The local correlation treatment
is switched on by preceding the command name by an L, i.e., by using the
LMP2, LMP3, LMP4, LQCISD, LCCSD, LDCSD, or LCISD commands.
The LQCISD and LCCSD commands can be appended by a specification for
the perturbative treatment of triple excitations (e.g., LCCSD(T0)):
- Use the default triples method. Currently this is T0.
- Non-iterative local triples. This is the fastest triples option. It is usually sufficiently accurate
and recommended to be used in most cases.
- T0 plus one perturbative update of the triples amplitudes. If the accuracy of T0 is insufficient (very rarely the case!), this can
be used to improve the accuracy. The computational cost is at least twice as large as for T0. In contrast to
T0, the triples amplitudes must be stored on disk, which can be a bottleneck in calculations for large
molecules. Also the memory requirements are substantially larger than for T0.
- As T1, but a caching algorithm is used which avoids the simultaneous storage of
all triples amplitudes on disk (as is the case for (T1) or (TF)). Hence,
T1C requires less disk space but more CPU-time than T1. The more disk space is made available
for the caching algorithm (using the T1DISK option on the local card, see below), the less CPU time is used.
- Full iterative triples calculation. With full domains and without weak pair approximations this
gives the same result as a canonical (T) calculation. Typically, 3-5 iterations are needed, and therefore the
computational effort is 2-3 times larger than for (T1). The disk and memory requirements are the same as for T1.
The T0 energy is also computed and printed. TFULL and FULL are aliases for TF.
- As TF, but the T1 energy is also computed.
Since the first iteration is different for T1,
the convergence of the triples iterations is slightly different with TF and TA (TF being somewhat
faster in most cases). TALL and ALL are aliases for TA.
Density fitting can be invoked by prepending the command name by DF-, e.g.
DF-LMP2, DF-LCCSD(T0) etc. In density fitting calculations
an additional auxiliary basis set is needed. Details about choosing such basis sets
and other options for density fitting are described in sections 32.10 and 15.
The general input for local coupled LMP2 or coupled cluster calculations is:
- Local MP2 calculation
- Local CCSD calculation
- Local DCSD calculation
- Local CCSD(T0) calculation
The same options as on the command line can also be given on subsequent
LOCAL and MULTP directives. Instead of using the MULTP
directive, the MULTP option on the command line can also be used.
In the following, we will first give a summary of all options and directives.
These will be described in more detail in the subsequent sections.
For new users it is recommended to read section 32.9 at the
end of this chapter before starting calculations.