🔬 materials science

Scaling active spaces in simulations of surface reactions through sample-based quantum diagonalization

This paper demonstrates the potential of sample-based quantum diagonalization (SQD) and its extended variant (Ext-SQD) to accurately model oxygen reduction reactions in lithium batteries by scaling active space simulations up to 32 orbitals on an IBM quantum processor, outperforming classical reference methods at larger scales.

Original authors: Marco Antonio Barroca, Tanvi Gujarati, Vidushi Sharma, Rodrigo Neumann Barros Ferreira, Young-Hye Na, Maxwell Giammona, Antonio Mezzacapo, Benjamin Wunsch, Mathias Steiner

Published 2026-07-14

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Marco Antonio Barroca, Tanvi Gujarati, Vidushi Sharma, Rodrigo Neumann Barros Ferreira, Young-Hye Na, Maxwell Giammona, Antonio Mezzacapo, Benjamin Wunsch, Mathias Steiner

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to solve a massive, 3D puzzle of how atoms dance together during a chemical reaction. The puzzle is so huge that even the world's most powerful supercomputers get dizzy trying to figure out every single piece at once. This is the problem scientists face when trying to predict the energy of reactions, like the one happening inside a lithium battery when oxygen meets lithium.

In this study, the researchers tried a new trick: they asked a quantum computer to help pick out the most important puzzle pieces, and then used a clever math shortcut to solve the rest.

The "Needle in a Haystack" Problem
Think of a chemical reaction like a crowded dance floor. There are thousands of electrons (the dancers) moving around. To predict exactly how the dance ends (the energy of the reaction), you theoretically need to track every single dancer. But that's impossible. So, scientists usually try to focus only on the "active" dancers—the ones actually changing partners.

The paper argues that old ways of picking these active dancers (using standard computer simulations) often miss the subtle moves or get stuck in a "barren plateau," a term for a situation where the computer gets lost and can't find the right answer no matter how hard it tries. The authors explicitly reject the idea that we can just throw more classical computing power at this specific type of problem and expect a perfect solution; the math gets too heavy too fast.

The Quantum "Spotlight"
Instead of trying to see the whole dance floor, the team used a quantum computer as a super-powered spotlight. They used a method called Sample-based Quantum Diagonalization (SQD).

Here's how it works with an analogy: Imagine you are trying to guess the winning lottery numbers. Instead of checking every single possible combination (which would take forever), you ask a quantum computer to quickly generate a handful of "likely" tickets based on the rules of the game. Then, you take those specific tickets and run them through a super-fast classical calculator to see which one wins.

In this study, the "tickets" are electronic configurations (ways the electrons can arrange themselves). The quantum computer, using a specific circuit design called Local Unitary Cluster Jastrow (LUCJ), sampled these configurations. The researchers found that by using a "truncated" version of this circuit (cutting out the very last steps to save time and reduce errors), they could get a good mix of tickets without the quantum computer getting too noisy.

The "Super-Charged" Upgrade
The team didn't stop there. They introduced a new, upgraded version called Ext-SQD. Think of this as taking the lottery tickets the quantum computer gave you and asking, "What if we tweaked these just a little bit?" They applied "excitation operators," which are like small, calculated nudges to the electron arrangements. This allowed them to refine the list of tickets even further, making the final answer much sharper.

What They Found (and What They Didn't)
The researchers tested this on a lithium-oxygen reaction, which is crucial for making better batteries. They started with small groups of orbitals (the "dance moves") and slowly added more, up to 32 orbitals.

The Good News: When they used the upgraded Ext-SQD method, the results were incredibly accurate. For active spaces up to 12 orbitals, their quantum-assisted results matched perfectly with the best-known classical methods. Even more impressively, when they scaled up to 27 orbitals, the Ext-SQD method actually predicted the reaction energy better than the standard classical methods (like CCSD) that are usually considered the gold standard.
The Reality Check: The paper is very careful not to call this a "solved" problem for all chemistry. They explicitly state that as they tried to go beyond 32 orbitals, the quantum computer's noise (errors) started to mess up the results. At 34 orbitals, they couldn't even get the quantum computer to produce a single valid result in their massive sample of 6 × 10⁶ tries. The "noise" drowned out the signal.

The "Sweet Spot"
The authors suggest that there is a "sweet spot" right now. If you have a problem that needs about 20 to 32 orbitals to solve, this hybrid method (quantum sampling + classical math) is currently the best tool in the shed. It beats the old-school supercomputers in accuracy for this specific size range.

However, they are honest about the limits. The paper notes that while the quantum part is doing the heavy lifting of sampling, the "post-processing" (the classical math part) is getting huge. As they add more orbitals, the classical computer has to diagonalize a subspace that grows incredibly fast, becoming a bottleneck itself.

The Bottom Line
This study doesn't claim to have fixed all of chemistry. It explicitly rules out the idea that we can just run these massive simulations on current quantum hardware without help; the noise is still too high for very large systems. Instead, it suggests that by carefully selecting which parts of the problem to send to the quantum computer and using the Ext-SQD method to refine the answers, we can get highly accurate predictions for reactions that were previously too difficult to model.

The researchers found that for the lithium-oxygen reaction, their method predicted a reaction energy of about -4.3 eV (electron volts) for the larger active spaces, which is much closer to what we expect from real-world experiments than some older methods. They conclude that while we aren't there yet for every chemical reaction, this hybrid approach is a promising path forward, provided we can keep the quantum computers from getting too noisy and the classical computers from getting too overwhelmed.

Technical Summary: Scaling Active Spaces in Surface Reaction Simulations via Sample-Based Quantum Diagonalization

Problem Statement
Accurate prediction of chemical reaction energies requires solving the many-body Schrödinger equation for correlated electronic systems. Classical methods face significant limitations: Density Functional Theory (DFT) often fails to capture strong electron-electron correlations, while high-accuracy wavefunction methods like Complete Active Space Configuration Interaction (CASCI) become computationally unfeasible as the active space size increases. While quantum computing offers a pathway to simulate these systems efficiently, current hardware limitations (noise, qubit count) restrict the size of chemical systems that can be directly simulated. Specifically, there is a need for hybrid quantum-classical workflows that can scale active spaces beyond the limits of classical diagonalization while maintaining accuracy on near-term quantum devices.

Methodology
The authors propose a hybrid quantum-classical framework applied to the surface reaction of lithium and oxygen in lithium-metal batteries ( $4 \text{Li} + \text{O}_2 \rightarrow 2 \text{Li}_2\text{O}$ ). The workflow consists of four main stages:

Geometry Optimization and DFT: Initial structures are optimized using DFT (VASP) and PySCF with the GTH-DZV basis set and PBE functional. Reaction energies are defined as the difference between product and reactant states ( $\Delta E = E_{\text{prod}} - E_{\text{reac}}$ ).
Active Space Selection: A systematic selection of relevant orbitals is performed using:
- Density Difference Analysis (DDA): Computes the electronic density difference between reactant and product states to identify orbitals involved in charge redistribution.
- Natural Orbitals: Refined using Coupled Cluster Singles and Doubles (CCSD) calculations to rank orbitals by occupation numbers. The active space is constructed around the Highest Occupied Natural Orbital (HONO) and Lowest Unoccupied Natural Orbital (LUNO).
Quantum Sampling (SQD): The electronic Hamiltonian is mapped to qubits using Jordan-Wigner encoding. A Local Unitary Cluster Jastrow (LUCJ) ansatz is used to prepare quantum states. The circuit is initialized with parameters from classical CCSD simulations.
- State Preparation: A truncated, two-layer LUCJ ansatz is employed to balance gate depth and accuracy.
- Sampling: Configurations are sampled on the IBM Heron R2 processor (up to 80 qubits). The sampled configurations are post-processed to ensure correct electron counts and spin sectors.
Diagonalization and Extension (Ext-SQD):
- SQD: Projects the Hamiltonian onto the subspace of sampled configurations and diagonalizes it classically.
- Ext-SQD: An extension that applies excitation operators (singles and doubles) to the sampled configurations to refine the configuration space and improve accuracy without increasing quantum circuit depth. This mimics the "Extended Heat-bath Configuration Interaction" (Ext-HCI) approach.

Key Contributions

First Application to Reaction Prediction: This work represents the first application of Sample-based Quantum Diagonalization (SQD) and its extended version (Ext-SQD) to predict chemical reaction energies in surface reactions.
Scalability Demonstration: The authors successfully benchmarked quantum-computed reaction energies for active spaces up to 12 orbitals against CASCI. For larger active spaces (up to 32 orbitals), they benchmarked against Heat-Bath Configuration Interaction (HCI) and CCSD.
Hybrid Workflow Integration: The study integrates DDA for orbital selection, LUCJ for efficient state preparation, and Ext-SQD for accuracy enhancement, demonstrating a viable path for scaling quantum simulations on current hardware.
Performance "Sweet Spot": The study identifies a specific regime (around 20–32 orbitals) where the quantum-assisted Ext-SQD method outperforms standard classical reference methods (CCSD and HCI) in terms of accuracy, despite the noise limitations of current quantum processors.

Results

Accuracy vs. Classical Methods: For active spaces up to 12 orbitals, Ext-SQD results match CASCI benchmarks. For larger spaces (up to 32 orbitals), Ext-SQD results converge faster than CCSD and HCI.
Chemical Accuracy: At 27 orbitals, Ext-SQD results show prediction accuracy improvements over standard quantum-chemical reference methods that remain computationally feasible at that scale. The method achieves results within the 1.6 mHa chemical accuracy bound compared to Ext-HCI.
Hardware Limitations: The study highlights that as the number of orbitals increases (beyond 32), the quality of quantum-sampled configurations deteriorates due to noise (high rates of forbidden configurations/incorrect Hamming weights). At 34 orbitals, the team could not obtain configurations with correct Hamming weights within $6 \times 10^6$ samples.
Classical Bottleneck: While quantum sampling is feasible up to ~32 orbitals, the classical post-processing (diagonalization of the extended subspace) becomes the computational bottleneck for Ext-SQD as the number of orbitals increases, due to the quadratic and quartic growth of single and double excitations, respectively.

Significance and Claims
The paper claims that sample-based quantum diagonalization, particularly when enhanced with excitation operators (Ext-SQD), offers a promising path for high-accuracy reaction modeling in chemistry and materials science. The results suggest that hybrid quantum-classical methods can complement and extend classical approaches, potentially delivering improvements in both computational efficiency and accuracy for complex electronic structure problems.

The authors modestly note that while their approach is feasible on current devices, noise resilience and circuit depth limitations remain key obstacles. They posit that the method's "sweet spot" (20–32 orbitals) demonstrates the utility of quantum-assisted simulations before quantum hardware reaches fault tolerance. Future scalability to 40+ orbitals is contingent on lower-error quantum processors and optimized classical post-processing techniques (e.g., parallelization). The study does not claim to have solved the full problem of dynamic correlation outside the active space but suggests that Ext-SQD wavefunctions could serve as multi-reference states for subsequent perturbation theory (PT2) corrections, similar to CASPT2 or NEVPT2 strategies.

Technical Summary: Scaling Active Spaces in Surface Reaction Simulations via Sample-Based Quantum Diagonalization

More like this