{"id":"0589fdb1-ae79-47f5-9981-ebf8a1753777","arxiv_id":"1908.02163","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":6,"one_line_summary":"The authors introduce an O(N^4)-term Hamiltonian for lattice protein folding and demonstrate a CVaR-VQE plus genetic algorithm that folds small peptides, including a 7-amino-acid peptide on IBM Q hardware.","lead":"An IBM quantum computing group constructed a coarse-grained protein folding model on a tetrahedral lattice and encoded it as a quantum Hamiltonian whose lowest energy state is meant to be the folded chain. They used a hybrid quantum-classical optimizer to fold a 7-amino-acid peptide on a real 9-qubit IBM Q chip and a 10-amino-acid peptide in a noisy 22-qubit simulation.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Hamiltonian is a relaxed objective: setting all interaction qubits to 0 removes overlap penalties and repulsive contacts, so non-self-avoiding configurations can be ground states.","rationale":"The paper's load-bearing assertion is that the minimum of H(q) is the native fold. That assertion requires exact self-avoidance and an honest representation of all contact energies. The paper explicitly disclaims both: only local overlaps near active interaction pairs are penalized, and the q-register can switch off every interaction term. This is not a modeling convention I happen to dislike; it is an internal inconsistency between the Hamiltonian and the problem it claims to encode. The same two mechanisms were named by the reader's weakest_assumption, so my analysis agrees. A useful independent check is exhaustive enumeration for the paper's own 7-residue experiment: with only a few hundred basis states, one can decide unambiguously whether any invalid minimum exists. The hardware run and the O(N^4) fit are genuine contributions, but they validate the optimizer and the term-counting, not the mapping from H(q) to protein folding. Since the central mapping fails as stated, the rejection stands.","tokens_in":32026,"tokens_out":10130,"duration_ms":115728,"concrete_test":"Enumerate, for the small APRLRFY system reported in Fig. 3, all computational basis states of the 9-qubit register. Reconstruct each turn assignment into Cartesian coordinates on the tetrahedral lattice via Eqs. (SI-3)-(SI-13), flag any pair of non-bonded beads that share a lattice site, and compute the paper's H(q) for each basis state using its lambda ratios and MJ parameters. If the lowest-energy state contains an overlap, or if an overlapping state is degenerate with the lowest self-avoiding state, the ground-state-to-fold equivalence is refuted. The enumeration is exhaustive, so this check settles the concern for the reported experiment.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that 'the solution to the folding problem is the ground state of the Hamiltonian H(q)' (folding algorithm section). This requires that every global minimum of H(q), after optimizing over all registers, is a self-avoiding conformation with the correct contact energy. The construction does not enforce this. The paragraph after Eq. (1) states that 'we only prevent overlaps that occur in the vicinity of an interaction pair,' and the overlap penalties are multiplied by the interaction qubits q_ij^(l). Setting every q_ij^(l)=0 therefore removes those penalties and also removes any repulsive contact energy. In that case the energy of an overlapping configuration is governed only by Hgc and Hch, which (Eqs. SI-15 to SI-26) forbid immediate back-tracking and enforce side-chain chirality but do not prevent long-range self-intersection. Consequently, a path that crosses itself away from any chosen contact is energetically degenerate with a valid no-contact path, and for sequences whose optimized q is all-zero the global minimum is not a valid fold. The claim that the ground state 'lies in the 2^Ncf dimensional space of the configuration qubits' is also not justified because H(q) acts on the interaction register as well. The O(N^4) term count and the hardware demonstration do not repair this gap.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a coarse-grained quantum algorithm for protein folding on a tetrahedral lattice. The model encodes polymer turns in configuration qubits and pairwise contacts in a separate interaction-qubit register, and defines a Hamiltonian H(q) = H_gc + H_ch + H_in with O(N^4) Pauli terms and O(N^2) qubits. The authors combine CVaR-VQE with a differential-evolution optimizer and report (i) a noisy simulation of the 10-amino-acid Angiotensin peptide on 22 qubits and (ii) a hardware experiment folding a 7-amino-acid neuropeptide on 9 qubits of a 20-qubit IBM Q device. The central claim is that the minimum-energy self-avoiding lattice conformation is obtained as the ground state of H(q).","tokens_in":32305,"tokens_out":3754,"duration_ms":42338,"significance":"If the Hamiltonian construction were sound, the O(N^4) scaling, the quadratic qubit count, and the 9-qubit hardware demonstration would be a useful step toward NISQ-era protein-folding experiments. The use of Miyazawa-Jernigan contact energies for the Angiotensin simulation is a positive feature, as is the validation on real hardware of the CVaR-VQE/genetic-optimizer workflow. However, the central mapping from folding problem to ground state is not established: the Hamiltonian does not enforce global self-avoidance, and the interaction qubits allow the optimizer to remove all overlap penalties and repulsive terms. The secondary-structure demonstration in the SI is also constructed in a circular way. These issues affect the core scientific claim, not merely the presentation.","major_comments":[{"comment":"The central equivalence between the ground state of H(q) and the minimum-energy self-avoiding fold is not established. The overlap penalties appear only inside h^{(1)}_{ij} in Eq. (SI-30), multiplied by q^{(1)}_{ij}, and the main text explicitly states that 'we only prevent overlaps that occur in the vicinity of an interaction pair.' Since the interaction qubits are variational degrees of freedom, the optimizer can set every q^{(l)}_{ij}=0. In that case H(q) reduces to H_gc + H_ch, which penalizes only immediate back-tracking and chirality (Eqs. SI-17 and SI-26) and does not prevent long-range self-intersection. A self-intersecting configuration with no selected contacts is therefore energetically degenerate with a valid no-contact fold, and for sequences whose optimized q register is all-zero the ground state is not a valid protein conformation. This invalidates the claim in the folding-algorithm section that 'the solution to the folding problem is the ground state of H(q)'.","section":"Main text, 'The Hamiltonian' and 'The interaction energy terms'; SI Eq. (SI-30)"},{"comment":"The assertion that the ground state 'lies in the 2^{N_cf} dimensional space of the configuration qubits' is not correct as stated: H(q) acts on the joint register (q_cf, q_in), and the minimization is performed over both registers. The energy labels in Fig. 2(b) are reported in terms of contact-qubit strings rather than full configuration strings, so the low-energy states are not shown to correspond to self-avoiding conformations of the polymer. A valid ground-state computation must optimize over q_in and then verify that the resulting configuration is self-avoiding; the manuscript does neither.","section":"Main text, 'Folding algorithm' section and Fig. 2(b)"},{"comment":"The secondary-structure demonstration is circular as presented. The contact maps in Fig. S2 are explicitly designed to stabilize an alpha-helix (upper triangle) or an antiparallel beta-sheet (lower triangle), and the simulations then reproduce those expected structures. This does not provide evidence that the model predicts secondary structure from sequence; it only shows that the optimizer can find the designed minimum of a hand-built cost function. This section should be reframed as a consistency check rather than as a reproduction of secondary structure.","section":"SI, Section II, Fig. S2"},{"comment":"The model cannot represent repulsive interactions. The term q^{(l)}_{ij}(epsilon^{(l)}_{ij}+lambda(d(i,j)-l)) contributes only when the interaction qubit is 1. For a repulsive interaction with epsilon > 0, the optimizer can set q=0 and remove the energy penalty entirely, so configurations in which repulsive pairs are in contact are never penalized. The Discussion's statement that the model can account for Lennard-Jones-like interactions is therefore unsupported by the Hamiltonian as written.","section":"Main text, 'The interaction energy terms' and Discussion"}],"minor_comments":[{"comment":"There are typos in the text, including 'curently' in the Introduction and 'Hamltonain'/'Hamltonian' in the Discussion; these should be corrected.","section":"Introduction and Discussion, spelling"},{"comment":"The penalty parameters lambda_1, lambda_2, lambda_3, lambda_5 and the chirality/backtracking penalties lambda_back, lambda_chirality are not given numerical values for the reported simulations; please provide the values used and state how they were chosen to dominate the contact energies.","section":"SI Eq. (SI-30) and Materials and Methods"},{"comment":"The text reports max_p P0(p) = 42.2% for the Angiotensin simulation and later says max_p P0(p) peaks at 33% for the 7-amino-acid hardware run; clarify in the figure caption and text which probability is being reported for each system.","section":"Main text, 'Applications' and Fig. 3(e)"}],"recommendation":"reject","confidential_remarks":"The central gap is in the constraint structure: global self-avoidance is not enforced, and the interaction qubits allow the optimizer to bypass all overlap and repulsive penalties. Repairing this would require a fundamentally different Hamiltonian construction or an additional global constraint, which would change the O(N^4)/O(N^2) resource claims and the experimental interpretation. The hardware demonstration may still be of interest as a test of CVaR-VQE on a classical cost function, but it does not validate a solution to the protein-folding problem."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The real news here is the encoding and the 9-qubit IBM Q run. The interaction-qubit register, the tetrahedral lattice with chirality constraints, and the dense 2-qubit-per-turn variant are all new relative to the earlier quantum folding literature. The CVaR-VQE plus differential evolution combination also seems to work well in practice, and the hardware demonstration on a 7-residue peptide was a legitimate milestone for 2019. Give credit where it is due: the paper shows real craft in Hamiltonian construction and a sensible experimental pipeline.\n\nThe soft spot is load-bearing. The paper claims that the solution to the folding problem is the ground state of H(q), but the Hamiltonian is a relaxed objective. Overlap penalties are multiplied by the interaction qubits q_ij^(l), and the text admits that overlaps are only prevented in the vicinity of an interaction pair. If the optimizer sets every q to zero, all overlap penalties and any repulsive contact energy vanish, leaving only the geometric and chirality constraints. Those do not prevent long-range self-intersection. So a non-self-avoiding configuration can become globally optimal. That is not a minor gap; it breaks the stated equivalence between the ground state and the minimum-energy valid fold. The paper itself hints at this when it calls the treatment of overlaps \"unconventional,\" but it never addresses the q=0 switch.\n\nOther concerns are minor. The O(N^4) scaling is demonstrated by counting Pauli strings for small N and fitting a curve, not derived. The SI secondary-structure demonstration tunes the contact map to produce an alpha-helix or beta-sheet and then reports that structure, which is input-to-output rather than a prediction. Those are worth noting but not disqualifying by themselves.\n\nWho is this for? People working on NISQ optimization for classical cost functions and on lattice protein models. It is a good example of how an encoding choice can silently relax the problem you intended to solve. A serious referee should engage with it, because the encoding ideas are reusable even if the central claim needs repair. My recommendation: send it to review, but with the expectation that the authors either prove the q=0 case is harmless (I do not see how), modify the Hamiltonian to enforce self-avoidance globally, or honestly reframe the model as a heuristic relaxed objective rather than an exact reduction.","headline":"Genuinely new encoding and a real hardware demo, but the central reduction to a ground-state problem has a q=0 loophole that breaks self-avoidance, so it needs heavy revision or reframing.","tokens_in":32899,"tokens_out":2287,"would_cite":false,"duration_ms":26382,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper's central claim is that the folding problem reduces to finding the ground state of an O(N^4)-term qubit Hamiltonian, and that a variational quantum-classical optimizer can find that state for short peptides on NISQ hardware.","keywords":["protein folding","quantum Hamiltonian","tetrahedral lattice","variational quantum eigensolver","CVaR-VQE","NISQ","coarse-grained model","genetic algorithm"],"falsifier":"Enumerate all tetrahedral-lattice conformations of a short peptide, compute its exact minimal-energy self-avoiding fold, and compare it with the ground state of $H(q)$; any low-energy state containing a self-intersection far from a contact, or a repulsive contact with its interaction qubit set to zero, would show the claimed equivalence does not hold.","tokens_in":31766,"feed_emoji":"🧬","tokens_out":12458,"duration_ms":116655,"temperature":0.7,"pith_summary":"This paper tries to establish that protein folding on a tetrahedral lattice can be captured by a qubit Hamiltonian whose ground state is the minimum-energy fold, at a resource cost of O($N^{2}$) qubits and O($N^{4}$) Hamiltonian terms for N monomers. That cost matters because earlier lattice folding Hamiltonians grew much faster, sometimes exponentially, putting them out of reach of near-term quantum devices. The authors pair this Hamiltonian with a hybrid optimizer, CVaR-VQE steered by a genetic algorithm, and report two demonstrations: a noisy simulation of the 10-amino-acid peptide Angiotensin on 22 qubits, and a real-hardware run that folds a 7-amino-acid peptide on 9 qubits, which they describe as the largest variational folding calculation on a NISQ device to date. If the construction holds, it offers a concrete route for using quantum processors to optimize a classical NP-hard problem, not just to simulate quantum systems.","feed_headline":"Nine qubits fold a 7-amino-acid peptide on quantum hardware","feed_subtitle":"A variational algorithm maps folding to a Hamiltonian ground state, cutting resource needs to O(N^2) qubits.","key_machinery":"The machinery is the qubit Hamiltonian $H(q) = H_{gc} + H_{ch} + H_{in}$, defined over two registers. Configuration qubits encode the sequence of turns that grow the chain on the tetrahedral lattice; interaction qubits $q_{i,j}^{(l)}$ mark whether beads $i$ and $j$ form an $l$-th-nearest-neighbour contact. The interaction term $q_{i,j}^{(l)}(\\epsilon_{ij}^{(l)} + \\lambda(d(i,j)-l))$ gives the contact energy exactly when the distance equals $l$ and the contact qubit is 1, and otherwise applies a large penalty, so the ground state is pushed toward contact patterns with low total energy. Growth and chirality constraints are added as penalties, and the optimisation is done by CVaR-VQE, which minimises the tail of the energy distribution, with a differential-evolution genetic algorithm updating the circuit parameters.","core_discovery":"The paper's central claim is that the solution to the folding problem is the ground state of the Hamiltonian $H(q)$. The Hamiltonian is built from configuration qubits that encode the polymer's turns on a tetrahedral lattice and interaction qubits that certify pairwise contacts; energy is assigned only when a contact qubit is active and the geometric distance matches, with penalties for mismatches. The authors claim this construction scales as $\\mathcal{O}(N^2)$ qubits and $\\mathcal{O}(N^4)$ Pauli terms with locality independent of $N$, and they demonstrate it by folding a 7-amino-acid neuropeptide on 9 qubits of a 20-qubit superconducting processor, calling it the largest folding calculation on a NISQ device using a variational algorithm. They also simulate the folding of Angiotensin (10 amino acids) on 22 qubits under a realistic noise model.","pith_inferences":["The paper does not test whether the ground state of $H(q)$ coincides with the true self-avoiding minimum for arbitrary sequences, because overlaps are only penalised in the vicinity of an interaction pair; an exact enumeration for short chains would settle how often unpenalised self-intersections appear in low-energy states.","The same CVaR-VQE-plus-genetic-optimizer recipe applies to any diagonal Hamiltonian, so the demonstrated convergence suggests a generic method for classical cost-function optimisation on NISQ hardware, though the paper only shows protein folding.","The sparser 4-qubit-per-turn encoding is more faithful to the lattice but uses more qubits; the denser 2-qubit-per-turn encoding trades locality (5-local terms) for fewer qubits, and a systematic comparison of the two encodings on the same peptide would show which is more noise-resilient."],"forward_implications":["The $\\mathcal{O}(N^4)$ term count and $\\mathcal{O}(N^2)$ qubit count mean lattice protein folding avoids the exponential resource growth of earlier Hamiltonian encodings, provided the construction generalises beyond the demonstrated peptide sizes.","On real hardware the 7-amino-acid peptide folded on 9 qubits with the average ground-state probability over the population exceeding 20% and the best individual reaching 33%, which the authors present as evidence that noisy devices can run the algorithm.","In noisy simulations the 10-amino-acid Angiotensin system on 22 qubits converged so that at 1024 measurements 100% of the population produced low-energy conformations, and secondary-structure elements such as an alpha-helix and a beta-sheet could be selected by tuning the contact map.","Because the contact energies can be taken from pre-existing pairwise interaction tables, the same Hamiltonian can be re-parameterised for different amino-acid sequences without changing the qubit layout."],"supporting_citations":[{"why":"Earlier spin-Hamiltonian encoding of on-lattice protein folding; the resource-heavy baseline this work improves on.","marker":"20"},{"why":"Provides the original Hamiltonian formulation for heteropolymer folding on a lattice.","marker":"21"},{"why":"Coarse-grained lattice folding model with exponential qubit requirements; one of the cost baselines the paper compares against.","marker":"22"},{"why":"Resource-reduced polymer lattice model that still scales exponentially in qubits and gates; the immediate predecessor for this Hamiltonian.","marker":"23"},{"why":"Quantum-annealer folding experiments that required 81 and 200 qubits; the experimental qubit counts this model's 9-qubit run is measured against.","marker":"24"},{"why":"QAOA-based variational approach to lattice protein folding; the previous variational algorithm this work extends to a tetrahedral lattice.","marker":"26"},{"why":"Pairwise interaction energy table used to build the contact matrix for the simulated peptides.","marker":"27"},{"why":"Introduces the CVaR objective for variational quantum eigensolvers, which the folding algorithm uses as its cost function.","marker":"30"},{"why":"Differential evolution classical optimizer used to update the variational circuit parameters.","marker":"31"}],"fun_headline_variants":["9 qubits fold a 7-amino-acid peptide on IBM Q","Quantum folding with O(N^2) qubits demonstrated on NISQ","Hybrid quantum-genetic algorithm folds neuropeptide on 9 qubits","Resource-efficient quantum algorithm folds 7 amino acids on 9 qubits"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the Hamiltonian's ground state is the minimum-energy self-avoiding fold, but the penalty terms only prevent overlaps that occur near an interaction pair, and repulsive contacts can be ignored by setting the corresponding interaction qubit to zero, so self-intersections elsewhere are not truly forbidden.","fun_headline_variants_meta":{"raw":{"variants":["9 qubits fold a 7-amino-acid peptide on IBM Q","Quantum folding with O(N^2) qubits demonstrated on NISQ","Hybrid quantum-genetic algorithm folds neuropeptide on 9 qubits","Resource-efficient quantum algorithm folds 7 amino acids on 9 qubits"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00058,"raw_usage":{"total_tokens":2779,"prompt_tokens":1037,"completion_tokens":1742,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":653,"completion_tokens_details":{"reasoning_tokens":1661}},"tokens_in":653,"tokens_out":1742,"duration_ms":15919,"temperature":1.0,"reasoning_tokens":1661,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:53:30.925899+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Enumerate all tetrahedral-lattice conformations of a short peptide, compute its exact minimal-energy self-avoiding fold, and compare it with the ground state of $H(q)$; any low-energy state containing a self-intersection far from a contact, or a repulsive contact with its interaction qubit set to zero, would show the claimed equivalence does not hold.","supporting_citations":[{"cited_title":"Construction of Energy Functions for Lattice Heteropolymer Models: A Case Study in Constraint Satisfaction Programming and Adiabatic Quantum Optimization","cited_arxiv_id":"1211.3422","evidence_quote":"Earlier spin-Hamiltonian encoding of on-lattice protein folding; the resource-heavy baseline this work improves on."},{"cited_title":"Perdomo Ortiz , author C","cited_arxiv_id":null,"evidence_quote":"Provides the original Hamiltonian formulation for heteropolymer folding on a lattice."},{"cited_title":"Miyazawa \\ and\\ author R","cited_arxiv_id":null,"evidence_quote":"Pairwise interaction energy table used to build the contact matrix for the simulated peptides."}],"review_version":1}