Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

Designing lattice proteins with variational quantum algorithms

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read For lattice-protein sequence optimization, this paper shows that shallow problem-agnostic circuits outperform problem-informed QAOA circuits once device noise is included, because QAOA's circuit depth becomes prohibitive.

desk verdict A careful, reproducible QAOA-vs-HEA comparison on HP sequence optimization; the qualitative noise-tolerance ranking holds, but the hardware threshold N≤12 is one step beyond the data and the instance selection is narrow. read the letter →

arxiv 2508.02369 v1 pith:M2VXDZSM submitted 2025-08-04 quant-ph

Hanna Linn , Lucas Knuthson , Anders Irbäck , Sandipan Mohanty , Laura García-Álvarez , Göran Johansson This is my paper · ORCID
classification quant-ph
keywords latticeproteinsHPmodelproteindesignsequenceoptimizationvariationalquantumalgorithmsQAOAhardware-efficientansatzQUBO
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper asks whether variational quantum algorithms can carry out the sequence-optimization half of protein design: given a fixed target structure, find an amino-acid sequence that minimizes energy in that structure. Using the two-dimensional HP lattice model as a test bed, the authors compare problem-informed QAOA circuits with problem-agnostic hardware-efficient ansatz (HEA) circuits. In noiseless simulation, the QAOA variants find the known optimal sequences for chains up to N=16, but their success collapses when device noise is added because their circuits are very deep. The shallow HEA keeps higher success rates in noisy simulation, and the paper concludes it could be used on the Torino device for chain lengths up to N=12. The paper's central conclusion is that for this problem on today's hardware, circuit depth and noise tolerance matter more than building problem structure into the circuit.

What carries the argument

The carrying object is the QUBO reformulation of the HP contact energy. For a fixed target structure, the contact matrix $w_{ij}$ is known, so the only variables are the bead-type bits $s_i$; energy is a quadratic binary function with a Lagrange-penalty term $\lambda(\sum_i s_i - N_H)^2$ that fixes the number of hydrophobic beads. The paper contrasts two circuit families built on this cost: QAOA, whose phase separator and XY-mixers encode the problem and the composition constraint directly (with Dicke states preparing the feasible subspace), and HEA, which ignores problem structure and uses alternating parameterized $R_y/R_z$ rotations and CNOT entangling layers matched to the hardware topology. The metric that carries the comparison is the success rate—the fraction of runs that return the known unique optimal sequence, averaged over ten independent optimizations.

What would settle it

Run the same one-layer-HEA versus QAOA comparison on a randomized set of HP design instances that includes degenerately designable structures, multiple compositions per chain length, and targets without a known unique sequence; if QAOA matches or beats HEA on such instances, or if the N≈12 boundary shifts substantially, the paper's conclusion would not generalize. A second check is to repeat the Torino experiments with a noise model that includes temporal and correlated multi-qubit errors: if the simulation-hardware gap persists, the paper's stated explanation for the discrepancy is called into question.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central finding is a comparison: problem-informed QAOA variants are not usable for HP sequence optimization on noisy intermediate-scale hardware, whereas a minimal one-layer HEA is, at least for short chains. The authors encode each bead type in one qubit and minimize $E(s)=-\sum_{i<j} w_{ij}s_i s_j + \lambda(\sum_i s_i - N_H)^2$, with $s_i=1$ for H and $0$ for P. Five QAOA variants—standard X-mixer, fully connected and ring XY-mixers, each with basis or Dicke initial states—yield acceptable or high success rates in noiseless simulation, but all five drop sharply when a hardware-derived noise model is applied; the fully connected XY-mixer with Dicke states, the best performer, has circuit depth above 2000 at N=16. One- and two-layer HEAs, whose circuits are much shallower, achieve comparable noiseless success rates and substantially better noisy success rates, with one-layer HEA the most noise-tolerant. Hardware runs on the Torino device confirm the HEA approach works for N≤11, with the paper concluding it can be used for N≤12, while noting that simulated success rates overestimate hardware performance, an effect the authors attribute to temporal and correlated multi-qubit errors absent from the noise model.

Load-bearing premise

The load-bearing premise is that the chosen test instances—one target structure and one composition value per chain length, each with a unique known solution that folds correctly—are representative of protein sequence optimization; the irregular size dependence of the results shows that instance-specific effects may dominate, so the ordering of algorithms could change on typical or degenerate problems.

Editorial extensions

If this is right

  • One-layer HEA with warm-started parameters can be used to solve HP sequence optimization for short chains (N up to about 12) on current superconducting devices.
  • QAOA, including constraint-preserving XY-mixer variants, is not competitive on this problem until circuit depth is reduced or error mitigation becomes effective.
  • Hardware-tailored but problem-agnostic circuits can outperform problem-informed circuits under noise, so hardware compatibility is a key design criterion for NISQ-era variational optimization.
  • Success rates from existing noise models are optimistic: ignoring temporal and correlated multi-qubit errors leads to systematic overestimation of hardware performance.
  • Parameter donation across increasing chain lengths makes HEA optimization work for larger systems where random initialization fails.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Extension: Because the depth penalty comes from the fully connected quadratic form, the QAOA-versus-HEA ordering may change under alternative encodings, such as domain-wall or parity encodings, that reduce entangling-layer depth.
  • Extension: The success-rate metric, measured on uniquely solvable instances, is a good algorithm benchmark but not the criterion for real design, where near-optimal sequences that still fold to the target would be useful; a folding-aware evaluation could rank the algorithms differently.
  • Extension: The same tradeoff likely applies to other fixed-backbone design problems with pairwise contact energies, not only HP lattice proteins, because the resource bottleneck is the quadratic all-to-all structure rather than protein-specific detail.
  • Extension: The observed simulation-hardware gap makes hardware-aware noise models with temporal correlations, or noise-tailored variational training, a natural next test; if such a model closes the gap, it would support the paper's stated explanation.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. This paper investigates the use of variational quantum algorithms (VQAs) for the first step of lattice-protein design: finding HP sequences that minimize the energy in a fixed target structure. The authors formulate the problem as a QUBO (Eq. 2) and compare five QAOA variants (Table I) with hardware-efficient ansatz (HEA) circuits, using noiseless simulations, simulations with the IBM Torino noise model, and hardware experiments on Torino. Problem instances are taken from Appendix A; for each chain length N (4 to 28) one target structure and one NH are selected so that the sequence-optimization problem has a unique known solution that folds to the target. The main findings are that QAOA variants perform well only in noiseless simulations and are defeated by circuit-depth-induced noise, whereas one-layer HEA has better noise tolerance in simulation but shows a sharp drop on hardware for N ≥ 12. The paper concludes that HEA could be used to solve the sequence optimization problem on Torino for short chains (N ≤ 12).

Significance. If the results are taken with appropriate qualifications, the paper provides a useful empirical comparison of problem-informed and problem-agnostic VQAs on a biologically motivated combinatorial optimization problem. Its strengths include the use of instances with externally verified unique ground states, statistics over 10 runs with standard errors, the combination of noiseless, noisy, and hardware measurements, and the public availability of the code. The qualitative conclusion that QAOA's depth makes it less noise-tolerant than shallow HEA is well supported. However, the quantitative hardware claim and its generalization to 'short chains' are not fully supported by the single-instance data, as detailed in the major comments.

major comments (3)
  1. [Sec. V versus Sec. III B / Fig. 8a] The concluding claim that HEA 'could be used to solve the sequence optimization problem on IBM's Torino device for short chains (N ≤ 12)' is one step beyond the reported hardware data. The text in Sec. III B states that one-layer HEA success rates are significant or high for N ≤ 11 but tiny for N ≥ 12, and Fig. 8a shows a sharp decline already at N = 12. The stated boundary should be N ≤ 11, or additional N = 12 hardware data (preferably on more than one instance) should be supplied.
  2. [Appendix A and Sec. III] The generality of the central claim is not established because only one hand-selected instance is used per chain length. For each N, Appendix A fixes one target structure and one NH, chosen so that the instance has a unique solution and known folding; these are the most designable structures, while only about 2% of HP sequences have a unique ground state. The irregular N-dependence visible in Figs. 3 and 7 shows that instance-specific effects likely dominate the observed success-rate behavior. Either test multiple instances per N (including instances with degenerate energy landscapes) or explicitly restrict the conclusion to the specific instances studied.
  3. [Sec. V and Sec. III B] The phrase 'could be used to solve' is not quantified against any classical baseline on the same instances. Because the instances are small and have known unique solutions, a comparison with a standard classical optimizer on the identical instances is needed to establish that the measured success rates represent a useful capability rather than trivial success on selected easy instances.
minor comments (4)
  1. [Sec. III A] The sentence 'Second, for a given initial state, the fully connected XY-mixer gives the results.' is missing a word; it should presumably read 'gives the best results.'
  2. [Sec. V] 'VQAs o ffers a promising approach' should be 'VQAs offer a promising approach', and the ligature should be removed.
  3. [Sec. II D] The phrase 'both angles initialized toπ' should have a space before π ('initialized to π').
  4. [Sec. III B and Fig. 8] The hardware results in Fig. 8 are shown for N up to 26, while the text discusses simulations up to N = 28; please clarify the range of the hardware experiments in the main text.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: benchmarks are external exhaustive enumerations, and no prediction reduces to a fitted value.

full rationale

The paper's central claims are empirical success-rate measurements of QAOA and HEA variants against known ground states. Equation (2) is a standard QUBO formulation whose only hand-set constant, λ = 1.1, is declared and justified as robust once above a threshold, not fitted to the observed success rates. The ground-truth solutions and the guarantee that chosen instances have unique solutions that fold to the target structure are taken from exhaustive enumerations [33,34]. Although those references have author overlap with the present paper, they are parameter-free external results that supply the benchmark answers; they do not incorporate or depend on the VQA performance being reported, so the citation is not load-bearing in a circular way. Parameter donation is described as an initialization heuristic: optimized parameters from smaller instances seed larger ones, but the reported success rates are measured after optimizing on the target instance, so the measurement is not the fit itself. The paper's own caveats about irregular N-dependence, a single instance per chain length, and the overestimation of hardware success by noisy simulations are limitations on generality and model fidelity, not evidence that any claimed result is equivalent to its inputs by construction. No self-definitional step, fitted-input-as-prediction, or uniqueness-imported-from-authors pattern is present. Therefore the derivation chain is self-contained as an empirical benchmark study, and the appropriate circularity score is 0.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central claims rest on the HP model formulation, external enumeration ground truths, the QUBO encoding, and the assumed noise model. No new physical entities are introduced. The single free parameter λ is disclosed.

free parameters (1)
  • Lagrange multiplier λ = 1.1
    Penalty weight in Eq. (2) that biases the total number of H beads toward NH. Chosen by hand and stated to work for all instances; robustness to changes asserted but not systematically swept.
assumptions (4)
  • domain assumption The 2D HP model with contact energy EHP = -NHH is an adequate test bed for protein design.
    The entire study is framed within the Lau-Dill HP model (Sec. II B), so conclusions apply only to this simplified representation.
  • standard math Exhaustive enumerations in Refs. [33,34] correctly identify unique ground states and designability for N ≤ 30.
    Used to select instances with unique solutions and to verify folding (Appendix A). This is prior computational evidence taken as ground truth.
  • standard math The QUBO objective in Eq. (2) with the λ penalty is equivalent to minimizing EHP subject to the composition constraint NH.
    Standard conversion of constrained binary optimization to QUBO; the penalty term enforces the constraint in the limit of large λ.
  • domain assumption The IBM Torino noise model (gate errors, readout errors, thermal relaxation) captures the dominant error sources in noisy simulations.
    The paper states the model omits non-Markovian effects and crosstalk (Sec. II G), and the hardware results show it overestimates success rates for N ≥ 12, so this assumption is only partially valid.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Designing lattice proteins with variational quantum algorithms." pith.science (2026). https://pith.science/paper/M2VXDZSM

@misc{pith2026250802369,
  author       = {Pith},
  title        = {Pith review of: Designing lattice proteins with variational quantum algorithms},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/M2VXDZSM}},
  note         = {Machine review of arXiv:2508.02369}
}
read the original abstract

Quantum heuristics have shown promise in solving various optimization problems, including lattice protein folding. Equally relevant is the inverse problem, protein design, where one seeks sequences that fold to a given target structure. The latter problem is often split into two steps: (i) searching for sequences that minimize the energy in the target structure, and (ii) testing whether the generated sequences fold to the desired structure. Here, we investigate the utility of variational quantum algorithms for the first of these two steps on today's noisy intermediate-scale quantum devices. We focus on the sequence optimization task, which is less resource-demanding than folding computations. We test the quantum approximate optimization algorithm and variants of it, with problem-informed quantum circuits, as well as the hardware-efficient ansatz, with problem-agnostic quantum circuits. While the former algorithms yield acceptable results in noiseless simulations, their performance drops under noise. With the problem-agnostic circuits, which are more compatible with hardware constraints, an improved performance is observed in both noisy and noiseless simulations. However, the results deteriorate when running on a real quantum device. We attribute this discrepancy to features not captured by the simulated noise model, such as the temporal aspect of the hardware noise.

Figures

Figures reproduced from arXiv: 2508.02369 by the authors.

Figure 1
Figure 1. FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. shows the problem size dependence of the simu￾lated success probabilities. In the absence of noise (Fig. 3a), it is possible to obtain significant success probabilities with the XY-mixers even for the largest problem size, N = 16. With the X-mixer, the success probability stays <0.2 for N ≥ 12. The N-dependence is somewhat irregular, probably in part due to the fact that we consider a single instance for each N. How… view at source ↗
Figures from the paper (7 more)
Figure 6
Figure 6. Figure 6: FIG. 6 [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 7
Figure 7. Figure 7: FIG. 7 [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: FIG. 8 [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: FIG. 9 [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: FIG. 10 [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Penalty-free quantum optimization applied to lattice protein folding

    quant-ph 2026-06 unverdicted novelty 7.0 of 10

    A QAOA variant without quadratic penalties, using independent sets in a conflict graph, is applied to lattice protein folding and validated on proteins up to length 14 via simulation and heuristic search.

Reference graph

Works this paper leans on

51 extracted references · 43 canonical work pages · cited by 1 Pith paper

  1. [1]

    K. Wang, Z. Lu, C. Zhang, et al. , Demonstration of low-overhead quantum error correction codes (2025), arXiv:2505.09684 [quant-ph]

    arXiv 2025

  2. [2]

    Krinner, N

    S. Krinner, N. Lacroix, A. Remm, et al., Nature 605, 669–674 (2022)

    work page 2022

  3. [3]

    Acharya, I

    R. Acharya, I. Aleiner, R. Allen, et al. , Nature 614, 676–681 (2023)

    work page 2023

  4. [4]

    Acharya, D

    R. Acharya, D. A. Abanin, L. Aghababaie-Beni, et al., Nature 638, 920–926 (2024)

    work page 2024

  5. [5]

    Cerezo, A

    M. Cerezo, A. Arrasmith, R. Babbush, et al., Nat. Rev. Phys. 3, 625–644 (2021)

    work page 2021

  6. [6]

    Farhi, J

    E. Farhi, J. Goldstone, and S. Gutmann, arXiv:1411.4028 (2014)

    arXiv 2014

  7. [7]

    Hadfield, Z

    S. Hadfield, Z. Wang, B. O’Gorman, et al., Algorithms 12, 34 (2019)

    work page 2019

  8. [8]

    Kandala, A

    A. Kandala, A. Mezzacapo, K. Temme, et al., Nature 549, 242 (2017)

    work page 2017

Show all 51 references
  1. [9]

    Preskill, Quantum 2, 79 (2018)

    J. Preskill, Quantum 2, 79 (2018)

    2018

  2. [10]

    J. R. McClean, S. Boixo, V . N. Smelyanskiy, et al., Nat. Com- mun. 9, 4812 (2018)

    2018

  3. [11]

    Langfitt, J

    Q. Langfitt, J. Falla, I. Safro, and Y . Alexeev, in2023 IEEE In- ternational Conference on Quantum Computing and Engineer- ing (QCE), V ol. 02 (2023) pp. 300–301

    2023

  4. [12]

    Montañez-Barrera, D

    A. Montañez-Barrera, D. Willsch, A. Maldonado-Romo, and K. Michielsen, Quantum Sci. Technol. 9, 025022 (2024)

    2024

  5. [13]

    Galda, E

    A. Galda, E. Gupta, J. Falla, et al., Front. Quantum Sci. Tech- nol. 2, 1200975 (2023)

    2023

  6. [14]

    Lyngfelt and L

    I. Lyngfelt and L. García-Álvarez, Phys. Rev. A 111, 022418 (2025)

    2025

  7. [15]

    Kosen, H.-X

    S. Kosen, H.-X. Li, M. Rommel, et al., Quantum Sci. Technol. 7, 035018 (2022)

    2022

  8. [16]

    Fingerhuth, T

    M. Fingerhuth, T. Babej, and C. Ing, arXiv:1810.13411 (2018)

    2018 arXiv

  9. [17]

    Robert, P

    A. Robert, P. K. Barkoutsos, S. Woerner, and I. Tavernelli, Npj Quantum Inf. 7, 38 (2021)

    2021

  10. [18]

    Boulebnane, X

    S. Boulebnane, X. Lucas, A. Meyder, et al., Npj Quantum Inf. 9, 70 (2023)

    2023

  11. [19]

    Perdomo-Ortiz, N

    A. Perdomo-Ortiz, N. Dickson, M. Drew-Brook, et al. , Sci. Rep. 2, 248 (2012)

    2012

  12. [20]

    Outeiral, G

    C. Outeiral, G. M. Morris, J. Shi,et al., New J. Phys.23, 103030 (2021)

    2021

  13. [21]

    Irbäck, L

    A. Irbäck, L. Knuthson, S. Mohanty, and C. Peterson, Phys. Rev. Res. 4, 043013 (2022)

    2022

  14. [22]

    H. Linn, I. Brundin, L. García-Álvarez, and G. Johansson, Phys. Rev. Res. 6, 033112 (2024)

    2024

  15. [23]

    Kuhlman, G

    B. Kuhlman, G. Dantas, G. C. Ireton, et al., Science 302, 1364 (2003)

    2003

  16. [24]

    Bhardwaj, V

    G. Bhardwaj, V . K. Mulligan, C. D. Bahl, et al. , Nature 538, 329 (2016)

    2016

  17. [25]

    K. K. Yang, Z. Wu, and F. H. Arnold, Nat. Methods 16, 687 (2019)

    2019

  18. [26]

    Kuhlman and P

    B. Kuhlman and P. Bradley, Nat. Rev. Mol. Cell Biol. 20, 681 (2019)

    2019

  19. [27]

    L. Cao, I. Goreshnik, B. Coventry, et al. , Science 370, 426 (2020)

    2020

  20. [28]

    V . K. Mulligan, H. Melo, H. I. Merritt, et al. , bioRxiv:10.1101/752485 (2020)

    2020 doi

  21. [29]

    Irbäck, L

    A. Irbäck, L. Knuthson, S. Mohanty, and C. Peterson, Phys. Rev. Res. 6 (2024)

    2024

  22. [30]

    Panizza, P

    V . Panizza, P. Hauke, C. Micheletti, and P. Faccioli, PRX Life 2, 043012 (2024)

    2024

  23. [31]

    M. H. Khatami, U. C. Mendes, N. Wiebe, and P. M. Kim, PLOS Comput. Biol. 19, 1 (2023)

    2023

  24. [32]

    K. F. Lau and K. A. Dill, Macromolecules 22, 3986 (1989)

    1989

  25. [33]

    Irbäck and C

    A. Irbäck and C. Troein, J. Biol. Phys. 28, 1 (2002)

    2002

  26. [34]

    Holzgräfe, A

    C. Holzgräfe, A. Irbäck, and C. Troein, J. Chem. Phys. 135, 195101 (2011)

    2011

  27. [35]

    Irbäck, C

    A. Irbäck, C. Peterson, F. Potthast, and E. Sandelin, Structure 7, 347 (1999)

    1999

  28. [36]

    Aina and S

    A. Aina and S. Wallin, J. Chem. Phys. 147, 095102 (2017)

    2017

  29. [37]

    Nilsson and A

    D. Nilsson and A. Irbäck, Phys. Rev. E 101, 022413 (2020)

    2020

  30. [38]

    Statt, H

    A. Statt, H. Casademunt, C. P. Brangwynne, and A. Z. Pana- giotopoulos, J. Chem. Phys. 152, 075101 (2020)

    2020

  31. [39]

    Bornberg-Bauer and H

    E. Bornberg-Bauer and H. S. Chan, Proc. Natl. Acad. Sci. USA 96, 10689 (1999)

    1999

  32. [40]

    Aguirre, P

    J. Aguirre, P. Catalán, J. A. Cuesta, and S. Manrubia, Open Biol. 8, 180069 (2018)

    2018

  33. [41]

    Kadowaki and H

    T. Kadowaki and H. Nishimori, Phys. Rev. E 58, 5355 (1998)

    1998

  34. [42]

    Farhi, J

    E. Farhi, J. Goldstone, S. Gutmann, et al. , Science 292, 472 (2001)

    2001

  35. [43]

    Bärtschi and S

    A. Bärtschi and S. Eidenbenz, in Fundamentals of Computation Theory, edited by L. A. G˛ asieniec, J. Jansson, and C. Levcopou- los (Springer, Cham, 2019) pp. 126–139

    2019

  36. [44]

    Bittel and M

    L. Bittel and M. Kliesch, Phys. Rev. Lett. 127, 120502 (2021)

    2021

  37. [45]

    Zhou, S.-T

    L. Zhou, S.-T. Wang, S. Choi, et al., Phys. Rev. X 10, 021067 (2020)

    2020

  38. [46]

    J. A. Montañez-Barrera, D. Willsch, and K. Michielsen, Quan- tum Inf. Process. 24, 129 (2025)

    2025

  39. [47]

    C. R. Harris, K. J. Millman, S. J. van der Walt, et al. , Nature 585, 357 (2020)

    2020

  40. [48]

    Virtanen, R

    P. Virtanen, R. Gommers, T. E. Oliphant, et al., Nat. Methods 17, 261 (2020)

    2020

  41. [49]

    J. D. Hunter, Computing in Science & Engineering9, 90 (2007)

    2007

  42. [50]

    Javadi-Abhari, M

    A. Javadi-Abhari, M. Treinish, K. Krsulich, et al. , Quantum computing with Qiskit (2024), arXiv:2405.08810 [quant-ph]

    2024 arXiv

  43. [51]

    M. J. D. Powell, A direct search optimization method that mod- els the objective and constraint functions by linear interpo- lation, in Advances in Optimization and Numerical Analysis , edited by S. Gomez and J.-P. Hennart (Springer Netherlands, Dordrecht, 1994) pp. 51–67

    1994

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.