Quantum chemistry with provable convergence via randomized sample-based Krylov quantum diagonalization

arXiv:2508.02578 · quant-ph, physics.chem-ph · Submitted 2025-08-04 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: Today's paper: "Quantum chemistry with provable convergence via randomized sample-based Krylov quantum diagonalization".

Mira: Quantum algorithms based on classical processing of individual samples have recently emerged as the most effective and robust methods to approximate ground-state wave functions of manybody quantum systems on pre-fault-tolerant…

Kai: First, who's behind it and why it matters.

Title and authors: Kai: So we're discussing the paper now, "Quantum chemistry with provable convergence via randomized sample-based Krylov quantum diagonalization," and looking at the authors, Samuele Piccinelli and his collaborators. Mira The title itself tells us a lot; it’s not just about doing quantum chemistry, but specifically about achieving provable convergence through a randomized sample-based Krylov quantum diagonalization method.

Lev: I'm thinking about who these authors are and what their backgrounds might bring to this work, especially since we're talking about methods that need rigorous mathematical backing for hardware execution.

Kai: Well, the authors come from institutions like IBM Quantum and EPFL, which suggests a strong foundation in both the theoretical quantum algorithm side and actual experimental implementation. Mira I see that their background is multidisciplinary; they've got people from chemistry departments alongside quantum physics and error correction research groups.

Lev: That mix is exactly what this paper seems to require; you need someone who understands the deep theory of many-body systems, someone who can build the actual quantum circuits, and someone who knows how to handle the noise inherent in physical hardware.

Kai: It’s interesting how they've managed to bridge that gap by focusing on a method that is inherently designed to be compatible with NISQ devices, which is where this research sits. Mira The implication here is that they are pushing the boundary of what's feasible for current quantum platforms by making sure the algorithm scales and converges predictably even when running on hardware with limitations.

Lev: If we look at the context of other work, like those papers on universal recovery in approximate quantum error correction, this suggests that their convergence proofs might be robust enough to withstand some level of noise, which is a big concern for anyone trying to run these algorithms reliably.

Kai: I agree; they are focusing on creating methods where the sampling process itself provides a mechanism for robustness, rather than relying solely on perfect hardware fidelity. Mira The main point of the title is that this isn't just another heuristic approach; it’s a method with established theoretical convergence properties for ground-state energy calculations.

Lev: That level of theoretical underpinning makes it much more attractive for researchers who need to move from simulation to something that can actually be verified experimentally.

The paper's summary: Kai: Now, let's look at the actual summary of this paper, "Quantum chemistry with provable convergence via randomized sample-based Krylov quantum diagonalization." Essentially, they are describing how SqDRIFT works step-by-step. Mira They explain that the core idea is to use samples from a quantum circuit to identify bitstrings that contribute most significantly to the ground state wave function, and then diagonalize the Hamiltonian classically in that subspace.

Lev: So they are essentially using sampling as a way to intelligently prune the Hilbert space, which is much better than brute force diagonalization, but I need to know how they ensure that this pruning actually leads us to the right answer reliably.

Kai: They rely on a concentration hypothesis here, which guarantees that if the ground state wave function is well-approximated by a concentrated wavefunction, then diagonalizing in the subspace generated by samples from a quantum circuit will give an accurate approximation without needing an exponentially large subspace. Mira That concentration hypothesis is the theoretical backbone of their method, suggesting that this approach can work for complex molecular systems.

Lev: But I recall something earlier about the paper mentioning that a given bitstring b i might be sampled with probability phi 0b i squared, which can be far from uniform over the important bitstrings, which is a key detail for understanding the sampling overhead <ref:2508.02578#pg2,2$, which can be far from uniform over the>.

Kai: That non-uniform sampling is something they address by aiming for a quantum circuit that generates a constant probability distribution over those important bitstrings, which minimizes the amount of sampling overhead needed to extract all of them. Mira So they are trying to design the quantum part to be as efficient as possible during this sampling phase.

Lev: That makes sense; if you can control the distribution of samples, you control how much time and resources you need for that classical diagonalization step later on.

Kai: And they also mentioned that this works successfully with implementations utilizing up to seventy-seven qubits in their tests, which shows it’s already being tested on systems bigger than what's classically feasible for exact methods <ref:2508.02578#pg1,utilizing up to 77 qubits>. Mira So the summary paints a picture of a method that tackles the scaling problem head-on by focusing on finding concentrated regions of the Hilbert space.

Lev: It seems like they are proving that even with limited resources, if you use this specific sampling strategy, you can get results comparable to much larger simulations.

The paper's improvements: Kai: Shifting our focus now to the improvements suggested by the paper concerning "Quantum chemistry with provable convergence via randomized sample-based Krylov quantum diagonalization." Mira The authors suggest a few ways we can enhance this method, and they are focused on making it more versatile.

Lev: I'm curious about the practical enhancements; what do they actually propose that moves this from a theoretical concept to something useful for actual computation?

Kai: They suggest improving the circuit choice by using time-evolution circuits instead of variational ansatzes, which eliminates the heuristic component because time-evolution circuits are guaranteed to sample configurations on which the ground state has large support. Mira That shifts the burden away from finding a good initial guess and onto designing a quantum circuit that samples well.

Lev: That’s a strong methodological improvement because it removes one of the major unknowns in many variational approaches, which is how well those ansätze actually sample the true ground state wave function.

Kai: Another key suggestion is optimizing the fermion-to-qubit mapping by minimizing the distance between qubit indices p q and q q to maximize locality, which reduces the Pauli weight significantly. Mira That directly tackles circuit depth on NISQ hardware by making the circuits as shallow as possible.

Lev: Reducing that Pauli weight is vital; every extra Pauli term adds complexity and noise, so if you can drop it down, you’re working with a much more stable result on real hardware.

Kai: And finally, they introduce Extended-SQD or Ext-SqD by applying low-order electronic excitation operators E to the sampled configurations to enlarge the subspace through i = Eb i to improve accuracy for excited states. Mira That gives us a pathway to not just find the ground state, but also get better estimates for higher energy states too.

Lev: If we can explore those excited states accurately, then the utility of this method expands beyond just finding the lowest energy molecule; it becomes a more comprehensive tool for characterizing chemical dynamics.

Conclusion: Kai: So, to wrap up on this paper, we've seen that SqDRIFT is a method built on combining SKQD and qDRIFT compilation that offers provable convergence guarantees. Mira The main implication is that it shows we can achieve high-fidelity ground-state energy calculations for complex molecular systems using current quantum hardware.

Lev: From an error correction perspective, the convergence bounds they provide are what really matter because they give us a mathematical assurance that the noise won't just blow up unpredictably as we scale up the algorithm.

Kai: Exactly; it provides a framework for running these kinds of experiments on near-term devices with much higher confidence than we have before. Mira It suggests that by intelligently managing the sampling and compilation, we can get reliable chemical accuracy even when constrained by NISQ constraints.

Lev: I just think the practical path forward involves figuring out how to map those theoretical convergence guarantees directly onto the actual hardware noise profiles they are seeing in their experiments.

Kai: That’s a fair point; the next step is moving from this successful algorithm description to demonstrating its real-world performance on physical quantum chips. Mira Ultimately, the work on "Quantum chemistry with provable convergence via randomized sample-based Krylov quantum diagonalization" provides a very concrete blueprint for how we can tackle the challenge of simulating complex molecules using near-term quantum computers today.

Samuele Piccinelli, *Alberto Baiardi, Stefano Barison, *Max Rossmannek, Almudena Carrera Vazquez, Francesco Tacchino, Stefano Mensa

IBM Quantum Research Europe - Zurich, Switzerland · Institute of Physics, Ecole Polytechnique Fédérale de Lausanne (EPFL), Switzerland · Institute for Theoretical Physics, ETH Zürich, Switzerland · The Hartree Centre, STFC

quant-ph, physics.chem-ph

Submitted: 2025-08-04

Updated: 2026-10-05

Comments: Added new quantum-hardware experiments; Clarified the scope and novelty of our convergence guarantee; Strengthened the classical benchmarks; Added a direct comparison against sampling from the exact ground state; Expanded the discussion of the concentration hypothesis

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 83/100

The gist: Quantum algorithms based on classical processing of individual samples have recently emerged as the most effective and robust methods to approximate ground-state wave functions of manybody quantum

Key concepts

SqDRIFT
This is the proposed algorithm that combines Sample-based Krylov Quantum Diagonalization (SKQD) with a qDRIFT randomization technique for compiling the Hamiltonian. It allows researchers to perform quantum chemistry experiments, like calculating ground state energies, on noisy quantum processors by using classical sampling methods.
Krylov Subspace
This refers to a set of states generated by repeatedly applying the time-evolution circuit to an initial state. The method uses these subspaces because they are guaranteed to sample configurations where the true ground state has significant support, making them efficient for approximating the full system's behavior.
qDRIFT Compilation Protocol
Instead of standard Trotter formulas, qDRIFT randomly samples terms in the Hamiltonian evolution. This randomization creates an approximation of the unitary time-evolution operator. This approach improves accuracy for sparse Hamiltonians because its error scales better than conventional methods.

Terminology

Summary

Quantum algorithms based on classical processing of individual samples have recently emerged as the most effective and robust methods to approximate ground-state wave functions of manybody quantum systems on pre-fault-tolerant and early-fault-tolerant quantum devices. The resulting algorithm, termed SqDRIFT, enables quantum chemistry experiments on quantum processors while preserving the convergence guarantees similar to phase estimation algorithms.

The gist

SqDRIFT is a practical, quantum-chemistry friendly variant of Sample-based Krylov Quantum Diagonalization (SKQD) that combines SKQD with a qDRIFT randomized compilation strategy of the Hamiltonian propagator to enable accurate ground-state energy calculations on quantum processors.

Theoretical Foundation and Convergence Guarantees

The method builds upon the concept that if a target Hamiltonian's ground state is well-approximated by a concentrated wave function, then diagonalizing it in a subspace generated by samples from a quantum circuit can yield an accurate approximation without requiring an exponentially large subspace. The core of the approach involves constructing Krylov subspaces using time-evolution circuits, which are guaranteed to sample configurations on which the ground state has large support.

The theoretical framework establishes rigorous convergence bounds for SqDRIFT. Lemma A.1 proves that if all states in a Krylov subspace are prepared using Nr qDRIFT randomizations of length N, the resulting approximate ground state energy estimate satisfies an error bound:

**)&E˜ − E0 ≤ ξ (A2) with probability at least 1 − δ, where ξ is given by Eq. (A3). **

This convergence relies on bounding the difference between the actual Hamiltonian and the approximated one, specifically showing that with probability at least 1 − δ, H − H′′ ≤ ϵQH (A19), where εQ is defined by Eq. (A7). This demonstrates that the quality of solutions can be systematically and efficiently improved by increasing the Krylov dimension d and the length of qDRIFT sequences N.

The SqDRIFT Algorithm Workflow

The SqDRIFT protocol integrates SKQD and qDRIFT into a hardware-friendly pipeline. The workflow involves several key steps:

  1. Compile the time-evolution circuit associated with propagation time tk by sampling Nr qDRIFT randomized sequences, collecting S bitstrings from each realization.

  2. The union of all collected samples defines the set of samples associated with the target time-step tk.

  3. The target Hamiltonian is then diagonalized classically in the subspace defined by this union of samples collected for all time-steps to obtain an approximation of the ground state energy and wavefunction supported in that subspace.

qDRIFT Compilation Protocol

Instead of conventional Trotter formulas, SqDRIFT leverages qDRIFT, a randomized time-evolution approach. The protocol aims to compile the unitary operator e−iHt for a Hamiltonian H = XΣcihi (6) by constructing Vk = YΣj=1 e−ihkj tλ/N (7), where the sequence of indices k is obtained by randomly sampling terms hi from a distribution defined by ci/λ. The average of this channel, EqDRIFT[ρ] = Xk pkVkρV †k (8), yields an approximation to the exact time-evolution channel Ut[·] = eiHt(·)e−iHt. A key advantage is that the approximation error ε grows as O(λ2t2/N), and this error does not depend on the number of terms N in Eq. (6) but rather on the L1-norm of H, yielding an improvement over conventional Trotter formulas for sparse Hamiltonians.

Computational Pipeline and Optimization

The computational pipeline involves several optimizations to make circuits amenable to quantum hardware execution:

**- Particle-number conservation is enforced by sampling Hamiltonian terms in the fermionic basis rather than the qubit basis, which preserves particle-number conservation crucial for SQD success. This allows for optimization of the fermion-to-qubit (F2Q) mapping to minimize circuit depth. **

**- The F2Q mapping is optimized by minimizing the distance between qubit indices pq and qq to maximize locality, which minimizes Pauli weight. For 1-body excitations, this leads to Pauli terms of weight 2 in the optimal case. **

**- Circuit synthesis utilizes Qiskit v2.1.0 with its default transpiler pipeline set to the highest level of optimization for targeted backend connectivity, enhanced by a high-level synthesis plugin leveraging rustiq v0.0.8 to preserve the order of Pauli operators and avoid biasing the qDRIFT protocol. **

**- An optional step, Extended-SQD (Ext-SqD), can be performed by applying low-order electronic excitation operators E ∈ O to sampled configurations bi, enlarging the subspace through ˜bi⟩ = E bi⟩ to improve accuracy for excited states.

Improvements for AI systems

Based on the provided scientific paper, here are specific improvements to AI systems that can be derived from the SqDRIFT algorithm and its theoretical framework:

The core improvement is in enabling high-fidelity, quantum chemistry simulations for complex many-body systems (like molecular electronic structures) on near-term quantum hardware by mitigating the circuit depth bottleneck.

Here are specific improvements and capabilities:

  1. The AI system can perform ground-state energy calculations for polycyclic aromatic hydrocarbons (PAHs) and other molecules beyond the reach of exact diagonalization, specifically targeting systems with up to 48 qubits (e.g., coronene).

  2. The system can achieve chemical accuracy in energy estimates for molecular systems by intelligently managing the trade-off between circuit complexity and sampling overhead through the SqDRIFT protocol.

Specific mechanisms driving these improvements:

  1. The AI system utilizes a combination of Sample-based Krylov Quantum Diagonalization (SKQD) and the qDRIFT randomized compilation strategy to generate quantum circuits for time evolution. This allows it to sample configurations that are highly likely to support the true ground-state wave function, even when using shallower circuits than standard Trotter methods.

  2. The system employs a hardware-aware fermion-to-qubit mapping optimization (F2Q optimization). This reduces the weight of Pauli terms in the generated quantum circuits, leading to significantly shallower circuits and lower gate counts on current quantum processors (as demonstrated by Fig. 3), thereby minimizing noise accumulation during execution.

  3. The system incorporates a configuration recovery method (CR) within the Extended-SQD (Ext-SQD) framework. This allows the AI to refine initial samples by applying low-order electronic excitation operators to previously sampled configurations, effectively exploring a richer, more accurate subspace and improving energy estimates beyond what is achievable with raw sampling.

  4. The system can dynamically optimize its computational resources based on convergence analysis:

(a) It can increase the Krylov dimension (number of excitations sampled from the Hamiltonian) to improve accuracy when using localized orbitals (LO), while being aware that this increases circuit depth for canonical Hartree-Fock (HF) orbitals.

(b) It can increase the number of qDRIFT randomizations and samples per circuit to improve success probability, allowing it to trade circuit depth for sampling overhead when targeting deeper circuits.

Specific capabilities derived from these improvements:

  1. The AI system can perform scalable quantum chemistry experiments on quantum processors, enabling the study of complex molecular systems whose exact solution is classically intractable (e.g., large PAHs).

  2. It can provide reliable energy estimates for these systems, achieving accuracy comparable to or better than high-level classical methods like Coupled Cluster (CCSD) and Configuration Interaction (CI) methods, even when limited by the constraints of NISQ devices and hardware noise.

  3. The system can identify predominant configurations iteratively by using a first SqDRIFT run (e.g., starting with an HF determinant) to guide subsequent iterations, mimicking classical selected CI methods for more efficient subspace exploration in strongly correlated systems.

Sources

Related papers