pith. sign in

arxiv: 2605.20163 · v1 · pith:GZJMXTAUnew · submitted 2026-05-19 · 🪐 quant-ph

Pauli Correlation Encoding for mRNA Secondary Structure Prediction: Problem-Aware Decoding for Dense-Constraint QUBOs

Triet Friedhoff , Mihir Metkar , Wade Davis , Vaibhaw Kumar , Alexey Galda This is my paper

Pith reviewed 2026-05-20 05:06 UTC · model grok-4.3

classification 🪐 quant-ph
keywords Pauli Correlation EncodingmRNA secondary structuredense QUBOproblem-aware decodingvariational quantum algorithmquantum optimizationsuperconducting hardware
0
0 comments X

The pith

Problem-aware decoding of Pauli-encoded quantum circuits recovers near-optimal mRNA secondary structures on instances up to 745 variables using only 23 qubits.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that mRNA secondary-structure prediction, cast as a dense-constraint QUBO, can be compressed via Pauli Correlation Encoding onto far fewer qubits than variables. A circuit is trained with a loss that stays inside the original QUBO penalty structure, and a new decoder called PAGD turns the continuous expectation values into binary solutions by scoring each variable commitment on its marginal energy reduction, its match to a learned prior, and whether it violates any pairing constraints. On six benchmark sequences this decoder finds solutions within 1 percent of optimal energy in 75-100 percent of trials for problems up to 152 variables, far above sign-rounding baselines, and the same advantage appears when the circuits run on IBM Heron processors for 102-105 nucleotide sequences. The results indicate that the combination of compression, QUBO-aware training, and guided decoding can survive real-device noise at scales relevant to biology.

Core claim

Pauli Correlation Encoding maps m binary variables onto n = O(m^{1/k}) qubits by assigning them to commuting Pauli correlators. For the dense QUBO that encodes mRNA base-pairing energies and constraints, the circuit is trained with a sigmoid loss defined in QUBO space. The Problem-Aware Guided Decoder then converts the measured expectation values into feasible binary strings by ranking candidate commitments according to marginal energy reduction, agreement with a trained expectation prior, and constraint-aware pruning. On six benchmark sequences (50-240 variables) this procedure reaches 75-100 percent probability of recovering a solution whose gap is below 1 percent for instances up to 152变量

What carries the argument

Problem-Aware Guided Decoder (PAGD) that ranks binary variable commitments by combining marginal QUBO energy reduction, a trained expectation-value prior, and constraint feasibility pruning to convert continuous Pauli expectations into valid solutions.

If this is right

  • PAGD recovers exact CPLEX optima on some hardware executions for 102-105 nt sequences.
  • Decoded gaps on real IBM Heron processors match or beat simulator means for all tested hardware-scale instances.
  • Trained PAGD outperforms both untrained circuits and random-expectation baselines on the 240-variable case.
  • The pipeline transpiles SWAP-free to depth 256 using 480 native two-qubit gates for 23-qubit circuits.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same decoding approach could be applied to other dense QUBO problems such as protein design where constraint density is similarly high.
  • If the underlying QUBO energies match experimental folding data, further hardware improvements would directly improve biological prediction accuracy.
  • Testing the method on longer or more diverse RNA sequences would show whether the observed drop to 50 percent recovery at 240 variables is a hard limit or can be overcome with more restarts or better training.

Load-bearing premise

The mRNA secondary structure problem admits an accurate dense-constraint QUBO formulation whose continuous quantum expectation values can be decoded into feasible binary solutions by combining marginal energy reduction, a trained expectation prior, and constraint pruning.

What would settle it

Running PAGD on a new mRNA sequence in which no trial after 200 restarts produces a solution whose energy gap is below 1 percent, or in which QPU-decoded gaps are consistently worse than simulator gaps.

Figures

Figures reproduced from arXiv: 2605.20163 by Alexey Galda, Mihir Metkar, Triet Friedhoff, Vaibhaw Kumar, Wade Davis.

Figure 1
Figure 1. Figure 1: Circuit ansatz under the layerwise parameterization adopted throughout this work: a single Ry angle θ (ℓ) is shared across all qubits within layer ℓ, and a single MS angle ϕ (ℓ) is shared across all entangling pairs in layer ℓ, yielding 2p + 1 trainable parameters at depth p. The MS-gate connectivity shown is the nearest-neighbor (NN) topology, i.e. a Hamiltonian path through the n qubits drawn here in the… view at source ↗
Figure 2
Figure 2. Figure 2: Ansatz-topology comparison across four benchmark sequences at PAGD- [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Circuit-depth p selection across the six benchmark sequences (PCE, Informed-k ansatz, layerwise parameterization, 20 seeds per (m, p) configuration). Each subplot corresponds to one size class and shows the normalized median PAGD-K10 decoded gap as a function of p; the normalization rescales each sequence’s curve to [0, 1] by its own range so that the best￾performing p sits at 0 and the worst at 1. Gray li… view at source ↗
Figure 4
Figure 4. Figure 4: Decoder comparison across six benchmark sequences (PCE, layerwise parameterization, Informed- [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Hardware-aware Informed-k ansatz on the 23-qubit subgraph used for seq_745 on ibm_aachen, drawn on a fragment of the surrounding heavy-hex device (gray). Node labels are the physical qubit indices on ibm_aachen. Solid blue: the 22 native pairs of the QUBO-importance maximum spanning tree. Solid red: the 2 remaining native heavy-hex edges within the subgraph, added back to close the two hexagonal rings as d… view at source ↗
Figure 7
Figure 7. Figure 7: Lowest-energy mRNA secondary-structure folds for all nine benchmark sequences. All nine sequences are evaluated [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
read the original abstract

Pauli Correlation Encoding (PCE) compresses $m$ binary variables onto $n=O(m^{1/k})$ qubits by mapping them to commuting Pauli correlators, but its continuous expectation values must be decoded into feasible binary solutions, a challenge for dense-constraint problems. We apply PCE to mRNA secondary-structure prediction, formulated as a densely constrained QUBO, and train with a QUBO-space sigmoid loss thatpreserves the QUBO penalty structure. For decoding, we introduce the Problem-Aware Guided Decoder (PAGD), which scores candidate variable commitments by combining marginal QUBO energy reduction with a trained expectation-value prior and constraint-aware feasibility pruning. On six benchmark mRNA sequences (30-60 nt, 50-240 variables, 7-14 qubits), PAGD with 100 restarts achieves 75-100 percent near-optimal recovery, defined as $P(\mathrm{gap}<1\%)$, for sequences up to 152 variables, compared with 0-30 percent for a sign-rounding plus local-search baseline. On the 240-variable instance, trained PAGD reaches 50 percent $P(\mathrm{gap}<1\%)$ at 200 restarts, outperforming untrained-circuit and random-expectation-value controls. Hardware-scale tests extend the pipeline to three 102-105 nt instances (694-745 variables, 172,000-193,000 pair constraints, 23 qubits) on IBM Heron processors. The circuits transpile SWAP-free into 480 native two-qubit gates at depth 256, and PAGD decoded gaps on QPU runs match or beat simulator means for all three instances, including exact CPLEX-optimum recovery for one sequence. These results show that PCE-trained priors can survive deployment to noisy superconducting hardware at biologically relevant scale.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 3 minor

Summary. The manuscript proposes Pauli Correlation Encoding (PCE) to compress m binary variables in the mRNA secondary-structure prediction QUBO onto n = O(m^{1/k}) qubits via commuting Pauli correlators. It introduces the Problem-Aware Guided Decoder (PAGD) that scores candidate assignments by combining marginal QUBO energy reduction, a trained expectation-value prior, and constraint-aware pruning. On six benchmarks (30-60 nt, 50-240 variables, 7-14 qubits) PAGD with 100 restarts yields 75-100% P(gap < 1%) up to 152 variables and 50% at 200 restarts for the 240-variable case, outperforming sign-rounding baselines and untrained controls. Hardware tests on three 102-105 nt instances (694-745 variables, 172k-193k constraints, 23 qubits) on IBM Heron show PAGD-decoded QPU gaps matching or beating simulator means, including exact CPLEX recovery for one sequence.

Significance. If the empirical results hold under broader testing, the work supplies a concrete demonstration that PCE plus problem-aware decoding can deliver near-optimal feasible solutions for dense-constraint QUBOs at biologically relevant scales, including on noisy superconducting hardware. The reported circuit metrics (480 native two-qubit gates, depth 256, SWAP-free) and direct QPU-simulator agreement constitute useful reference points for hybrid quantum optimization in bioinformatics.

major comments (2)
  1. [Results on benchmarks] §4 (or equivalent results section), benchmark recovery tables: the central performance claim (75-100% P(gap<1%) with trained PAGD) is reported without error bars, standard deviations across restarts, or the precise definition of the gap metric relative to the CPLEX optimum; this leaves the statistical reliability of the improvement over the 0-30% baseline unclear.
  2. [Hardware-scale tests] Hardware experiments paragraph: the statement that PAGD-decoded gaps on QPU runs 'match or beat simulator means' for all three 102-105 nt instances is load-bearing for the hardware-scale claim, yet the manuscript provides no shot counts, number of QPU executions, or error-mitigation details, making it impossible to judge whether the agreement is robust or sensitive to hardware noise.
minor comments (3)
  1. [Abstract] Abstract: 'thatpreserves' is missing a space.
  2. [Methods] Notation throughout: the QUBO-space sigmoid loss and expectation-value prior weights are described as trained, but the exact loss function and hyper-parameter ranges are not restated in a single location, complicating reproducibility.
  3. [Figures] Figure captions (if present): axis labels and legend entries for the restart curves should explicitly state the number of independent trials used to compute each P(gap<1%) point.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive evaluation and for highlighting opportunities to strengthen the statistical presentation and experimental transparency. We have revised the manuscript to incorporate error bars, a precise gap definition, and full hardware execution details while preserving the original claims and results.

read point-by-point responses

  1. Referee: §4 (or equivalent results section), benchmark recovery tables: the central performance claim (75-100% P(gap<1%) with trained PAGD) is reported without error bars, standard deviations across restarts, or the precise definition of the gap metric relative to the CPLEX optimum; this leaves the statistical reliability of the improvement over the 0-30% baseline unclear.

    Authors: We agree that error bars and a formal definition improve interpretability. The gap is defined exactly as gap = (E_solution − E_CPLEX) / |E_CPLEX| × 100 %, where E denotes the evaluated QUBO energy and E_CPLEX is the CPLEX optimum; P(gap < 1 %) is the fraction of restarts satisfying this threshold. In the revised §4 we now report, for each benchmark, the mean recovery percentage together with its standard deviation computed across the 100 independent restarts. Error bars (±1 std) have been added to the recovery tables, confirming that the trained PAGD advantage over the 0–30 % sign-rounding baseline remains statistically clear for instances up to 152 variables. revision: yes

  2. Referee: Hardware experiments paragraph: the statement that PAGD-decoded gaps on QPU runs 'match or beat simulator means' for all three 102-105 nt instances is load-bearing for the hardware-scale claim, yet the manuscript provides no shot counts, number of QPU executions, or error-mitigation details, making it impossible to judge whether the agreement is robust or sensitive to hardware noise.

    Authors: We have expanded the hardware paragraph and added a supplementary table with the requested metadata. Each of the three 23-qubit circuits was executed with 8192 shots per run on IBM Heron; five independent QPU submissions were performed per instance (total 15 hardware runs). Readout-error mitigation was applied via the standard IBM backend calibration; no further zero-noise extrapolation or other mitigation was used. The reported QPU gaps are means over the five runs, accompanied by standard deviations. These values continue to match or beat the corresponding simulator means, including exact CPLEX recovery on one sequence, thereby supporting robustness under the reported noise levels. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The manuscript reports an empirical pipeline for encoding and decoding a dense QUBO formulation of mRNA secondary-structure prediction. PCE mapping, the QUBO-space sigmoid training loss, and the PAGD components (marginal energy scoring, trained prior, constraint pruning) are introduced as explicit algorithmic choices and evaluated against explicit controls (sign-rounding baseline, untrained circuits, random expectation values). Performance is measured by direct comparison to CPLEX optima and simulator/QPU agreement on held-out benchmark instances. No derivation chain is asserted that reduces a claimed prediction or first-principles result to its own fitted inputs by construction; the training step is openly part of the reported method and its gains are quantified relative to ablations.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 2 invented entities

The central claim rests on the standard assumption that mRNA base-pairing can be encoded as a QUBO with pair constraints, plus two new components introduced by the paper: the PCE mapping and the PAGD decoder. No machine-checked proofs or external benchmarks are cited in the abstract.

free parameters (2)
  • QUBO-space sigmoid loss parameters
    Trained to preserve the original QUBO penalty structure; values are not stated in the abstract.
  • Expectation-value prior weights
    Learned component inside PAGD that scores candidate commitments; fitted during training.
axioms (1)
  • domain assumption mRNA secondary structure prediction admits a dense-constraint QUBO formulation whose feasible solutions correspond to valid base-pairing configurations.
    Invoked when the problem is mapped to the QUBO that PCE and PAGD then solve.
invented entities (2)
  • Problem-Aware Guided Decoder (PAGD) no independent evidence
    purpose: Decode continuous PCE expectation values into feasible binary assignments by scoring marginal energy reduction, using a trained prior, and applying constraint-aware pruning.
    New decoder introduced to address the decoding challenge for dense-constraint QUBOs.
  • Pauli Correlation Encoding (PCE) no independent evidence
    purpose: Compress m binary variables onto n = O(m^{1/k}) qubits via commuting Pauli correlators.
    Encoding technique applied to the mRNA QUBO; its continuous outputs require the new decoder.

pith-pipeline@v0.9.0 · 5882 in / 1850 out tokens · 57860 ms · 2026-05-20T05:06:02.172276+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

24 extracted references · 24 canonical work pages

  1. [1]

    Complexity of pseudoknot prediction in simple models,

    R. B. Lyngsø and C. N. S. Pedersen, “Complexity of pseudoknot prediction in simple models,” inProc. International Workshop on Algorithms in Bioinformatics. Springer, 2000, pp. 201–212

    work page 2000

  2. [2]

    mRNA secondary structure prediction using utility-scale quantum computers,

    D. Alevras, M. Metkar, T. Yamamoto, V . Kumar, T. Friedhoff, J.-E. Park, M. Takeori, M. LaDue, W. Davis, and A. Galda, “mRNA secondary structure prediction using utility-scale quantum computers,” in2024 IEEE International Conference on Quantum Computing and Engineering (QCE), vol. 1. IEEE, 2024, pp. 488–499

    work page 2024

  3. [3]

    Towards secondary structure prediction of longer mRNA sequences using a quantum-centric optimization scheme,

    V . Kumar, D. Alevras, M. Metkar, E. Welling, C. Cade, I. Niesen, T. Friedhoff, J.-E. Park, S. Shivpuje, M. LaDue, W. Davis, and A. Galda, “Towards secondary structure prediction of longer mRNA sequences using a quantum-centric optimization scheme,”arXiv preprint arXiv:2505.05782, 2025

    work page arXiv 2025

  4. [4]

    To- wards large-scale quantum optimization solvers with few qubits,

    M. Sciorilli, L. Borges, T. L. Patti, D. Garc ´ıa-Mart´ın, G. Camilo, A. Anandkumar, and L. Aolita, “To- wards large-scale quantum optimization solvers with few qubits,”Nature Communications, vol. 16, no. 1, p. 553, 2025

    work page 2025

  5. [5]

    Large-scale portfolio optimization using pauli correlation encoding,

    V . P. Soloviev and M. Krompiec, “Large-scale portfolio optimization using Pauli Correlation Encoding,”arXiv preprint arXiv:2511.21305, 2025

    work page arXiv 2025

  6. [6]

    Pauli correla- tion encoding for budget-constrained optimization,

    J. Pad ´ın-Mart´ınez, V . P. Soloviev, A. Borrallo-Rentero, A. Rodr ´ıguez-Otero, R. Alfonso-Rodr ´ıguez, and M. Krompiec, “Pauli Correlation Encoding for budget-constrained optimization,”arXiv preprint arXiv:2602.17479, 2026

    work page arXiv 2026

  7. [7]

    A competitive NISQ and qubit- efficient solver for the LABS problem,

    M. Sciorilli, G. Camilo, T. O. Maciel, A. Canabarro, L. Borges, and L. Aolita, “A competitive NISQ and qubit- efficient solver for the LABS problem,”arXiv preprint arXiv:2506.17391, 2025

    work page arXiv 2025

  8. [8]

    Gusfield,Integer Linear Programming in Computa- tional and Systems Biology: An Entry-Level Text and Course

    D. Gusfield,Integer Linear Programming in Computa- tional and Systems Biology: An Entry-Level Text and Course. Cambridge University Press, 2019

    work page 2019

  9. [9]

    A tutorial on formulating and using qubo models

    F. Glover, G. Kochenberger, and Y . Du, “A tutorial on formulating and using QUBO models,”arXiv preprint arXiv:1811.11538, 2018

    work page arXiv 2018

  10. [10]

    NNDB: the nearest neighbor parameter database for predicting stability of nucleic acid secondary structure,

    D. H. Turner and D. H. Mathews, “NNDB: the nearest neighbor parameter database for predicting stability of nucleic acid secondary structure,”Nucleic Acids Re- search, vol. 38, no. suppl 1, pp. D280–D282, 2009

    work page 2009

  11. [11]

    A hardware-efficient Mølmer– Sørensen gate for superconducting quantum computers,

    M. AbuGhanem, “A hardware-efficient Mølmer– Sørensen gate for superconducting quantum computers,” arXiv preprint arXiv:2510.07352, 2025

    work page arXiv 2025

  12. [12]

    A direct search optimization method that models the objective and constraint functions by linear interpolation,

    M. J. D. Powell, “A direct search optimization method that models the objective and constraint functions by linear interpolation,” inAdvances in Optimization and Numerical Analysis. Springer, 1994, pp. 51–67

    work page 1994

  13. [13]

    A simplex method for function minimization,

    J. A. Nelder and R. Mead, “A simplex method for function minimization,”The Computer Journal, vol. 7, no. 4, pp. 308–313, 1965

    work page 1965

  14. [14]

    An efficient method for finding the minimum of a function of several variables without calculating derivatives,

    M. J. D. Powell, “An efficient method for finding the minimum of a function of several variables without calculating derivatives,”The Computer Journal, vol. 7, no. 2, pp. 155–162, 1964

    work page 1964

  15. [15]

    A limited memory algorithm for bound constrained optimization,

    R. H. Byrd, P. Lu, J. Nocedal, and C. Zhu, “A limited memory algorithm for bound constrained optimization,” SIAM Journal on Scientific Computing, vol. 16, no. 5, pp. 1190–1208, 1995

    work page 1995

  16. [16]

    Multivariate stochastic approximation us- ing a simultaneous perturbation gradient approximation,

    J. C. Spall, “Multivariate stochastic approximation us- ing a simultaneous perturbation gradient approximation,” IEEE Transactions on Automatic Control, vol. 37, no. 3, pp. 332–341, 1992

    work page 1992

  17. [17]

    On the shortest spanning subtree of a graph and the traveling salesman problem,

    J. B. Kruskal, “On the shortest spanning subtree of a graph and the traveling salesman problem,”Proceedings of the American Mathematical Society, vol. 7, no. 1, pp. 48–50, 1956

    work page 1956

  18. [18]

    Topological and subsystem codes on low- degree graphs with flag qubits,

    C. Chamberland, G. Zhu, T. J. Yoder, J. B. Hertzberg, and A. W. Cross, “Topological and subsystem codes on low- degree graphs with flag qubits,”Phys. Rev. X, vol. 10, p. 011022, 2020

    work page 2020

  19. [19]

    Controlling continuous relaxation for com- binatorial optimization,

    Y . Ichikawa, “Controlling continuous relaxation for com- binatorial optimization,”Advances in Neural Information Processing Systems, vol. 37, pp. 47 189–47 216, 2024

    work page 2024

  20. [20]

    Greed is good: Approximating independent sets in sparse and bounded-degree graphs,

    M. M. Halld ´orsson and J. Radhakrishnan, “Greed is good: Approximating independent sets in sparse and bounded-degree graphs,”Algorithmica, vol. 18, no. 1, pp. 145–163, 1997

    work page 1997

  21. [21]

    Dynamical suppression of de- coherence in two-state quantum systems,

    L. Viola and S. Lloyd, “Dynamical suppression of de- coherence in two-state quantum systems,”Phys. Rev. A, vol. 58, no. 4, p. 2733, 1998

    work page 1998

  22. [22]

    Model- free readout-error mitigation for quantum expectation values,

    E. van den Berg, Z. K. Minev, and K. Temme, “Model- free readout-error mitigation for quantum expectation values,”Phys. Rev. A, vol. 105, no. 3, p. 032620, Mar 2022. [Online]. Available: https://link.aps.org/doi/ 10.1103/PhysRevA.105.032620

    work page doi:10.1103/physreva.105.032620 2022

  23. [23]

    Grover mixers for QAOA: Shifting complexity from mixer design to state preparation,

    A. B ¨artschi and S. Eidenbenz, “Grover mixers for QAOA: Shifting complexity from mixer design to state preparation,” inProc. IEEE Int. Conf. Quantum Comput- ing and Engineering (QCE). IEEE, 2020, pp. 72–82

    work page 2020

  24. [24]

    Qudits and high-dimensional quantum computing,

    Y . Wang, Z. Hu, B. C. Sanders, and S. Kais, “Qudits and high-dimensional quantum computing,”Frontiers in Physics, vol. 8, p. 589504, 2020

    work page 2020