PASPT2: a size-extensive and size-consistent partial-active-space multi-state multi-reference second-order perturbation theory for strongly correlated electrons

The paper introduces PASPT2, a novel partial-active-space multi-state multi-reference second-order perturbation theory derived from IN-GMS-SU-CCSD that achieves strict size-extensivity and size-consistency by eliminating disconnected terms through a specialized reference-specific zeroth-order Hamiltonian.

The Gist

The Big Picture: Solving the "Traffic Jam" of Electrons

Imagine a molecule as a busy city. The electrons are the cars, and the orbitals are the roads. In most simple molecules, the traffic flows smoothly; there is one main route that almost all the cars take. This is easy to predict.

However, in strongly correlated systems (like complex transition metal complexes), the traffic is a total nightmare. There are many roads that are equally good, and the cars are constantly switching lanes, merging, and splitting up in a massive, chaotic superposition. No single route dominates. Predicting the energy of this system is like trying to predict the exact flow of a massive traffic jam where every car's movement depends on every other car's movement.

For decades, chemists have struggled to calculate the energy of these "traffic jams" accurately without the math breaking down or becoming impossibly slow.

The Problem with Previous Methods

The paper introduces a new method called PASPT2. To understand why it's special, we have to look at the problems with the old ways of doing things:

1. The "Complete" Approach (CAS): Imagine trying to map every single possible traffic pattern in the city. This is accurate but computationally impossible for large cities because the number of patterns grows exponentially. It's like trying to count every grain of sand on a beach.
2. The "Partial" Approach (sCI): To save time, scientists started picking only the "most important" traffic patterns. This is faster, but it has a flaw: if you split the city into two separate, non-interacting towns, the math sometimes fails to add up correctly. It's like calculating the cost of two separate parties and getting a total that is higher or lower than the sum of the two because the math got confused.
3. The "Universal" Approach (IN-GMS-SU-CC): There was a sophisticated theory (IN-GMS-SU-CCSD) that tried to fix this, but the authors of this paper discovered a hidden flaw: it wasn't actually "size-extensive." In simple terms, this means that as the system gets bigger, the error doesn't stay constant; it grows, making the results unreliable for large molecules.

The Solution: PASPT2

The authors, Chunzhang Liu, Ning Zhang, and Wenjian Liu, have developed PASPT2. Here is how it works, using an analogy:

The "Partial Active Space" (PAS) Strategy
Instead of trying to map the whole city, PASPT2 focuses on a specific "downtown district" (the active space) where the traffic is most chaotic. It selects a smart subset of the most important traffic patterns (configurations) to focus on, rather than every single possibility.

The "Special Zero-Order Hamiltonian" (The Traffic Cop)
The core innovation is a new way of setting up the rules of the road (the math).

• The Old Way: The old math had "disconnected terms." Imagine a traffic report that said, "Car A is moving, and separately, Car B is moving," but the report didn't account for the fact that Car A's movement actually changes Car B's path. This led to errors that piled up as the city got bigger.
• The PASPT2 Way: The authors introduced a special "Traffic Cop" (a reference-specific zeroth-order Hamiltonian). This cop ensures that every calculation is "connected." It forces the math to acknowledge that every part of the system is linked. By doing this, they eliminated the "disconnected terms" that plagued the previous theories.

The Result: A Perfectly Balanced Scale
Because of this new "Traffic Cop," PASPT2 is Size-Extensive and Size-Consistent.

• Size-Extensive: If you double the size of the molecule, the energy calculation doubles perfectly. The error doesn't grow.
• Size-Consistent: If you have two separate molecules far apart, the total energy calculated for them together is exactly the sum of their energies calculated separately. The math doesn't get confused by the distance.

How They Tested It

The authors didn't just write the math; they tested it on real-world "traffic jams":

1. Helium Chains: They lined up Helium atoms like a row of houses. They showed that as they added more houses, the energy calculation grew in a perfectly straight line, proving the method is reliable for larger systems.
2. Water Molecule (H2O): They calculated the energy needed to jump electrons to higher energy levels (excitation). They compared their results to "near-exact" benchmarks. They found that while their method was very good, the accuracy depended heavily on how well they chose the initial "traffic patterns" (the model space). If they picked a better starting point, the results were nearly perfect.
3. Nitrogen Molecule (N2): This is a classic test for breaking chemical bonds. As the Nitrogen atoms move apart, the "traffic" gets very chaotic. PASPT2 successfully tracked the energy curve smoothly, matching the most accurate benchmarks available, even when the atoms were far apart.

The Bottom Line

The paper claims that PASPT2 is a breakthrough because it is the first method of its kind that is:

1. Based on a Partial Active Space (making it fast enough for complex molecules).
2. Multi-State (able to handle multiple energy levels at once).
3. Strictly Size-Extensive and Size-Consistent (mathematically reliable for large systems).
4. Intruder-Free (avoids the mathematical "glitches" that happen when energy levels get too close).

The authors conclude that while the method is currently a major step forward, the next challenge is to make it "spin-adapted" (handling the magnetic spin of electrons even more perfectly), which they plan to do in future work. For now, PASPT2 offers a robust, accurate, and scalable tool for understanding the most difficult electronic systems in chemistry.

Technical Summary

1. Problem Statement

Accurate quantum chemical descriptions of strongly correlated systems (e.g., polynuclear transition metal complexes) remain a significant challenge. These systems require handling a massive superposition of electron configurations where no single configuration dominates.

• Limitations of Existing Methods:
  • CASCI/CASSCF: While foundational, they scale combinatorially with the number of active orbitals, limiting them to $\sim$12 orbitals. They also treat static correlation exactly but dynamic correlation approximately, leading to unbalanced treatments and "intruder states" (singularities in perturbation theory) for high-lying states.
  • Selected CI + PT2 (sCIPT2): While effective, standard Epstein-Nesbet perturbation theory used in sCIPT2 is not size-extensive when applied to a selected space that is not the full Hilbert space.
  • IN-GMS-SU-CCSD: The parent theory (Intermediate Normalization-based General-Model-Space State-Universal Coupled Cluster with Singles and Doubles) proposed by Li and Paldus aims to be general but suffers from disconnected terms in its amplitude equations. This renders the method not size-extensive, contrary to previous claims.
• The Gap: There is a need for a Multistate (MS), Multireference (MR), Second-Order Perturbation Theory (PT2) method based on a Partial Active Space (PAS) that is strictly size-extensive and size-consistent, while avoiding intruder states.

2. Methodology: PASPT2

The authors formulate PASPT2 by linearizing the IN-GMS-SU-CCSD theory. The core innovation lies in the specific construction of the zeroth-order Hamiltonian and the handling of the model space.

A. Theoretical Framework

• Wave Operator: Based on the Jeziorski-Monkhorst (JM) Ansatz, mapping a Partial Active Space (PAS, $M_0$) to the exact manifold.
• Reference-Specific Partitioning: Unlike standard approaches using a single global Hamiltonian, PASPT2 employs a reference-specific zeroth-order Hamiltonian ($H_0^\alpha$) for each reference determinant $|\alpha\rangle$.
  • $H_0^\alpha = H_{0,I}^\alpha + H_{0,A}^\alpha$
  • Inactive Part ($H_{0,I}^\alpha$): A class-diagonal operator (Fock-like) acting on inactive orbitals.
  • Active Part ($H_{0,A}^\alpha$): A non-Hermitian operator acting on active orbitals. Crucially, it is constructed to exclude specific off-diagonal couplings that generate disconnected terms in the amplitude equations.
• Amplitude Equations: The first-order cluster amplitudes ($t$) are obtained by solving a linear system $At = V$. Due to the specific choice of $H_0^\alpha$, the right-hand side of these equations becomes termwise connected, eliminating the disconnected terms found in IN-GMS-SU-CCSD.

B. Handling Intruder States and Closure

• Intermediate Hamiltonian: The effective Hamiltonian ($\bar{H}_{eff}^{[2]}$) is constructed to be connected and closed.
• Buffer Space ($\bar{M}_0$): To ensure the Hamiltonian is closed (i.e., does not excite reference states to the external space $Q$), the model space is automatically extended from $M_0$ to an intermediate space $M_X = M_0 \cup \bar{M}_0$.
  • States in $\bar{M}_0$ act as "buffers." They are included to close the Hamiltonian but are not perturbed (their $T$-clusters are set to zero).
  • This mechanism effectively resolves the intruder state problem without ad-hoc level shifts.

C. Size-Extensivity and Size-Consistency

• Size-Extensivity: Proven by demonstrating that the $t$-amplitudes and the effective Hamiltonian are strictly connected. The specific form of $H_{0,A}^\alpha$ ensures that common-orbital terms cancel out correctly, a feature missing in the parent IN-GMS-SU-CCSD.
• Size-Consistency: Proven for a supermolecule $AB$ where the PAS is the direct product of the PAS of non-interacting fragments $A$ and $B$. The additive separability of $H_0^\alpha$ ensures the energy of $AB$ equals $E_A + E_B$.

3. Key Contributions

1. Correction of Parent Theory: The authors rigorously demonstrate that the previously proposed IN-GMS-SU-CCSD is not size-extensive due to disconnected terms, a flaw rectified in PASPT2.
2. Novel Zeroth-Order Hamiltonian: Introduction of a reference-specific, non-Hermitian active part ($H_{0,A}^\alpha$) that enforces termwise connectedness in the amplitude equations.
3. Automatic Intruder-State Resolution: Development of a closed intermediate Hamiltonian ($\bar{H}_{eff,X}^{[2]}$) via an automatically determined buffer space ($\bar{M}_0$), eliminating the need for manual level shifting.
4. First PAS-based MS-MRPT2: PASPT2 is presented as the first size-extensive, size-consistent, and intruder-free multistate MRPT2 method based on a partial active space under intermediate normalization.

4. Results and Validation

The method was implemented in the BDF program package and tested on three prototypical systems:

• Linear He Chains ($He_n$):
  • Test: Size-extensivity.
  • Result: The correlation energy scales linearly with the number of atoms, matching MP2 and near-exact iCIPT2 results.
• Water ($H_2O$) Vertical Excitations:
  • Test: Accuracy of excitation energies (VEE) against iCIPT2 benchmarks.
  • Findings:
    • Using ground-state HF orbitals yielded moderate errors.
    • Using semi-canonical SA-CASSCF orbitals improved low-lying states but worsened high-lying ones.
    • Optimal Strategy: Using a manually selected model space ($M_0$) extended to a closed space ($M_X$) with the $\bar{H}_{eff,X}^{[2]}$ variant provided the best balance.
    • The method successfully captured 43 excited states with Mean Absolute Errors (MAE) comparable to or better than IN-GMS-SU-CCSD and CASSCF-SDSPT2, provided the model space was of high quality.
• Nitrogen ($N_2$) Potential Energy Curves (PEC):
  • Test: Dissociation behavior and spectroscopic constants ($R_e, \omega_e, D_e$) for $^1\Sigma_g^+$ and $^3\Sigma_u^+$ states.
  • Challenge: $N_2$ dissociation requires a CAS(6,6), which is often too large for standard MRPT2 without selection.
  • Findings:
    • PASPT2 results agreed well with iCIPT2 benchmarks.
    • Refinement: In the bonding region, some uncoupled amplitudes were unphysically large (violating Brillouin conditions). Removing external excitations with amplitudes $|t| > 0.3$ significantly improved the PEC smoothness and non-parallelity errors (NPE).
    • Spectroscopic constants for both states were highly accurate, outperforming standard CASSCF and matching CASSCF-SDSPT2.

5. Significance

• Theoretical Rigor: PASPT2 resolves the long-standing issue of size-extensivity in general-model-space multireference perturbation theories.
• Practical Applicability: By utilizing a Partial Active Space (PAS) rather than a full Complete Active Space (CAS), the method can handle systems with larger active spaces than traditional CASSCF-based methods, provided a good initial selection of configurations is made.
• Robustness: The automatic construction of the buffer space ($\bar{M}_0$) makes the method robust against intruder states, a common failure point in MRPT2.
• Future Outlook: The authors note that while the current implementation uses spin-orbitals (class-diagonal inactive part), future work will focus on spin adaptation to further improve accuracy for open-shell systems.

In summary, PASPT2 represents a significant advancement in electronic structure theory, offering a rigorous, size-extensive, and computationally feasible approach for strongly correlated systems that balances the accuracy of multireference methods with the efficiency of perturbation theory.