Strange MBIS charges for molecules with Iodine when using DDX and def2-TZVPPD basis

[jthorton]

For the below molecule I get very strange MBIS charges with the combination of DDX implicit solvent and the def2-TZVPPD basis set compared to using def2-TZVP. I also only see this happen for molecules containing Iodine.

from qcelemental.models.common_models import Model
from qcelemental.models import AtomicInput, Molecule
import qcengine
import numpy

qc_mol = Molecule.from_data(
    """
    0 1
    C                    -2.770566610000     0.138041700000     0.070178320000
    C                    -1.574766790000    -2.182402580000     0.041630990000
    C                     1.033304570000    -2.323434390000    -0.024422900000
    C                     2.435037490000    -0.121299970000    -0.061679550000
    C                     1.259017560000     2.209348470000    -0.033641890000
    C                    -1.350150920000     2.328080550000     0.032456990000
    I                     6.389923580000    -0.318582790000    -0.161855610000
    H                    -4.794725900000     0.238941070000     0.121450260000
    H                    -2.664958600000    -3.891956620000     0.070604800000
    H                     1.951364020000    -4.127647160000    -0.046322210000
    H                     2.351710440000     3.913369620000    -0.062674850000
    H                    -2.265188840000     4.137542080000     0.054275640000
    units bohr
    no_com
    no_reorient
    """
)
# create the psi4 job
spec = Model(method="PBE0", basis="def2-TZVPPD")
task = AtomicInput(
    molecule=qc_mol,
    driver="properties",
    keywords={
        "function_kwargs": {
            "properties": [
                # "dipole_polarizabilities",
                "mbis_charges",
                "quadrupole",
                "mulliken_charges",
                "dipole",
                "lowdin_charges",
                "wiberg_lowdin_indices",
                "mayer_indices",
            ]
        },
        "ddx": True,
        "ddx_model": "pcm",
        "ddx_solvent_epsilon": 4.0,
        "ddx_solvent": "water",
    },
    model=spec,
)
result = qcengine.compute(
    input_data=task, raise_error=True, task_config={"ncores": 4}, program="psi4"
)
print(result.extras["qcvars"]["MBIS CHARGES"])

def2-TZVPPD

[[ 2.5896714 ]
[ 0.9083468 ]
[ 0.96984483]
[ 1.82049044]
[ 1.84275851]
[ 1.27477756]
[-0.33630084]
[ 0.9999906 ]
[-0.6920741 ]
[-7.79483929]
[-2.0360328 ]
[ 0.45360942]]

def2-TZVP

[[-0.14122935]
[-0.13601096]
[-0.12667168]
[-0.07069663]
[-0.12664356]
[-0.13593726]
[-0.05171043]
[ 0.15172393]
[ 0.15317262]
[ 0.16536587]
[ 0.16536188]
[ 0.15326172]]

conda output

# packages in environment at /Users/joshua/mambaforge/envs/qcsubmit-test-psi4:
#
# Name                    Version                   Build  Channel
psi4                      1.9             py311h5db8dd7_1    conda-forge
pyddx                     0.6.0           py311h6a48a4f_1    conda-forge

[JGrantHill]

Does the overall energy look sensible when you've got the additional diffuse functions?

I occasionally find that def2-TZVPD converges to some odd state with strange properties (energy, dipole etc). I'd put it down to some form of numerical difficulties as I could usually use a mixed basis of def2-TZVP on some atoms and def2-TZVPD on others and get logical values (identity of which atoms to change was system dependent, but often H is the main culprit).

[susilehtola]

Which version are you using? IIRC older versions have had some issues with ECPs.

[bismuthadams1]

Currently we're using
pyddx 0.6.0
and
psi4 1.8.1

To add to this issue further, a lot of my tests with def2-TZVPPD either a) fail to converge entirely or b) hang indefinitely.

[jthorton]

To add to this issue further, a lot of my tests with def2-TZVPPD either a) fail to converge entirely or b) hang indefinitely.

I am also seeing this with psi4 1.9 as well.