Despite its simplicity and relatively low computational cost, second-order Møller-Plesset perturbation theory (MP2) is well-known to overbind noncovalent interactions between polarizable monomers and some organometallic bonds. In such situations, the pairwise-additive correlation energy expression in MP2 is inadequate. Although energy-gap dependent amplitude regularization can substantially improve the accuracy of conventional MP2 in these regimes, the same regularization parameter worsens the accuracy for small molecule thermochemistry and density-dependent properties. Recently, we proposed a repartitioning of Brillouin-Wigner perturbation theory that is size-consistent to second order (BW-s2), and a free parameter (α) was set to recover the exact dissociation limit of H2 in a minimal basis set. Alternatively α can be viewed as a regularization parameter, where each value of α represents a valid variant of BW-s2, which we denote as BW-s2(α). In this work, we semi-empirically optimize α for noncovalent interactions, thermochemistry, alkane conformational energies, electronic response properties, and transition metal datasets, leading to improvements in accuracy relative to the ab initio parameterization of BW-s2 and MP2. We demonstrate that the optimal α parameter (α = 4) is more transferable across chemical problems than energy-gap-dependent regularization parameters. This is attributable to the fact that the BW-s2(α) regularization strength depends on all of the information encoded in the t amplitudes rather than just orbital energy differences. While the computational scaling of BW-s2(α) is iterative O(N5), this effective and transferable approach to amplitude regularization is a promising route to incorporate higher-order correlation effects at second-order cost.

Møller-Plesset perturbation theory (MP2) is a remarkable theoretical model chemistry with a simple pairwise additive form of the electron correlation energy and relatively low O(N5) compute cost scaling. It is the simplest ab initio theory that can approximately describe many forms of weak electron correlations, most notably dispersion but also short-ranged exchange effects. The correlation energy in the canonical molecular orbital basis can be written
Ec=14ijab|Iijab|2εa+εbεiεj=14ijab|Iijab|2Δijab,
(1)
where Iijab=(ijab) are antisymmetrized two-electron integrals and ɛp is the p-th orbital eigenvalue. Throughout this text we apply the standard notation where {i, j, k…} refer to occupied orbitals, {a, b, c…} to unoccupied orbitals, and {p, q, r…} to arbitrary (occupied or virtual) orbitals.

Formally, MP2 has many desirable properties. For example, it is free of delocalization errors, unlike the widely popular density functional theory (DFT).1–3 In contrast to both DFT and the direct random phase approximation (RPA),4 there is no self-correlation error. Consistent with Pople’s high standards for an approximate model chemistry5 (at the heart of which are formal principles which are practically useful in chemical predictions), MP2 is size-consistent, size-extensive, and orbital invariant.6 A model is size-consistent if the total energy of a supersystem comprised of noninteracting subsystems is the same as the sum of the energies of the isolated subsystems; this is an essential property when studying phenomena such as bond breaking. Second, a method is size-extensive if the total correlation energy in a linear chain of atoms grows linearly with number of electrons, which is essential for reaching the thermodynamic limit. Third, a method that yields the same correlation energy despite arbitrary orbital rotations in the occupied (or virtual) subspace is considered to be orbital invariant – a property that enables transformations to chemically-relevant bases such as the natural orbital or localized orbital representations.

MP2 routinely outperforms Hartree–Fock (HF) theory across myriad test sets with respect to experimental or near-exact numerical reference values.7 The accuracy of MP2 can be very high in the case of closed-shell and small organic molecules, and can exceed the accuracy of popular DFT functionals for important chemical properties such as reaction barrier heights (which are sensitive to delocalization errors).8 Indeed, MP2 is the most popular wavefunction component to be incorporated into double-hybrid density functionals, with promising results in many chemically-relevant situations.9–16 

However, over the years many shortcomings of MP2 have been found. It is well-known that perturbation theory in general is not suitable for multi-reference states, in which higher order (connected) excitations are required for a qualitatively correct description of the wavefunction. In addition, MP2 (and even higher orders of perturbation theory) can fail in certain cases where the reference determinant is severely spin-contaminated.17–22 In strongly correlated cases, this is a fatal issue; however, in weakly correlated systems, where the spin-symmetry breaking is artificial (i.e., due to deficiencies in the model chemistry’s treatment of dynamic correlation), MP2 with a restricted open-shell (RO) reference determinant can at times remedy this situation.21 Approximate Brueckner orbital approaches,23,24 e.g., orbital optimized (MP2) methods,25–32 can also clean up spin-symmetry breaking at the level of the Hartree–Fock orbitals.

Yet even after putting the above issues (stemming from multireference character and open-shell situations) aside, there are still serious difficulties that have historically limited the use and accuracy of MP2 approaches. For example, when bonds are stretched, the denominator of Eq. (1) can become zero, causing the correlation energy to diverge. This severely complicates the calculation of smooth potential energy surfaces. Interestingly, pair energies (corresponding to occupied orbitals i and j) can be overestimated even at equilibrium geometries, most notably in the cases of dispersion-dominated noncovalent interaction energies (NC) among polarizable monomers (e.g., those with conjugated π systems)33–35 and organometallic bonds involving, e.g., metal-carbonyl moieties.25,36,37 Physically-motivated regularization schemes that aim to remove divergences due to the energy denominator in Eq. (1) offer a promising approach to ameliorating the above problems. One example, κ-MP2, takes the form,
Ec=14ijab|Iijab|2Δijab1eκΔijab2,
(2)
and improves upon conventional MP2 for large-molecule NC and closed-shell transition-metal thermochemistry (TMTC) by factors of 5 and 2, respectively.7 

Despite such notable improvements over conventional MP2, energy-gap dependent protocols for MP2 regularization lack the desired level of transferability required to be widely used in a black-box fashion. For example, with regularization parameters optimized for NC and transition metal systems, the accuracy for main-group thermochemistry (TC) and electronic response properties is notably deteriorated – at times these regularized MP2 approaches are worse than conventional MP2 by factors of 2 or 3.7 Similarly, κ-MP2 demonstrated very promising improvements in accuracy relative to MP2 for nuclear magnetic resonance (NMR) chemical shifts only when element-specific κ values were employed.38 The prospect of developing a more transferable approach to regularized second-order perturbation theory, which preserves high accuracy for NC and TM datasets, is the primary motivation for the present work.

Brillouin-Wigner perturbation theory (BWPT)39–42 is an alternative to the Rayleigh-Schrödinger approach (the latter gives rise to MP2). We recently proposed a size-consistent variant of second-order BWPT, which naturally regularizes the t-amplitudes by shifting the occupied orbital energies in the denominator to lower values, thus increasing the effective orbital energy gaps and damping artificially overestimated amplitudes.43 The single free parameter in our BW-s2 approach was determined such that the dissociation limit of a system with two electrons in two orbitals (e.g., the H2 molecule in a minimal basis set) is exact. Importantly, this model, which we refer to as BW-s2, is size-consistent, size-extensive, and orbital invariant. While BW-s2 was found to be less accurate than (optimally parameterized) κ-MP2 in cases where exceptionally strong regularization was required, for a wide variety of main group TC its performance is superior to κ-MP2 and conventional MP2. In this work we aim to explore the landscape of the free parameter, which we will call α, by investigating many different data sets representative of NC, large-gap TMTC, main-group TC, barrier heights, and molecular dipoles and polarizabilities.

Recently, we have shown that with the following repartitioning of the Hamiltonian (with R̂, a one-electron regularizer operator):
Ĥ=H̄̂0+λV̄̂,
(3)
where
H̄̂0=Ĥ0+R̂V̄̂=V̂R̂
(4)
the second-order BWPT correction becomes
E(2)=k0Φ0|V̄̂|ΦkΦk|V̄̂|Φ0(Ē0Ēk)+E(2),
(5)
where Ē0 and Ēk are eigenvalues of the shifted zero-order Hamiltonian, H̄̂0. The energy gap Δ̄=Ē0Ēk generally satisfies the relationship Δ̄Δ where Δ = E0Ek is the gap derived from the usual eigenvalues of the unshifted Ĥ0. The use of these barred quantities is the only difference between our approach and typical BWPT.
The above expressions are general, and while there are infinitely many partitions of H̄̂0, there are finitely many that are size-consistent. We have chosen a particular form of R̂ that is size-consistent, and is represented in an arbitrary molecular orbital basis as,
Rijklabcd=α2(Wikδjl+δikWjl)δacδbd.
(6)
with,
Wij=12kabtikab(jkab)+tjkab(ikab).
(7)
The generalized tensor formulation of the second-order amplitude equation then reads,44–46 
klcdΔijklabcd+Rijklabcdtklcd=Iijab.
(8)
where
Δijklabcd=(Facδbd+δacFbd)δikδjl(Fikδjl+δikFjl)δacδbd.
(9)
Given the definition in Eqs. (6) and (8) (which is orbital invariant) can be solved by rotating the occupied subspace from the canonical basis into a basis where the matrix Foo+α2W is diagonal. To do this, we solve the following eigenvalue equation,
Foo+α2WU=ε̃U
(10)
to obtain a set of dressed occupied orbital eigenvalues. In this dressed-orbital basis, Eq. (8) can be written as,
(εa+εbε̃iε̃j)t̃ijab=Ĩijab
(11)
leading to,
t̃ijab=Ĩijab(εa+εbε̃iε̃j)
(12)
and,
Ẽc=14ijab|Ĩijab|2(εa+εbε̃iε̃j)
(13)
Thus, the dressed eigenvalues ε̃p have the effect of augmenting the original denominator, Δijab, by adding a correlation contribution to the occupied orbital energies. The undetermined parameter α was set to 1 based on making the theory exact for the two electron in two orbital problem. Similar ideas have recently been presented in Green’s function based perturbation theories,48–50 but unlike these methods, our BW-s2 approach retains the crucial property of orbital invariance. There are also notable similarities between BW-s2 and the perturbation-adapted perturbation theory (PAPT) of Knowles, which seeks to optimize the partitioning of Ĥ.51 Whereas PAPT costs O(N6) already at second order, BW-s2 scales much more favorably at iterative O(N5).

In our original set of benchmarks,43 we found that BW-s2 consistently outperforms MP2 across myriad chemical problems, which is very encouraging. However, it was evident that specific, optimal choices of κ in κ-MP2 could significantly outperform BW-s2 in problems where strong regularization was required (such as transition metal thermochemistry). How much improvement is possible if we lift the restriction of α = 1, and instead view α as a parameter that controls regularization strength? That is the question that we will investigate here.

In this work, we benchmark the performance of various values of α against a variety of data sets in an effort to tune the accuracy of BW-s2 [henceforth, the empirical variant will be referred to as BW-s2(α)]. Notably, the particular value of α does not influence the size-consistency of the method, but it may be a determining factor in the overall quality of the results. We will assess the transferability of the α parameter across various chemical problems, and attempt to make a recommendation for a broadly applicable α value.

The results for all benchmark sets apart from electronic properties are shown in Table I and are plotted individually as a function of α for each data set in Figs. S6–S9. These data include NC for sets of small dimers such as A24,54 S22,55 S66,56 and the non-I-containing subset of X40 (hereafter referred to as X31),57 along with the large π-stacked dimers of L7.58 TC is assessed on H-atom transfer (HTBH38) and non-H-atom transfer (NHTBH38) sets,59,60 along with the more comprehensive single-reference subset of W4-11.61 As compared to our original work, we extend our coverage of TMTC with reaction energies from MOR3936 (a subset of MOR41 with triple-ζ reference values),43 MC09,62 and a set of 13 Au, Pt, and Ir reaction energies that we call AuPtIr13.63 Finally, we also include the ACONFL set of relative alkane conformational isomer energies.64 

TABLE I.

Root-mean-square error in kcal/mol across chemical benchmark sets.

 
 
a

κ = 1.1.

b

heavy-aug-cc-pVDZ/heavy-aug-cc-pVTZ extrapolation to the complete basis set limit.

c

Noncovalent interaction energies, aug-cc-pVDZ/aug-cc-pVTZ extrapolation to the complete basis set limit, unless otherwise noted.

d

Using the reference data from Ref. 47.

e

Thermochemistry, aug-cc-pVTZ/aug-cc-pVQZ extrapolation to the complete basis set limit.

f

Transition metal thermochemistry, MOR39: Def2-TZVPP/Def2-ECP, MC09: Def2-QZVPP/Def2-ECP, AuIrPt13: cc-pVTZ/cc-pVTZ-PP.

g

Alkane conformational isomer energies, heavy-aug-cc-pVDZ/heavy-aug-cc-pVTZ extrapolation to the complete basis set limit.

The MP2 results always improve on those obtained from HF, and gap-regularized κ-MP2 improves further on these results in all benchmark sets, apart from TC where the results degrade by up to 3.2 kcal/mol. A similar trend emerges for BW-s2(α), with noticeable improvements over MP2 for NC, TMTC, and ACONFL data sets, but the results for barrier heights degrade by only about half as much as κ-MP2 (0.03–1.9 kcal/mol less accurate for modest parameters in the range 1 ≤ α ≤ 4). Furthermore, BW-s2(α) performs roughly 1 kcal/mol better than MP2 on the W4-11 benchmark set regardless of the particular value of α, whereas κ-MP2 performs slightly (0.7 kcal/mol) worse. The improvements in NC, TMTC, and ACONFL sets with minimal degradation in the results for TC suggest that the BW-s2(α) α-parameter is more transferable than the κ in κ-MP2.7,43

Regarding the transferability argument, it is instructive to consider electronic properties such as dipole moments and polarizabilities that are shown in Table II. Whereas κ-MP2 doubles the errors relative to MP2 for both dipoles and polarizabilities, BW-s2(α) exhibits an exceptional flatness in the errors as a function of α, peaking at 4% for the most severe α = 8.0 where the errors nonetheless remain lower than κ-MP2. Reference 7 reports κ-MP2 errors (1.6 ≥ κ ≥ 1) for dipoles that span the range 4.7%–7.5% while polarizability errors span 4.2%–5.9%. Not only are these close to the largest errors that we report for BW-s2(α), but BW-s2(α = 1) actually improves the results for dipole moments relative to MP2, whereas κ-MP2 errors monotonically increase as κ decreases. These results for electronic properties suggest that the α parameter in BW-s2(α) is indeed much more transferable between classes of chemical problem than gap-dependent regularizers.

TABLE II.

Root-mean-square relative error in % for electronic properties.

 
 
a

aug-cc-pCVQZ basis set.

b

aug-cc-pCVTZ/aug-cc-pCVQZ extrapolation to complete basis set limit.

c

κ = 1.1.

d

Dipole benchmark data from Ref. 52.

e

Polarizability benchmark data from Ref. 53.

Comparison of mean root-mean square deviation (MRMSD) values in Fig. 1(a) emphasizes the enhanced transferability of BW-s2(α) relative to κ-MP2. The MRMSD, evaluated as a simple average over RMSD values of each data set in Table I, reveals that BW-s2(α) outperforms the previously suggested7 optimal κ-MP2(κ = 1.1) over the wide range of 3 ≤ α ≤ 6. While this data weighs the error of each data set on equal footing, we also report the weighted total RMSD (WTRMSD2) statistics in Fig. 1(b), which accounts for the different sizes and energy scales of each data set. Specifically, WTRMSD2 is analogous to the type-2 weighted total mean absolute deviation metric proposed in Ref. 65, and it is calculated as,
WTRMSD2=78.29iNiiNiRMSDi|Ēi|
(14)
where Ni is the number of values in set i, |Ēi| is the average absolute value of the benchmark energies in set i, and the constant 78.29 kcal/mol was determined as the average of all Ē values for all sets. The WTRMSD2 data reinforce the idea that BW-s2(α) is more flexible than κ-MP2, with an even wider range of 1 ≤ α ≤ 5 that outperform κ-MP2(κ = 1.1). Notably, WTRMSD2 suggests that even the original BW-s2(α = 1) outperforms the best κ value. WTRMSD2 is likely to skew the results towards nominal performance on TC properties due to the relative enormity of the W4-11 set (which contains 745 reactions), hence the preference for lower α values in this case. While the optimal value of α shifts depending on the particular averaging scheme used, a value of α = 4 is roughly optimal relative to both MRMSD and WTRMSD2 metrics, and is likely a sensible compromise value that performs well for most chemical problems.
FIG. 1.

A comparison of errors across all of the data sets in Table I using (a) the mean root-mean square deviation (MRMSD) and (b) the weighted total root-mean square deviation – type 2 (WTRMSD2) for the most successful range of κ-MP2 κ values from Ref. 7 contrasted with BW-s2(α) α values from this work. X-axes are oriented in the direction of increasing regularization strength.

FIG. 1.

A comparison of errors across all of the data sets in Table I using (a) the mean root-mean square deviation (MRMSD) and (b) the weighted total root-mean square deviation – type 2 (WTRMSD2) for the most successful range of κ-MP2 κ values from Ref. 7 contrasted with BW-s2(α) α values from this work. X-axes are oriented in the direction of increasing regularization strength.

Close modal

Some particularly interesting highlights are that BW-s2(α) can reduce errors relative to MP2 in the L7 data set from 9.5 to 1.3 kcal/mol. TMTC data can also be improved by a factor of 2–3 relative to the MP2 results, reducing errors from 14 kcal/mol to 4–6 kcal/mol for MOR39 and MCO9 sets if moderate to large α parameters are applied. Finally, errors in alkane conformational energies can be reduced from 1 kcal/mol with MP2 to just 0.1 kcal/mol with BW-s2(α), achieving something close to chemical accuracy. Of course, excellent performance for particular kinds of chemical problem does not suggest a “universal” α value, and there is likely no α parameter that is entirely satisfactory in all chemical contexts. However, we make the recommendation of α = 4 based on the analysis presented alongside Fig. 1. Taking a closer look, the BW-s2(α = 4) error statistics suggest considerable improvements relative to MP2 for NC, main-group TC (W4-11), TMTC, and ACONFL sets, while minimal damage is done to the results for H-atom/non-H-atom transfer barrier heights and electronic properties.

While there is no universal parameter, BW-s2(α) stands out from gap-dependent regularizers like κ-MP2 (and the similarly-performing σ-MP2 and σ2-MP2 methods)7 in the sense that it is clearly more transferable across different chemical problems. This may be due to the fact that BW-s2(α) defines a valid second order BW correction for each α. As a consequence it incorporates the full set of t amplitudes in the regularizer, whereas gap-dependent schemes rely only on the orbital energy gaps. The self-consistent nature of BW-s2(α) may also act to further refine the orbital energy gap, introducing a feedback loop that fine-tunes the resultant amplitudes.

As a final test for the robustness of our parameterization, we consider a secondary free parameter, β, that directly modulates the amount of BW-s2(α) correlation energy such that, E = EHF + βEBW-s2(α). The results in Sec. S1 show that the optimal β parameter generally hovers in the range 0.9 ≤ β ≤ 1.1. Furthermore, when α nears its optimal value, β → 1.0 with the exception of non-H-atom barrier heights in NHTBH38 (Fig. S2) where β = 1.1 when α = 1. A β > 1 implies systematic under-correlation, and points to an optimal α for NHTBH38 that is less than 1. In stark contrast to this, the landscape of the parameter space for TMTC in Fig. S3 features an optimal β = 0.7 at low α = 1, which increases to β = 1 only when α → 8. This implies a significant over-correlation for transition-metal systems that is tempered only by larger α parameters.

The NC, W4-11, and ACONFL data sets in Figs. S1, S2, and S4, respectively, show a relatively flat slope defined by the line tracing minα,βError(α,β). For these sets, the optimal β is very close to 1 across α parameters, suggesting that BW-s2(α) offers a balanced description of correlation for NC, main-group TC, and conformational isomers. Overall, the relatively low slopes across the parameter space and the proximity of β to 1 across various α both speak to the transferability of the BW-s2(α) approach. Thus, moving forward we suggest the single parameter BW-s2(α = 4) approach for general chemical applications.

All calculations were performed in a development version of Q-Chem v6.0.2.66 All calculations (aside from evaluations of electronic properties) feature SCF convergence thresholds that were set to 10−8 root-mean-square error. The correlation energy was considered to be converged at a change of 10−8 Ha between iterations for all calculations except for those of the L7 dataset, where this was relaxed to 10−5 Ha. Relevant derivatives with respect to electric fields for properties such as dipoles and polarizabilities were evaluated via finite difference. Because finite difference results are especially sensitive to numerical errors, the SCF convergence and correlation energy thresholds were set to 10−11. To achieve complete basis set limit extrapolations for NC, TC, and ACONFL we follow the protocol in Ref. 67, which has been verified to perform well with the heavy-aug-cc-pVDZ/heavy-aug-cc-pVTZ basis sets used for L7.68 For electronic response properties, we use the same extrapolation method reported in Ref. 7.

We use restricted open-shell orbitals which are separately pseudocanonicalized in the α and β spaces before computing the correlation energy in all open-shell systems.17–22 For such systems, non-Brillouin singles (NBS) contributions are included via,
ENBS=ia|Fia|2εaεi
(15)
where Fia are off-diagonal Fock matrix elements.

Since W depends on the t amplitudes, which themselves depend on the modulation of the energy gap supplied by the W matrix, the BW-s2 equations must be solved self-consistently. We begin each BW-s2 calculation with canonical Hartree–Fock orbitals and an MP2 guess for the initial t amplitudes, though we note the possibility of obtaining a strictly non-divergent initial guess by means of Davidson’s repartitioning of the one-electron Fock operator.69 To accelerate these calculations, our implementation uses the resolution-of-the-identity (RI) approximation for the two-electron integrals,70,71 resulting in a formal scaling of m×O(N5), where m is the number of iterations (typically between 4 and 6) and N is the number of basis functions. Due to computational limitations, the I functions in the auxiliary RI basis sets were removed for transition-metal calculations in the MOR39, MCO9, and AuIrPt13 data sets.

Additional figures pertaining to scans over α, and the secondary β parameter in BW-s2(α). Detailed data are provided for NC, main-group TC, TMTC, ACONFL, and electronic response properties.

This work was supported by the Director, Office of Science, Office of Basic Energy Sciences, of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231. K.C.-F. acknowledges support from the National Institute Of General Medical Sciences of the National Institutes of Health under Award No. F32GM149165. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Martin Head-Gordon is a part-owner of Q-Chem, which is the software platform used to perform the developments and calculations described in this work.

Kevin Carter-Fenk: Data curation (lead); Investigation (lead); Methodology (equal); Software (lead); Writing – original draft (equal); Writing – review & editing (equal). James Shee: Methodology (equal); Writing – original draft (equal); Writing – review & editing (equal). Martin Head-Gordon: Conceptualization (lead); Funding acquisition (lead); Supervision (lead); Writing – review & editing (equal).

The data that support the findings of this study are available within the article and its supplementary material.

1.
Y.
Zhang
and
W.
Yang
, “
A challenge for density functionals: Self-interaction error increases for systems with a noninteger number of electrons
,”
J. Chem. Phys.
109
,
2604
2608
(
1998
).
2.
P.
Mori-Sánchez
,
A. J.
Cohen
, and
W.
Yang
, “
Many-electron self-interaction error in approximate density functionals
,”
J. Chem. Phys.
125
,
201102
(
2006
).
3.
P.
Mori-Sánchez
,
A. J.
Cohen
, and
W.
Yang
, “
Localization and delocalization errors in density functional theory and implications for band-gap prediction
,”
Phys. Rev. Lett.
100
,
146401
(
2008
).
4.
G. P.
Chen
,
V. K.
Voora
,
M. M.
Agee
,
S. G.
Balasubramani
, and
F.
Furche
, “
Random-phase approximation methods
,”
Annu. Rev. Phys. Chem.
68
,
421
445
(
2017
).
5.
J. A.
Pople
, “
Nobel lecture: Quantum chemical models
,”
Rev. Mod. Phys.
71
,
1267
1274
(
1999
).
6.
A.
Szabo
and
N. S.
Ostlund
, “
The correlation energy in the random phase approximation: Intermolecular forces between closed-shell systems
,”
J. Chem. Phys.
67
,
4351
4360
(
1977
).
7.
J.
Shee
,
M.
Loipersberger
,
A.
Rettig
,
J.
Lee
, and
M.
Head-Gordon
, “
Regularized second-order Møller-Plesset theory: A more accurate alternative to conventional MP2 for noncovalent interactions and transition metal thermochemistry for the same computational cost
,”
J. Phys. Chem. Lett.
12
,
12084
12097
(
2021
).
8.
M.
Martínez González
,
F. G. D.
Xavier
,
J.
Li
,
L. A.
Montero-Cabrera
,
J. M.
Garcia de la Vega
, and
A. J. C.
Varandas
, “
Role of augmented basis sets and quest for ab initio performance/cost alternative to Kohn–Sham density functional theory
,”
J. Phys. Chem. A
124
,
126
134
(
2019
).
9.
J.-D.
Chai
and
M.
Head-Gordon
, “
Long-range corrected double-hybrid density functionals
,”
J. Chem. Phys.
131
,
174105
(
2009
).
10.
K.
Sharkas
,
J.
Toulouse
, and
A.
Savin
, “
Double-hybrid density-functional theory made rigorous
,”
J. Chem. Phys.
134
,
064113
(
2011
).
11.
J. C.
Sancho-García
and
C.
Adamo
, “
Double-hybrid density functionals: Merging wavefunction and density approaches to get the best of both worlds
,”
Phys. Chem. Chem. Phys.
15
,
14581
14594
(
2013
).
12.
L.
Goerigk
and
S.
Grimme
, “
Double-hybrid density functionals
,”
Wiley Interdiscip. Rev.: Comput. Mol. Sci.
4
,
576
600
(
2014
).
13.
E.
Brémond
,
I.
Ciofini
,
J. C.
Sancho-García
, and
C.
Adamo
, “
Nonempirical double-hybrid functionals: An effective tool for chemists
,”
Acc. Chem. Res.
49
,
1503
1513
(
2016
).
14.
N.
Mardirossian
and
M.
Head-Gordon
, “
Survival of the most transferable at the top of Jacob’s ladder: Defining and testing the ωB97M(2) double hybrid density functional
,”
J. Chem. Phys.
148
,
241736
(
2018
).
15.
C.
Kalai
and
J.
Toulouse
, “
A general range-separated double-hybrid density-functional theory
,”
J. Chem. Phys.
148
,
164105
(
2018
).
16.
J. M. L.
Martin
and
G.
Santra
, “
Empirical double-hybrid density functional theory: A ‘third way’ in between WFT and DFT
,”
Isr. J. Chem.
60
,
787
804
(
2020
).
17.
I.
Hubač
and
P.
Čársky
, “
Correlation energy of open-shell systems. Application of the many-body Rayleigh-Schrödinger perturbation theory in the restricted Roothaan-Hartree-Fock formalism
,”
Phys. Rev. A
22
,
2392
2399
(
1980
).
18.
C.
Murray
and
E. R.
Davidson
, “
Perturbation theory for open shell systems
,”
Chem. Phys. Lett.
187
,
451
454
(
1991
).
19.
W. J.
Lauderdale
,
J. F.
Stanton
,
J.
Gauss
,
J. D.
Watts
, and
R. J.
Bartlett
, “
Many-body perturbation theory with a restricted open-shell Hartree–Fock reference
,”
Chem. Phys. Lett.
187
,
21
28
(
1991
).
20.
R. D.
Amos
,
J. S.
Andrews
,
N. C.
Handy
, and
P. J.
Knowles
, “
Open-shell Møller–Plesset perturbation theory
,”
Chem. Phys. Lett.
185
,
256
264
(
1991
).
21.
P. J.
Knowles
,
J. S.
Andrews
,
R. D.
Amos
,
N. C.
Handy
, and
J. A.
Pople
, “
Restricted Møller–Plesset theory for open-shell molecules
,”
Chem. Phys. Lett.
186
,
130
136
(
1991
).
22.
T. J.
Lee
and
D.
Jayatilaka
, “
An open-shell restricted Hartree–Fock perturbation theory based on symmetric spin orbitals
,”
Chem. Phys. Lett.
201
,
1
10
(
1993
).
23.
R. C.
Lochan
and
M.
Head-Gordon
, “
Orbital-optimized opposite-spin scaled second-order correlation: An economical method to improve the description of open-shell molecules
,”
J. Chem. Phys.
126
,
164101
(
2007
).
24.
Y.
Akinaga
,
Y.
Kawashima
, and
S.
Ten-no
, “
Use of Brueckner orbitals in second-order approximate coupled-cluster with singles and doubles (CC2) model
,”
Chem. Phys. Lett.
506
,
276
281
(
2011
).
25.
F.
Neese
,
T.
Schwabe
,
S.
Kossmann
,
B.
Schirmer
, and
S.
Grimme
, “
Assessment of orbital-optimized, spin-component scaled second-order many-body perturbation theory for thermochemistry and kinetics
,”
J. Chem. Theory Comput.
5
,
3060
3073
(
2009
).
26.
D.
Stück
and
M.
Head-Gordon
, “
Regularized orbital-optimized second-order perturbation theory
,”
J. Chem. Phys.
139
,
244109
(
2013
).
27.
S. M.
Sharada
,
D.
Stück
,
E. J.
Sundstrom
,
A. T.
Bell
, and
M.
Head-Gordon
, “
Wavefunction stability analysis without analytical electronic Hessians: Application to orbital-optimised second-order Møller–Plesset theory and VV10-containing density functionals
,”
Mol. Phys.
113
,
1802
1808
(
2015
).
28.
E.
Soydaş
and
U.
Bozkaya
, “
Assessment of orbital-optimized MP2.5 for thermochemistry and kinetics: Dramatic failures of standard perturbation theory approaches for aromatic bond dissociation energies and barrier heights of radical reactions
,”
J. Chem. Theory Comput.
11
,
1564
1573
(
2015
).
29.
R. M.
Razban
,
D.
Stück
, and
M.
Head-Gordon
, “
Addressing first derivative discontinuities in orbital-optimised opposite-spin scaled second-order perturbation theory with regularisation
,”
Mol. Phys.
115
,
2102
2109
(
2017
).
30.
J.
Lee
and
M.
Head-Gordon
, “
Regularized orbital-optimized second-order Møller–Plesset perturbation theory: A reliable fifth-order-scaling electron correlation model with orbital energy dependent regularizers
,”
J. Chem. Theory Comput.
14
,
5203
5219
(
2018
).
31.
U.
Bozkaya
,
A.
Ünal
, and
Y.
Alagöz
, “
Energy and analytic gradients for the orbital-optimized coupled-cluster doubles method with the density-fitting approximation: An efficient implementation
,”
J. Chem. Phys.
153
,
244115
(
2020
).
32.
S.
Behnle
and
R. F.
Fink
, “
UREMP, RO-REMP, and OO-REMP: Hybrid perturbation theories for open-shell electronic structure calculations
,”
J. Chem. Phys.
156
,
124103
(
2022
).
33.
M. O.
Sinnokrot
,
E. F.
Valeev
, and
C. D.
Sherrill
, “
Estimates of the ab initio limit for ππ interactions: The benzene dimer
,”
J. Am. Chem. Soc.
124
,
10887
10893
(
2002
).
34.
M. O.
Sinnokrot
and
C. D.
Sherrill
, “
Highly accurate coupled cluster potential energy curves for the benzene dimer: Sandwich, T-shaped, and parallel-displaced configurations
,”
J. Phys. Chem. A
108
,
10200
10207
(
2004
).
35.
B.
Nguyen
,
G. P.
Chen
,
M. M.
Agee
,
A. M.
Burow
,
M.
Tang
, and
F.
Furche
, “
Divergence of many-body perturbation theory for noncovalent interactions of large molecules
,”
J. Chem. Theory Comput.
16
,
2258
2273
(
2020
).
36.
S.
Dohm
,
A.
Hansen
,
M.
Steinmetz
,
S.
Grimme
, and
M. P.
Checinski
, “
Comprehensive thermochemical benchmark set of realistic closed-shell metal organic reactions
,”
J. Chem. Theory Comput.
14
,
2596
2608
(
2018
).
37.
J.
Shee
,
M.
Loipersberger
,
D.
Hait
,
J.
Lee
, and
M.
Head-Gordon
, “
Revealing the nature of electron correlation in transition metal complexes with symmetry breaking and chemical intuition
,”
J. Chem. Phys.
154
,
194109
(
2021
).
38.
J.
Wong
,
B.
Ganoe
,
X.
Liu
,
T.
Neudecker
,
J.
Lee
,
J.
Liang
,
Z.
Wang
,
J.
Li
,
A.
Rettig
,
T.
Head-Gordon
, and
M.
Head-Gordon
, “
An in-silico NMR laboratory for nuclear magnetic shieldings computed via finite fields: Exploring nucleus-specific renormalizations of MP2 and MP3
,”
J. Chem. Phys.
158
,
164116
(
2023
).
39.
J. E.
Lennard-Jones
, “
Perturbation problems in quantum mechanics
,”
Proc. R. Soc. London, Ser. A
129
,
598
615
(
1930
).
40.
L.
Brillouin
, “
Les problèmes de perturbations et les champs self-consistents
,”
J. Phys. Radium
3
,
373
389
(
1932
).
41.
E. P.
Wigner
,
Math. Naturwiss. Anz. Ungar Akad. Wiss.
53
,
477
(
1935
).
42.
I.
Hubač
and
S.
Wilson
,
Brillouin-Wigner Methods for Many-Body Systems
(
Springer
,
2010
).
43.
K.
Carter-Fenk
and
M.
Head-Gordon
, “
Repartitioned Brillouin-Wigner perturbation theory with a size-consistent second-order correlation energy
,”
J. Chem. Phys.
158
,
234108
(
2023
).
44.
M.
Head-Gordon
,
P. E.
Maslen
, and
C. A.
White
, “
A tensor formulation of many-electron theory in a nonorthogonal single-particle basis
,”
J. Chem. Phys.
108
,
616
625
(
1998
).
45.
M. S.
Lee
,
P. E.
Maslen
, and
M.
Head-Gordon
, “
Closely approximating second-order Møller–Plesset perturbation theory with a local triatomics in molecules model
,”
J. Chem. Phys.
112
,
3592
3601
(
2000
).
46.
R. A.
DiStasio
,
Y.
Jung
, and
M.
Head-Gordon
, “
A resolution-of-the-identity implementation of the local triatomics-in-molecules model for second-order Møller–Plesset perturbation theory with application to alanine tetrapeptide conformational energies
,”
J. Chem. Theory Comput.
1
,
862
876
(
2005
).
47.
M. S.
Marshall
,
L. A.
Burns
, and
C. D.
Sherrill
, “
Basis set convergence of the coupled-cluster correction, δMP2CCSD(T): Best practices for benchmarking non-covalent interactions and the attendant revision of the S22, NBC10, and HSG databases
,”
J. Chem. Phys.
135
,
194102
(
2011
).
48.
T.
Nguyen Lan
,
A. A.
Kananenka
, and
D.
Zgid
, “
Rigorous ab initio quantum embedding for quantum chemistry using Green’s function theory: Screened interaction, nonlocal self-energy relaxation, orbital basis, and chemical accuracy
,”
J. Chem. Theory Comput.
12
,
4856
4870
(
2016
).
49.
D.
Neuhauser
,
R.
Baer
, and
D.
Zgid
, “
Stochastic self-consistent second-order Green’s function method for correlation energies of large electronic systems
,”
J. Chem. Theory Comput.
13
,
5396
5403
(
2017
).
50.
C. J. N.
Coveney
and
D. P.
Tew
, “
A regularized second-order correlation method from Green’s function theory
,”
J. Chem. Theory Comput.
19
,
3915
3928
(
2023
).
51.
P. J.
Knowles
, “
Perturbation-adapted perturbation theory
,”
J. Chem. Phys.
156
,
011101
(
2022
).
52.
D.
Hait
and
M.
Head-Gordon
, “
How accurate is density functional theory at predicting dipole moments? An assessment using a new database of 200 benchmark values
,”
J. Chem. Theory Comput.
14
,
1969
1981
(
2018
).
53.
D.
Hait
and
M.
Head-Gordon
, “
How accurate are static polarizability predictions from density functional theory? An assessment of over 132 species at equilibrium geometry
,”
Phys. Chem. Chem. Phys.
20
,
19800
19810
(
2018
).
54.
J.
Řezáč
and
P.
Hobza
, “
Describing noncovalent interactions beyond the common approximations: How accurate is the ‘gold standard,’ CCSD(T) at the complete basis set limit?
,”
J. Chem. Theory Comput.
9
,
2151
2155
(
2013
).
55.
P.
Jurečka
,
J.
Šponer
,
J.
Černý
, and
P.
Hobza
, “
Benchmark database of accurate (MP2 and CCSD(T) complete basis set limit) interaction energies of small model complexes, DNA base pairs, and amino acid pairs
,”
Phys. Chem. Chem. Phys.
8
,
1985
1993
(
2006
).
56.
J.
Řezác
,
K. E.
Riley
, and
P.
Hobza
, “
S66: A well-balanced database of benchmark interaction energies relevant to biomolecular structures
,”
J. Chem. Theory Comput.
7
,
2427
2438
(
2011
);
[PubMed]
erratum: 10,
1359
1360
(
2014
).
57.
J.
Řezác
,
K. E.
Riley
, and
P.
Hobza
, “
Benchmark calculations of noncovalent interactions of halogenated molecules
,”
J. Chem. Theory Comput.
8
,
4285
4292
(
2012
).
58.
R.
Sedlak
,
T.
Janowski
,
M.
Pitoňák
,
J.
Řezáč
,
P.
Pulay
, and
P.
Hobza
, “
Accuracy of quantum chemical methods for large noncovalent complexes
,”
J. Chem. Theory Comput.
9
,
3364
3374
(
2013
).
59.
Y.
Zhao
,
N.
González-García
, and
D. G.
Truhlar
, “
Benchmark database of barrier heights for heavy atom transfer, nucleophilic substitution, association, and unimolecular reactions and its use to test theoretical methods
,”
J. Phys. Chem. A
109
,
2012
2018
(
2005
);
[PubMed]
erratum:
110
,
4942
(
2006
).
60.
J.
Zheng
,
Y.
Zhao
, and
D. G.
Truhlar
, “
Representative benchmark suites for barrier heights of diverse reaction types and assessment of electronic structure methods for thermochemical kinetics
,”
J. Chem. Theory Comput.
3
,
569
582
(
2007
).
61.
A.
Karton
,
S.
Daon
, and
J. M. L.
Martin
, “
W4-11: A high-confidence benchmark dataset for computational thermochemistry derived from first-principles W4 data
,”
Chem. Phys. Lett.
510
,
165
178
(
2011
).
62.
B.
Rudshteyn
,
D.
Coskun
,
J. L.
Weber
,
E. J.
Arthur
,
S.
Zhang
,
D. R.
Reichman
,
R. A.
Friesner
, and
J.
Shee
, “
Predicting ligand-dissociation energies of 3d coordination complexes with auxiliary-field quantum Monte Carlo
,”
J. Chem. Theory Comput.
16
,
3041
3054
(
2020
).
63.
R.
Kang
,
W.
Lai
,
J.
Yao
,
S.
Shaik
, and
H.
Chen
, “
How accurate can a local coupled cluster approach be in computing the activation energies of late-transition-metal-catalyzed reactions with Au, Pt, and Ir?
,”
J. Chem. Theory Comput.
8
,
3119
3127
(
2012
).
64.
S.
Ehlert
,
S.
Grimme
, and
A.
Hansen
, “
Conformational energy benchmark for longer n-alkane chains
,”
J. Phys. Chem. A
126
,
3521
3535
(
2022
).
65.
L.
Goerigk
,
A.
Hansen
,
C.
Bauer
,
S.
Ehrlich
,
A.
Najibi
, and
S.
Grimme
, “
A look at the density functional theory zoo with the advanced GMTKN55 database for general main group thermochemistry, kinetics and noncovalent interactions
,”
Phys. Chem. Chem. Phys.
19
,
32184
32215
(
2017
).
66.
E.
Epifanovsky
,
A. T. B.
Gilbert
et al, “
Software for the frontiers of quantum chemistry: An overview of developments in the Q-Chem 5 package
,”
J. Chem. Phys.
155
,
084801
(
2021
).
67.
F.
Neese
and
E. F.
Valeev
, “
Revisiting the atomic natural orbital approach for basis sets: Robust systematic basis sets for explicitly correlated and conventional correlated ab initio methods?
,”
J. Chem. Theory Comput.
7
,
33
43
(
2011
).
68.
F.
Ballesteros
,
S.
Dunivan
, and
K. U.
Lao
, “
Coupled cluster benchmarks of large noncovalent complexes: The L7 dataset as well as DNA-ellipticine and buckycatcher-fullerene
,”
J. Chem. Phys.
154
,
154104
(
2021
).
69.
E. R.
Davidson
, “
Selection of the proper canonical Roothaan-Hartree-Fock orbitals for particular applications. I. Theory
,”
J. Chem. Phys.
57
,
1999
2005
(
1972
).
70.
M.
Feyereisen
,
G.
Fitzgerald
, and
A.
Komornicki
, “
Use of approximate integrals in ab initio theory. An application in MP2 energy calculations
,”
Chem. Phys. Lett.
208
,
359
363
(
1993
).
71.
D. E.
Bernholdt
and
R. J.
Harrison
, “
Large-scale correlated electronic structure calculations: The RI-MP2 method on parallel computers
,”
Chem. Phys. Lett.
250
,
477
484
(
1996
).

Supplementary Material