RUNTYP=COMP does not work with parallel run

[shoubhikraj]

I am trying to run a composite method run using the hard coded methods accessed by RMETHD in $COMP. However, this fails when the computation is run in parallel. This should not happen as all the individual calculation steps in the method can be done in parallel (i.e. Hessian, MP2, CCSD(T)).

My input file is:

 $CONTRL SCFTYP=RHF RUNTYP=COMP ICHARG=-1 MULT=1 $END
 $COMP RMETHD=CCCA-S4 $END
 $SYSTEM MWORDS=10 MEMDDI=2 $END

 $DATA 
Title
C1
O     8.0    -1.46541     0.60542     0.00000
H     1.0    -0.49541     0.60542     0.00000
 $END

(hydroxyl anion, with CCCA-S4 method)

The optimization and zero-point energy calculation(hessian), and MP2 energy calculation parts run okay but at the end, when the CCSD(T) energy has to be calculated the program fails:

    REFERENCE ENERGY,   CESCF(3) =   -75.43400142
    CORRELATION ENERGY, CECOR(3) =    -0.26529982
    TOTAL ENERGY,       CETOT(3) =   -75.69930124
    ------------------------------------------------------------------
      RESTART INFO FOR COMPOSITE METHOD CALCULATIONS
      ADDING CETOT(X) WILL STOP CALCULATION X
      IF RERUNNING SAVE THE ORIGINAL DATA AT THIS POINT IF NEEDED
  CXETH =      0.00000000
  CXHTH =      0.00000000
  CXGTH =      0.00000000
  CCXZPE =      0.00000000
  CZPSCL =      1.00000000
 $DATA

 C1
 $END
  CETOT(1)=    -75.70282585
  CETOT(2)=    -75.72485441
  CETOT(3)=    -75.69930124
    ------------------------------------------------------------------

    ==COMPOSITE ENERGY CALCULATION, 3 ,DONE==

 BAD $CCINP READ

job aborted:
[ranks] message

[0] terminated

[1] application aborted
aborting MPI_COMM_WORLD (comm=0x44000000), error 911, comm rank 1

[2] application aborted
aborting MPI_COMM_WORLD (comm=0x44000000), error 911, comm rank 2

[3-5] terminated

---- error analysis -----

[1-2] on LAPTOP-V556K7BI
C:\Users\Public\gamess-64\gamess.2019.R1.P1.mkl.exe aborted the job. abort code 911

---- error analysis -----
           Disk usage (in bytes)
 ========================================
 ----------------------------------------
            Local date and time
 ========================================
11-10-2020 - 19:58:35.19

I am running 2019 R1 pre-built binary of GAMESS for windows. The machine specifications are intel core i3 6100u, 8GB RAM, Windows 10. The parallel version was run on 3 cores using MS-MPI that was given with the pre-built binary.

However, this issue disappears when the same input file is run on one core. The program runs without any problem.

[saromleang]

This is the observed behavior of GAMESS (version May 2011) with your input. I would suggest posting a question on the GAMESS Google Group.
The code's author (Sean Nedd) does frequent that forum.

https://groups.google.com/g/gamess

First question I would ask Sean, is the method meant to run in parallel?

[shoubhikraj]

Thanks!

I found a 5-year old post on GAMESS google group about the same issue. In that post, Sean wrote that the code had not yet been optimized for parallelization, and that the only workaround (at the time) would be to do the calculations by removing the CCSD(T) code and doing that step manually. There is no mention of whether this will be fixed in future or not.

[saromleang]

The fix would likely have to come from the community. I do not see it being addressed internally.