---
title: Size-Consistent BW-s2 Perturbation Theory
url: https://www.emergentmind.com/topics/size-consistent-brillouin-wigner-approach-bw-s2
type: topic
---

# Size-Consistent BW-s2 Perturbation Theory

The **Size-Consistent Brillouin-Wigner approach**, usually abbreviated **BW-s2**, is a repartitioned second-order Brillouin-Wigner perturbation theory designed to preserve the regular, energy-denominator-dressed character of Brillouin-Wigner perturbation theory while repairing its traditional lack of size consistency and size extensivity. In its original formulation, BW-s2 is a second-order, self-consistent, orbital-invariant method built from a modified Hamiltonian partitioning in which a specially chosen one-electron regularizer is embedded in the zeroth-order Hamiltonian; in practical form, it resembles MP2 with self-consistently dressed occupied orbital energies rather than bare Hartree-Fock orbital energies [2303.06271]. Subsequent work reinterpreted BW-s2 as a one-parameter family BW-s2($\alpha$), where $\alpha$ controls the strength of the amplitude-dependent regularization [2309.01376], and later extended the approach to periodic solids, where it was assessed as a regularized second-order alternative to MP2 for metals, semiconductors, molecular crystals, and rare-gas solids [2508.15744].

## 1. Historical and conceptual setting

BW-s2 was introduced against a specific background: conventional MP2 is simple and low cost, but its denominator structure makes it unreliable in small-gap and near-degenerate regimes, and it can severely overbind noncovalent interactions and some transition-metal problems. Ordinary second-order Brillouin-Wigner perturbation theory offers an appealing counterpoint because the correlation energy appears in the denominator and regularizes the expansion, but that same energy dependence destroys size consistency and size extensivity in truncated form [2303.06271].

The original BW-s2 construction addresses exactly this tradeoff. It seeks a second-order Brillouin-Wigner-like theory that remains regular in small-gap situations while becoming size-consistent, size-extensive, and orbital invariant through second order. The later BW-s2($\alpha$) formulation retains the same formal structure but interprets the method’s scaling factor as a regularization strength rather than solely as an exact-condition parameterization [2309.01376].

A central conceptual distinction is therefore between **ordinary BW2**, whose denominator contains a global extensive correlation-energy shift, and **BW-s2**, which replaces that global shift by occupied-space energy dressing generated from the amplitudes themselves. This suggests that BW-s2 is best viewed neither as a conventional level shift nor as a purely gap-dependent damping formula, but as a self-consistent denominator renormalization derived from a repartitioned Brillouin-Wigner framework [2303.06271].

## 2. Repartitioned Hamiltonian and defining equations

The formal starting point is a repartitioning of the Hamiltonian,
$$
\hat H=\hat{\bar H}_0+\lambda \hat{\bar V},
$$
with
$$
\begin{aligned}
\hat{\bar H}_0 &= \hat H_0+\hat R,\\
\hat{\bar V}   &= \hat V-\hat R,
\end{aligned}
$$
where $\hat H_0$ is the Fock operator and $\hat R$ is a one-electron regularizer,
$$
\hat R=\sum_{ij} r_{ij} a_j^\dagger a_i.
$$
Under this repartitioning, the modified second-order Brillouin-Wigner expression is
$$
E^{(2)}= \sum_{k\neq0} \frac{\langle\Phi_0|\hat V|\Phi_k\rangle \langle\Phi_k|\hat V|\Phi_0\rangle} {(E_0-E_k)+(E_{\mathrm R,0}-E_{\mathrm R},k)+E^{(2)}}.
$$
This is the core formal object from which BW-s2 is derived [2303.06271].

To enforce orbital invariance and obtain a size-consistent second-order theory, the regularizer is written in tensor form. In the original BW-s2 paper,
$$
R_{ijkl}^{abcd} = \frac{1}{2} \left( W_{ik}\delta_{jl}+\delta_{ik}W_{jl} \right)\delta_{ac}\delta_{bd},
$$
with
$$
W_{ij}= \frac{1}{2}\sum_{kab} \left[ t_{ik}^{ab}(jk||ab)+t_{jk}^{ab}(ik||ab) \right].
$$
A crucial identity is
$$
\mathrm{tr}(\mathbf W)=E^{(2)}.
$$
The later BW-s2($\alpha$) formulation generalizes the regularizer to
$$
R_{ijkl}^{abcd} = \frac{\alpha}{2}(W_{ik}\delta_{jl}+\delta_{ik}W_{jl})\delta_{ac}\delta_{bd},
$$
so that each value of $\alpha$ defines a valid variant denoted BW-s2($\alpha$) [2303.06271, 2309.01376].

This construction is the formal reason BW-s2 differs from ordinary BW2. In ordinary BW2, the denominator depends explicitly on the total second-order correlation energy of the full system. In BW-s2, the trace contribution generated by $\mathbf W$ cancels that global term through second order, leaving only orbital-resolved occupied-space shifts [2303.06271].

## 3. Practical working form and self-consistent solution

In computation, BW-s2 is solved through dressed occupied orbital energies. The occupied-space generalized eigenproblem is
$$
\left(\mathbf F_{\mathrm{oo}+\frac{\alpha}{2}\mathbf W\right)\mathbf U = \tilde\varepsilon\,\mathbf U,
$$
which reduces to
$$
\left(\mathbf F_{\mathrm{oo}+\frac{1}{2}\mathbf W\right)\mathbf U = \tilde\varepsilon\,\mathbf U
$$
for the original $\alpha=1$ form. After rotation of the occupied orbitals, the amplitudes satisfy
$$
(\varepsilon_a+\varepsilon_b-\tilde\varepsilon_i-\tilde\varepsilon_j)\tilde t_{ij}^{ab} = -\tilde{\mathbb I}_{ijab},
$$
so that
$$
\tilde t_{ij}^{ab} = -\frac{\tilde{\mathbb I}_{ijab}}{\varepsilon_a+\varepsilon_b-\tilde{\varepsilon}_i-\tilde{\varepsilon}_j},
$$
and the correlation energy is evaluated as
$$
\tilde E_c = -\frac{1}{4}\sum_{ijab} \frac{|\tilde{\mathbb I}_{ijab}|^2}{\varepsilon_a+\varepsilon_b-\tilde{\varepsilon}_i-\tilde{\varepsilon}_j}.
$$
Thus the practical distinction from MP2 is entirely in the denominator: the occupied Hartree-Fock energies are replaced by self-consistently dressed occupied energies [2303.06271, 2309.01376].

Because $\mathbf W$ depends on the amplitudes and the amplitudes depend on the dressed energies derived from $\mathbf W$, BW-s2 is intrinsically iterative. The chemistry implementation reports formal cost
$$
m \times \mathcal{O}(N^5),
$$
where $m$ is the number of iterations, and states that $m$ is typically 4–6. That work used MP2 amplitudes as the starting guess, was implemented in a development version of **Q-Chem v6.0.2**, and used the **resolution-of-the-identity (RI)** approximation for two-electron integrals [2309.01376].

For periodic solids, the same idea appears as an occupied-space self-consistent dressing problem at each crystal momentum. The periodic occupied-occupied block of the regularizer is constructed from the first-order amplitudes, and the dressed occupied energies are obtained from
$$
\left(\mathbf{F}_{\mathbf{k},\mathrm{oo} + \alpha \mathbf{W}_{\mathbf{k}}\right)\mathbf{U}_{\mathbf k} = \tilde{\varepsilon}_{\mathbf{k},o}\mathbf{U}_{\mathbf k}.
$$
The resulting BW-s2 correlation energy per cell is
$$
E_{c}^{\textrm{BW-s2}} = \frac{1}{4N_k} \sum_{ijab} \sum_{\mathbf{k}_1\mathbf{k}_2\mathbf{q}} \frac{|\langle \tilde{i}_{\mathbf{k}_1}\tilde{j}_{\mathbf{k}_2+\mathbf{q}}||a_{\mathbf{k}_1+\mathbf{q}}b_{\mathbf{k}_2}\rangle|^2}{\tilde{\varepsilon}_{i_{\mathbf{k}_1}} + \tilde{\varepsilon}_{j_{\mathbf{k}_2+\mathbf{q}}} - \varepsilon_{a_{\mathbf{k}_1+\mathbf{q}}} - \varepsilon_{b_{\mathbf{k}_2}}}.
$$
The solid-state paper states that BW-s2($\alpha$) “scales the same as MP2, albeit with an extra factor of $N_{\mathrm{iter}}$ required to reach self-consistency,” and reports periodic MP2 scaling as
$$
\mathcal O(N_k^3 N^5).
$$
[2508.15744]

## 4. Formal properties

The defining formal claim of BW-s2 is that, **through second order**, it is size-consistent, size-extensive, and orbital invariant. The size-consistency proof proceeds by considering two infinitely separated closed-shell subsystems $A$ and $B$ and showing that the regularizer matrix becomes block diagonal,
$$
\mathbf W= \begin{bmatrix} \mathbf W_{AA} & \mathbf 0\\ \mathbf 0 & \mathbf W_{BB} \end{bmatrix},
$$
so that the BW-s2 energy expression
$$
E_c^{(2)}= -\frac{1}{4}\sum_{ijab} \frac{|(ij||ab)|^2}{\Delta_{ij}^{ab}+\frac{1}{2}(W_{ii}+W_{jj})}
$$
decomposes additively over subsystems. The resulting relation is
$$
E_c^{(2)}(A+B)=E_c^{(2)}(A)+E_c^{(2)}(B).
$$
This is the formal basis for the method’s size consistency and size extensivity through second order [2303.06271].

The same work gives numerical tests of these properties. For a separated $\mathrm{H}_2$ dimer, BW-s2 gives identical energies in canonical and Edmiston-Ruedenberg localized orbitals, whereas IEPA/BGE2 changes by **6.3 meV**. For He···Xe at **40 Å**, BW-s2 gives zero interaction energy, while BW2 leaves **111 meV** and xBW2 leaves **1 meV**. For He chains, the BW-s2 correlation energy per electron has zero slope versus chain length, like MP2 and xBW2 [2303.06271].

BW-s2 also satisfies a notable exact-condition result. In minimal-basis $\mathrm{H}_2$, the unscaled choice $\alpha=1$ is fixed by requiring exact recovery of the dissociation limit of the two-electron/two-orbital problem, and BW-s2 reaches the exact FCI dissociation limit regardless of whether the starting orbitals are RHF or UHF [2303.06271]. Later work explicitly recast $\alpha$ as a regularization parameter and emphasized that there is likely no single value satisfactory for all chemical contexts, even though each $\alpha$ defines a legitimate BW-s2 variant [2309.01376].

A necessary qualification is that the second-order guarantee is exactly that: **size-inconsistent terms re-enter at third and higher orders**. BW-s2 is therefore a second-order cure, not a general all-order linked-cluster reformulation [2303.06271].

## 5. Relation to ordinary BW, MP2, and other BW-like methods

BW-s2 is most clearly understood by contrast with neighboring perturbative constructions.

| Method | Denominator logic | Size-consistency status in the cited literature |
|---|---|---|
| MP2 | Bare orbital gaps $\Delta_{ij}^{ab}$ | Size-consistent reference second order |
| BW2 | Includes global $E^{(2)}$ shift | Not size-consistent or size-extensive |
| BW-s2 | Uses occupied-space dressing through $\mathbf W$ | Size-consistent through second order |
| $\kappa$-MP2 / $\sigma$-MP2 | Explicit gap-dependent damping | Regularized, but not BW-derived |

Ordinary BW2 can be written as
$$
E^{(2)}_{\text{BW}}= \sum_{k\neq 0} \frac{\langle\Phi_0|\hat V|\Phi_k\rangle \langle\Phi_k|\hat V|\Phi_0\rangle}{E_0-E_k+E^{(2)}_{\text{BW}}},
$$
whereas MP2 uses fixed Hartree-Fock gaps. BW-s2 keeps the Brillouin-Wigner-style self-consistent denominator idea but replaces the global denominator shift by local occupied-space dressing [2303.06271].

This distinction matters because several Brillouin-Wigner-related schemes are **not** BW-s2. A variant of Brillouin-Wigner perturbation theory with Epstein-Nesbet partitioning explicitly states that it “does not satisfy the size-consistency requirement,” so it should not be identified with BW-s2 [1307.4238]. Hybrid RS/BW schemes such as RSBW, iter-RSBW, multi-step RSBW, and SS-RSBW use Rayleigh-Schrödinger effective-Hamiltonian preconditioning followed by state-specific Brillouin-Wigner corrections; these methods may reduce size-consistency error or improve conditioning, but they do not present a formal size-consistent BW-s2 construction [2311.08356, 2408.16505, 2509.16152].

The same caution applies in nuclear many-body perturbation theory. Closed-shell and open-shell nuclear Brillouin-Wigner formulations with optimized partitioning parameter $\xi$ develop convergence criteria and efficient $\hat K$-box machinery for energy-dependent Bloch-Horowitz/Brillouin-Wigner perturbation theory, but they do **not** discuss size consistency, size extensivity, separability, or any method named BW-s2 [2306.13629, 2401.12691]. A common misconception is therefore to equate “convergent BW” with “size-consistent BW”; the cited nuclear literature supports the former, not the latter.

## 6. Benchmark performance in molecules and solids

In molecular electronic-structure benchmarks, BW-s2 was introduced as a parameter-free alternative to MP2 that improves behavior in bond dissociation, noncovalent interactions, and thermochemistry. The original paper reports that BW-s2 is exact for minimal-basis $\mathrm{H}_2$, gives correct single-bond dissociation behavior for ethane, and, for the full W4-11 set, reaches
$$
\mathrm{RMSE}=6.2~\text{kcal/mol},
$$
which is about **1.5 kcal/mol** better than MP2 and $\kappa$-MP2, while rivaling overall CCSD performance [2303.06271].

The later BW-s2($\alpha$) study systematically optimized $\alpha$ across noncovalent interactions, thermochemistry, alkane conformational energies, electronic response properties, and transition-metal datasets. It examined
$$
\alpha = 1.0,\ 2.0,\ 3.0,\ 3.5,\ 4.0,\ 4.5,\ 5.0,\ 6.0,\ 8.0,
$$
and concluded that
$$
\boxed{\alpha = 4}
$$
is a good compromise and “roughly optimal” with respect to both MRMSD and WTRMSD2. Representative improvements include **L7**, where the RMSD changes from **9.49** for MP2 to **1.47** for BW-s2(4), **X31**, where BW-s2(4) gives **0.26**, and **W4-11**, where BW-s2(4) gives **6.01** compared with **7.55** for MP2. For transition-metal chemistry, **MOR39** improves from **14.13** for MP2 to **6.23** for BW-s2(4), and **AuIrPt13** from **4.30** to **2.58** [2309.01376].

The same optimization study also clarifies transferability limits. BW-s2(4) substantially improves large and polarizable noncovalent problems and alkane conformers, but some barrier-height datasets degrade relative to MP2: for **HTBH38**, MP2 gives **5.03** while BW-s2(4) gives **5.54**; for **NHTBH38**, MP2 gives **2.40** and BW-s2(4) gives **4.01**. Even so, the authors state that BW-s2(4) damages barrier heights and electronic properties much less than gap-only regularizers such as $\kappa$-MP2 [2309.01376].

For periodic solids, BW-s2($\alpha$) was benchmarked as a regularized alternative to periodic MP2. In **BCC lithium**, where MP2 diverges, BW-s2 at $\alpha=1$ gives a cohesive energy of **1.67 eV/atom**, only **0.01 eV/atom** from the ZPE-corrected experimental value of **1.66 eV/atom**, and a lattice constant of **3.44 Å** versus experimental **3.45 Å**. In **diamond**, BW-s2($\alpha=2$) gives a cohesive energy of **7.50 eV/atom**, only **0.05 eV** below the experimental reference of **7.55 eV**. For the **benzene crystal**, BW-s2($\alpha=2$) is reported to give cohesive energies “more or less exact,” within the experimental uncertainty. By contrast, in **neon**, BW-s2 performs poorly: with $\alpha=2$ it predicts **non-binding**, and even with $\alpha=0.5$ it still severely underestimates cohesion relative to MP2 [2508.15744].

These results support a fairly sharp physical interpretation. BW-s2 performs best where MP2 fails because of excessive low-denominator pair correlation—metallic divergence, narrow-gap pathology, or overbinding in molecular crystals and some chemical datasets. It performs less well where MP2 already underbinds because important beyond-second-order many-body effects are missing, as in rare-gas solids [2508.15744].

## 7. Scope, limitations, and current interpretation

BW-s2 is a second-order method, not a replacement for systematically improvable higher-order many-body theories. The original paper is explicit that size-consistency errors re-enter at third and higher orders, and that the method remains only a second-order HF-based theory [2303.06271]. The optimization study likewise emphasizes that BW-s2 incorporates some higher-order correlation effects only implicitly through self-consistent denominator dressing, and that there is likely no universal $\alpha$ satisfactory for all chemical contexts [2309.01376].

The performance record also shows that regularization strength is context dependent. The original exact-condition choice is $\alpha=1$, fixed from minimal-basis $\mathrm{H}_2$ dissociation. For broad molecular chemistry, the later benchmark study recommends $\alpha=4$ as the most transferable compromise. For solids, the periodic paper identifies BW-s2($\alpha=2$) as particularly promising, while explicitly noting that $\alpha=2$ is neither derived from first principles nor claimed to be universal [2309.01376, 2508.15744].

A further limitation is that BW-s2 is self-consistent and iterative. The chemistry implementation reports several self-consistent cycles, and the solid-state implementation retains MP2-like formal scaling only up to an additional iteration factor [2309.01376, 2508.15744]. The method should therefore be understood as a practical regularized perturbation theory rather than a one-shot MP2 replacement.

Taken together, the cited literature defines BW-s2 as a specific second-order Brillouin-Wigner reformulation with a carefully chosen occupied-space regularizer, not as a generic label for self-consistent BW, hybrid RS/BW, or convergence-optimized Bloch-Horowitz methods. Its central achievement is narrow but important: it shows that one can retain the useful Brillouin-Wigner denominator logic while restoring size consistency through second order and preserving orbital invariance, with empirically useful consequences across molecular chemistry and, in adapted form, periodic electronic-structure calculations [2303.06271, 2508.15744].

Source: https://www.emergentmind.com/topics/size-consistent-brillouin-wigner-approach-bw-s2