Frequency calculation fails with "Too much data, unit 2"

JohnKendrick

Hi,

I am doing a frequency with intensities calculation on an Iron complex with four Fe atoms in the unit cell. I am trying to do the ferromagnetic state with HSE06-D3 functional and the POB-tzvp-rev2 basis. I have optimised the cell, copied the optimised cell to fort.34 and copied the fort.9 file of the optimised calculation to fort.20, ready for a frequency calculation with intensities.

The calculation is being run on 24 cores on a single node, with up to 4GBs of memory /core. The run uses intel mpi for interprocess communication.

After reporting the hyperpolarisability and electric susceptibility tensors (Beta and Chi), the program crashes with:

COMPUTING IR TENSOR ALONG DIRECTION X
forrtl: severe (67): input statement requires too much data, unit 2, file /mnt/scratch/chmjke/SpinCrossover/CRYSTAL/HSE06/AFM2/fort.2.pe15

I attach the input file: frequencies.d12
A tar file of the output and fort.34: info.tgz

Any help would be grateful received.
Thanks John

aerba

Hi,

Thank you for reporting this.

While we run some tests on our cluster, may I suggest switching from a coupled-perturbed Kohn-Sham (CPKS) approach to a Berry phase (BP) approach for the IR intensities? The latter is way less computationally demanding than the former and in this case could be beneficial to the success of the calculation.

You are currently using CPKS as per your input file:

FREQCALC
NOECKART
INTENS
INTCPHF
FMIXING
60
ANDERSON
MAXCYCLE
300
ENDCPHF
ENDFREQ

To switch to BP, you can use instead:

FREQCALC
NOECKART
INTENS
ENDFREQ

Let me know how this goes,

andrejsc

JohnKendrick
I have had the same problem in a slightly different task. I have found a solution, but it may not work for your particular task: I have reduced the number of k-points.
I am modelling Mn5+ ion in a AlPO4 cell in which one P is replaced by Mn.
My input is

EXTERNAL
FREQCALC
ANALYSIS
INTENS
INTRAMAN
INTCPHF
END
END
BASISSET
POB-TZVP-REV2
DFT
WC1LYP
SPIN
END
EXCHSIZE 
44000000
BIPOSIZE
44000000
SHRINK
6 6
TOLINTEG
9 9 9 12 20
MAXCYCLE
400
TOLDEE
10
SPINLOCK
2
-6
ATOMSPIN
1
10 1
SLOSHING
END

The same input with

SHRINK
8 8

Produced the same error as yours, regardless of convergence tools (DIIS / NODIIS (in CPHF block) / buffer sizes / etc):

ELECTRIC FIELD APPLIED ALONG CARTESIAN DIRECTIONS XX
[some lines omitted]

BECKE WEIGHT FUNCTION
RADSAFE =     2.00
TOLERANCES - DENSITY:10**- 6; POTENTIAL:10**- 9; GRID WGT:10**-14

RADIAL INTEGRATION  - INTERVALS (POINTS,UPPER LIMIT):            1( 75,  4.0*R)

ANGULAR INTEGRATION - INTERVALS (ACCURACY LEVEL [N. POINTS] UPPER LIMIT):
 1(  4[  86]   0.2)  2(  8[ 194]   0.5)  3( 12[ 350]   0.9)  4( 16[ 974]   3.5)
 5( 12[ 350]9999.0)
TTTTTTTTTTTTTTTTTTTTTTTTTTTTTT MOQGAD      TELAPSE     7695.33 TCPU     7664.63
forrtl: severe (67): input statement requires too much data, unit 81, file /scratch/tmp_p267436_student/fort.81.pe11

All running on a single machine: Dell EMC C6400 Server (2x20-Core Intel XEON Gold 6148 2.40GHz, 192GB RAM, 3x480GB SSD).