Possible issue with linear dependencies in the basis when calculating dynamic polarizabilities

The input:

$molecule
  0      1
  C       0.0000000000     1.3936420000     0.5000000000
  C       1.2069290000     0.6968210000     0.5000000000
  C       1.2069290000    -0.6968210000     0.5000000000
  C       0.0000000000    -1.3936420000     0.5000000000
  C      -1.2069290000    -0.6968210000     0.5000000000
  C      -1.2069290000     0.6968210000     0.5000000000
  H       0.0000000000     2.4750900000     0.5000000000
  H       2.1434910000     1.2375450000     0.5000000000
  H       2.1434910000    -1.2375450000     0.5000000000
  H       0.0000000000    -2.4750900000     0.5000000000
  H      -2.1434910000    -1.2375450000     0.5000000000
  H      -2.1434910000     1.2375450000     0.5000000000
  C       0.0000000000     1.3936420000    -0.5000000000
  C       1.2069290000     0.6968210000    -0.5000000000
  C       1.2069290000    -0.6968210000    -0.5000000000
  C       0.0000000000    -1.3936420000    -0.5000000000
  C      -1.2069290000    -0.6968210000    -0.5000000000
  C      -1.2069290000     0.6968210000    -0.5000000000
  H       0.0000000000     2.4750900000    -0.5000000000
  H       2.1434910000     1.2375450000    -0.5000000000
  H       2.1434910000    -1.2375450000    -0.5000000000
  H       0.0000000000    -2.4750900000    -0.5000000000
  H      -2.1434910000    -1.2375450000    -0.5000000000
  H      -2.1434910000     1.2375450000    -0.5000000000
$end

$rem
  MOPROP              DYN_POLAR
  METHOD              wB97M-V
  BASIS               aug-pcseg-1
  MEM_TOTAL           120000
  MEM_STATIC          1000
  SCF_CONVERGENCE     8
  THRESH              14
$end

$fdpfreq
  400
  nm
$end

produces blown-up numbers for the polarizability. In the first cycle,

      ====================================================
      ===                                              ===
      ===      MO-based Calculation of Properties      ===
      ===                                              ===
      ===          J. Kussmann, C. Ochsenfeld          ===
      ===                                              ===
      ====================================================

WARNING: SYMMETRY USAGE disabled in MO-CPSCF for now...

 --> frequency-dependent polarizability


  Warnings: (diff < 1.e-4)
  alpha-spectrum: 0 warning(s) -- smallest element:    0.0855165335 (Freq.:   0)

  Calculate frequency-dependent polarizabilities for 1 frequencies:
  0.113908

Now forming Dipole-matrix (explicitly)...
...done!

  DIIS-Info:
  starting crit.:                        1.0e+00
  Max. dimension of DIIS-subspace:            15
  Mode: cyclic vector sampling, shifting vectors if max. dimension is reached.

  ++++++++++++++++++++++++++++
  ++++ TDSCF  - 1st order ++++
  ++++++++++++++++++++++++++++

  ConvCrit:  1.0e-06       Max. Iter:    50
  Treat Ex/Ey/Ez together...

    FREQUENCY   1:         0.1139083808 a.u.

  ----------------------------------------------------------------------------------------
     Iter              Error Ex           Error Ey          Error Ez          Converged
  ----------------------------------------------------------------------------------------

 =======================================================
   Intermediate polarization tensor
 =======================================================
   ISOTROPIC:   -29245185903.29356384
   ANISOTROPIC: 116698387570.63948059
   Trace =      -87735557709.88069153
   Full Tensor:
     -37511085838.26757812  -1472920902.62760854  19857443087.98506546
     62939458360.70392609  -50195159217.86652374  -16766512973.97079277
     25695190463.47547531  -49183265887.87603760  -29312653.74659226
 =======================================================

I suspect it is an issue with linear dependencies in the basis set since replacing “aug-pcseg-1” for the one without diffuse functions, “pcseg-1”, produces sensible results.

      ====================================================
      ===                                              ===
      ===      MO-based Calculation of Properties      ===
      ===                                              ===
      ===          J. Kussmann, C. Ochsenfeld          ===
      ===                                              ===
      ====================================================

WARNING: SYMMETRY USAGE disabled in MO-CPSCF for now...

 --> frequency-dependent polarizability


  Warnings: (diff < 1.e-4)
  alpha-spectrum: 0 warning(s) -- smallest element:    0.0999386384 (Freq.:   0)

  Calculate frequency-dependent polarizabilities for 1 frequencies:
  0.113908

Now forming Dipole-matrix (explicitly)...
...done!

  DIIS-Info:
  starting crit.:                        1.0e+00
  Max. dimension of DIIS-subspace:            15
  Mode: cyclic vector sampling, shifting vectors if max. dimension is reached.

  ++++++++++++++++++++++++++++
  ++++ TDSCF  - 1st order ++++
  ++++++++++++++++++++++++++++

  ConvCrit:  1.0e-06       Max. Iter:    50
  Treat Ex/Ey/Ez together...

    FREQUENCY   1:         0.1139083808 a.u.

  ----------------------------------------------------------------------------------------
     Iter              Error Ex           Error Ey          Error Ez          Converged
  ----------------------------------------------------------------------------------------

 =======================================================
   Intermediate polarization tensor
 =======================================================
   ISOTROPIC:       49.37612367
   ANISOTROPIC:     22.51476645
   Trace =         148.12837101
   Full Tensor:
         56.88351172       0.00000000       0.00000000
          0.00000000      56.87857965       0.00000000
          0.00000000       0.00000000      34.36627964
 =======================================================

and converges nicely.

What happens if you set THRESH=14 and S2THRESH=16 for the diffuse basis set? Those are settings that can often handle linear dependencies.

Still very large numbers with THRESH=14 and S2THRESH=16.

can you point me to the documentation for this feature?