Dear all
I have to calculate SPEs for some (25-30 molecules of max 15 atoms of 60-65 electrons in one molecule) molecules and applied CCSD(T)/6-311+g(3df, 2p) theory to do that but it is taking a lot of time and above that, for reading & writing file (rwf) it is consuming a lot of disk space also.
I gave 25-30 SPE simulations to one server at a time and it took all the disk space of around 660 GBs in rwf files and gave "link died" for all the simulations.
1) What should I do?
2) Which another level of theory should I apply for SPE (which can give appx. results) and which is balanced b/w time and accuracy?
Dear AM Verma,
unfortunately I don't think that CCSD(T) can be replaced by another level without losing some accuracy, but CCSD should be able to handle most of the important effects concerning energy since the main contribution to it come from the double excitations (D).
But a point that you should evaluate more carefully is your basis set. I have some doubts if CCSD(T) can exploit all its potential with this basis, which is not correlation-consistent. I would prefer using CCSD with a Dunning's basis set of higher quality than a so expensive method as CCSD(T) with a Pople's basis set.
Moreover, which code are you using at the moment? If you are using Gaussian, for example, changing to a code which was specially developed for Coupled-Cluster calculations, as CFOUR, should give you a much better disk and time management.
Hi Dr. Wagner
Thanks for comment......
I use Gaussian 09 software and GaussView for GPU.
Suppose if I use Dunning's basis set then also, I think it would expensive one. Upto cc-pvdz, it's ok but if I use triple zeta then would it be that accurate as of Pople's taken basis set?
If you perform CCSD(T) calculations with G09 on a regular basis, I advice a change in the hardware of your computer, i.e. to change a SATA disk by a solid state drive (SSD). For this type of calculations and software this is the only way to speed up the calculations substantially.
Gaussian is known as not the best choice for Coupled Cluster calculations. Instead, I would prefer CFOUR, and after it GAMESS or DALTON to do that. In terms of accuracy for energy calculations, I think that Dunning's basis sets are way better than Pople's, but you can properly check this in the literature for molecules that more resemble your system under study.
Dear Verma,
In one part, you should give to the calculation the maximum amount of RAM memory that it is available. In the output you will see a line with the name "Integral buffers will be" and also "Would need an additional 25699233073 words for in-memory AO integral storage". With this numbers you will be able to get an idea of the total amount of memory that it will be required (see the link for transformation to memory GB ).
Also, I have notice that with the increase of number of CPU the memory required slightly decrease. Also, as explained above, I think that Dunning basis set must be used with CC.
Other alternative, with less accuracy, is the use of MP2 with complete basis set extrapolation or, the other possibility is the MP4 calculation, single basis o CBS too.
I hope it helps you,
Joaquim Rius
http://deviceanalytics.com/memcalc.php
Thanks everybody.
Thanks Dr. Bartra
These all are HF and post HF theories which takes much time than DFT. Does DFT have any new launched functional (other than B3LYP, M05, M06 etc.) which can be up to that accurate as of MP4 or CCSD? Sorry if I asked wrong question.
Dear Verma,
It is difficult to say "this dft works extremely good, better than traditional ab initio". Even so, few time ago, I read a paper that says that some functionals had the same or even better accuracy than CCSD(T). The problem I can not remember the name and I can not find it.
On the other hand, it is possible that a functional gives you a very good results but every one will give you different results, then you should try different ones, make comparison and accept some kind of approximation to accept DFT as a "high accurate method". The problem in this case, is that this acceptance depends on each researcher, and there are a very different opinions with this.
- Badges
- Science topic