CCL:G: OPT fails in a big tungsten-phosphors compound and others

 Sent to CCL by: "YUNPENG  LU" [yunpeng2000-x-gmail.com]
 Dear CCL Users,
 Recently, I have helped an research group in my department to calculate an
 organometallic compound's geometry and its absorption spectra in solvent. The
 compound contains three tungsten atoms (W), and carbon, hydrogen, oxygen,
 phosphorus, and sulfur.But I have a lot of problems:
 1) PBE1PBE converge problem
 The geometry optimization was firstly done with BP86 and B3LYP in Gaussian 03,
 the basis set I use for tungsten is LANL2DZ, and 6-31G(d) for the rest atoms.
 Both optimization were done successfully. Because the crystal structure of the
 compound has been determined, I can compare the geometrical values from the
 calculations to assess the quality of model chemistry. To my disappointment,
 both calculations are not good. The calculated P-P bond is 2.348 from BP86, and
 2.318 from B3LYP as the P-P distance in crystal is 2.138. The P-W bond are
 2.609, 2.754 from BP86,and 2.640, 2.762 from B3LYP as the P-W bond in crystal
 are 2.512, 2.611. (Bond distance are all in Angstrong.)
 My first questions is at what reasonable difference between crystal data and
 theoretical value one can say that the geometry optimization is of good quality?
 I read several papers recently about the validations of DFT functionals but all
 are using gas phase measurements for comparisons.I read the paper by Dr. P.
 Jeffrey Hay, in Journal of Physical Chemistry A 2002, 106, 1634,
 "Theoretical Studies of the Ground and Excited Electronic States in
 Cyclometalated Phenylpyridine Ir(III) Complexes Using Density Functional
 Theory", as he has provided some comparison about the geometry optimization
 and the crystal data for different Ir(III) compounds.While his calculation
 results shows that the difference of bond distance is usually very small.I just
 wander how that can happen as one is gas phase calculation but the other is
 condensed phase structure.So what is a reasonable criteria for comparison?
 I realize that B3LYP and BP86 may not be good and I tested the same basis with
 PBE1PBE in Gaussian 03. The final result from PBE1PBE calculation make me
 difficult to conclude it is converged or not. From the energy values for the
 intermediate runs at the last several steps, they don't change but converge to a
 value. But the log file from Gaussian 03 finally shows:
 ===================================================================
  Maximum Force            0.001708     0.000002     NO
  RMS     Force            0.000193     0.000001     NO
  Maximum Displacement          NaN     0.000006     YES
  RMS     Displacement          NaN     0.000004     YES
  Predicted change in Energy=          NaN
  GradGradGradGradGradGradGradGradGradGradGradGradGradGradGradGradGradGrad
  Leave Link  103 at Tue Apr 13 12:10:39 2010, MaxMem= 2031616000 cpu:
 11.2 (Enter /opt/gaussian/g03/l202.exe)
  Small interatomic distances encountered:
      2    1
      3    1
      3    2
      4    1
      4    2
      4    3
      5    1
      5    2
      5    3
      5    4
      6    1
      6    2
      6    3
      6    4
      6    5
      7    1
      7    2
      7    3
      7    4
      7    5
  Problem with the distance matrix.
  Error termination via Lnk1e in /opt/gaussian/g03/l202.exe at Tue Apr 13
 12:10:54 2010.
  Job cpu time:  0 days  0 hours 52 minutes 59.4 seconds.
  File lengths (MBytes):  RWF=    265 Int=      0 D2E=      0 Chk=     37 Scr=
 1
 =====================================================================
 The job fails. But when I check the final geometry from this failure run, I
 found it is actually quite good. The P-P distance is 2.193 and P-W distance is
 2.533, 2.651, very close to the crystal data P-P 2.138 and P-W 2.512, 2.629. The
 rest parts of geometry are also reasonable.
 I think PBE1PBE is better than B3LYP and BP86 for the calculations based on the
 current basis set. But the question is how to converge the calculation?
 I have tried to use Opt=verytight, SCF=tight and INT=ultrafinegrid. But they
 don't work. The job still failed.
 2) Optimization issue in Gaussian 09
 After I know there are many new functional available in Gaussian 09, I port my
 calculation work to use Gaussian 09. But something very strange happened, the
 work running with Gaussian 03 can not proceed to the first step of geometry
 optimization in Gaussian 09. The error message shows:
 =========================================================================
  GradGradGradGradGradGradGradGradGradGradGradGradGradGradGradGradGradGrad
  Berny optimization.
  NTrRot=    -1 NTRed=   260 NAtoms=    51 NSkip=   113 IsLin=F
  Error in internal coordinate system.
 =========================================================================
 Something has been changed from Gaussian 03 to Gaussian 09. What is that?
 How to turn off the change so that my job can run some steps in Gaussian 09?
 As I can not make my job run in Gaussian09, I removed two tungsten atoms and run
 geometry optimizatons with 21 functionals. My optimization results give me a
 feeling that PBE1PBE is still the best functional for the compound as the P-P
 bonnd and P-W bond from geometry optimization based on PBE1PBE are most close to
 crystal data.
 3) Basis set problem
 As all DFT calculations provide longer bond distances than the crystal data, I
 try to make the basis set larger. I use 6-311g(d) for C, H, O, P, S and
 LANLTZ(f) for W in Gaussian 03. But all these calculations failed and the W-C
 bond distance became unreasonably short.Basis set used also include other basis
 set such as Def2-TZVP and its ECP for W and Def2-TZVP for C, H, O, P, S
 and the LANL08 basis set. But none of them can work.
 4) Relativity effect
 > From all my previous testings, I was naively thinking that the relativity
 effect should be quite important in the geometry optimization as there are three
 tungsten atoms in the compound.And maybe it is also important in the absorption
 spectra. As I am searching for free ab initio package to tackle this issue, I
 choose NWChem 5.1.1. Is my choice reasonable?
 Any other free ab initio packages has integrated the relativity effect with
 geometry optimization calculations at ZORA or DK theory?
 I am very sorry for my long e-mail. But I really have quite some puzzles.
 I really appreciate any help or suggestions from you.
 sincerely yours,
 Yunpeng Lu
 yunpeng2000 : gmail.com