Open-shell frozen natural orbital approach for quantum eigensolvers

This paper introduces an open-shell frozen natural orbital approach based on ZAPT2 that significantly reduces the virtual orbital space for quantum eigensolvers while maintaining high accuracy, enabling efficient and precise simulation of singlet-triplet energy gaps in large open-shell systems like the Ir(ppy)$_3$ complex.

The Gist

The Big Picture: The "Too Many Ingredients" Problem

Imagine you are trying to bake the perfect cake (which represents a molecule) to understand how it tastes (its energy). To get the recipe right, you need to know exactly how every single ingredient interacts with every other ingredient.

In the world of atoms, these "ingredients" are electrons.

• The Problem: For complex molecules (like the ones used in OLED screens for your phone), there are hundreds of electrons. If you try to calculate how all of them interact at once, the math becomes so massive that even the world's fastest supercomputers (and future quantum computers) would crash. It's like trying to count every grain of sand on a beach to figure out the weight of the beach.
• The Current Solution: Scientists usually pick a "small group" of the most important electrons (the active space) and ignore the rest. They assume the ignored ones don't matter much.
• The Flaw: The current way of picking this group is like choosing ingredients based on their price tag (orbital energy). You pick the "cheapest" or "most expensive" ones. But in chemistry, the most important ingredients aren't always the ones with the highest price tags. Sometimes, a cheap ingredient is actually the secret sauce that makes the cake rise. If you pick the wrong group, your cake (the calculation) tastes terrible, or you have to use a massive group of ingredients just to get it right, which defeats the purpose of saving time.

The New Solution: The "Smart Chef" (ZAPT-FNO)

The authors of this paper have invented a new way to pick the ingredients. They call it ZAPT-FNO (Open-Shell Frozen Natural Orbital).

Instead of looking at the price tag, this method looks at how much the ingredient actually contributes to the flavor (correlation energy).

Here is how it works, step-by-step:

1. The "Frozen" Concept: Imagine you have a giant pantry with 1,000 spices. You only have room in your kitchen for 40 spices to cook with at once.

   • Old Way (CMO): You grab the 40 spices that are sitting on the top shelf (highest energy). You ignore the rest.
   • New Way (ZAPT-FNO): You taste-test the spices first. You realize that 90% of the spices on the top shelf are just "filler" (like salt or sugar that doesn't change the dish much). But there are 40 specific spices hidden in the back that are critical for the flavor. You grab those 40, and you "freeze" (ignore) the other 960.

2. The "Open-Shell" Twist: Most molecules are like a balanced team where electrons come in pairs. But some molecules (like the ones in phosphorescent lights) have "lonely" electrons that don't have a partner. These are called open-shell systems.

   • Previous methods for picking ingredients worked great for balanced teams but got confused and messy when lonely electrons were involved.
   • This new method uses a special mathematical tool called ZAPT2 (a type of perturbation theory) that is specifically designed to handle these "lonely" electrons without getting confused.

Why This Matters: The "Magic" Results

The authors tested this new "Smart Chef" method on three difficult scenarios:

1. Hydrogen Peroxide (H₂O₂):

   • The Test: They tried to calculate the energy difference between two states of the molecule.
   • The Result: The old method needed almost all the ingredients (orbitals) to get the answer right. The new method got the same accurate answer using only 20% of the ingredients. It's like getting a Michelin-star meal using a tiny spice rack instead of a warehouse.

2. Oxygen (O₂):

   • The Test: Oxygen is tricky because it has unpaired electrons.
   • The Result: The old method gave answers that jumped around wildly (erratic) as they added more ingredients. The new method gave a smooth, steady path to the correct answer. It showed that the old method was only getting the right answer by "luck" (canceling out errors), while the new method actually understood the chemistry.

3. The Big Boss: Ir(ppy)₃ (A Phosphorescent Complex):

   • The Test: This is a giant molecule with 260 electrons (61 atoms). It's used in high-end displays. To get an accurate result, you need a huge number of "ingredients" (a large basis set).
   • The Result: Using the old method, the calculation was impossible or wildly inaccurate. Using the new ZAPT-FNO method, they were able to use a massive, high-quality set of ingredients but only keep the 40 most important ones for the actual calculation.
   • The Outcome: The result was incredibly close to the real-world experimental value. They successfully simulated a massive, complex molecule that was previously too hard to handle accurately.

The Takeaway

Think of Quantum Eigensolvers (like the iQCC method used in the paper) as a very powerful but expensive car engine.

• The Problem: The engine is so powerful it can handle a Formula 1 car, but we are trying to drive a massive 18-wheeler truck (complex molecules). The truck is too heavy; the engine stalls.
• The Fix: The ZAPT-FNO method is like a smart suspension system. It doesn't make the truck lighter physically, but it tells the engine exactly which wheels need to bear the weight and which ones can be lifted off the ground.
• The Benefit: Suddenly, that same Formula 1 engine can drive the 18-wheeler truck efficiently.

In simple terms: This paper gives us a smarter way to filter out the "noise" in complex molecules. It allows scientists to use powerful quantum computers (or simulations of them) to study large, difficult materials—like the ones that make our screens glow—without needing a supercomputer the size of a city. It opens the door to designing better materials for energy, medicine, and electronics.

Technical Summary

1. Problem Statement

Accurate ab initio modeling of open-shell systems (e.g., transition metal complexes, phosphorescent emitters) requires large, extended atomic basis sets (including diffuse and polarization functions) to capture electronic structure and Rydberg states. However, these large basis sets generate a massive number of virtual molecular orbitals, making high-accuracy correlation calculations computationally prohibitive, especially for quantum eigensolvers like the Variational Quantum Eigensolver (VQE) or iterative Qubit Coupled Cluster (iQCC).

Existing methods to reduce the virtual space, such as Frozen Natural Orbitals (FNO), have been primarily limited to closed-shell systems using Restricted Hartree-Fock (RHF) references. Attempts to extend FNO to open-shell systems using Unrestricted Hartree-Fock (UHF) references (e.g., Equation-of-Motion FNO) result in inconsistent truncation of $\alpha$ and $\beta$ virtual spaces, leading to erratic energies and incompatibility with standard fermion-to-qubit mappings required for single-fermionic Hamiltonians.

2. Methodology

The authors propose the ZAPT-FNO (Z-averaged Perturbation Theory Frozen Natural Orbital) approach, designed specifically for open-shell systems compatible with quantum computing.

• Theoretical Foundation: The method utilizes Second-order Z-averaged Perturbation Theory (ZAPT2) as the reference for correlation. Unlike UHF-based approaches, ZAPT2 uses a Restricted Open-Shell Hartree-Fock (ROHF) reference. This ensures a single set of spatial orbitals, maintaining spin-restriction and compatibility with standard fermion-to-qubit transformations (e.g., Jordan-Wigner).
• Algorithm Workflow:
  1. Perform a self-consistent field (SCF) calculation using ROHF.
  2. Compute the virtual-virtual block of the ZAPT2 one-particle density matrix ($P^{(2)}$).
  3. Diagonalize this block to obtain Natural Orbitals (NOs) sorted by occupation number.
  4. Select a subset of NOs to form the active space (Freezing those with low occupation).
  5. Semicanonicalize the active virtual orbitals to preserve spin-restriction.
  6. Transform integrals and map the Hamiltonian to qubits for the eigensolver (iQCC).
• Energy Correction ($\Delta E_{FNO}$): To recover the correlation energy lost by freezing orbitals, the authors apply a correction term:
  $$\Delta E_{FNO} = E_{MP/ZAPT}^{MO} - E_{MP/ZAPT}^{NO}$$
  This difference between the full MO basis and the truncated NO basis correlation energy is added to the final eigensolver energy.

3. Key Contributions

• First Open-Shell FNO for Quantum: Introduces the first FNO scheme compatible with open-shell systems that uses a spin-restricted reference, making it directly applicable to quantum eigensolvers (iQCC, VQE) without the spin-channel inconsistencies of UHF-based methods.
• Systematic Convergence: Demonstrates that ZAPT-FNO provides systematic, monotonic convergence of correlation energies and excitation gaps (specifically Singlet-Triplet $T_1-S_0$ gaps) with respect to active space size, unlike Canonical Molecular Orbital (CMO) truncation which relies on orbital energy and often suffers from fortuitous error cancellation.
• Handling Diffuse Basis Sets: Shows that ZAPT-FNO effectively filters out diffuse Rydberg orbitals that contribute little to correlation energy but dominate the virtual space in augmented basis sets, allowing for the use of high-quality basis sets within manageable active space sizes.

4. Results and Benchmarks

The method was validated across four distinct systems using the iQCC method (and CASCI for comparison):

• Hydrogen Peroxide ($H_2O_2$):
  • In the high-spin triplet state, ZAPT-FNO recovered a significantly larger fraction of correlation energy with fewer active orbitals compared to CMO truncation.
  • The $T_1-S_0$ gap converged to within chemical accuracy (1 mEh) using only 20% of the virtual space, whereas CMO required nearly the full space to converge.
• Oxygen ($O_2$):
  • ZAPT-FNO showed smooth, systematic convergence of the $T_1-S_0$ gap as the Complete Active Space (CAS) size increased.
  • CMO selection showed erratic behavior; while it appeared accurate at specific CAS sizes due to error cancellation, it failed to converge systematically.
  • The corrected ZAPT-FNO total energies matched CCSD(T) reference values within chemical accuracy.
• Methylene ($CH_2$) Bond Dissociation:
  • Tested across weak to strong correlation regimes (stretching C-H bonds).
  • ZAPT-FNO with the aug-cc-pVTZ basis set achieved chemical accuracy (within 1 mEh of CASCI) across the entire dissociation path, including regions of strong static correlation where CMO failed (diverging by up to 30 mEh).
  • Limitation Identified: The $\Delta E_{FNO}$ correction becomes unreliable at extreme bond stretches ($>2r_e$) where ZAPT2 perturbation theory overestimates virtual orbital occupations.
• Iridium Complex ($Ir(ppy)_3$):
  • A challenging 260-electron system requiring large basis sets (LANL2TZ//6-31+G(d)).
  • ZAPT-FNO reduced the virtual space by 97% (from 601 to 40 active orbitals) while maintaining accuracy.
  • The computed $T_1-S_0$ gap was 2.412 eV (vs. experimental 2.525 eV), a threefold improvement in error over the CMO approach (2.897 eV).
  • Analysis revealed that CMOs retained diffuse Rydberg orbitals with low correlation contribution, while ZAPT-FNO prioritized localized orbitals with high natural occupation, crucial for accurate correlation modeling.

5. Significance

The ZAPT-FNO approach represents a significant advancement in resource-efficient quantum chemistry. It enables the simulation of large, open-shell materials (such as phosphorescent OLED emitters) with chemical accuracy using compact active spaces on current and near-term quantum hardware.

By decoupling the quality of the basis set from the size of the active space, this method allows researchers to:

1. Use extended basis sets (essential for diffuse states) without exploding the qubit count.
2. Avoid the "fortuitous error cancellation" pitfalls of energy-based orbital selection.
3. Bridge the gap between classical high-accuracy methods (like CCSD(T)) and quantum eigensolvers, providing a robust pathway for simulating complex excited states and transition metal complexes.

The work confirms that ZAPT-FNO is a versatile, generalizable strategy suitable for various quantum eigensolvers (iQCC, VQE) and is critical for scaling quantum simulations to industrially relevant molecular sizes.