Putting PASPT2 on a Firmer Basis

arXiv:2607.22152 · physics.chem-ph · Submitted 2026-07-24 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: We're looking at the paper: "Putting PASPT2 on a Firmer Basis".

Mira: Here's the gist of the paper. The paper introduces and analyzes the partial-active-space second-order perturbation theory (PASPT2) method, focusing on its properties regarding size-extensivity and intruder states to establish a more robust framework for strongly correlated systems.

Kai: The core finding is that the original variant of PASPT2, denoted as PASPT2H, is near-extensive for the target states supported by M0 and its weak violation of size-extensivity has hardly numerical consequence on those target states Near-extensive for the states with major projections on the chosen PAS M0.

Mira: Methodological Development The authors introduce several variants of PASPT2, including PASPT2M, which is defined by Eq.

Kai: (38), and PASPT2X, which is defined by Eq.

Title and authors: Kai: So we're looking at this paper today, "Putting PASPT2 on a Firmer Basis," which is about making the partial active space second order perturbation theory, or PASPT2, more reliable for strongly correlated systems.

Mira: Right. It’s trying to address some of the fundamental issues with the original version of PASPT2 by adding some new criteria and variants to it.

Lev: From an engineering standpoint, I’m interested in how this affects our ability to run these calculations on actual hardware, because those methods need to be stable for real physical systems.

Kai: Well, what the paper is saying is that they've developed several variants of PASPT2—like PASPT2M and PASPT2X—and they’re showing that while some are strictly size-extensive and size-consistent, they still have problems with intruder states.

Mira: And the main focus here seems to be on the original variant, which they call PASPT2H, because it’s apparently intruder free for all target states supported by a specific choice of the partial active space M0.

Lev: But if it's not strictly size-extensive, how do we know if our results will stay accurate when we try to model something larger than what was in the initial calculation?

Kai: That’s where they introduce a new concept called generalized size-extensivity, which is an operational definition to test the method numerically. They found that for PASPT2H, the size-extensivity errors for the lowest thirteen states are only between one and forty meV when compared against a specific model space M0.

Mira: That’s a relatively small error, which suggests that even though it's not strictly size-extensive in the strict sense, its impact on those target states is minimal.

Lev: So what this means for us running these calculations, it suggests we can use PASPT2H with some confidence when modeling strongly correlated solid states because the energy scaling isn't completely broken.

Kai: Exactly. The paper also introduces several improvements to the original idea, focusing on how to treat static and dynamic correlations in a more adaptive way, which they say is necessary because any fixed reference space is inherently unbalanced.

Mira: That points toward a method that can handle the entanglement between different types of correlation components better than what was there before.

Lev: And another thing they suggest is using hybrid model spaces, where you combine the M0 and the orthogonal complement RX within a larger closed space MX as a buffer to treat states supported by RX more soundly.

Kai: This hybrid approach helps manage states that might not be fully supported by the initial one-particle basis set M0, making them easier to handle in practice.

Mira: It’s also important for machine learning applications because the paper argues that because PASPT2 is state-universal, it becomes a very good model for training machine learning models to find those dressed active space Hamiltonians.

Lev: From an error control perspective, the authors emphasize that PASPT2H can be used to construct an effective Hamiltonian that is "manifestly connected," which helps ensure the energies are connected upon diagonalization, stabilizing calculations for states with major projections on M0.

Kai: So, to wrap up this paper on "Putting PASPT2 on a Firmer Basis," the main conclusion is that PASPT2H is near-extensive for target states strongly projected onto M0, and this weak size-extensivity violation doesn't cause big numerical problems for those states.

Mira: They justify the original claim that this variant is up to date because it’s intruder free for all target states supported by the chosen partial active space M0, which can be generated through methods like sCI.

Lev: And if we consider what this means practically, it suggests we have a method that works well for strongly correlated solid state calculations where size-extensivity is a big deal.

Kai: So, to summarize the core findings of "Putting PASPT2 on a Firmer Basis," they’ve shown that PASPT2H is near-extensive for states with major projections on M0, and this weak violation of size-extensivity has hardly any numerical consequence for those states.

Mira: The paper also lays out some really useful improvements, like the adaptive treatment of static and dynamic correlations, which guides future development because a fixed reference space is just too inflexible to handle everything.

Lev: And that hybrid model space utilization idea is interesting; using the RX space as a buffer means we can get a physically sound approximate treatment for those states with major projections on RX.

Kai: Ultimately, the state-universal nature of PASPT2 makes it an ideal model for training machine learning models to find dressed active space Hamiltonians across different molecular states efficiently.

Mira: It seems like they’ve given us a much more transparent framework for handling the complex entanglement of static and dynamic correlations in these strongly correlated systems.

Lev: So, when we think about putting this on real hardware, PASPT2H offers a stabilized path forward for calculations where size-extensivity isn't perfectly achieved but is at least near enough for our needs.

Kai: We’re wrapping up our discussion on "Putting PASPT2 on a Firmer Basis." It sounds like this work gives us a much more reliable tool for modeling these difficult quantum systems.

The paper's summary: Kai: So, we’ve seen how they tested PASPT2H numerically, and now we’re looking at what they actually conclude about this paper, "Putting PASPT2 on a Firmer Basis."

Mira: Basically, they say that while some other variants of PASPT2 are strictly size-extensive and size-consistent, they still have those pesky intruder state problems.

Kai: But the main focus is on this original version, PASPT2H. And what they find is that for the states that project strongly onto their chosen partial active space M0, it’s near-extensive.

Mira: That means even if it's not perfectly size-extensivity compliant in every single case, the error doesn't really matter numerically for the specific target states they care about. It’s just a small one.

Kai: Right. They use this operational definition of generalized size-extensivity to test it, and they found that for those lowest thirteen states, the error is only around one to forty meV when compared against their chosen model space M0.

Mira: That level of error is what makes them say that it’s practically fine for strongly correlated solid state calculations. It’s not a perfect solution, but it works well enough in practice for those specific states.

Kai: And they point out that this near-extensivity is actually a big plus because the original PASPT2H variant is intruder-free for all target states supported by the partial active space M0, which can be generated using methods like sCI.

Mira: That’s a huge piece of information. If you’re using something like sCI to pick your model space, and PASPT2H doesn't have intruder states in those spaces, that makes it much more reliable for whatever you're trying to calculate.

Kai: So, the paper concludes that the original claim about PASPT2 being the only size-consistent and intruder-free one is actually justified when you look at this variant. It’s a solid foundation.

Mira: And they also highlight how this method has these other advantages, like spin adaptation and being state-universal, which makes it a great starting point for training machine learning models to figure out those dressed active space Hamiltonians.

Kai: It seems like the big implication here is that we can use PASPT2H with more confidence in strongly correlated systems because its error scaling is manageable for the states we are interested in.

Mira: So, it’s not a perfect method yet, but it’s a much firmer basis for moving forward with calculations where size-extensivity isn't perfectly achieved but is at least near enough.

Kai: And this leads us to think about those improvements they suggest, like the adaptive treatment of correlation components and using hybrid model spaces as buffers for states in the orthogonal space.

Mira: Yeah, that hybrid idea is interesting because it lets you treat states supported by the orthogonal space RX more soundly when dealing with complex systems.

Kai: So we’ve seen what they built and where it stands now—it’s a near-extensivity win for PASPT2H, but the real work for hardware is still figuring out how to keep those calculations stable on real superconducting qubits.

The paper's improvements: Kai: So we’ve looked at how they tested PASPT2H numerically, and now we’re going to talk about what improvements this paper suggests for making the method even better.

Mira: Basically, they suggest ways to handle those static and dynamic correlations in a more adaptive way, because a fixed reference space just isn't flexible enough anymore.

Kai: That points toward a method that can adjust how it treats different kinds of correlation components depending on what’s happening in the system right now. It needs to be more dynamic than before.

Mira: And they propose using hybrid model spaces, where you mix the M0 and the orthogonal complement space RX within a larger closed space MX as a buffer to treat those states in RX more soundly.

Kai: That hybrid approach helps manage states that aren't fully supported by the initial one-particle basis set M0, which makes it easier to get physically sound approximate results for those states projecting onto RX.

Mira: And on the machine learning side, they emphasize that because PASPT2 is state-universal, it becomes an ideal model for training machine learning models to find those dressed active space Hamiltonians across different molecular states efficiently.

Kai: That’s a big idea; using the method as a template for AI training means you can capture the physics across many different molecular states without having to re-train everything from scratch.

Mira: It seems like they’re pushing for a more sophisticated framework that allows the theory to evolve with the complexity of strongly correlated systems rather than just being stuck in one fixed approximation.

Kai: So, the implication is that we might see a future where these methods aren't just static calculations but become part of an adaptive system capable of handling those complex entanglement issues better.

Mira: Right. It’s about moving from a set of fixed rules to a more flexible framework that can respond to how correlated the system actually gets.

Kai: That flexibility is exactly what we need when we’re trying to model things that are highly dynamic, like things in condensed matter or even high-energy physics where you have rapid changes.

Mira: And this opens up new avenues for understanding those materials, since they can adapt to the actual correlation structure instead of just being forced into a pre-defined box.

Kai: So we’ve seen the improvements—the adaptive treatment and hybrid spaces—and it looks like the next step is really about building systems that can respond to that adaptability in real time.

Mira: It’s about giving the theory more tools to handle the messiness of real-world correlation structures without just throwing up another complicated, rigid approximation.

Conclusion: Kai: So to wrap up this session on "Putting PASPT2 on a Firmer Basis," the main thing is that they’ve validated their original claim about PASPT2H by showing its near-extensivity for states with major projections on M0.

Mira: It means the method is much more trustworthy for strongly correlated solid state calculations because even though it’s not perfectly size-extensive everywhere, the numerical consequences are minimal for the specific states they are targeting.

Kai: Right. They’ve shown that this variant is intruder-free across all target states supported by their chosen partial active space M0, which is a significant piece of evidence.

Lev: From an error control standpoint, it’s good to see that even with these caveats, the method has a defined scaling for the size-extensivity error, which is helpful when you start thinking about running this on actual hardware where overhead matters.

Mira: I agree. And those other improvements they suggest—like adaptive correlation treatment and hybrid model spaces—show them a clear path for how to evolve this theory beyond just fixing the original version.

Kai: So, the implication is that we can treat these Hamiltonians as more robust starting points for machine learning, which is huge because it allows us to train models that capture physics across different molecular states efficiently.

Lev: And from an engineering view, if you can get a method this stable for theory, it gives us a better target when we try to map those complex interactions onto actual quantum hardware like superconducting circuits.

Mira: So, the paper "Putting PASPT2 on a Firmer Basis" gives us a more transparent and reliable tool for handling the entanglement of static and dynamic correlations in these strongly correlated systems.

Kai: It really does. We’ve seen that this work lays down a much firmer mathematical foundation for using PASPT2H in our research.

Lev: Moving forward, I think we need to see how quickly these adaptive features can be implemented with low overhead so it actually becomes usable on real-world quantum computers.

Mira: That’s the next big question for me—how much flexibility do we actually gain by adding these new components versus just the computational cost of those extra terms?

Kai: Right, so we’ve got this solid update on PASPT2H, but now we need to see how fast we can build systems that can actually utilize these new adaptive features.

Chunzhang Liu, Ning Zhang, Wenjian Liu

Qingdao Institute for Theoretical and Computational Sciences · Center for Optics Research and Engineering, Shandong University

physics.chem-ph

Submitted: 2026-07-24

Updated: 2026-10-30

Comments: 3 figures,1 table

DOI: 10.1021/prechem.6c00106

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 83/100

The gist: The paper introduces and analyzes the partial-active-space second-order perturbation theory (PASPT2) method…

Key concepts

PASPT2
A method used in quantum chemistry to calculate energy corrections by considering only a subset of orbitals (the active space). It uses second-order perturbation theory to account for electron correlation effects, which are crucial for accurately describing strongly interacting electrons.
Size-Extensivity
A property where the calculated energy scales correctly when the system size is doubled. If you double the number of electrons in a system, the calculated energy should scale predictably. This property is important for ensuring that calculations are physically meaningful for large systems like solids.
Intruder States
These are states that appear in a calculation but do not correspond to physical states of the actual system. They cause problems because they can introduce unphysical results into the energy calculations, making the method less trustworthy.

Terminology

Summary

The paper introduces and analyzes the partial-active-space second-order perturbation theory (PASPT2) method, focusing on its properties regarding size-extensivity and intruder states to establish a more robust framework for strongly correlated systems. p. 4

Where the paper says it

fills as many of them as possible. Motivated by the desire of having an adaptive and size-extensive many-body theory for multiple states of strongly correlated electrons, we have recently developed a partial-active-space (PAS) second-order perturbation theory (PASPT2).

The core finding is that the original variant of PASPT2, denoted as PASPT2H, is near-extensive for the target states supported by M0 and its weak violation of size-extensivity has hardly numerical consequence on those target states Near-extensive for the states with major projections on the chosen PAS M0. p. 4

Where the paper says it

However, it is near-extensive and the weak violation of size-extensivity has hardly numerical consequence on the target states. To this end, we will derive two new variants (PASPT2M and PASPT2X) of PASPT2 and demonstrate that our original variant (to be renamed to 5

Methodological Development

The authors introduce several variants of PASPT2, including PASPT2M, which is defined by Eq. (38), and PASPT2X, which is defined by Eq. p. 0

Where the paper says it

Nevertheless, PASPT2 is near-extensive for the states with major projections on the chosen PAS M0. This becomes more transparent upon introducing PASPT2X, a strictly size-extensive variant.

(37). Both the variant defined by Eq. p. 0
Where the paper says it

Nevertheless, PASPT2 is near-extensive for the states with major projections on the chosen PAS M0. This becomes more transparent upon introducing PASPT2X, a strictly size-extensive variant.

(46) along with Eqs. p. 1
Where the paper says it

This issue can largely be alleviated by choosing a sufficiently large CAS. However, a large CAS is itself problematic in many aspects:1 for instance, how to select which and how many orbitals as active orbitals, how to maintain a consistent CAS during orbital optimization and along potential energy surface scans, not to mention the combinatorial explosion in computational cost.

(39) and (47) for the amplitudes as well as Eq. p. 0
Where the paper says it

cn Abstract The recently proposed partial-active-space (PAS) multi-state second-order perturbation theory (PASPT2) [Precis. 4, 997 (2026)] features connected amplitudes and a connected and closed intermediate Hamiltonian.

(52) for the effective Hamiltonian H eff[2]M, both the cluster operators and effective Hamiltonian are defined on a non-closed PAS M0. p. 0
Where the paper says it

The weak violation of size-extensivity is therefore numerically insignificant for the target states supported by M0, reinforcing the theoretical basis of PASPT2H. Keywords: strong correlation, partial active space, closed operator, quasi-open operator, open operators, connectivity of cluster amplitudes, connectivity and closure of effective Hamiltonian, size-extensive, near-extensive 1 arXiv:2607.

In contrast, in the variant defined by Eq. p. 0
Where the paper says it

Nevertheless, PASPT2 is near-extensive for the states with major projections on the chosen PAS M0. This becomes more transparent upon introducing PASPT2X, a strictly size-extensive variant.

(37) along with Eqs. p. 1
Where the paper says it

This issue can largely be alleviated by choosing a sufficiently large CAS. However, a large CAS is itself problematic in many aspects:1 for instance, how to select which and how many orbitals as active orbitals, how to maintain a consistent CAS during orbital optimization and along potential energy surface scans, not to mention the combinatorial explosion in computational cost.

(39) and (42) for the amplitudes as well as Eq. p. 0
Where the paper says it

cn Abstract The recently proposed partial-active-space (PAS) multi-state second-order perturbation theory (PASPT2) [Precis. 4, 997 (2026)] features connected amplitudes and a connected and closed intermediate Hamiltonian.

(21) for the effective Hamiltonian H eff[2]X, both the cluster operators and effective Hamiltonian are defined on MX, the closed counterpart of M0. p. 0
Where the paper says it

The weak violation of size-extensivity is therefore numerically insignificant for the target states supported by M0, reinforcing the theoretical basis of PASPT2H. Keywords: strong correlation, partial active space, closed operator, quasi-open operator, open operators, connectivity of cluster amplitudes, connectivity and closure of effective Hamiltonian, size-extensive, near-extensive 1 arXiv:2607.

Defining Effective Hamiltonians

The derivation involves starting from a multi-exponential Ansatz and expanding the Schrödinger equation to obtain a generalized Bloch equation. The effective Hamiltonian H eff is determined by the condition ⟨βe-T αHeT αα⟩ = sum γ∈M0⟨βe-T αe T γγH eff(γα). p. 7

Where the paper says it

Inserting the wave operator (1) into the Schrödinger equation leads to the generalized Bloch equation40,41 e −T α HeT α |α⟩ = ∑ γ∈M0 e −T α e T γ |γ⟩H eff γα, (8) H eff γα = ⟨γ|H eff|α⟩. (9) The effective Hamiltonian Heff is determined by the condition (without assuming the IN) ⟨β|e −T α HeT α |α⟩ = ∑ γ∈M0 ⟨β|e −T α e T γ |γ⟩H eff γα, (10) which can be expanded as ⟨β|H + [H, T α ] + 1 2 [[H, T α ], T α ] + · · · |α⟩ = ∑ γ∈M0 ⟨β|1 + (T γ − T α ) + 1 2 [T γ , T α ] + 1 2 (T γ − T α ) 2 + · · · |γ⟩H eff γα = H eff βα + ∑ γ∈M0,γ̸=α ⟨β|(T γ − T α ) + 1 2 [T γ , T α ] + 1 2 (T γ − T α ) 2 + · · · |γ⟩H eff γα.

The second-order effective Hamiltonian H eff[2] is derived as H eff[2] = P(H + sum α∈M0 [H, T α(1)]Pα)P. p. 9
Where the paper says it

This term should therefore be ignored, thereby leading to the following effective Hamiltonian correct to second order6 H eff[2] = P(H + ∑ α∈M0 [H, T α(1) ]Pα)P, (21) which is manifestly connected. The diagonal elements are simply the energies of the corresponding reference functions by (non-Hartree-Fock) second-order Møller-Plesset perturbation theory (MP2),42 viz.

Comparison of Variants

Table 1 compares the variants of PASPT2, showing that PASPT2M and PASPT2X are both strictly size-extensive (and size-consistent). However, both are plagued by the intruder-state problem. p. 2

Where the paper says it

Given a GMS, the intruder-state problem7 that plagues nonvariational methods can persist. In particular, methods that are immune to this problem for lowlying states may encounter it again for high-lying states.

p. 2
Where the paper says it

Given a GMS, the intruder-state problem7 that plagues nonvariational methods can persist. In particular, methods that are immune to this problem for lowlying states may encounter it again for high-lying states.

In contrast, the original variant of PASPT2, PASPT2H, is intruder-free for all target states supported by the chosen PAS (which can readily be generated by sCI). p. 0
Where the paper says it

Compared with the intruder-prone PASPT2X, the intruder-free PASPT2H merely neglects second-order corrections that are important only for those states with major projections on the orthogonal complement RX of M0 within the closed space MX (= M0 ⊕ RX). Such states are usually not supported by the chosen finite one-particle basis set.

p. 0
Where the paper says it

Compared with the intruder-prone PASPT2X, the intruder-free PASPT2H merely neglects second-order corrections that are important only for those states with major projections on the orthogonal complement RX of M0 within the closed space MX (= M0 ⊕ RX). Such states are usually not supported by the chosen finite one-particle basis set.

Testing Size-Extensivity

The operational definition of size-extensivity (GSE) is introduced to test the method numerically. Testing shows that PASPT2M is indeed strictly size-extensive, whereas PASPT2H is not size-extensive, see Table 4. p. 0

Where the paper says it

4, 997 (2026)] features connected amplitudes and a connected and closed intermediate Hamiltonian. Despite these hallmarks, PASPT2 (denoted as PASPT2H from now on) is not strictly size-extensive, as originally thought (and numerically confirmed), albeit strictly size-consistent.

However, the size-extensivity errors for the lowest 13 states are merely 1 40 meV for the lowest 13 states (ordered according to ER/S). p. 4
Where the paper says it

It will be shown that PASPT2 is not strictly size-extensive, as originally thought (and numerically confirmed), albeit strictly size-consistent. However, it is near-extensive and the weak violation of size-extensivity has hardly numerical consequence on the target states.

Conclusion and Merits

The paper concludes that PASPT2H is near-extensive for the target states with major projections on M0. The combination of spin-adaptation, its combination with relativistic-QED Hamiltonians, and its state-universality renders it an ideal model for machine learning of dressed activespace Hamiltonians. p. 3

Where the paper says it

13 Therefore, the method should readily be combined with relativistic-QED Hamiltonians14,15 or spin-separated exact two-component (X2C) relativistic Hamiltonians16,17 for treating spin-orbit coupling and correlation on an equal footing. They require the use of spin-dependent and spin-free configuration state functions as the manyelectron basis, respectively.

Test of generalized size-extensivity

The GSE procedure involves partitioning the whole set of orbitals into two disjoint subsets R and S in an arbitrary manner to split the two-body operator V α (67) into three terms V α = V α R + V α S + V α RS. The calculation of the extensivity error ∆GSE = ERS − (ER/S0 + ER0/S − E0R/S) is performed for a given method. p. 2

Where the paper says it

Given a GMS, a method that is size-extensive is still highly desired. Note that size-extensivity (or connectedness of the energy)2,3 is a mathematical property of the method itself.

Additional Merits

The additional merits of PASPT2 include spin-adaptation can readily be achieved, its combination with relativistic-QED Hamiltonians14,15 or spin-separated X2C Hamiltonians16,17 is straightforward following our previous works (3). Its state-universality renders it an ideal model for machine learning of dressed activespace Hamiltonians. p. 1

Where the paper says it

To guide future development of ab initio quantum chemical methods for accurately describing such systems, we introduce the following criteria for a good wave function Ansatz: (1) Adaptive treatment of static and dynamic correlations. This is necessitated by the strong entanglement and mutual interchangeability of the two components, which renders any structured reference space (e.

Its near-extensivity enables its application to strongly correlated solid states. p. 1
Where the paper says it

1 Introduction Strongly correlated systems, especially those with many unpaired electrons, feature both strong static and strong dynamic correlations. The former stems from strong couplings among near-degenerate configurations and demands a multi-configurational description even just for a qualitative description, whereas the latter originates from contributions of high-lying configurations characterized by small coefficients in the target wave function.

Table 1

Table 1 Comparison of variants of PASPT2 variant H α0 T α(1) Heff[2] size-extensive size-consistent intruder-free rigorous rigorous no domaina full M0 M0 property additively separable connected connected & closed PASPT2M equation (38), (39), (47) (46) (52) rigorous rigorous no domaina full MX MX property additively separable connected connected & closed PASPT2X equation (38), (39), (42) (37) rigorous rigorous no domaina full MX property additively separable connected connected & closed a M0 ⊕ RX = MX.

Figure 2

Figure 2 Structures of A matrices in Eq. (59) for PASPT2H and PASPT2X/PASPT2M Unfilled areas are null. p. 4

Where the paper says it

However, it is near-extensive and the weak violation of size-extensivity has hardly numerical consequence on the target states. To this end, we will derive two new variants (PASPT2M and PASPT2X) of PASPT2 and demonstrate that our original variant (to be renamed to 5

p. 4
Where the paper says it

However, it is near-extensive and the weak violation of size-extensivity has hardly numerical consequence on the target states. To this end, we will derive two new variants (PASPT2M and PASPT2X) of PASPT2 and demonstrate that our original variant (to be renamed to 5

Figure 3

Figure 3 Structures of the A matrices in Eq. (59) for PASPT2H and PASPT2X/PASPT2M Unfilled areas are null.

Table 2

Table 2 Model spaces M0 for equilibrium water with MS = 0 Irrep A1 A2 B1 B2 oCFGa HF⟩ b Sizes of M0, MX and CAS(8,8) (10, 104, 1234) (12, 22, 1234) (8, 10, 1216) (14, 104, 1216) (M'0, MX, MC) c M'0 is selected from MX with absolute coefficients larger than 0.05

Table 3

Table 3 Test of generalized size-extensivity of PASPT2M/cc-pVDZ for 22 lowest singlet states of equilibrium water with model spaces M0 in Table 2 The energies for the ground and excited states are given in Eh and eV, respectively. ∆GSE are given in meV (mean absolute error: 3. p. 2

Where the paper says it

(2) Size-extensive. Given a GMS, a method that is size-extensive is still highly desired.

p. 2
Where the paper says it

(2) Size-extensive. Given a GMS, a method that is size-extensive is still highly desired.

044 meV)

Table 4

Table 4 Test of generalized size-extensivity of PASPT2H/cc-pVDZ for 22 lowest singlet states of equilibrium water with model spaces M0 in Table 2 The energies for the ground and excited states are given in Eh and eV, respectively. ∆GSE are given in meV (mean absolute error: 21. p. 2

Where the paper says it

(2) Size-extensive. Given a GMS, a method that is size-extensive is still highly desired.

090 meV)

Table 5

Table 5 Test of generalized size-extensivity of PASPT2H/cc-pVDZ for 22 lowest singlet states of equilibrium water with enlarged model spaces M'0 in Table 2 The energies for the ground and excited states are given in Eh and eV, respectively. ∆GSE are given in meV (mean absolute error: 3.044 meV)

Table 6

Table 6 Test of generalized size-extensivity of PASPT2X/cc-pVDZ for 20 lowest singlet states of equilibrium water with CAS(4,4) as MX The energies for the ground and excited states are given in Eh and eV, respectively. ∆GSE are given in meV. p. 2

Where the paper says it

(2) Size-extensive. Given a GMS, a method that is size-extensive is still highly desired.

p. 2
Where the paper says it

(2) Size-extensive. Given a GMS, a method that is size-extensive is still highly desired.

Figure 4

Figure 4 Size-extensivity errors of PASPT2H with different model spaces (see Table 2)

Conclusion

Three variants of PASPT2 have been formulated. Both PASPT2M and PASPT2X are strictly size-extensive (and size-consistent). p. 4

Where the paper says it

However, it is near-extensive and the weak violation of size-extensivity has hardly numerical consequence on the target states. To this end, we will derive two new variants (PASPT2M and PASPT2X) of PASPT2 and demonstrate that our original variant (to be renamed to 5

However, both are plagued by the intruder-state problem. In particular, PASPT2M has only limited accuracy. p. 4
Where the paper says it

However, it is near-extensive and the weak violation of size-extensivity has hardly numerical consequence on the target states. To this end, we will derive two new variants (PASPT2M and PASPT2X) of PASPT2 and demonstrate that our original variant (to be renamed to 5

In contrast, the original variant of PASPT2, PASPT2H, is intruder-free for all target states supported by the chosen PAS (which can readily be generated by sCI). As confirmed by comparing with PASPT2X and by numerical evidence as well, PASPT2H is near-extensive and the weak violation of size-extensivity has hardly numerical consequence on the target states. p. 4
Where the paper says it

It will be shown that PASPT2 is not strictly size-extensive, as originally thought (and numerically confirmed), albeit strictly size-consistent. However, it is near-extensive and the weak violation of size-extensivity has hardly numerical consequence on the target states.

Therefore, the original claim that “PASPT2 is up to date the only size-extensive, size-consistent and intruder-free PAS-based multi-state MRPT2” is fully justified. p. 0
Where the paper says it

Despite these hallmarks, PASPT2 (denoted as PASPT2H from now on) is not strictly size-extensive, as originally thought (and numerically confirmed), albeit strictly size-consistent. Nevertheless, PASPT2 is near-extensive for the states with major projections on the chosen PAS M0.

Acknowledgments

This work was supported by the National Natural Science Foundation of China (Grant Nos. 22373057 and 22503051).

Data availability

No new data were generated or analyzed in this study.

Conflicts of interest

There are no conflicts to declare.

References

(1) Lei, Y.; Suo, B.; Liu, W. iCAS: Imposed Automatic Selection and Localization of Complete Active Spaces. p. 1

Where the paper says it

This is necessitated by the strong entanglement and mutual interchangeability of the two components, which renders any structured reference space (e. , complete active space (CAS) or generalized valence bond) inherently inflexible and unbalanced.

J. Chem. Theory Comput. p. 0
Where the paper says it

Putting PASPT2 on a Firmer Basis Chunzhang Liu, Ning Zhang, and Wenjian Liu∗ Qingdao Institute for Theoretical and Computational Sciences and Center for Optics Research and Engineering, Shandong University, Qingdao, Shandong 266237, China E-mail: liuwj@sdu. cn Abstract The recently proposed partial-active-space (PAS) multi-state second-order perturbation theory (PASPT2) [Precis.

p. 0
Where the paper says it

Putting PASPT2 on a Firmer Basis Chunzhang Liu, Ning Zhang, and Wenjian Liu∗ Qingdao Institute for Theoretical and Computational Sciences and Center for Optics Research and Engineering, Shandong University, Qingdao, Shandong 266237, China E-mail: liuwj@sdu. cn Abstract The recently proposed partial-active-space (PAS) multi-state second-order perturbation theory (PASPT2) [Precis.

2021, 17, 4846–4859

(2) Bartlett, R. J.; Purvis, G. D. Many-body perturbation theory, coupled-pair manyelectron theory, and the importance of quadruple excitations for the correlation problem. p. 4

Where the paper says it

fills as many of them as possible. Motivated by the desire of having an adaptive and size-extensive many-body theory for multiple states of strongly correlated electrons, we have recently developed a partial-active-space (PAS) second-order perturbation theory (PASPT2).

Int. J. Quantum Chem. p. 0
Where the paper says it

22152v3 [physics. chem-ph] 8 Oct 2026

p. 0
Where the paper says it

22152v3 [physics. chem-ph] 8 Oct 2026

p. 0
Where the paper says it

22152v3 [physics. chem-ph] 8 Oct 2026

1978, 14, 561–581

(3) Nooijen, M.; Shamasundar, K.; Mukherjee, D. Reflections on size-extensivity, sizeconsistency and generalized extensivity in many-body theory. p. 4

Where the paper says it

fills as many of them as possible. Motivated by the desire of having an adaptive and size-extensive many-body theory for multiple states of strongly correlated electrons, we have recently developed a partial-active-space (PAS) second-order perturbation theory (PASPT2).

Mol. Phys. 2005, 103, 2277–2298

(4) Pople, J. A.; Binkley, J. S.; Seeger, R. Theoretical models incorporating electron correlation. p. 0

Where the paper says it

The weak violation of size-extensivity is therefore numerically insignificant for the target states supported by M0, reinforcing the theoretical basis of PASPT2H. Keywords: strong correlation, partial active space, closed operator, quasi-open operator, open operators, connectivity of cluster amplitudes, connectivity and closure of effective Hamiltonian, size-extensive, near-extensive 1 arXiv:2607.

Int. J. Quantum Chem. Symp. 1976, 10, 1–19

(5) Battaglia, S.; Galván, I. F.; Lindh, R. Theoretical and Computational Photochemistry; Elsevier, 2023; pp 135–162

(6) Liu, C.; Zhang, N. p. 0

Where the paper says it

Putting PASPT2 on a Firmer Basis Chunzhang Liu, Ning Zhang, and Wenjian Liu∗ Qingdao Institute for Theoretical and Computational Sciences and Center for Optics Research and Engineering, Shandong University, Qingdao, Shandong 266237, China E-mail: liuwj@sdu. cn Abstract The recently proposed partial-active-space (PAS) multi-state second-order perturbation theory (PASPT2) [Precis.

p. 0
Where the paper says it

Putting PASPT2 on a Firmer Basis Chunzhang Liu, Ning Zhang, and Wenjian Liu∗ Qingdao Institute for Theoretical and Computational Sciences and Center for Optics Research and Engineering, Shandong University, Qingdao, Shandong 266237, China E-mail: liuwj@sdu. cn Abstract The recently proposed partial-active-space (PAS) multi-state second-order perturbation theory (PASPT2) [Precis.

; Liu, W. PASPT2: a novel size-extensive and size-consistent partial-active-space multi-state multi-reference second-order perturbation theory for strongly correlated electrons. p. 4
Where the paper says it

fills as many of them as possible. Motivated by the desire of having an adaptive and size-extensive many-body theory for multiple states of strongly correlated electrons, we have recently developed a partial-active-space (PAS) second-order perturbation theory (PASPT2).

p. 4
Where the paper says it

fills as many of them as possible. Motivated by the desire of having an adaptive and size-extensive many-body theory for multiple states of strongly correlated electrons, we have recently developed a partial-active-space (PAS) second-order perturbation theory (PASPT2).

Precis. Chem. 2026, 4, 997–1015

(7) Schucan, T.; Weidenmüller, H. The effective interaction in nuclei and its perturbation expansion: An algebraic approach. p. 3

Where the paper says it

(9) Near-exact. If desired, the method can be made to approach full configuration interaction (FCI) systematically, by enlarging the GMS.

Ann. Phys. 1972, 73, 108–135

(8) Hayashi, M.; Saitow, M.; Uemura, K.; Yanai, T. Quasi-degenerate extension of local N-electron valence state perturbation theory with pair-natural orbital method based on localized virtual molecular orbitals. J. Chem. Phys. 2024, 160, 194105

(9) Li, Y.; Luo, S.; Wu, P.; Lei, Y. The application of Dyall Hamiltonian-based MRPT2 in high-lying electronically excited state calculations. p. 2

Where the paper says it

Given a GMS, the intruder-state problem7 that plagues nonvariational methods can persist. In particular, methods that are immune to this problem for lowlying states may encounter it again for high-lying states.

Chem. Phys. Lett. 2025, 871, 142096

(10) Roos, B. O.; Andersson, K. Multiconfigurational perturbation theory with level shift—the Cr2 potential revisited. p. 3

Where the paper says it

Nevertheless, it is still possible for a GMSbased method to yield continuous potential energy surfaces in two ways. One is to stick to the same set of configurations across geometries, and the other is to retain all configurations with appreciable coefficients in the GMS, so that the coefficients of the outer configurations can be estimated by perturbation with sufficient accuracy relative to those by diagonalization.

Chem. Phys. Lett. 1995, 245, 215–223

(11) Forsberg, N.; Malmqvist, P.-Å. Multiconfiguration perturbation theory with imaginary level shift. p. 0

Where the paper says it

Putting PASPT2 on a Firmer Basis Chunzhang Liu, Ning Zhang, and Wenjian Liu∗ Qingdao Institute for Theoretical and Computational Sciences and Center for Optics Research and Engineering, Shandong University, Qingdao, Shandong 266237, China E-mail: liuwj@sdu. cn Abstract The recently proposed partial-active-space (PAS) multi-state second-order perturbation theory (PASPT2) [Precis.

Chem. Phys. Lett. 1997, 274, 196–204

(12) Kong, L. Orbital invariance issue in multireference methods. p. 4

Where the paper says it

It turns out that PASPT2 satisfies all the above criteria except for the energy invariance under rotation of active orbitals, which is inherited from the Jeziorski-Monkhorst (JM) Ansatz31 for the wave operator. As a matter of fact, all methods rooted in this Ansatz lack orbital invariance, even when a CAS is taken as the reference space.

Int. J. Quantum Chem. 2010, 110, 2603–2613

(13) Liu, W. Perspective: Simultaneous treatment of relativity, correlation, and QED. p. 3

Where the paper says it

(10) Readily combined with relativity and even quantum electrodynamics (QED). The ultimate goal of electronic structure calculations is to make the left- and high-hand sides of the equation “relativity + correlation + QED = experiment” as close as possible.

WIRES Comput. Mol. Sci. 2023, 13, e1652

(14) Liu, W.; Lindgren, I. Going beyond “no-pair relativistic quantum chemistry”. p. 3

Where the paper says it

The ultimate goal of electronic structure calculations is to make the left- and high-hand sides of the equation “relativity + correlation + QED = experiment” as close as possible. 13 Therefore, the method should readily be combined with relativistic-QED Hamiltonians14,15 or spin-separated exact two-component (X2C) relativistic Hamiltonians16,17 for treating spin-orbit coupling and correlation on an equal footing.

J. Chem. Phys. 2013, 139, 014108 (E)144, 049901 (2016).

(15) Liu, W. p. 0

Where the paper says it

Putting PASPT2 on a Firmer Basis Chunzhang Liu, Ning Zhang, and Wenjian Liu∗ Qingdao Institute for Theoretical and Computational Sciences and Center for Optics Research and Engineering, Shandong University, Qingdao, Shandong 266237, China E-mail: liuwj@sdu. cn Abstract The recently proposed partial-active-space (PAS) multi-state second-order perturbation theory (PASPT2) [Precis.

Essentials of relativistic quantum chemistry. p. 3
Where the paper says it

The ultimate goal of electronic structure calculations is to make the left- and high-hand sides of the equation “relativity + correlation + QED = experiment” as close as possible. 13 Therefore, the method should readily be combined with relativistic-QED Hamiltonians14,15 or spin-separated exact two-component (X2C) relativistic Hamiltonians16,17 for treating spin-orbit coupling and correlation on an equal footing.

J. Chem. Phys. 2020, 152, 180901

(37) Li, X.; Paldus, J. General-model-space state–universal coupled-cluster theory: Connectivity conditions and explicit equations

(38) Paldus, J.; Li, X.; Petraco, N. D. General-model-space state–universal coupled-cluster method: Diagrammatic approach

(39) Lindgren, I.; Mukherjee, D. On the connectivity criteria in the open-shell coupledcluster theory for general model spaces

(40) Bloch, C. Sur la théorie des perturbations des états liés. Nuc. Phys. 1958, 6, 329–347

(41) Lindgren, I.; Morrison, J. Atomic many-body theory; Springer Science & Business Media, 2012; Vol. p. 4

Where the paper says it

fills as many of them as possible. Motivated by the desire of having an adaptive and size-extensive many-body theory for multiple states of strongly correlated electrons, we have recently developed a partial-active-space (PAS) second-order perturbation theory (PASPT2).

3

(42) Møller, C.; Plesset, M. p. 9

Where the paper says it

This term should therefore be ignored, thereby leading to the following effective Hamiltonian correct to second order6 H eff[2] = P(H + ∑ α∈M0 [H, T α(1) ]Pα)P, (21) which is manifestly connected. The diagonal elements are simply the energies of the corresponding reference functions by (non-Hartree-Fock) second-order Møller-Plesset perturbation theory (MP2),42 viz.

S. Note on an approximation treatment for many-electron systems. p. 2
Where the paper says it

Note that size-extensivity (or connectedness of the energy)2,3 is a mathematical property of the method itself. It means that the calculated energy scales asymptotically linearly with respect to the number of interacting electrons, which is a necessary condition for uniform accuracy as the system size increases–only size-extensive methods can be applied to extended systems.

Phys. Rev. 1934, 46, 618–622

(43) Meissner, L.; Kucharski, S. A.; Bartlett, R. J. A multireference coupled-cluster method for special classes of incomplete model spaces. p. 6

Where the paper says it

They further showed37,38 that the resulting IN-GMS-SUCC theory is size-extensive. However, this declaration is disapproved during the course of formulating PASPT26 by perturabtive expansion of IN-GMS-SU-CCSD (coupled-cluster with singles and doubles): there exists a disconnected term in the amplitude equation, which can only be removed to first order (by introducing a special zeroth-order Hamiltonian; vide post), but shows up again beyond the first order.

J. Chem. Phys. 1989, 91, 6187–6194

(44) Chaudhuri, R.; Sinha, D.; Mukherjee, D. On the extensivity of the roots of effective Hamiltonians in many-body formalisms employing incomplete model spaces. p. 4

Where the paper says it

Motivated by the desire of having an adaptive and size-extensive many-body theory for multiple states of strongly correlated electrons, we have recently developed a partial-active-space (PAS) second-order perturbation theory (PASPT2). 6 Here, the term PAS is regarded as equivalent to GMS18–20 (also called incomplete model space (IMS) or incomplete active space (IAS) in the literature) that emphasizes the complete lack of structure.

Chem. Phys. Lett. 1989, 163, 165–170

(45) Balkova, A.; Kucharski, S.; Meissner, L.; Bartlett, R. J. The multireference coupledcluster method in Hilbert space: An incomplete model space application to the LiH molecule. p. 4

Where the paper says it

Motivated by the desire of having an adaptive and size-extensive many-body theory for multiple states of strongly correlated electrons, we have recently developed a partial-active-space (PAS) second-order perturbation theory (PASPT2). 6 Here, the term PAS is regarded as equivalent to GMS18–20 (also called incomplete model space (IMS) or incomplete active space (IAS) in the literature) that emphasizes the complete lack of structure.

J. Chem. Phys. 1991, 95, 4311–4316

(46) Mukhopadhyay Jr, D.; Mukherjee, D. Molecular applications of size-extensive quasiHilbert-and quasi-Fock-space coupled-cluster formalisms using incomplete model spaces. p. 4

Where the paper says it

Motivated by the desire of having an adaptive and size-extensive many-body theory for multiple states of strongly correlated electrons, we have recently developed a partial-active-space (PAS) second-order perturbation theory (PASPT2). 6 Here, the term PAS is regarded as equivalent to GMS18–20 (also called incomplete model space (IMS) or incomplete active space (IAS) in the literature) that emphasizes the complete lack of structure.

Chem. Phys. Lett. 1991, 177, 441–446

(47) Liu, C.; Zhang, N.; Liu, W. PASPT2: a novel size-extensive and size-consistent partial-active-space multi-state multi-reference second-order perturbation theory for strongly correlated electrons. arXiv preprint arXiv:2512. p. 0

Where the paper says it

The weak violation of size-extensivity is therefore numerically insignificant for the target states supported by M0, reinforcing the theoretical basis of PASPT2H. Keywords: strong correlation, partial active space, closed operator, quasi-open operator, open operators, connectivity of cluster amplitudes, connectivity and closure of effective Hamiltonian, size-extensive, near-extensive 1 arXiv:2607.

16212v6 2026

(48) Zhang, N.; Liu, W. Unified Implementation of Relativistic Wave Function Methods: 4C-iCIPT2 as a Showcase. p. 1

Where the paper says it

The major challenge of such systems is that the static and dynamic components of correlation are strongly entangled and even interchangeable, thereby rendering a preset partitioning of the two generally invalid. To guide future development of ab initio quantum chemical methods for accurately describing such systems, we introduce the following criteria for a good wave function Ansatz: (1) Adaptive treatment of static and dynamic correlations.

J. Chem. Theory Comput. 2024, 20, 9003–9017

(49) Zhang, N.; Xiao, Y.; Liu, W. SOiCI and iCISO: combining iterative configuration interaction with spin–orbit coupling in two ways. p. 3

Where the paper says it

13 Therefore, the method should readily be combined with relativistic-QED Hamiltonians14,15 or spin-separated exact two-component (X2C) relativistic Hamiltonians16,17 for treating spin-orbit coupling and correlation on an equal footing. They require the use of spin-dependent and spin-free configuration state functions as the manyelectron basis, respectively.

J. Phys.: Condens. Matter 2022, 34, 224007

(50) Zhang, N.; Wang, Q.; Liu, W. MetaWave: A Platform for Unified Implementation of Nonrelativistic and Relativistic Wave Functions. p. 3

Where the paper says it

13 Therefore, the method should readily be combined with relativistic-QED Hamiltonians14,15 or spin-separated exact two-component (X2C) relativistic Hamiltonians16,17 for treating spin-orbit coupling and correlation on an equal footing. They require the use of spin-dependent and spin-free configuration state functions as the manyelectron basis, respectively.

J. Phys. Chem. A 2025, 129, 5170–5188

(51) Li, X.; Paldus, J. Performance of the general-model-space state-universal coupledcluster method. p. 1

Where the paper says it

Treating the former exactly but the latter approximately is inherently imbalanced. All these pinpoint to the use of a general model space (GMS) devoid of any special structure, which can readily be generated by selected configuration interaction (sCI).

J. Chem. Phys. 2004, 120, 5890–5902

(52) Dunning Jr, T. H. Gaussian basis sets for use in correlated molecular calculations. p. 2

Where the paper says it

Ideally, computed energies are independent of the choice of molecular orbitals. Since different orbital sets are related by unitary transformations (or orbital rotations), it is highly desirable for the method to be invariant under such transformations within each orbital class.

I The atoms boron through neon and hydrogen. J. Chem. Phys. 1989, 90, 1007–1023

Table 3 Test of generalized size-extensivity of PASPT2M/cc-pVDZ for 22 lowest singlet states of equilibrium water with model spaces M0 in Table 2 The energies for the ground and excited states are given in Eh and eV, respectively. ∆GSE are given in meV.

Figure 1

Figure 1 Illustration of closed, quasi-open and open operators for non-closed model space M0 M0 ⊕ RX = MX P = M0 Q = 1 − P = RX ⊕M⊥ X Closed operators always stay in M0, open operators always go from M0 to M⊥ X, whereas quasi-open operators can stay in M0 but can also go from M0 to RX.

Figure 2 Structures of (a) PASPT2M, (b) PASPT2X and (c) PASPT2H Hamiltonians. M0 ⊕ RX = MX Note that the block upper triangular structure (a) is typical for PASPT2M, although not universal.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improved Hamiltonian Construction for Strongly Correlated Systems The PASPT2H variant can be used to construct an effective Hamiltonian H eff[2] that is manifestly connected, which ensures the connectedness of the energies upon diagonalization by enforcing a condition on qRefs, thereby stabilizing calculations for states with major projections on M0.

  2. Near-Extensive Energy Predictions For target states supported by M0, PASPT2H provides near-extensive results, as confirmed by numerical evidence showing that the weak violation of size-extensivity has hardly numerical consequence on the target states, making it suitable for strongly correlated solid state calculations.

  3. Adaptive Treatment of Correlation Components For systems with strong static and dynamic correlations, the method guides future development by allowing an adaptive treatment of static and dynamic correlations, which is necessary because any structured reference space (e.g., complete active space (CAS) or generalized valence bond) [is] inherently inflexible and unbalanced.

  4. Hybrid Model Space Utilization For complex systems, the PASPT2H variant can be used as a hybrid of PASPT2M and PASPT2X by taking the RX space complementary to M0 within MX as a buffer, allowing for physically sound approximate treatment of states with major projections on RX.

  5. Machine Learning Model Training For State-Universal Hamiltonians The state-universal nature of PASPT2 makes it an ideal model for machine learning of dressed activespace Hamiltonians, enabling the training of ML models that capture the physics across different molecular states efficiently.

Abstract

The recently proposed partial-active-space (PAS) multi-state second-order perturbation theory (PASPT2) [Precis. Chem. 4, 997 (2026)] features connected amplitudes and a connected, closed intermediate Hamiltonian. Despite these hallmarks, PASPT2 (denoted as PASPT2H from now on) is not strictly size-extensive, as originally thought (and numerically confirmed), albeit strictly size-consistent. Nevertheless, PASPT2 is near-extensive for the states with major projections on the chosen PAS M 0. This becomes more transparent upon introducing PASPT2X, a strictly size-extensive variant. Compared with the intruder-prone PASPT2X, the intruder-free PASPT2H merely neglects the second-order corrections that are important only for those states with major projections on the orthogonal complement R X of M 0 within the closed space M X (= M 0 plus R X); however, such states are not supported by the chosen finite one-particle basis set. The weak violation of size-extensivity is therefore numerically insignificant for the target states supported by M 0, reinforcing the theoretical basis of PASPT2H.

Sources

Related papers