Choosing a CIDER Functional

CiderPress 0.5.0 packages CIDER23X, CIDER24X, and CIDER26XC model files. Pass the model’s short name to the PySCF or GPAW constructor. The families learn different parts of the functional, and the learned quantity determines the initializer arguments described below.

Family selection

Packaged functional families
Family	Learned quantity	Electronic features	Main use	Backend
CIDER23X	Exchange	Semilocal or NLDF	Surrogate-hybrid molecular and solid calculations	PySCF and classic GPAW
CIDER24X	Exchange	Semilocal and SDMX	Molecular energetics and orbital-energy-sensitive applications	PySCF; PyTorch required
CIDER26XC	Full exchange-correlation	Semilocal and NLDF	Molecular chemistry, solids, surfaces, and adsorption	PySCF and classic GPAW, subject to the model-specific restrictions below

CIDER23X exchange models

All CIDER23X files learn exchange and are normally evaluated in the PBE0/CIDER surrogate-hybrid form. The semilocal models are useful references for separating the value of nonlocality from the regression method; the nonlocal meta-GGA variants are the main application models from that work.

CIDER23X model files
Name	Feature level	Distinguishing role
`CIDER23X_SL_GGA`	Semilocal GGA	Learned semilocal exchange reference using the reduced gradient
`CIDER23X_SL_MGGA`	Semilocal meta-GGA	Adds kinetic-energy-density information
`CIDER23X_NL_GGA`	Nonlocal GGA	Adds three version-j NLDF descriptors
`CIDER23X_NL_MGGA`	Nonlocal meta-GGA	Combines meta-GGA and version-j NLDF descriptors
`CIDER23X_NL_MGGA_PBE`	Nonlocal meta-GGA	Uses a PBE exchange baseline in the learned exchange model
`CIDER23X_NL_MGGA_DTR`	Nonlocal meta-GGA	The PBE-baseline model retrained on the diverse-training (DTR) partition of GMTKN55 for improved numerical stability

Initialize one of these models with the exchange fraction and external semilocal terms stated explicitly:

mf = make_cider_calc(
    dft.RKS(mol),
    "CIDER23X_NL_MGGA_DTR",
    xmix=0.25,
    xkernel="GGA_X_PBE",
    ckernel="GGA_C_PBE",
)

The corresponding GPAW construction uses the same three composition arguments. Changing xmix defines a different surrogate-hybrid functional.

CIDER24X SDMX exchange models

The CIDER24X models use smoothed density-matrix exchange descriptors and a mapped neural evaluator. Install their optional PyTorch dependency before loading them:

pip install 'ciderpress[cider24]'

For systems requiring a platform-specific CPU or CUDA build, install PyTorch using its platform instructions before installing CiderPress.

CIDER24Xne: Trained on energy labels.
CIDER24Xe: Trained on total-energy and orbital-energy labels.

Both are exchange models. Use the explicit PBE0/CIDER composition shown for CIDER23X. They are supported through the molecular PySCF interface. The periodic PySCF SDMX implementation supports methodological reproduction of the CIDER24X work. Packaged periodic NLDF calculations use GPAW.

CIDER26XC full-XC models

CIDER26XC model files
Name	Intended domain	Dispersion	Recommended backend
`CIDER26XCCHEM`	Molecular chemistry	Electronic model	PySCF
`CIDER26XCCHEMD4`	Molecular chemistry where an explicit D4 asymptote is desired	Post-density D4	PySCF only
`CIDER26XCSURFSCI`	Molecules, solids, surfaces, and adsorption	Electronic model	PySCF and classic GPAW

These files already contain the full XC energy form and feature vectors described in From Exchange Models to Full XC. PySCF’s defaults are correct:

mf = make_cider_calc(dft.RKS(mol), "CIDER26XCCHEM")

GPAW’s initializer serves both exchange and full-XC models. Its defaults are xmix=1.0 with the PBE kernel strings "GGA_X_PBE" and "GGA_C_PBE"; leaving ckernel set would add a second PBE correlation term to the full XC already stored in CIDER26XC. Supply the full-XC composition explicitly:

xc = get_cider_functional(
    "CIDER26XCSURFSCI",
    xmix=1.0,
    xkernel=None,
    ckernel=None,
)

get_cider_functional validates these arguments against the stored model composition. Pass all three explicitly for GPAW full-XC models.

PySCF is the intended molecular representation for CIDER26XCCHEM; the electronic model can also be evaluated in an isolated GPAW PAW box. CIDER26XCCHEMD4 uses the PySCF D4 energy interface, and selecting it in GPAW raises an error.

D4 with `CIDER26XCCHEMD4`

D4 is evaluated from the geometry after the density SCF, so it changes the total energy but not the CIDER potential or density. CiderPress adds the term the model was trained with exactly once, whether or not a dispersion wrapper is already attached to the SCF object. The calculation attributes that record this are documented in Molecular Calculations with PySCF.

The analytical molecular gradient includes the corresponding D4 derivative; the complete gradient interface is documented in Energies and Derivative Properties.

Loading rules and model trust

All packaged names accept an optional .yaml suffix. Explicit paths to other YAML or joblib models remain supported, and a real file at the supplied path takes precedence over a packaged name.

Mapped CIDER YAML and joblib files reconstruct Python model objects and therefore require a trusted source. Packaged models have the fixed checksums listed below. The checksum provides an exact identifier for an external model artifact.

Checksums

SHA-256 of packaged models
Model	SHA-256
`CIDER23X_SL_GGA`	`2e518727b836cd806c4f27c5566e9401b8c15db77182e34380728efdea24f0ce`
`CIDER23X_SL_MGGA`	`fdd60d58ae0f981dcdf64ec24437454bdc334e65b162e379e3630bb99dc82bba`
`CIDER23X_NL_GGA`	`b4e29d9530eaa7c94a3c61cc5e942a0cf7cfd0dd3c4a265dc35395123ae26c86`
`CIDER23X_NL_MGGA`	`f49060978575ffeb8b18c64df8b9eb917fddf2e0cb8b4cc120e4a11f8a01a537`
`CIDER23X_NL_MGGA_PBE`	`9ce3303986f80aee859c00f8973c7947f02bf3837dfc356c57d7d82fa36660f4`
`CIDER23X_NL_MGGA_DTR`	`93312ddde97a8b8e88a9f875df221a44f65a1a1b7c955a55d9214bd449e363ea`
`CIDER24Xne`	`ab76620ad504ae2934f2f1d599c9c5457c3b882315cbbca710358d641b3e505a`
`CIDER24Xe`	`f2650a9416ca13e0967d932e3b8d43e232fe87bd7bd2b27eb46fe6c82fd3aae1`
`CIDER26XCCHEM`	`fd2e0b5cd7408cd4b0ff09495bb026b109b5066bc2f1267c011ce5f0408bdf1d`
`CIDER26XCCHEMD4`	`ee6824e258625246180efb75fdec7c1e23310a6fe2fcdf2ab836ec90d29bb00c`
`CIDER26XCSURFSCI`	`e141a998359da9a64f3c5d06b4804e06762ab2e53dc6409979ff3eb0eacd793e`

See Citing CIDER and CiderPress for family-specific citations.