS of the computer, how many CPUs are actually working for the calculation.

Takuji Ogawa,

"Normal termination" is appearing repeatedly in the same output (see example above).

We are doing some tests to check how many CPUs are being used. Soon I'll post the result.

Thanks..

In Gaussian 09, you can use pure DFT with the density fitting approximation.

# BLYP/TZVP/TZVPFit

 B. I. Dunlap, “Robust and variational fitting: Removing the four-center integrals from center stage in quantum chemistry,” J. Mol. Struct. (Theochem), 529 (2000) 37-40.

To run the Gaussian (Windows version) on 4 CPUs, add %CPUNumber=0,1,2,3 command while NProcShared=4 (check following URL). Now, you can go up to 50% one a core-i7 CPU.

http://www.gaussian.com/g_tech/g_ur/m_linda.htm.

According to the Guassian website, the windows version is limited to use 4 Processors/Cores.

First you try make high symmetry point of structure then you have to run .

First, optimize at a low-level theory and basis set (e.g. mp2/6-31g(d)). Make sure convergence criteria is met. Then, convert the output of the job into input and submit at the required level of theory.

Specify a minimum of 2GB and 3GB memory for jobs submitted to “parallel8” and “parallel”. More memory will not have much effect for energy and gradient calculations unless you use “SCF=NoInCore” key word in your input files.