[cc]

Available in OpenQP 1.3.0

Energy-only CCSD and CCSD(T) are included in OpenQP 1.3.0.

The [cc] section controls the frozen core and the coupled-cluster solver for [input] method=ccsd and [input] method=ccsd(t). The defaults are usable as they stand; the section is only needed to freeze core orbitals, to change the convergence behaviour, or to override how the ladder integrals are stored.

Coupled cluster is a post-SCF energy workflow. It requires an HF reference, so [input] functional must be empty and [input] runtype must be energy. In concise input, select the reference with ccsd_t(reference=rhf|rohf|uhf)/BASIS; the value lowers to [scf] type and is not a keyword in this section.

Minimal Examples

CCSD(T) in .oqp:

ccsd_t/6-31g
geom="h2o.xyz"

Python:

from oqp.openqp import OpenQP

job = OpenQP("h2o_ccsd_t")
job.molecule(geometry="water")
job.theory.ccsd_t(basis="6-31g")
mol = job.run()

Legacy .inp:

[input]
method=ccsd(t)
runtype=energy
functional=

To freeze core orbitals, add an exact .oqp section call:

ccsd_t/6-31g cc(nfzc=1)
geom="h2o.xyz"

The legacy section is:

[cc]
nfzc=1

Keywords

nfzc

Field	Value
Type	integer
Default	0
Used by	CCSD and CCSD(T)

Number of lowest orbitals excluded from the correlation treatment. Must be non-negative and smaller than the number of occupied orbitals; the module aborts otherwise.

For an ROHF reference the core is removed before semicanonicalisation, so the correlated space is the span of the reference orbitals that were kept. See the coupled-cluster workflow.

conv

Field	Value
Type	float
Default	1e-7
Used by	CCSD and CCSD(T)

Convergence threshold. The iteration stops when both the amplitude RMS change and the change in the correlation energy fall below this value.

maxit

Field	Value
Type	integer
Default	50
Used by	CCSD and CCSD(T)

Maximum number of CCSD iterations. A run that does not converge within this many iterations aborts rather than reporting an unconverged energy.

ndiis

Field	Value
Type	integer
Default	8
Used by	CCSD and CCSD(T)

DIIS subspace size for the amplitude iteration. Set 0 to disable DIIS; the same tolerance then typically needs two to three times as many iterations.

cholesky

Field	Value
Type	auto, true, or false
Default	auto
Used by	CCSD and CCSD(T), closed-shell reference only

Whether to hold the ladder integrals (ab|cd) as an explicit v^4 array or to rebuild them from Cholesky vectors as the iteration needs them.

Factorising trades arithmetic for memory. The vectors are far smaller than v^4, but every block of ladder integrals has to be reassembled from them, and that assembly costs nchol / no^2 times the ladder contraction it feeds. The number of vectors tracks the basis set while no^2 tracks the correlated electrons, so a small molecule in a large basis is the worst case: H₂O in cc-pVQZ has 1680 vectors and four correlated occupied orbitals, where factorising made the CCSD iterations about five times slower for the same energy.

auto therefore takes the vectors only when the explicit route will not fit, sizing it against the memory actually available (see cholesky_direct for how that is probed). On a 500 GB node the crossover is near 460 basis functions; on a 16 GB laptop it is near 200. Set true to factorise regardless, or false to force the explicit array and let the memory guard refuse the job if it does not fit.

cholesky_tol

Field	Value
Type	float
Default	1e-10
Used by	CCSD and CCSD(T), when Cholesky vectors are in use

Truncation threshold for the decomposition. The pivoted Cholesky stops once the largest remaining diagonal falls below this value. Must be positive and finite; the module aborts otherwise, because a non-positive or infinite threshold either never terminates or silently produces zero vectors.

Looser values trade accuracy for both memory and time. The default is tight enough that the correlation energy matches the explicit ladder to better than 1e-9 Hartree. If the vector cap is reached before the threshold is met, the log warns that the result is less accurate than requested.

cholesky_direct

Field	Value
Type	auto, true, or false
Default	auto
Used by	CCSD and CCSD(T), when Cholesky vectors are in use

Whether to build the vectors straight from recomputed AO integrals instead of from the packed AO integral store.

The direct route never allocates the packed store, which is the larger of the two for a big basis, but it sweeps the shell-pair list once per pivot block and is measurably slower wherever both fit. It is therefore chosen on memory, never on speed: auto takes it only when the packed store would not fit.

Available memory is the tightest of physical RAM, the kernel's MemAvailable, and the cgroup limit — the last being what actually binds under SLURM or in a container. OQP_MEMORY_LIMIT_GB overrides the probe when the automatic answer is wrong.

Notes

• method=ccsd and method=ccsd(t) accept runtype=energy only.
• Both reject non-empty [input] functional values.
• The reference is selected through [scf] type; RHF, UHF, and ROHF are supported. Open-shell references run through a spin-orbital solver whose storage grows as (2*nmo)^4, so keep those systems small.
• The integrals are held in memory on both paths. The module prints the storage it needs and refuses when that exceeds what the machine can give, measured at run time rather than against a fixed ceiling — see cholesky_direct for what is probed and how to override it.

See the coupled-cluster workflow for complete input and Python examples.