632-Like Correction in gRPA+

• 632-Like Correction is a semilocal, meta-GGA-style modification applied to RPA+ to correct self-interaction errors in one-electron, spin-polarized systems.
• It uses a damping factor g(ζ, B) to suppress unphysical correlation in problematic regions while leaving homogeneous electron gas and spin-unpolarized cases unchanged.
• With an optimized parameter c≈0.2, the method significantly improves ionization potentials, electron affinities, and total correlation energies compared to RPA and RPA+.

Searching arXiv for the cited paper and closely related RPA+/self-interaction-correction work. “632-Like Correction” is not a standard named method in the cited literature. In the context of RPA-like correlation functionals, the closest precise referent is the generalized RPA+ (gRPA+) introduced as a simple self-interaction correction to RPA+. Its defining role is to multiply the RPA+ correlation-energy density by a semilocal damping factor that suppresses unphysical correlation in one-electron-like, spin-polarized regions, while preserving the exact RPA+ description of the homogeneous electron gas and leaving spin-unpolarized systems unchanged (Gould et al., 2019).

1. Nomenclature and conceptual placement

The term “632-Like Correction” is best understood as an informal label for a correction that “looks like” a local or meta-GGA-style modification layered onto an RPA-like correlation functional, rather than a full redesign through an approximate exchange-correlation kernel. In that sense, gRPA+ is the relevant construction: it is a computationally efficient generalized RPA+ that changes RPA+ only for spin-polarized systems and is designed to be exact for all one-electron densities (Gould et al., 2019).

The underlying target of the correction is the self-interaction or, more precisely in the paper’s language, the one-electron-density problem of RPA and RPA+. RPA is exact for the exchange energy of a many-electron ground state, but its correlation energy is too negative by about $0.5$ eV per electron. RPA+ addresses this mainly short-range error by adding a local LSDA-based correction and is exact for the homogeneous electron gas. However, RPA and RPA+ remain very poor for spin-polarized one-electron systems and for spin-polarization energies of atoms. The “632-like” characterization therefore points to a lightweight semilocal correction that removes those failures without disturbing the nonlocal physics that RPA already captures well.

A plausible implication is that the label emphasizes architectural style rather than formal nomenclature: the correction is “like” a semilocal meta-GGA dressing of an RPA backbone, not a separate many-body formalism.

2. Motivation: the failure modes of RPA and RPA+

The motivation begins from the known short-range deficiency of RPA correlation energies. The paper states that RPA correlation energies are systematically too negative by about $0.5$ eV per electron, roughly $11.5$ kcal/mol, and that this large short-range error often cancels in isoelectronic energy differences. This explains why RPA can remain useful for dispersion, surfaces, and some structural properties even when its absolute correlation energies are poor (Gould et al., 2019).

RPA+ was introduced to correct the short-range part by adding the difference between the exact LSDA correlation energy density and the LSDA-based RPA correlation energy density of the uniform electron gas. By construction, RPA+ is exact for jellium and accurate for the jellium surface. It also often gives realistic total energies for atoms or solids in which spin-polarization corrections are absent or small, and it yields realistic singlet binding energy curves for $\mathrm{H}_2$ and $\mathrm{N}_2$.

The central defect is more specific. RPA and RPA+ can be very wrong for spin-polarized one-electron systems, especially stretched $\mathrm{H}_2^+$, and for spin-polarization energies of atoms. Those spin-polarization contributions are often a small part of the total energy, but they are important for ionization energies, electron affinities, and atomization energies. The gRPA+ correction is therefore designed to alter RPA+ only where those failures occur: in spin-polarized, one-electron-like regions.

This suggests a sharply targeted philosophy. The correction does not attempt to replace the RPA/RPA+ long-range correlation framework; it suppresses the semilocally identifiable sectors in which the RPA+ correlation density should vanish but does not.

3. Formal definition of the correction

The paper writes the RPA correlation energy in density-integral form as

$$E_c^{\mathrm{RPA}} = \int d\mathbf r\, n(\mathbf r)\, \epsilon_c^{\mathrm{RPA}}(\mathbf r).$$

RPA+ is then defined by

$$E_c^{\mathrm{RPA+}} = \int d\mathbf r\, n(\mathbf r)\, \epsilon_c^{\mathrm{RPA+}}(\mathbf r),$$

with

$$\epsilon_c^{\mathrm{RPA+}}(\mathbf r) = \epsilon_c^{\mathrm{RPA}}(\mathbf r) + \epsilon_c^{\mathrm{LSDA}}(n_\uparrow(\mathbf r),n_\downarrow(\mathbf r)) - \epsilon_c^{\mathrm{LRPA}}(n(\mathbf r)).$$

Here $n=n_\uparrow+n_\downarrow$, $0.5$0 is the exact uniform-gas correlation energy density, and $0.5$1 is the uniform-gas RPA correlation energy density parameterization. The generalized correction then multiplies the RPA+ correlation-energy density by a damping factor: $0.5$2

The spin polarization variable is

$0.5$3

and the meta-GGA-like switching variable is

$0.5$4

with

$0.5$5

The interpretation given in the paper is:

• $0.5$6: one- or two-electron-like region,
• $0.5$7: homogeneous electron-gas-like region,
• $0.5$8: shell-overlap or more metallic-like region.

The damping factor is modeled as

$0.5$9

subject to

$11.5$0

A convenient choice is

$11.5$1

This yields the limiting behaviors explicitly emphasized in the paper:

• if $11.5$2, then $11.5$3;
• if $11.5$4 and $11.5$5, then $11.5$6;
• if $11.5$7, then $11.5$8 and $11.5$9 (Gould et al., 2019).

4. Exactness constraints and preserved properties

The decisive exactness condition is that the correlation energy must vanish for any one-electron density. The paper enforces this through the combined limits $\mathrm{H}_2$0 and $\mathrm{H}_2$1, which imply

$\mathrm{H}_2$2

This is the self-interaction correction property that distinguishes gRPA+ from both bare RPA and ordinary RPA+ (Gould et al., 2019).

At the same time, the construction is designed not to disturb regimes where RPA+ already performs well. Three preservation properties are stated explicitly.

First, for spin-unpolarized densities, $\mathrm{H}_2$3, so $\mathrm{H}_2$4, and gRPA+ reduces exactly to RPA+.

Second, for homogeneous-electron-gas or jellium-like regions, $\mathrm{H}_2$5, so the damping vanishes and gRPA+ remains exact for jellium.

Third, the paper notes that intermediate-range van der Waals interactions occur where $\mathrm{H}_2$6 is close to $\mathrm{H}_2$7, where $\mathrm{H}_2$8. The intended consequence is that the correction does not spoil the nonlocal physics that RPA already captures well.

These constraints are central to understanding the method’s narrow scope. It is not a universal rescaling of correlation. It acts only when both spin polarization and one-electron-like structure indicate that the RPA+ correlation density is physically inappropriate.

5. Parameterization and benchmark performance

The only free parameter in the damping model is $\mathrm{H}_2$9. The paper optimizes $\mathrm{N}_2$0 using atomic ionization potentials and finds the formal minimum at $\mathrm{N}_2$1, but rejects that limit as unphysical because it produces a sharp step-like function. The practical choice is

$\mathrm{N}_2$2

described as nearly optimal and smooth (Gould et al., 2019).

For the correlation contributions to the ionization potentials of Be, C, N, O, F, and Na, the paper reports:

• formal minimum $\mathrm{N}_2$3: MAE $\mathrm{N}_2$4 kcal/mol,
• chosen practical value $\mathrm{N}_2$5: gRPA+ MAE $\mathrm{N}_2$6 kcal/mol,
• RPA+ MAE $\mathrm{N}_2$7 kcal/mol,
• RPA MAE $\mathrm{N}_2$8 kcal/mol.

For correlation contributions to electron affinities, the reported MAEs are:

• gRPA+: $\mathrm{N}_2$9 kcal/mol,
• RPA+: $\mathrm{H}_2^+$0 kcal/mol,
• RPA: $\mathrm{H}_2^+$1 kcal/mol.

For total correlation energies of atoms and ions, including neutral atoms and ions up to several charge states and elements such as Mg, Al, and P, the paper reports approximate mean absolute errors of:

• RPA: about $\mathrm{H}_2^+$2 kcal/mol,
• RPA+: about $\mathrm{H}_2^+$3 kcal/mol,
• gRPA+: about $\mathrm{H}_2^+$4 kcal/mol.

The qualitative interpretation given in the paper is equally important. RPA+ gives the best raw absolute correlation energies. gRPA+ slightly worsens some absolute energies relative to RPA+ for spin-unpolarized systems, but it smooths out jagged errors across cationic sequences and improves reaction-like differences and spin-polarization-sensitive quantities. This is consistent with the design principle that gRPA+ should primarily improve differences between systems with different spin polarization or electron number, rather than uniformly lower absolute-energy errors across all cases (Gould et al., 2019).

6. Relation to other RPA-like corrections and methodological significance

Relative to RPA, gRPA+ addresses the fact that RPA correlation is generally too negative and performs poorly for one-electron spin-polarized cases and for atomic ionization potentials and electron affinities. Relative to RPA+, it retains the short-range uniform-gas correction already built into RPA+ but adds a self-interaction-oriented damping in the problematic spin-polarized, low-electron-number sectors (Gould et al., 2019).

The paper also contrasts this approach with beyond-RPA methods that introduce an approximate exchange-correlation kernel $\mathrm{H}_2^+$5. Such methods can improve accuracy, but they are more computationally expensive, and many fail to be exact for all one-electron densities. The authors emphasize that those methods can also, in principle, be self-interaction corrected by the same kind of damping factor, though perhaps with a different optimized $\mathrm{H}_2^+$6.

In that methodological landscape, the “632-like” interpretation is precise only at the level of structure: a lightweight, semilocal, meta-GGA-style damping correction layered onto an RPA-like correlation functional. The essential content of the proposal is the factor

$\mathrm{H}_2^+$7

used with $\mathrm{H}_2^+$8, so that correlation is suppressed in one-electron spin-polarized regions while remaining unchanged in spin-unpolarized and homogeneous-electron-gas-like regions.

A plausible implication is that the main significance of the correction lies in reconciling two constraints that are often in tension in RPA-like functional design: preservation of jellium and long-range nonlocal behavior, and exactness for one-electron densities. In the cited formulation, that reconciliation is achieved by semilocal damping rather than by changing the underlying response kernel or abandoning the RPA+ framework altogether (Gould et al., 2019).