Hello everybody:
I am trying to optimize a big structure (348 atoms including: N, S, O, Cl, P, C, H and Pd). I started with a crystal structure and I preoptimized it using: b3lyp/3-21g. Once I have this as input, I tried to optimize using b3lyp/ 6-31g(d) for non metallic atoms and lanl2dz (with an ECP) for the Pd. My route line is: opt b3lyp/gen geom=check pseudo=cards.
When I send the structure to g09, the program takes the structure information of the chk and then it ends with the message:
FoFCou: FMM=F IPFlag= 0 FMFlag= 100000 FMFlg1= 2001
NFxFlg= 0 DoJE=T BraDBF=F KetDBF=T FulRan=T
wScrn= 0.000000 ICntrl= 500 IOpCl= 0 I1Cent= 200000004 NGrid= 0
NMat0= 1 NMatS0= 1 NMatT0= 0 NMatD0= 1 NMtDS0= 0 NMtDT0= 0
Petite list used in FoFCou.
Requested convergence on RMS density matrix=1.00D-08 within 128 cycles.
Requested convergence on MAX density matrix=1.00D-06.
Requested convergence on energy=1.00D-06.
No special actions if energy rises.
Defaulting to unpruned grid for atomic number 46.
Defaulting to unpruned grid for atomic number 46.
Defaulting to unpruned grid for atomic number 46.
Which I do not understand because the error is not specified. Moreover, I have checked the memmory consumed and it is less than the one I asked for. Other thing that happens is that I look for the convergence criteria (the four YES or NO) in the log file and there is none.
I think one of the solutions could be trying to optimize at a higher level of theory than 3-21 but lower than 6-31g(d)/lanldz, however I do not know which one would be appropiated.
Thank you very much for your help.
Dear Fatima Lucio,
Compounds with maximum atom number of 60-70 can DFT handle. Your compound is huge. Your job which was submitted to G09 crashed because of the length of your compound which needs a huge memory. There is NO way that you can optimize this large molecule with DFT or ab initio methods.
You can optimize your compound by molecular mechanics methods (MM2 and etc), or you can try to optimize it with semi-empirical method such as AM1 or PM3.
Hoping this will be helpful,
Rafik
According to the properties in which you are interested, a QM/MM approach using the n-layer ONIOM method could be indicated in the case of system with a huge amount of atoms.
Two approaches mentioned above seem like a good choice, since DFT is out of the questions for such a large system.
1. ONIOM (implemented in gaussian), if the system has a well-defined core with Pd atom(s) and large, organic parts without Pd atoms. You can partition the system into two layers, and treat the one with Pd atom(s) at the DFT level (b3lyp/ 6-31g(d) + lanl2dz are fine) and the rest at the semiempirical PM6 level. GaussView can do the partitioning efficiently.
2. Semiempirical. PM6 in gaussian (if it can handle Pd) or free MOPAC (which definitely can handle Pd). While this method can fail in predicting proper electronic states of transition metal complexes, it works in many cases and will give you the optimized structure FAST. Such optimized geometry could also be a good starting point for ONIOM.
Maybe i write quite late, but today i had the same problem. The reason was not enough spase memory on computational cluster.
It is still not cheap, but pretty possible nowadays to calculate 300 atoms on DFT level with double-zeta quality basis-set. DLPNO- ab initio with Normal PNO and double-zeta basis set also would work but i would not really believe to abinitio with such a small basis set, and you still need huge memory. For optimisation on DFT of 300 atoms i would believe 128 or may be even 64 Gb of RAM and 16-32 cores will be sufficient.
- Badges
- Science topic
- Similar topics
- Chemistry