Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add prepare_segmented function #373

Merged
merged 2 commits into from
Jul 11, 2024
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion docs/example_scripts/convert_fchk_molden.py
Original file line number Diff line number Diff line change
Expand Up @@ -5,4 +5,4 @@
mol = load_one("water.fchk")
# Here you may put some code to manipulate mol before writing it the data
# to a different file.
dump_one(mol, "water.molden")
dump_one(mol, "water.molden", allow_changes=True)
tovrstra marked this conversation as resolved.
Show resolved Hide resolved
10 changes: 0 additions & 10 deletions iodata/basis.py
Original file line number Diff line number Diff line change
Expand Up @@ -239,13 +239,3 @@ class MolecularBasis:
def nbasis(self) -> int:
"""Number of basis functions."""
return sum(shell.nbasis for shell in self.shells)

def get_segmented(self):
"""Unroll generalized contractions."""
shells = []
for shell in self.shells:
for angmom, kind, coeffs in zip(shell.angmoms, shell.kinds, shell.coeffs.T):
shells.append(
Shell(shell.icenter, [angmom], [kind], shell.exponents, coeffs.reshape(-1, 1))
)
return attrs.evolve(self, shells=shells)
34 changes: 32 additions & 2 deletions iodata/convert.py
Original file line number Diff line number Diff line change
Expand Up @@ -23,18 +23,20 @@
Most of them are used by the prepare module.
"""

import attrs
import numpy as np
from numpy.typing import NDArray

from .basis import MolecularBasis
from .basis import MolecularBasis, Shell
from .orbitals import MolecularOrbitals

__all__ = (
"convert_to_unrestricted",
"convert_conventions",
"iter_cart_alphabet",
"HORTON2_CONVENTIONS",
"CCA_CONVENTIONS",
"convert_to_unrestricted",
"convert_to_segmented",
)


Expand Down Expand Up @@ -253,3 +255,31 @@ def convert_to_unrestricted(mo: MolecularOrbitals) -> MolecularOrbitals:
None if mo.energies is None else np.concatenate([mo.energies, mo.energies]),
None if mo.irreps is None else np.concatenate([mo.irreps, mo.irreps]),
)


def convert_to_segmented(obasis: MolecularBasis, keep_sp: bool = False) -> MolecularBasis:
tovrstra marked this conversation as resolved.
Show resolved Hide resolved
"""Convert basis with generalized contractions to one with only single contractions.

Parameters
----------
obasis
The basis set to convert.
keep_sp
If True, SP shells are not split up.
This can be useful for file formats only support
SP-generalized contractions and no other ones.

Returns
-------
A new ``MolecularBasis`` instance with separate single contractions.
"""
shells = []
for shell in obasis.shells:
if (shell.ncon == 1) or (keep_sp and shell.ncon == 2 and (shell.angmoms == [0, 1]).all()):
shells.append(shell)
else:
for angmom, kind, coeffs in zip(shell.angmoms, shell.kinds, shell.coeffs.T):
shells.append(
Shell(shell.icenter, [angmom], [kind], shell.exponents, coeffs.reshape(-1, 1))
)
return attrs.evolve(obasis, shells=shells)
tovrstra marked this conversation as resolved.
Show resolved Hide resolved
10 changes: 7 additions & 3 deletions iodata/formats/fchk.py
Original file line number Diff line number Diff line change
Expand Up @@ -31,7 +31,8 @@
from ..docstrings import document_dump_one, document_load_many, document_load_one
from ..iodata import IOData
from ..orbitals import MolecularOrbitals
from ..utils import DumpError, LineIterator, LoadError, LoadWarning, PrepareDumpError, amu
from ..prepare import prepare_segmented
from ..utils import LineIterator, LoadError, LoadWarning, PrepareDumpError, amu

__all__ = ()

Expand Down Expand Up @@ -592,7 +593,7 @@ def prepare_dump(data: IOData, allow_changes: bool, filename: str) -> IOData:
"followed by fully virtual ones.",
filename,
)
return data
return prepare_segmented(data, True, allow_changes, filename, "FCHK")


@document_dump_one(
Expand Down Expand Up @@ -671,7 +672,10 @@ def dump_one(f: TextIO, data: IOData):
elif shell.ncon == 2 and (shell.angmoms == [0, 1]).all():
shell_types.append(-1)
else:
raise DumpError("Cannot identify type of shell!", f)
raise RuntimeError(
"Generalized contractions other than SP are not supported. "
"Call prepare_dump first."
tovrstra marked this conversation as resolved.
Show resolved Hide resolved
)

num_pure_d_shells = sum([1 for st in shell_types if st == 2])
num_pure_f_shells = sum([1 for st in shell_types if st == 3])
Expand Down
42 changes: 23 additions & 19 deletions iodata/formats/molden.py
Original file line number Diff line number Diff line change
Expand Up @@ -40,7 +40,7 @@
from ..orbitals import MolecularOrbitals
from ..overlap import compute_overlap, gob_cart_normalization
from ..periodic import num2sym, sym2num
from ..prepare import prepare_unrestricted_aminusb
from ..prepare import prepare_segmented, prepare_unrestricted_aminusb
from ..utils import DumpError, LineIterator, LoadError, LoadWarning, PrepareDumpError, angstrom

__all__ = ()
Expand Down Expand Up @@ -793,7 +793,8 @@ def prepare_dump(data: IOData, allow_changes: bool, filename: str) -> IOData:
raise PrepareDumpError("The Molden format requires an orbital basis set.", filename)
if data.mo.kind == "generalized":
raise PrepareDumpError("Cannot write Molden file with generalized orbitals.", filename)
return prepare_unrestricted_aminusb(data, allow_changes, filename, "Molden")
data = prepare_unrestricted_aminusb(data, allow_changes, filename, "Molden")
return prepare_segmented(data, False, allow_changes, filename, "Molden")


@document_dump_one("Molden", ["atcoords", "atnums", "mo", "obasis"], ["atcorenums", "title"])
Expand Down Expand Up @@ -826,16 +827,19 @@ def dump_one(f: TextIO, data: IOData):
# to be relevant. If it happens, an error is raised.
angmom_kinds = {}
for shell in obasis.shells:
for angmom, kind in zip(shell.angmoms, shell.kinds):
if angmom in angmom_kinds:
if kind != angmom_kinds[angmom]:
raise DumpError(
"Molden format does not support mixed pure+Cartesian functions for one "
"angular momentum.",
f,
)
else:
angmom_kinds[angmom] = kind
if shell.ncon != 1:
raise RuntimeError("Generalized contractions not supported. Call prepare_dump first.")
kind = shell.kinds[0]
angmom = shell.angmoms[0]
if angmom in angmom_kinds:
if kind != angmom_kinds[angmom]:
raise DumpError(
"Molden format does not support mixed pure+Cartesian functions for one "
"angular momentum.",
f,
)
else:
angmom_kinds[angmom] = kind

# Fill in some defaults (Cartesian) for angmom kinds if needed.
angmom_kinds.setdefault(2, "c")
Expand Down Expand Up @@ -863,12 +867,12 @@ def dump_one(f: TextIO, data: IOData):
f.write("\n")
last_icenter = shell.icenter
f.write("%3i 0\n" % (shell.icenter + 1))
# Write out as a segmented basis. Molden format does not support
# generalized contractions.
for iangmom, angmom in enumerate(shell.angmoms):
f.write(f" {angmom_its(angmom):1s} {shell.nexp:3d} 1.00\n")
for exponent, coeff in zip(shell.exponents, shell.coeffs[:, iangmom]):
f.write(f"{exponent:20.10f} {coeff:20.10f}\n")
# Write out the basis.
# It is guaranteed to be segmented when reaching this part of the code.
angmom = shell.angmoms[0]
f.write(f" {angmom_its(angmom):1s} {shell.nexp:3d} 1.00\n")
for exponent, coeff in zip(shell.exponents, shell.coeffs[:, 0]):
f.write(f"{exponent:20.10f} {coeff:20.10f}\n")
f.write("\n")

# Get the permutation to convert the orbital coefficients to Molden conventions.
Expand Down Expand Up @@ -913,7 +917,7 @@ def dump_one(f: TextIO, data: IOData):
irreps,
)
else:
raise RuntimeError("This should not happen because of prepare_dump")
raise RuntimeError("Generalized orbitals are not support. Call prepare_dump first.")


def _dump_helper_orb(f, spin, occs, coeffs, energies, irreps):
Expand Down
22 changes: 13 additions & 9 deletions iodata/formats/molekel.py
Original file line number Diff line number Diff line change
Expand Up @@ -34,7 +34,7 @@
from ..docstrings import document_dump_one, document_load_one
from ..iodata import IOData
from ..orbitals import MolecularOrbitals
from ..prepare import prepare_unrestricted_aminusb
from ..prepare import prepare_segmented, prepare_unrestricted_aminusb
from ..utils import DumpError, LineIterator, LoadError, LoadWarning, PrepareDumpError, angstrom
from .molden import CONVENTIONS, _fix_molden_from_buggy_codes

Expand Down Expand Up @@ -308,7 +308,8 @@ def prepare_dump(data: IOData, allow_changes: bool, filename: str) -> IOData:
raise PrepareDumpError("The Molekel format requires an orbital basis set.", filename)
if data.mo.kind == "generalized":
raise PrepareDumpError("Cannot write Molekel file with generalized orbitals.", filename)
return prepare_unrestricted_aminusb(data, allow_changes, filename, "Molekel")
data = prepare_unrestricted_aminusb(data, allow_changes, filename, "Molekel")
return prepare_segmented(data, False, allow_changes, filename, "Molekel")


@document_dump_one("Molekel", ["atcoords", "atnums", "mo", "obasis"], ["atcharges"])
Expand Down Expand Up @@ -346,15 +347,18 @@ def dump_one(f: TextIO, data: IOData):
f.write("$BASIS\n")
iatom_last = 0
for shell in data.obasis.shells:
if shell.ncon != 1:
raise RuntimeError("Generalized contractions not supported. Call prepare_dump first.")
iatom_new = shell.icenter
if iatom_new != iatom_last:
f.write("$$\n")
for iangmom, (angmom, kind) in enumerate(zip(shell.angmoms, shell.kinds)):
iatom_last = shell.icenter
nbasis = len(CONVENTIONS[(angmom, kind)])
f.write(f" {nbasis} {angmom_its(angmom).capitalize():1s} 1.00\n")
for exponent, coeff in zip(shell.exponents, shell.coeffs[:, iangmom]):
f.write(f"{exponent:20.10f} {coeff:17.10f}\n")
angmom = shell.angmoms[0]
kind = shell.kinds[0]
iatom_last = shell.icenter
nbasis = len(CONVENTIONS[(angmom, kind)])
f.write(f" {nbasis} {angmom_its(angmom).capitalize():1s} 1.00\n")
for exponent, coeff in zip(shell.exponents, shell.coeffs[:, 0]):
f.write(f"{exponent:20.10f} {coeff:17.10f}\n")
f.write("\n")
f.write("$END\n")
f.write("\n")
Expand Down Expand Up @@ -388,7 +392,7 @@ def dump_one(f: TextIO, data: IOData):
_dump_helper_occ(f, data, spin="b")

else:
raise RuntimeError("This should not happen because of prepare_dump")
raise RuntimeError("Generalized orbitals are not support. Call prepare_dump first.")


# Defining help dumping functions
Expand Down
19 changes: 10 additions & 9 deletions iodata/formats/wfn.py
Original file line number Diff line number Diff line change
Expand Up @@ -39,7 +39,7 @@
from ..orbitals import MolecularOrbitals
from ..overlap import gob_cart_normalization
from ..periodic import num2sym, sym2num
from ..prepare import prepare_unrestricted_aminusb
from ..prepare import prepare_segmented, prepare_unrestricted_aminusb
from ..utils import LineIterator, LoadError, PrepareDumpError

__all__ = ()
Expand Down Expand Up @@ -537,7 +537,8 @@ def prepare_dump(data: IOData, allow_changes: bool, filename: str) -> IOData:
raise PrepareDumpError(
"The WFN format only supports Cartesian MolecularBasis.", filename
)
return prepare_unrestricted_aminusb(data, allow_changes, filename, "WFN")
data = prepare_unrestricted_aminusb(data, allow_changes, filename, "WFN")
return prepare_segmented(data, False, allow_changes, filename, "WFN")


@document_dump_one(
Expand All @@ -550,13 +551,13 @@ def dump_one(f: TextIO, data: IOData) -> None:
# get shells for the de-contracted basis
shells = []
for shell in data.obasis.shells:
for i, (angmom, kind) in enumerate(zip(shell.angmoms, shell.kinds)):
for exponent, coeff in zip(shell.exponents, shell.coeffs.T[i]):
shells.append(
Shell(
shell.icenter, [angmom], [kind], np.array([exponent]), coeff.reshape(-1, 1)
)
)
if shell.ncon != 1:
raise RuntimeError("Generalized contractions not supported. Call prepare_dump first.")
tovrstra marked this conversation as resolved.
Show resolved Hide resolved
# Decontract the shell
angmom = shell.angmoms[0]
kind = shell.kinds[0]
for exponent, coeff in zip(shell.exponents, shell.coeffs[:, 0]):
shells.append(Shell(shell.icenter, [angmom], [kind], [exponent], [[coeff]]))
# make a new instance of MolecularBasis with de-contracted basis shells; ideally for WFN we
# want the primitive basis set, but IOData only supports shells.
obasis = MolecularBasis(shells, data.obasis.conventions, data.obasis.primitive_normalization)
Expand Down
19 changes: 10 additions & 9 deletions iodata/formats/wfx.py
Original file line number Diff line number Diff line change
Expand Up @@ -33,7 +33,7 @@
from ..iodata import IOData
from ..orbitals import MolecularOrbitals
from ..periodic import num2sym
from ..prepare import prepare_unrestricted_aminusb
from ..prepare import prepare_segmented, prepare_unrestricted_aminusb
from ..utils import LineIterator, LoadError, LoadWarning, PrepareDumpError
from .wfn import CONVENTIONS, build_obasis, get_mocoeff_scales

Expand Down Expand Up @@ -369,7 +369,8 @@ def prepare_dump(data: IOData, allow_changes: bool, filename: str) -> IOData:
raise PrepareDumpError(
"The WFX format only supports Cartesian MolecularBasis.", filename
)
return prepare_unrestricted_aminusb(data, allow_changes, filename, "WFX")
data = prepare_unrestricted_aminusb(data, allow_changes, filename, "WFX")
return prepare_segmented(data, False, allow_changes, filename, "WFX")


@document_dump_one(
Expand All @@ -390,13 +391,13 @@ def dump_one(f: TextIO, data: IOData):
# get shells for the de-contracted basis
shells = []
for shell in data.obasis.shells:
for i, (angmom, kind) in enumerate(zip(shell.angmoms, shell.kinds)):
for exponent, coeff in zip(shell.exponents, shell.coeffs.T[i]):
shells.append(
Shell(
shell.icenter, [angmom], [kind], np.array([exponent]), coeff.reshape(-1, 1)
)
)
if shell.ncon != 1:
raise RuntimeError("Generalized contractions not supported. Call prepare_dump first.")
tovrstra marked this conversation as resolved.
Show resolved Hide resolved
# Decontract the shell
angmom = shell.angmoms[0]
kind = shell.kinds[0]
for exponent, coeff in zip(shell.exponents, shell.coeffs[:, 0]):
shells.append(Shell(shell.icenter, [angmom], [kind], [exponent], [[coeff]]))
# make a new instance of MolecularBasis with de-contracted basis shells; ideally for WFX we
# want the primitive basis set, but IOData only supports shells.
obasis = MolecularBasis(shells, data.obasis.conventions, data.obasis.primitive_normalization)
Expand Down
6 changes: 3 additions & 3 deletions iodata/overlap.py
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,7 @@

from .basis import MolecularBasis, Shell
from .convert import HORTON2_CONVENTIONS as OVERLAP_CONVENTIONS
from .convert import convert_conventions, iter_cart_alphabet
from .convert import convert_conventions, convert_to_segmented, iter_cart_alphabet
from .overlap_cartpure import tfs

__all__ = ("OVERLAP_CONVENTIONS", "compute_overlap", "gob_cart_normalization")
Expand Down Expand Up @@ -105,7 +105,7 @@ def compute_overlap(
raise ValueError("The overlap integrals are only implemented for L2 normalization.")

# Get a segmented basis, for simplicity
obasis0 = obasis0.get_segmented()
obasis0 = convert_to_segmented(obasis0)
tovrstra marked this conversation as resolved.
Show resolved Hide resolved

# Handle optional arguments
if obasis1 is None:
Expand All @@ -126,7 +126,7 @@ def compute_overlap(
"array of atomic coordinates is expected."
)
# Get a segmented basis, for simplicity
obasis1 = obasis1.get_segmented()
obasis1 = convert_to_segmented(obasis1)
identical = False

# Initialize result
Expand Down
Loading