ccECP PSEUDOPOTENTIALS FOR CASINO
=================================

The 98 correlation-consistent ECPs (ccECPs) of pseudopotentiallibrary.org, for 74 elements from
H to Rn, converted to CASINO's tabulated format and each one checked.

WHAT IS HERE

  <El>/<recipe>/<el>_pp.data     one file per ECP, named as CASINO expects it
  MANIFEST.txt                   one line per file: source, versions, checksums, test results
  MANIFEST.json                  the same, machine-readable

<recipe> is the library's own directory name: ccECP for the standard core, and where the library
offers more than one, ccECP_He_core (only 1s in the core), ccECP_28_core and so on (the number of
core electrons), ccECP.S, ccECP_reg (an all-electron potential with a regularised nucleus) and
ccECP-soft (softer 3d-metal potentials). Choose the one your basis set was made for and copy it into
the run directory as <el>_pp.data (the manual's section on xx_pp.data has the rest).

Each file is the output of CASINO's utils/pseudo_converters/nwchem/nwchem_pp_convert.py
(version 3.1.532) for the library's <El>.ccECP.nwchem (or <El>.ccECP-soft.nwchem) file, renamed. Its first line, which CASINO
only prints, is the converter's generic title, "eCEPP Pseudopotential in real space".

SPIN-ORBIT TERMS. Forty of the ccECPs, those of 38 heavier elements, carry a spin-orbit part as
well as the averaged potential. CASINO cannot yet use spin-orbit terms, so these files hold the
averaged (scalar-relativistic) potential alone, and their title line ends ", spin-orbit terms left
out". MANIFEST.txt marks them so:yes.

HOW EACH FILE WAS CHECKED

1. The tabulated potential against the library's analytic Gaussian form, on the radial grid CASINO
   reads, the local channel with its -Z_eff/r and each non-local channel, reproducing CASINO's own
   reading of the file and evaluating the Gaussians independently in 50-digit arithmetic. Every
   channel of every file agrees to round-off at every grid point CASINO uses (it ignores the r = 0
   row, extrapolating there): to the rounding of the 13 printed figures, and near the nucleus,
   where r V is a small difference of large terms, to the double-precision rounding of that
   difference. Column T1 is the largest error as a fraction of that bound; all are below 1.

2. The atom. A Hartree-Fock calculation of the neutral atom's ground configuration with the same
   ccECP in its cc-pVDZ basis (PySCF 2.14.0), its orbitals taken into CASINO by molden2qmc, and a
   variational Monte Carlo calculation in CASINO with no Jastrow factor and this file. Such a
   calculation must reproduce the Hartree-Fock energy exactly, up to its statistical error, if the
   potential, the orbitals and CASINO's non-local integration are all right. Every atom does, within
   three standard errors (the largest deviation is 2.3): the difference and its error bar are in
   MANIFEST.txt.

   An atom cannot test a channel above its highest occupied angular momentum (its projection onto
   that channel is zero), so for most atoms the f channel, and a local g channel, are proved by the
   first check alone.

The library files used are those of github.com/QMCPACK/pseudopotentiallibrary at the commit named in
MANIFEST.txt; please cite the ccECP papers listed there for each element.
