Reducing quantum measurements in qubit-based overlapping grouping methods for quantum energy estimation through better initializations

arXiv:2607.02794 · quant-ph · Submitted 2026-07-02 · 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: We're looking at the paper: "Reducing quantum measurements in qubit-based overlapping grouping methods for quantum energy estimation through better initializations".

Mira: Here's the gist of the paper. Reducing quantum measurements in qubit-based overlapping grouping methods for quantum energy estimation through better initializations is addressed by introducing variance-aware SI (VarSI), a family of covariance-informed non-overlapping Pauli grouping heuristics to reduce measurement counts The proposed algorithms leverage covariance dictionaries, already required by overlapping methods, to construct better non-overlapping groups.

Kai: Molecular benchmarks with 130 Hamiltonians demonstrate consistent, non-overlapping measurement improvements over SI of 38% and enhanced downstream ICS results when initialized from VarSI groups.

Mira: We find that the initializations considered here achieve up to 70% measurement reductions for ICS, compared to the standard SI initialization with mean reductions of 9–15.3% depending on qubit mappings and covariance dictionaries used.

Kai: These results show that non-overlapping grouping remains a consequential design step even when the final estimator uses overlapping fragments.

Title and authors: Kai: So we're looking at this paper now called "Reducing quantum measurements in qubit-based overlapping grouping methods for quantum energy estimation through better initializations". It sounds like they're trying to cut down on how many times we have to measure our qubits when doing these kinds of calculations.

Mira: Exactly. The authors are focusing on how we group the Pauli words before we even start measuring, which is a big deal because measurement cost is a real bottleneck in variational quantum algorithms. They mention that overlapping grouping methods already need some non-overlapping grouping from the start, usually done by an algorithm called Sorted Insertion or SI, and then they use covariance dictionaries to help guide the next steps.

Lev: And what interests me about this is that it’s not just tweaking the standard SI approach; they introduce a whole family of heuristics called variance-aware SI, or VarSI, which uses those covariance dictionaries to build better non-overlapping groups.

Kai: So VarSI is this new set of rules for grouping Pauli words that takes the covariance information into account to try and make the groups more efficient in terms of measurement counts. What's the main idea behind these new rules?

Mira: The core idea is that you use those covariance dictionaries, which are already necessary for overlapping methods, to construct better non-overlapping groups than what a standard SI algorithm gives you. They propose three variants: a global greedy grouping insertion rule, a variance-informed SI analog, and then a local refinement step starting from either the standard SI or their new VarSI method.

Lev: That sounds like they’re creating some kind of iterative improvement process rather than just one fixed way to group everything upfront. It suggests that you can start with a baseline grouping, maybe the standard one, and then use these other methods to refine it for better measurement reduction.

Kai: And the results they show are pretty substantial; for molecular benchmarks with one hundred thirty Hamiltonians, they see consistent, non-overlapping measurement improvements over the standard SI approach of about thirty-eight percent.

Mira: That's significant because it shows that this isn't just a theoretical exercise; it works on real molecular systems. They also show that when you initialize your overlapping measurement strategy using these VarSI groups, the downstream results for calculating expectation values improve too.

Lev: And the numbers they put forward are interesting—they found that the initializations they considered can achieve up to seventy percent measurement reductions for calculating those expectation values, compared to just using the standard SI initialization which gave mean reductions of about nine to fifteen point three percent depending on how you mapped your qubits and what covariance dictionaries you used.

Title and authors: Kai: That jump from around fifteen percent down to seventy percent is a huge difference in terms of hardware time saved. It really shows that this non-overlapping grouping step is still a consequential design choice even when the final estimator uses overlapping fragments.

Mira: That’s the key point there—it proves that optimizing those initial groupings isn't just some academic exercise; it directly impacts how much QPU time you actually save, and it suggests these groupings can serve as effective initializations for overlapping measurement strategies, reducing measurement costs by an average of nine to fifteen point three percent relative to SI-ICS.

Lev: And the paper notes that this improvement happens across different types of covariances used for scoring and fermion-to-qubit mappings, which adds a layer of robustness to the finding.

Kai: It sounds like they’ve shown that these VarSI methods aren't just better than standard SI in a vacuum; they hold up regardless of some of those other settings. What does this mean for someone who is actually trying to run an experiment on real hardware?

Mira: Well, it means if you are setting up your overlapping measurement initialization for something like iterative coefficient splitting or ICS, adopting these VarSI methods could give you a substantial reduction in the number of measurements you need to take. The authors suggest that the refinement variants, specifically VarSI-R and VarSI-OR, emerge as the most promising options for achieving those significant reductions.

Lev: From my side building for real hardware, I have to think about how this translates practically. They found that these refinement variants yield reductions in measurement requirements by around thirty-five to thirty-nine percent across the dataset regardless of whether you use different covariance types for scoring or different fermion-to-qubit mappings.

Kai: That means even if your specific hardware setup doesn't perfectly align with their ideal scenario, you should still be looking at using these refinement methods. So, what is the real impact on a quantum chemist who is running an energy estimation task?

Mira: The implication is that by using these initial groupings, you can make the ICS method much more competitive against more expensive frameworks like SPP. And they estimate that with VarSI-OR-ICS initialization, you can see reductions with respect to SI-ICS of negative one point eight percent or nine point two percent, with maximum reductions reaching thirty seven point five percent for Hamiltonians mapped using JW methods.

Lev: It’s worth remembering that even the refinement step from the standard SI groupings in VarSI-R still yields appreciable reductions in measurement costs, which is a strong signal that this approach is viable for real-world applications on noisy hardware.

Title and authors: Kai: So, to wrap up this paper on "Reducing quantum measurements in qubit-based overlapping grouping methods for quantum energy estimation through better initializations," the main thing we’re hearing is that optimizing the non-overlapping grouping step using variance awareness can lead to massive measurement savings, potentially saving up to one hour of QPU time in common chemistry applications when using the VarSI-OR-ICS initialization.

Mira: Exactly. They show that this initial grouping strategy is a consequential design step in the measurement reduction pipeline, and for anyone looking at improving how we initialize overlapping measurements, this paper suggests that VarSI-R and VarSI-OR are the most promising variants to try first.

Lev: I just want to add that while they show these gains across various systems with one hundred thirty Hamiltonians, the practical overhead of building those covariance dictionaries for the scoring rule is something you have to consider when running this on actual hardware, which is where my experience comes in.

Kai: That’s a fair point about the overhead, Lev. But honestly, looking at these numbers—like that up to seventy percent reduction—it makes me think that we need to prioritize building those better initializations because the payoff in terms of measurement count is so much better than the setup cost for the dictionaries.

Mira: It does sound promising, but we still have to remember what they say about their limitations. They note that they are focusing on qubit-based overlapping grouping methods, so this specific technique might not directly apply to every single kind of quantum estimation problem out there without some modification.

Lev: Right, it’s specific to the qubit architecture and the way the Hamiltonian is structured for these methods. But even with those limitations, if you are working on a problem where you already have that structure, they suggest VarSI-R and VarSI-OR are definitely worth testing out for better measurement scaling.

Kai: So we’ve covered how this paper on "Reducing quantum measurements in qubit-based overlapping grouping methods for quantum energy estimation through better initializations" shows that using variance awareness to guide the non-overlapping grouping step can lead to very large reductions in measurement counts, up to seventy percent in some cases, which is a big win for variational algorithms.

Mira: And it confirms that these groupings are a crucial part of the pipeline for overlapping techniques, suggesting that if you want to compete with more expensive methods like SPP, improving this initialization is definitely something worth investigating further.

Lev: I think the biggest thing here is that VarSI-R and VarSI-OR seem like the most viable options right now because they show consistent reductions across different system types, which gives us a good starting point for what to try next on real hardware.

The paper's summary: Kai: So, to recap, this paper is about taking quantum energy estimation and making it faster by fixing how we group our Pauli operators at the beginning of the process.

Mira: Right, they look at this whole pipeline for variational algorithms—specifically overlapping methods—and they found that a better starting point makes all the difference in how much measurement data you need.

Lev: They’re talking about these non-overlapping groups, which are usually done by something called Sorted Insertion, and then they use covariance dictionaries to make those groups smarter.

Kai: And what they did is introduce this family of heuristics called VarSI, which is variance-aware SI. It uses that covariance info to build better groupings than the standard way.

Mira: It’s a family of rules with three variants, including a global greedy rule and a local refinement step, which you can start from either the standard sorted insertion or this new variance-aware one.

Lev: The math shows that the worst case cost for the standard sorted insertion is big O of n log n plus n squared q. But their VarSI scales as big O of n log n plus two to the k times (n q plus one).

Kai: That scaling difference suggests that even when you are building up these complex Hamiltonian structures, this grouping method actually makes a difference in how fast the computation runs on real hardware.

Mira: They tested it on one hundred thirty molecular Hamiltonians and found consistent measurement improvements over the standard sorted insertion of about thirty-eight percent. That’s a solid gain for their benchmark set.

Lev: But what’s really interesting is the comparison with overlap estimation, like iterative coefficient splitting, which they call ICS. They found that using these VarSI groups as an initialization reduces measurement costs by about nine to fifteen point three percent relative to standard sorted insertion over ICS.

Kai: That number seems small compared to the big thirty-eight percent gain they saw over standard sorted insertion, but it’s still a real reduction when you factor in the whole overlapping process.

Mira: They argue that this initial grouping step is a consequential design choice even when you are using overlapping fragments for your final estimator. It means you don't just optimize the measurement part, you have to optimize the setup too.

Lev: If we take their best refinement variants, VarSI-R and VarSI-OR, they consistently cut down on measurement requirements by around thirty-five to thirty-nine percent across all those different molecular systems.

Kai: So it looks like this isn't just some theoretical tweak; it’s something that translates directly into fewer shots on the quantum computer.

Mira: And the authors suggest that these improved initial groupings can serve as effective starting points for overlapping measurement strategies, which is a huge deal for making ICS more competitive against frameworks like SPP.

Lev: They even estimate that with this VarSI-OR initialization, you can see reductions compared to standard sorted insertion over ICS of negative one point eight percent or nine point two percent, and at most thirty-seven point five percent for those Hamiltonians mapped using the JW method.

Kai: So we’re looking at a path where we can get substantial savings on QPU time, potentially saving up to one hour of runtime in common quantum chemistry tasks when using this VarSI initialization for energy estimation.

Mira: It really shows that focusing on improving how you group those Pauli words before the measurement starts is just as important as optimizing the measurement strategy itself.

Lev: The code for building that parallel covariance dictionary is actually available on GitHub, which means we can test this out and see if it holds up when we move from theory to real hardware implementation.

The paper's improvements: Kai: So, to recap, this paper introduces these variance aware SI methods as a way to dramatically cut down on how many times we have to measure our qubits when doing quantum energy estimation using overlapping techniques.

Mira: It’s about showing that because of how you organize your initial groups, you can make the entire measurement process much more efficient for things like iterative coefficient splitting.

Lev: They are proposing these better initializations—the VarSI groups—to give the downstream estimator a better foundation to work with, which is important when we're dealing with noisy hardware.

Kai: They show that these non-overlapping groupings, even though they are just a setup step, can cut measurement counts by around thirty-five to forty percent over the standard sorted insertion method.

Mira: That's where it gets interesting because they say this is consistent across different molecular systems and even when you change the way you map your spins onto the qubits.

Lev: The caveat here is that building those covariance dictionaries for scoring, which is part of their setup, takes time and resources on a real quantum computer.

Kai: I see that, but what they’re emphasizing is that this initial grouping strategy can reduce measurement costs by nine to fifteen point three percent relative to standard sorted insertion over ICS.

Mira: That lower number seems less impressive than the thirty-five percent gain they see over the standard method, but it’s a real reduction when you look at how it compares to other complex frameworks like SPP.

Lev: If we use their most promising options, VarSI-R and VarSI-OR, they consistently reduce measurement requirements by about thirty-five to thirty-nine percent across the whole dataset regardless of what kind of covariance you use for scoring or how you map those fermions to qubits.

Kai: So it sounds like this gives us a really solid path forward—we can start with a variance aware initialization and expect some guaranteed savings on our measurement budget.

Mira: They are suggesting that these refined groupings are the most promising variants for overlapping measurement strategies, which means if you want to compete with more expensive methods, you should definitely try those two options first.

Lev: The implication for us building hardware is that this means we can potentially save up to one hour of QPU time in common quantum chemistry applications when we use the VarSI-OR initialization for energy estimation tasks.

Kai: That sounds like a significant amount of time saved, and it makes this paper really relevant for anyone working on practical quantum chemistry applications right now.

Conclusion: Kai: So we’re wrapping up on this paper, "Reducing quantum measurements in qubit-based overlapping grouping methods for quantum energy estimation through better initializations."

Mira: Basically, the authors are proving that improving how you organize your Pauli groups at the very start of a calculation can make a huge difference in how much measurement data you need.

Lev: They’re showing that these variance aware SI methods can cut measurement counts by thirty-five to forty percent over standard sorted insertion, which is pretty significant for reducing runtime on actual quantum hardware.

Kai: I mean, the fidelity of our qubits matters, and if we can cut down on the number of shots needed for those energy estimations, that’s a big win for running more complex calculations.

Mira: They stress that this improvement isn't just theoretical; it holds up across different molecular systems and even when you change how you map your spins onto the qubits.

Lev: The overhead is building those covariance dictionaries, which is something we have to account for on real hardware, but the savings they show are substantial enough to justify that setup cost for many applications.

Kai: It sounds like this gives us a really practical way to improve the measurement pipeline right out of the gate without having to completely overhaul our entire algorithm structure.

Mira: They suggest that focusing on these refined groupings, specifically VarSI-R and VarSI-OR, is the most promising path if you want to compete with more intensive measurement schemes like SPP.

Lev: I think those refinement variants are the best bets because they show consistent savings across different covariance types for scoring and different fermion to qubit mappings.

Kai: So, it seems the main point is that optimizing that initial grouping step is a key design feature when you're building overlapping measurement strategies today.

Mira: They’ve shown we can get these substantial reductions, potentially saving up to one hour of QPU time in common chemistry applications when using the VarSI-OR initialization for energy estimation tasks.

Lev: The code is available, so we can actually try implementing this on our simulators and see how it performs before we commit to a full hardware run.

Department of Chemistry and Chemical Biology, McMaster University · Brockhouse Institute for Materials Research, McMaster University

quant-ph

Submitted: 2026-07-02

Updated: 2026-11-09

Comments: Main paper: 9 pages, 2 figures, 2 tables. Supplementary Material: 22 pages, 4 figures, 5 tables

Code: https://github.com/ishume94/VarSI

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

Importance score: 79/100

The gist: Reducing quantum measurements in qubit-based overlapping grouping methods for quantum energy estimation through better initializations is addressed by introducing variance-aware SI (VarSI)…

Key concepts

Overlapping Grouping Methods
These are techniques used in quantum algorithms where the Hamiltonian is split into overlapping sets of Pauli operators. This approach helps reduce the total number of measurements needed to estimate molecular energy, which is a major bottleneck in variational quantum computing.
Sorted Insertion (SI)
SI is a standard method used to create an initial non-overlapping grouping of Pauli operators from a Hamiltonian. The paper compares its performance against VarSI, showing that VarSI's groupings are superior for reducing measurement costs.
Variance-aware SI (VarSI)
VarSI is a new heuristic that uses covariance dictionaries to construct better non-overlapping Pauli groups than standard SI. By considering the variance of operator contributions, it aims to select groups that minimize the total measurement cost more effectively.
Covariance Dictionaries
These are data structures derived from approximate wavefunctions used in quantum chemistry calculations. They provide information about how different parts of a molecule interact, which VarSI uses to inform and improve the construction of optimal Pauli groupings.

Terminology

Summary

Reducing quantum measurements in qubit-based overlapping grouping methods for quantum energy estimation through better initializations is addressed by introducing variance-aware SI (VarSI), a family of covariance-informed non-overlapping Pauli grouping heuristics to reduce measurement counts The proposed algorithms leverage covariance dictionaries, already required by overlapping methods, to construct better non-overlapping groups. p. 0

Where the paper says it

Motivated by these findings, we introduce variance-aware SI (VarSI), a family of covariance-informed non-overlapping Pauli grouping heuristics to reduce measurement counts. VarSI grouping leverages the covariance dictionaries, already required by overlapping methods, to construct better non-overlapping groups.

Molecular benchmarks with 130 Hamiltonians demonstrate consistent, non-overlapping measurement improvements over SI of 38% and enhanced downstream ICS results when initialized from VarSI groups. p. 0
Where the paper says it

We showcase the use of groupings generated by our VarSI heuristic algorithms to initialize overlapping methods using the iterative coefficient-splitting (ICS) algorithm. Molecular benchmarks with 130 Hamiltonians demonstrate consistent, non-overlapping measurement improvements over SI of 38% and enhanced downstream ICS results when initialized from VarSI groups.

We find that the initializations considered here achieve up to 70% measurement reductions for ICS, compared to the standard SI initialization with mean reductions of 9–15. p. 0
Where the paper says it

Molecular benchmarks with 130 Hamiltonians demonstrate consistent, non-overlapping measurement improvements over SI of 38% and enhanced downstream ICS results when initialized from VarSI groups. We find that the initializations considered here achieve up to 70% measurement reductions for ICS, compared to the standard SI initialization with mean reductions of 9–15.

3% depending on qubit mappings and covariance dictionaries used. p. 0
Where the paper says it

We find that the initializations considered here achieve up to 70% measurement reductions for ICS, compared to the standard SI initialization with mean reductions of 9–15. 3% depending on qubit mappings and covariance dictionaries used.

These results show that non-overlapping grouping remains a consequential design step even when the final estimator uses overlapping fragments. p. 0
Where the paper says it

3% depending on qubit mappings and covariance dictionaries used. These results show that non-overlapping grouping remains a consequential design step even when the final estimator uses overlapping fragments.

The Problem Context

The measurement cost for estimating expectation values of Hamiltonians is a central bottleneck in variational quantum algorithms. Grouping strategies significantly reduce this cost, with overlapping techniques being the state of the art in the field. p. 0

Where the paper says it

Vargas-Hern´andez1, 2, † 1Department of Chemistry and Chemical Biology, McMaster University, Hamilton, ON L8S 4M1, Canada 2Brockhouse Institute for Materials Research, McMaster University, Hamilton, ON, Canada (Dated: September 11, 2026) The measurement cost for estimating expectation values of Hamiltonians is a central bottleneck in variational quantum algorithms. Grouping strategies significantly reduce this cost, with overlapping techniques being the state of the art in the field.

Overlapping grouping methods require i) a non-overlapping grouping of the Hamiltonian, typically obtained from the Sorted Insertion (SI) algorithm as initialization, and ii) the construction of covariance dictionaries from approximate wavefunctions to guide the optimization. p. 0
Where the paper says it

Grouping strategies significantly reduce this cost, with overlapping techniques being the state of the art in the field. Overlapping grouping methods require i) a non-overlapping grouping of the Hamiltonian, typically obtained from the Sorted Insertion (SI) algorithm as initialization, and ii) the construction of covariance dictionaries from approximate wavefunctions to guide the optimization.

The number of Pauli words in molecular Hamiltonians scales O(N 4q) with the number of qubits/spin-orbitals (Nq), making the direct measurement of each Pauli operator required to obtain the necessary expectation values impractical. p. 0
Where the paper says it

For these variational algorithms, the Hamiltonian or a generic operator is mapped to a qubit operator expanded as a linear combination, H = X NP k=1 ckPk; Pk = O Nq n=1 σ (k) n , (1) where Pk are Pauli words that correspond to the tensor product of Pauli operators and identities for each qubit σ (k) n ∈ {xˆn, yˆn, zˆn, ˆ1n} associated to the k-th Pauli word and ck the coefficients of the linear combination. For quantum chemistry applications, the number of Pauli words in molecular Hamiltonians scales O(N4 q ) [20] with the number of qubits/spin-orbitals (Nq), making the direct measurement of each Pauli operator required to obtain the necessary expectation values impractical [21, 22].

VarSI Heuristics and Variants

We introduce variance-aware SI (VarSI), a family of covariance-informed non-overlapping Pauli grouping heuristics to reduce measurement counts. VarSI grouping leverages the covariance dictionaries, already required by overlapping methods, to construct better non-overlapping groups. p. 0

Where the paper says it

Motivated by these findings, we introduce variance-aware SI (VarSI), a family of covariance-informed non-overlapping Pauli grouping heuristics to reduce measurement counts. VarSI grouping leverages the covariance dictionaries, already required by overlapping methods, to construct better non-overlapping groups.

We propose three variants: a global greedy grouping insertion rule, a variance-informed SI analog, and a local refinement step initialized from SI or our variance-informed variant. p. 0
Where the paper says it

VarSI grouping leverages the covariance dictionaries, already required by overlapping methods, to construct better non-overlapping groups. We propose three variants: a global greedy grouping insertion rule, a variance-informed SI analog, and a local refinement step initialized from SI or our variance-informed variant.

The total score for a grouping G is defined as S(G) = X G∈G p VG, ε2M(G) = S(G) 2. p. 2
Where the paper says it

We limit this study to FC groupings as they provide the lowest measurement counts [22, 29, 38]; however, VarSI can be employed with QWC and k-commuting groups [52]. We define the total score for a grouping G as, S(G) = X G∈G p VG, ε2M(G) = S(G) 2 .

Performance Metrics and Scaling

The worst-case grouping cost for SI is O(N log N + N 2nq), while VarSI-O scales as O(N log N + N squared (nq + 1)). The scoring rule used across all VarSI variants is based on the metric VGα∪k = VGα + c 2 kCkk + 2ck X i∈Gα c Cik. p. 2

Where the paper says it

(9), will serve as the basis for the scoring rule used across all VarSI variants presented here. (9), one can identify an extension that further reduces the classical cost of variance-based measurement allocation methods.

Ordered VarSI (VarSI-O) orders terms by their single-term variance contribution, vi = c 2 i Cii, and then processes this fixed list once. p. 2
Where the paper says it

SI orders terms by |ci |, and then inserts each Pauli term into the first compatible group. Instead, VarSI-O orders terms by their single-term variance contribution, vi = c 2 i Cii, (11) and then processes this fixed list once.

Global greedy VarSI (VarSI-G) starts with the largest single-term variance, then scans all remaining terms and compatible destinations. p. 2
Where the paper says it

If there is no compatible group, a new singleton group is opened. Global greedy VarSI (VarSI-G) starts with the largest single-term variance, then scans all remaining terms and compatible destinations.

Comparative Results

In Table II, we show the comparison of ε2M for a series of molecular systems, LiH, BeH2, H2O, and NH3. The proposed methods here reduce the number of measurements compared to the standard SI baseline by an average of 35–43% for the refinement variants. p. 3

Where the paper says it

For the results in Table II, we used the CISD wavefunction to construct the covariance dictionaries and report the ε 2M result with the exact ground-state wavefunction obtained from diagonalizing the qubit Hamiltonian. The proposed methods here, with the exception of VarSI-G, reduce the number of measurements compared to the standard SI baseline by an average of 35–43% for the refinement variants, even while using an approximate wavefunction for the scoring steps.

For overlapping ICS measurement initialization, VarSIO-ICS/VarSI-R-ICS/VarSI-OR-ICS obtain a reduction with respect to SI-ICS of-1. p. 4
Where the paper says it

For nonoverlapping groupings, we find that the refinement variants VarSI-R and VarSI-OR consistently reduce measurement requirements by around 35–39% across the dataset, regardless of the type of covariances used for scoring and fermion-to-qubit mappings. For overlapping ICS measurement initialization, we find that when using covariance dictionaries obtained from the exact ground-state wavefunction, VarSIO-ICS/VarSI-R-ICS/VarSI-OR-ICS obtain a reduction with respect to SI-ICS of -1.

8/10.1/9.2%, respectively, with maximum reductions of 37. p. 4
Where the paper says it

For overlapping ICS measurement initialization, we find that when using covariance dictionaries obtained from the exact ground-state wavefunction, VarSIO-ICS/VarSI-R-ICS/VarSI-OR-ICS obtain a reduction with respect to SI-ICS of -1. 2%, respectively, with maximum reductions of 37.

5/38.7/40.8% for JW mapped Hamiltonians. p. 0
Where the paper says it

INTRODUCTION Several variational quantum algorithms for problems like energy estimation of many-body Hamiltonians, including electronic structure problems [1–12], and quantum machine learning (QML) protocols [13–19], ultimately require extracting expectation values of observables from measurement outcomes on a quantum device. For these variational algorithms, the Hamiltonian or a generic operator is mapped to a qubit operator expanded as a linear combination, H = X NP k=1 ckPk; Pk = O Nq n=1 σ (k) n , (1) where Pk are Pauli words that correspond to the tensor product of Pauli operators and identities for each qubit σ (k) n ∈ {xˆn, yˆn, zˆn, ˆ1n} associated to the k-th Pauli word and ck the coefficients of the linear combination.

Conclusion and Impact

Our results demonstrate that the initial grouping used in qubit-based overlapping techniques remains a consequential step in the measurement-reduction pipeline. Relative to SI baselines, our non-overlapping groupings VarSI-O, VarSI-R, and VarSI-OR reduce measurement costs, with VarSI-R and VarSI-OR emerging as the most promising variants. p. 5

Where the paper says it

Our results demonstrate that the initial grouping used in qubit-based overlapping techniques remains a consequential step in the measurement-reduction pipeline, where additional gains can still be obtained. Relative to SI baselines, our non-overlapping groupings VarSI-O, VarSI-R, and VarSI-OR reduce measurement costs, with VarSI-R and VarSI-OR emerging as the most promising variants, with average reductions of approximately 35– 39%.

Importantly, even refinements to the SI groupings used in VarSI-R yield appreciable reductions in measurement costs. p. 5
Where the paper says it

Relative to SI baselines, our non-overlapping groupings VarSI-O, VarSI-R, and VarSI-OR reduce measurement costs, with VarSI-R and VarSI-OR emerging as the most promising variants, with average reductions of approximately 35– 39%. Importantly, even refinements to the SI groupings used in VarSI-R yield appreciable reductions in measurement costs, providing a straightforward path to resource savings.

We find that the refinement variants VarSI-R and VarSI-OR consistently reduce measurement requirements by around 35–39% across the dataset, regardless of the type of covariances used for scoring and fermion-to-qubit mappings. p. 4
Where the paper says it

We found that the VarSI-G variant generally fails to outperform the SI baseline in both non-overlapping and overlapping settings, see Table II and SM. For nonoverlapping groupings, we find that the refinement variants VarSI-R and VarSI-OR consistently reduce measurement requirements by around 35–39% across the dataset, regardless of the type of covariances used for scoring and fermion-to-qubit mappings.

The proposed groupings can serve as effective initializations for overlapping measurement strategies, reducing measurement costs by an average of 9–15. p. 5
Where the paper says it

We further showed that these improved nonoverlapping groupings can serve as effective initializations for overlapping measurement strategies. When used to initialize the ICS overlapping method, the proposed groupings reduce measurement costs by an average of 9– 15.

3% relative to SI-ICS. p. 3
Where the paper says it

From this small benchmark, we found that the VarSI-O-ICS, VarSI-R-ICS, and VarSI-OR-ICS variants consistently reduce measurement costs, with VarSI-ORICS reducing measurement costs by an average of 17. 1% relative to the commonly used SI-ICS method.

This allows ICS, when combined with improved initial groupings, to better compete with more expensive frameworks such as SPP. p. 5
Where the paper says it

3% relative to SI-ICS, depending on the mapping and covariance dictionaries employed, with maximum reductions of up to 70%. This allows ICS, when combined with improved initial groupings, to better compete with more expensive frameworks such as SPP.

The reductions found with VarSI amount to up to 1 hour of QPU time saved in common quantum chemistry applications. p. 5
Where the paper says it

This allows ICS, when combined with improved initial groupings, to better compete with more expensive frameworks such as SPP. Importantly, the reductions found with VarSI amount to up to 1 hour of QPU time saved in common quantum chemistry applications.

Code Availability

The code necessary to reproduce our results, including a parallel covariance dictionary builder, is available at https://github.com/ishume94/VarSI. p. 5

Where the paper says it

CODE AVAILABILITY The code necessary to reproduce our results, including a parallel covariance dictionary builder, is available at https://github. com/ishume94/VarSI.

Acknowledgments

ILHM acknowledges support from Sandbox AQ’s 2025 Research Excellence Scholarship. This research was enabled by support from the Digital Research Alliance of Canada and NSERC Discovery Grant No. p. 5

Where the paper says it

ACKNOWLEDGMENTS ILHM acknowledges support from Sandbox AQ’s 2025 Research Excellence Scholarship. This research was enabled by support from the Digital Research Alliance of Canada and NSERC Discovery Grant No.

RGPIN-2024-06594. p. 5
Where the paper says it

This research was enabled by support from the Digital Research Alliance of Canada and NSERC Discovery Grant No. RGPIN-2024- 06594.

This research was partly enabled by Compute Ontario (computeontario. p. 5
Where the paper says it

RGPIN-2024- 06594. This research was partly enabled by Compute Ontario (computeontario.

ca) and the Digital Research Alliance of Canada (alliancecan. p. 5
Where the paper says it

This research was partly enabled by Compute Ontario (computeontario. ca) and the Digital Research Alliance of Canada (alliancecan.

ca) support. p. 5
Where the paper says it

com/ishume94/VarSI. ACKNOWLEDGMENTS ILHM acknowledges support from Sandbox AQ’s 2025 Research Excellence Scholarship.

References

[1] M. Cerezo, A. Arrasmith, R. p. 5

Where the paper says it

ca) support. Arrasmith, R.

Babbush, S. p. 5
Where the paper says it

Arrasmith, R. Babbush, S.

C. Benjamin, S. p. 5
Where the paper says it

Babbush, S. Benjamin, S.

Endo, K. Fujii, J. R. McClean, K Mitarai, X Yuan, L Cincio and P J Coles. p. 5
Where the paper says it

Cincio, and P. Coles, Nature Reviews Physics 3, 625 (2021).

[2] A Peruzzo, J McClean P Shadbolt M-H Yung X-Q Zhou P J Love A Aspuru-Guzik and J L O’Brien. p. 5

Where the paper says it

Shadbolt, M. Aspuru-Guzik, and J.

[3] O Higgott D Wang and S Brierley. p. 5

Where the paper says it

Higgott, D. Wang, and S.

[4] H R Grimsley S E Economou E Barnes and N J Mayhall. p. 5

Where the paper says it

Grimsley, S. Economou, E.

[5] S E Smart and D A Mazziotti. p. 5

Where the paper says it

Smart and D. Mazziotti, Phys.

[6] J R McClean M E Kimchi-Schwartz J Carter and W A de Jong. p. 5

Where the paper says it

Kimchi-Schwartz, J. Carter, and W.

[7] C L Cortes and S K Gray. p. 5

Where the paper says it

Cortes and S. Gray, Phys.

[8] J R McClean J Romero R Babbush and A AspuruGuzik. p. 5

Where the paper says it

Babbush, and A. AspuruGuzik, New Journal of Physics 18, 023023 (2016).

[9] J Preskill Quantum 2. p. 5

Where the paper says it

AspuruGuzik, New Journal of Physics 18, 023023 (2016). Preskill, Quantum 2, 79 (2018).

[10] T Takeshita N C Rubin Z Jiang E Lee R Babbush and J R McClean. p. 5

Where the paper says it

Takeshita, N. Babbush, and J.

[11] K Seki and S Yunoki PRX Quantum 2. p. 5

Where the paper says it

Seki and S. Yunoki, PRX Quantum 2, 010333 (2021).

[12] P J Ollitrault A Kandala C-F Chen P K Barkoutsos A Mezzacapo M Pistoia S Sheldon S Woerner J M Gambetta and I Tavernelli.

[13] L Coopmans and M Benedetti Communications Physics 7. p. 6

Where the paper says it

Coopmans and M. Benedetti, Communications Physics 7, 274 (2024).

[14] K Mitarai M Negoro M Kitagawa and K Fujii PRX A 98. p. 6

Where the paper says it

Mitarai, M. Kitagawa, and K.

[15] S Y-C Chen H-H Tseng H-Y Lin and S Yoo Learning to measure quantum neural networks (2025) arXiv:2501. p. 6

Where the paper says it

Lin, and S. Yoo, Learning to measure quantum neural networks (2025), arXiv:2501.

05663.

[16] K Shen A Kurkin A P´erez-Salinas E Shishenina V Dunjko and H Wang npj Quantum Information 11. p. 6

Where the paper says it

Dunjko, and H. Wang, npj Quantum Information 11, 178 (2025).

[17] B Zhao and K Fujii Phys. p. 6

Where the paper says it

Zhao and K. Fujii, Phys.

Rev. Res. 7.

[18] X Pan Z Lu W Wang Z Hua Y Xu W Li W Cai X Li H Wang Y-P Song C-L Zou D-L Deng and L Sun Nature Communications 14. p. 6

Where the paper says it

Deng, and L. Sun, Nature Communications 14, 4006 (2023).

[19] A Kardashin Y Balkybek V V Palyulin and K Antipin Phys. p. 6

Where the paper says it

Palyulin, and K. Antipin, Phys.

Rev. Res. 7.

[20] J T Seeley M J Richard and P J Love The Journal of Chemical Physics 137. p. 6

Where the paper says it

Richard, and P. Love, The Journal of Chemical Physics 137, 224109 (2012).

[21] J F Gonthier M D Radin C Buda E J Doskocil C M Abuan and J Romero Phys. p. 6

Where the paper says it

Abuan, and J. Romero, Phys.

Rev. Res. 4.

[22] S Patel P Jayakumar T-C Yen and A F Izmaylov Chemical Reviews 125. p. 6

Where the paper says it

Yen, and A. Izmaylov, Chemical Reviews 125, 7490 (2025).

[23] V Verteletskyi T-C Yen and A F Izmaylov The Journal of Chemical Physics 152. p. 6

Where the paper says it

Yen, and A. Izmaylov, The Journal of Chemical Physics 152, 124114 (2020).

[24] B Wu J Sun Q Huang and X Yuan Quantum 7. p. 6

Where the paper says it

Huang, and X. Yuan, Quantum 7, 896 (2023).

[25] A Jena S Genin and M Mosca arXiv:1907. p. 6

Where the paper says it

Genin, and M. Mosca, arXiv:1907.

07859 arXiv:1907. p. 6
Where the paper says it

Mosca, arXiv:1907. 07859 (2019), arXiv:1907.

07859.

[26] P Gokhale O Angiuli Y Ding K Gui T Tomesh M Suchara M Martonosi and F T Chong IEEE Transactions on Quantum Engineering 1 1. p. 6

Where the paper says it

Martonosi, and F. Chong, IEEE Transactions on Quantum Engineering 1, 1 (2020).

[27] T-C Yen V Verteletskyi and A F Izmaylov Journal of Chemical Theory and Computation 16. p. 6

Where the paper says it

Verteletskyi, and A. Izmaylov, Journal of Chemical Theory and Computation 16, 2400 (2020).

[28] O Crawford B van Straaten D Wang T Parks E Campbell and S Brierley Quantum 5. p. 6

Where the paper says it

Crawford, B. van Straaten, D.

[29] T-C Yen A Ganeshram and A F Izmaylov npj Quantum Information 9. p. 6

Where the paper says it

Ganeshram, and A. Izmaylov, npj Quantum Information 9, 14 (2023).

[30] I L Huidobro-Meezs J Dai and R A VargasHern´andez Discrete flow-based generative models for measurement optimization in quantum computing (2025) arXiv:2607. p. 6

Where the paper says it

Dai, and R. VargasHern´andez, Discrete flow-based generative models for measurement optimization in quantum computing (2025), arXiv:2509.

02794.

[31] B DalFavero R Sarkar J Rowland D Camps N P D Sawaya and R LaRose Phys. p. 6

Where the paper says it

Sawaya, and R. LaRose, Phys.

Rev. A 112.

[32] E T Campbell B M Terhal and C Vuillot Nature 549. p. 6

Where the paper says it

Terhal, and C. Vuillot, Nature 549, 172 (2017).

[33] V Vandaele S Martiel S Perdrix and C Vuillot ACM Transactions on Quantum Computing 5. p. 6

Where the paper says it

Perdrix, and C. Vuillot, ACM Transactions on Quantum Computing 5, 10.

[41] A Uvarov J D Biamonte and D Yudin Phys. p. 6

Where the paper says it

Biamonte, and D. Yudin, Phys.

Rev. B 102.

[42] R R P Singh and D A Huse Phys. p. 6

Where the paper says it

Singh and D. Huse, Phys.

Rev. B 76.

[43] A N Ciavarella Phys. p. 6

Where the paper says it

B 76, 180407(R) (2007). Ciavarella, Phys.

Rev. D 108.

[44] J Zhang R Ferguson S K¨uhn J F Haase C Wilson K Jansen and C A Muschik Quantum 7. p. 6

Where the paper says it

Jansen, and C. Muschik, Quantum 7, 1148 (2023).

[45] D Wecker M B Hastings and M Troyer Phys. p. 6

Where the paper says it

Hastings, and M. Troyer, Phys.

Rev. A 92.

[46] J Romero R Babbush J R McClean C Hempel P J Love and A Aspuru-Guzik Quantum Science and Technology 4. p. 6

Where the paper says it

Love, and A. Aspuru-Guzik, Quantum Science and Technology 4, 014008 (2018).

[47] N C Rubin R Babbush and J McClean New Journal of Physics 20. p. 6

Where the paper says it

Babbush, and J. McClean, New Journal of Physics 20, 053020 (2018).

[53] P Jordan and E P Wigner Z Phys. p. 3

Where the paper says it

SPP and soSPP values taken from Ref. Non-overlapping groupings Overlapping methods System Nq NP SI VarSI-G VarSI-O VarSI-R; S = (100/500) VarSI-OR; S = (100/500) SI -ICS VarSI-G -ICS VarSI-O -ICS VarSI-R-ICS; S = (100/500) VarSI-OR-ICS; S = (100/500) SPP so-SPP Jordan–Wigner LiH 12 630 0.

47.

[54] S B Bravyi and A Y Kitaev Annals of Physics 298. p. 6

Where the paper says it

Bravyi and A. Kitaev, Annals of Physics 298, 210 (2002).

[55] J S Kottmann S Alperin-Lea T Tamayo-Mendoza A Cervera-Lierta C Lavigne T-C Yen V Verteletskyi P Schleich A Anand M Degroote S Chaney et al. p. 6

Where the paper says it

Alperin-Lea, T. Tamayo-Mendoza, A.

Quantum Science and Technology 6. p. 6
Where the paper says it

Love, and A. Aspuru-Guzik, Quantum Science and Technology 4, 014008 (2018).

[56] Q Sun T C Berkelbach N S Blunt G H Booth S Guo Z Li J Liu J D McClain E R Sayfutyarova S Sharma S Wouters and G K-L Chan WIREs Computational Molecular Science 8. p. 6

Where the paper says it

Wouters, and G. Chan, WIREs Computational Molecular Science 8, e1340 (2018).

[57] Q Sun X Zhang S Banerjee P Bao M Barbry N S Blunt N A Bogdanov G H Booth J Chen Z-H Cui J J Eriksen Y Gao S Guo J Hermann M R Hermes K Koh P Koval S Lehtola Z Li J Liu N Mardirossian J D McClain M Motta B Mussard H Q Pham A Pulkin W Purwanto P J Robinson E Ronca E R Sayfutyarova M Scheurer H F Schurkus J E T Smith C Sun S-N Sun S Upadhyay L K Wagner X Wang A White J D Whitfield M J Williamson S Wouters J Yang J M Yu T Zhu T C Berkelbach S Sharma A Y Sokolov and G K-L Chan The Journal of Chemical Physics 153.

[58] V Bergholm J Izaac M Schuld C Gogolin S Ahmed V Ajith M S Alam G Alonso-Linaje B AkashNarayanan A Asadi J M Arrazola U Azad S Banning C Blank T R Bromley B A Cordier J Ceroni A Delgado O D Matteo A Dusko T Garg D Guala A Hayes R Hill A Ijaz T Isacsson D Ittah S Jahangiri P Jain E Jiang A Khandelwal K Kottmann R A Lang C Lee T Loke A Lowe K McKiernan J J Meyer J A Monta˜nez-Barrera R Moyard Z Niu L J O’Riordan S Oud A Panigrahi C-Y Park D Polatajko N Quesada C Roberts N S´a I Schoch B Shi S Shu S Sim A Singh I Strandberg J Soni A Sz´ava S Thabet R A Vargas-Hern´andez T Vincent N Vitucci M Weber D Wierichs R Wiersema M Willmann V Wong S Zhang and N Killoran Pennylane: Automatic differentiation of hybrid quantumclassical computations (2022) arXiv:1811.04968.

[60] C Tornow N Kanazawa W E Shanks and D J Egger Phys. p. 7

Where the paper says it

Shanks, and D. Egger, Phys.

Rev. Appl. 17.

[61] IBM Quantum Platform Workload usage https://quantum. p. 7

Where the paper says it

17, 064061 (2022). [61] IBM Quantum Platform, Workload usage, https://quantum.

cloud.ibm.com/docs/en/guides/estimate-job-run-time (2026) iBM Quantum Documentation accessed 2 July 2026. p. 7
Where the paper says it

[61] IBM Quantum Platform, Workload usage, https://quantum. com/docs/en/guides/ estimate-job-run-time (2026), iBM Quantum Documentation, accessed 2 July 2026.

[62] J Rowland R Sarkar N P Sawaya N M Tubman and R LaRose Overlapped groupings for quantum energy estimation: Maximal variance reduction and deterministic algorithms for reducing variance (2026) arXiv:2604. p. 7

Where the paper says it

Tubman, and R. LaRose, Overlapped groupings for quantum energy estimation: Maximal variance reduction and deterministic algorithms for reducing variance (2026), arXiv:2604.

07156.

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. Variance-aware sorted insertion (VarSI) algorithms can generate non-overlapping Pauli groups that reduce measurement counts over standard Sorted Insertion (SI) by an average of 35 to 43 percent for refinement variants. This allows overlapping methods like iterative coefficient splitting (ICS) to compete with more expensive frameworks like Shared Pauli Products (SPP).

  2. The proposed grouping can serve as effective initializations for overlapping measurement strategies, reducing measurement costs by an average of 9 to 15.3 percent relative to SI-ICS, with maximum reductions of up to 70 percent when combined with the ICS method.

  3. The system can achieve savings equivalent to approximately one hour of QPU runtime in common quantum chemistry applications when using the VarSI-OR-ICS initialization for energy estimation tasks.

Abstract

The measurement cost for estimating expectation values of Hamiltonians is a central bottleneck in variational quantum algorithms. Grouping strategies significantly reduce this cost, with overlapping techniques as the state of the art. Overlapping grouping methods require i) a non-overlapping grouping of the Hamiltonian, typically obtained from the Sorted Insertion (SI) algorithm as initialization, and ii) the construction of covariance dictionaries from approximate wavefunctions to guide the optimization. It was recently shown that different initializations can potentially reduce measurement costs for overlapping methods. Motivated by these findings, we introduce variance-aware SI (VarSI), a family of covariance-informed non-overlapping Pauli grouping heuristics to reduce measurement counts. VarSI grouping leverages the covariance dictionaries, already required by overlapping methods, to construct better non-overlapping groups. We propose three variants: a global greedy grouping insertion rule, a variance-informed SI analog, and a local refinement step initialized from SI or our variance-informed variant. We showcase the use of the VarSI groupings as initialization of overlapping methods using the iterative coefficient-splitting (ICS) algorithm. Molecular benchmarks with 130 Hamiltonians demonstrate consistent, non-overlapping measurement improvements over SI of 36--39% and enhanced downstream ICS results when initialized from VarSI groups. We find that the initializations considered here achieve larger reductions of up to 48% for systems whose covariances deviate from idealized cases. Compared to the standard SI initialization, VarSI reaches mean reductions of 10.3--13.5% depending on qubit mappings and covariance dictionaries used. These results show that non-overlapping grouping remains a consequential design step even when the final estimator uses overlapping fragments.

Sources

Related papers