Alibaba Cloud Quantum Development Platform: Applications to Quantum Algorithm Design

Cupjin Huang; Fang Zhang; Jianxin Chen; Mario Szegedy; Xun Gao; Yaoyun Shi

REVIEW 4 major objections 5 minor 21 references

Alibaba Cloud Quantum Development Platform: Applications to Quantum Algorithm Design

T0 review · 4 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read A classical tensor-network engine evaluates QAOA energies for 1,000-qubit graphs.

desk verdict Solid QAOA simulation engineering with useful benchmark data, but the graph-isomorphism separation claim leans on an unproved conjecture and the abstract overstates scale. read the letter →

arxiv 1909.02559 v1 pith:AW3UE2G4 submitted 2019-09-05 quant-ph

Cupjin Huang , Mario Szegedy , Fang Zhang , Xun Gao , Jianxin Chen , Yaoyun Shi This is my paper · ORCID

classification quant-ph

keywords quantumapproximateoptimizationalgorithmtensornetworkcontractionclassicalsimulationofcircuitsgraphisomorphismsmall-cycle-freeregulargraphsMAX-CUTQAOAenergylandscapesimilarity

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 claims that a classical tensor-network simulator can evaluate the QAOA energy function for sparse graphs far beyond the reach of state-vector simulators: regular graphs with 1,000 vertices and up to depth 4 or 5. Because QAOA is a leading candidate for near-term quantum advantage, being able to test and optimize its angles classically at this scale turns it into a practically testable hybrid algorithm. The paper further claims that QAOA energies carry structural information about graphs: at depth one on 3-regular graphs the energy is determined by the number of triangles, and empirically the energy values separate all non-isomorphic 3-regular graphs up to 18 nodes, all strongly regular graphs up to 26 nodes, and hard-to-distinguish 20-node graph pairs. If these claims hold, QAOA energies are both a scalable classical benchmark and a promising graph invariant.

What carries the argument

The workhorse is the QAOA energy function $F_p(\vec\gamma,\vec\beta)$, the expectation of the cut Hamiltonian in the variational state. The platform evaluates each Hamiltonian term by contracting a tensor network built from the lightcone of that edge, exploiting the diagonality of the cost gates and the uniform-plus initial state to shrink the tensors. For small-cycle-free graphs, all edge lightcones are isomorphic to one tree whose size depends only on the degree $d$ and depth $p$, so the whole simulation collapses to a single template. For graph structure discovery, the machinery is the energy landscape: the vector of energies at several random angle sequences serves as a graph signature, and Theorem 1 identifies the depth-one invariant as the triangle count.

What would settle it

Run the energy-separation procedure on the 40-node hard graph pair described in the paper at depth 5 or higher over many random angle sequences; if their energies coincide for a non-negligible fraction of angle draws, the conjecture that QAOA energies separate all non-isomorphic graphs is false.

Watch

Extended reading notes

Core claim

On its own terms, the paper demonstrates that tensor-network contraction, applied one edge at a time within the lightcone of that edge, reduces QAOA energy evaluation to a set of tractable contractions, and that this reduction is powerful enough to handle regular graphs with 1,000 vertices at depth 4 (and degree 3 at depth 5), while remaining competitive with state-vector simulators at small sizes. The paper computes near-optimal angle sequences for small-cycle-free regular graphs, where the girth is at least 2p+2 and every edge's lightcone is the same tree, reporting the best values found for degrees 3 through 7. It then uses energy values at random angles as a graph signature: at depth 4 it separates all 4,060 non-isomorphic 3-regular graphs on 16 nodes and all 41,301 on 18 nodes, and at depth 3 it separates the strongly regular 26,10,3,4 family. The paper also states Theorem 1, attributed to a companion work, that depth-one QAOA energies of 3-regular graphs depend only on the degree sequence, the number of nodes, and the number of triangles; the observed clustering of the graph landscapes is decoded by triangle number.

Load-bearing premise

The graph-separation claims rest on an unproved conjecture: for any two different graphs, a randomly chosen set of QAOA angles will almost surely give different energy values; the paper checks only a finite list of graph families and angle draws.

Editorial extensions

If this is right

QAOA parameter optimization can be moved off the quantum device for sparse instances: the angle tables for small-cycle-free graphs can seed or replace the optimization loop on large random regular graphs.
The energy-separation results give a classical, circuit-free certificate of non-isomorphism for entire families of graphs, including 41,301 3-regular graphs on 18 nodes.
Because depth-one energies of 3-regular graphs are a function of triangle count, graphs with different triangle counts are immediately separated at depth one, making triangle count a cheap subclassifier.
Benchmarking at 1,000 vertices gives future QAOA simulators and hardware demonstrations a concrete scale target: single energy queries at depth 4 take tens of seconds on this platform.
The random-walk experiment suggests that QAOA energy separation grows with graph distance, so the landscape can be used as a rough dissimilarity measure for slowly changing graph sequences.

Reading between the lines

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

The authors do not test whether the small-cycle-free angle tables remain near-optimal on graphs with short cycles; a direct test would compare their performance against fresh optimization on random regular graphs of the same degree.
If QAOA energies form a hierarchy of local-subgraph counts, then depth $p$ should distinguish graphs differing in counts of small subgraphs of size roughly $p$; comparing energy values against explicit subgraph counts on random graphs would test this.
The clustering observed in 3-regular graph landscapes suggests that energy vectors could group graphs by structural similarity, not just decide isomorphism; the random-walk result points toward a metric on graph space that a clustering algorithm could consume.
Because the 40-node hard pair only fails at depth 4, it sets a concrete stress test: a simulator reaching depth 5 on that pair would resolve whether the separation phenomenon extends there.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

The paper is an engineering report with a real kernel of new results. The tensor-network QAOA simulator is benchmarked against Cirq, Qiskit, and qTorch on regular graphs up to 1000 vertices, which is a genuinely useful data point for anyone who needs classical QAOA energy evaluations. The small-cycle-free angle tables for degrees 3–7 are also practical, even if they are clearly heuristic rather than certified optima. The graph-separation experiments are the most interesting part, and the empirical separation of all 3-regular graphs up to 18 nodes is a striking demonstration if it holds up.

The soft spots are real but not fatal. The abstract says 'thousands of qubits' while the data show 1000 vertices; that is an overstatement, but a small one, and the rest of the text is more careful. Calling the angles in Table 2 'optimal' is loose—they are the best found values, with no optimality certificate. The running times in Table 1 have no error bars, which matters because Figure 1 shows high variance in some regimes.

The larger concern is the graph-structure claim. Conjecture 1, cited to the unpublished [17], says that non-isomorphic graphs give different QAOA energies with probability 1 over angle sequences. As the stress-test note points out, this conflicts with Theorem 1 (also from [17]): for 3-regular graphs at depth 1, energies depend only on triangle count, so two such graphs with the same triangle count have equal energies for every angle sequence. The conjecture might be intended only for sufficiently large depth, but the paper does not say that. The exhaustive separation sweep also lacks the sampling protocol, number of angle draws, and tolerances, so the claim is not independently reproducible as reported. These are fixable but need to be addressed.

I came away convinced the core benchmark is honest: they checked against external simulators, so the platform's capability is not circular. The paper is a reasonable demonstration of a practical tool, but the flashy conclusion about graph isomorphism should be read with caution until [17] appears and the numerics are documented.

Who is this for? Someone working on QAOA simulation, variational quantum algorithms, or graph-similarity heuristics will get value from the benchmark data and the angle tables. It is worth a serious referee pass, but not as-is; the abstract, the 'optimal' language, and the Conjecture 1/Theorem 1 tension all need revision. If I were an editor, I would send it to review, with the expectation that the authors tighten the claims and make the separation experiments reproducible.

Referee Report

4 major / 5 minor

Summary. The paper describes the Alibaba Cloud Quantum Development Platform (AC-QDP), a tensor-network-based classical simulator for evaluating QAOA energy functions. It benchmarks single energy-query times against Cirq, Qiskit, and qTorch on random regular graphs, reports optimization results for small-cycle-free regular graphs using several heuristic optimizers, and proposes using QAOA energies as a graph invariant for distinguishing non-isomorphic graphs. The main empirical claims are that AC-QDP can evaluate 1000-vertex regular graphs up to depth 5, that the platform found good (titled 'best') angle sequences for tree-like regular graphs, and that for all tested small graphs QAOA energies at modest depth separate non-isomorphic graphs in several families. The theoretical grounding for the graph-separation claims is cited to an unpublished companion paper [17].

Significance. If the benchmark numbers are reproducible, AC-QDP is a useful practical tool for QAOA studies at sizes beyond state-vector simulation; the comparison against three external simulators is an appropriate and non-circular evaluation. The graph-structure discovery results are potentially interesting, but as presented they rest on an unproved conjecture and on finite numerical sweeps, so the manuscript's strongest conclusion ('QAOA energies give a full characterization of isomorphism classes') is not established. The paper would be strengthened by clearly separating engineering results from conjectural statements and by supplying the missing experimental and theoretical support.

major comments (4)

[Abstract and Section 3.2] The abstract states that AC-QDP can 'simulate thousands of qubits for up to 4 layers', but the benchmark data in Table 1 report only n=1000 vertices, and for d=4 and d=5 the depth does not even reach p=4. If the 'thousands' claim is supported by additional experiments or by the earlier papers [8,20], it should be stated explicitly; otherwise the abstract should be corrected. In addition, Fig. 1 and Table 1 report average query times without error bars, standard deviations, or per-instance variation, although the text notes large variation for n=30 and n=50; please include the spread or the raw data so the comparison is reproducible.
[Section 5, Conjecture 1 and Theorem 1] As stated, Conjecture 1 conflicts with Theorem 1 of the same section. For 3-regular graphs, Theorem 1 says that depth-one QAOA values depend only on n and the number of triangles, so any two non-isomorphic 3-regular graphs with the same triangle count have identical depth-one energies for every angle sequence, making the probability of a difference 0, not 1. If the conjecture is meant only for sufficiently large p, or for generic (not all) degree sequences, that qualification is missing. Since this conjecture is the only theoretical support for the graph-separation algorithm, it must be stated precisely and proved, or explicitly labeled as an open problem, before the separation claims can be accepted.
[Section 4, Tables 2 and 3] The paper calls the angle sequences in Table 3 'best' and says the platform 'solved' the small-cycle-free cases, but Table 2's own heading says 'Best function values found', and the methods used (dlib, differential evolution, FOURIER, grid search plus local refinement) are heuristics with no optimality certificate. The claim of optimal QAOA sequences is therefore not established. Moreover, the sentence 'However, with the increase of the number of angles, optimization .' at the end of Section 4 is incomplete; it appears to acknowledge a limitation and must be completed, since an unfinished sentence cannot be assessed.
[Section 5, graph separation protocol] The exhaustive-separation claims (the rows 'All 4060 non-iso 3-regular graphs' and 'All 41301 non-iso 3-regular graphs on 18 nodes') are reported without the sampling protocol: the number of random angle draws per graph pair, the numerical tolerance for declaring two energies different, the floating-point precision used, and the specific depth at which each family is separated. If some non-isomorphic pairs have energy gaps at or below machine precision, the separation could be an artifact of rounding rather than a property of QAOA. Please specify the protocol, ideally with arbitrary-precision checks for near-degenerate pairs, and clarify that the 4060 row refers to 16-vertex graphs, which is not stated in the table. The broader statement that QAOA energies give 'a full characterization of isomorphism classes' goes beyond the evidence, since Conjecture 1 is unproved and the tests cover only finitely many small graphs.

minor comments (5)

[Throughout] There are several typographical errors, including 'Altough' (Section 3.1), 'whre' (Section 3.1), 'tensot entwork' (Section 3.1), 'chanllenging' (Sections 3.1 and 4), and 'supressed' (Section 3.2).
[Figure 1 caption] The caption says 'For the cased = 2' but the surrounding text says the figure shows degree 3 cases; this is likely a typo and should be corrected.
[References] Theorem 1 and Conjecture 1 are attributed to [17], which is described as 'in preparation'. Relying on an unpublished manuscript for the central theoretical statements makes the paper difficult to verify; either include the proofs in an appendix or state the results as conjectures with the evidence provided.
[Table 3] The angle sequences are reported to only four decimal places. Since the optimization landscape can be flat or have narrow features, please either provide more digits in an appendix or state that the full-precision sequences are available from the platform.
[Section 5, random walk] The random-walk observations in Fig. 3 are qualitative; no error bars, number of walk steps, or statistical measures of 'distance' versus 'QAOA separation' are given. This section should be framed as an exploratory observation.

Circularity Check

1 steps flagged · score 4.0 of 10

Graph-structure use case leans on an unproved conjecture from the authors' own in-preparation paper; platform benchmarks are externally validated and non-circular.

self citation load bearing [Section 5, Conjecture 1 and following paragraph (pages 6-7)]
"Conjecture 1 ([17]). With probability 1 over all degree sequences in [0, 2π]2d the above algorithm gives different energy values when G1 and G2 on n nodes are non-isomorphic. ... These findings and the theoretical results in [17] make us believe that QAOA energies give a full characterization of isomorphism classes, unlike many quantum walk based distinguishers that were considered earlier."

The paper's general graph-separation claim is not derived in this article; its only theoretical support is Conjecture 1, attributed to [17], an in-preparation paper by coauthor Mario Szegedy. The empirical separations are direct computations and are independent, but the claim that random degree sequences separate every non-isomorphic pair - the justification for treating QAOA energies as a full isomorphism-class invariant - rests on that self-citation rather than on a proof or external check presented here. Thus a load-bearing premise for the Section 5 use case is an unverified result from the same group, giving the graph-structure narrative a partially circular evidential structure.

full rationale

The platform-capability claim is externally anchored: Section 3.2 benchmarks AC-QDP against Cirq, Qiskit, and qTorch on identical random regular-graph instances, so the central simulator comparison is not circular. The small-cycle-free optimization section explicitly reports heuristic 'best function values found' and does not rename a fit as a prediction. Section 5's numerical separations are computed by the platform and are not constructed from the conjecture. The circularity concern is confined to the theoretical framing: Conjecture 1 and Theorem 1 are imported from [17], an in-preparation paper by the same group, and the paper uses those results to support the 'full characterization' belief. That is self-citation in a load-bearing position, though the central empirical content remains independent. The apparent tension between Conjecture 1 and Theorem 1 is a correctness or consistency risk, not an additional circularity. Hence score 4.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The paper introduces no new physical entities or free parameters. The central numerical claims rest on standard QAOA definitions, the tree-lightcone reduction, and two results (Theorem 1 and Conjecture 1) cited to an unpublished companion paper [17] by a coauthor, which is the main unverified load-bearing support.

assumptions (5)

domain assumption QAOA energy function F_p(gamma, beta) = <gamma,beta| C_hat |gamma,beta> is the correct performance measure for QAOA.
Used throughout to define optimization objective and graph invariants. Standard in QAOA literature (Farhi et al. [10]).
domain assumption For a graph with girth at least 2p+2, the QAOA lightcone of any edge is a tree and thus all such graphs of the same degree have identical F_p.
Section 4 basis for reducing simulation to a single tree-like subgraph. This is a mathematical consequence of girth, relying on no cycles within distance p.
domain assumption Theorem 1 ([17]): for 3-regular graphs, depth-one QAOA CUT energy depends only on degree sequence, number of nodes, and number of triangles.
Stated in Section 5 as Theorem 1 but proof is delegated to an unpublished companion paper [17] by a coauthor; this paper uses it to explain the observed clusters.
domain assumption Conjecture 1 ([17]): with probability 1 over random degree sequences, QAOA energies distinguish any two non-isomorphic graphs.
Section 5 lists this as Conjecture 1 and relies on it to claim that QAOA energies give a full characterization of isomorphism classes; the paper does not prove it.
domain assumption Uniform random angle sequences are sampled from [0,2*pi]^(2p) for graph separation tests.
Section 5 algorithm step 1; the empirical separation results depend on the randomness and number of draws, which are not fully specified.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Alibaba Cloud Quantum Development Platform: Applications to Quantum Algorithm Design." pith.science (2026). https://pith.science/paper/AW3UE2G4

@misc{pith2026190902559,
  author       = {Pith},
  title        = {Pith review of: Alibaba Cloud Quantum Development Platform: Applications to Quantum Algorithm Design},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AW3UE2G4}},
  note         = {Machine review of arXiv:1909.02559}
}

abstract

We report our work on the Alibaba Cloud Quantum Development Platform (AC-QDP). The capability of AC-QDP's computational engine was already reported in \cite{CZH+18, ZHN+19}.In this follow-up article, we demonstrate with figures how AC-QDP helps in testing large-scale quantum algorithms (currently within the QAOA framework). We give new benchmark results on regular graphs. AC-QDP's QAOA framework can simulate thousands of qubits for up to $4$ layers. Then we discuss two interesting use cases we have implemented on the platform: 1. Optimal QAOA sequences for small-cycle free graphs; 2. Graph structure discovery.

Figures

Figures reproduced from arXiv: 1909.02559 by the authors.

**Figure 2.** Different Landscapes of all 3-regular graphs on 16 and 18 nodes. Plots (b) and (c) are [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 3.** Random walks on the set of d-regular graphs map to walks on the corresponding landscapes. When we are at a graph G, a step of the walk constitutes of randomly picking two edges with four different end points and replacing the edge pair randomly with one of the two other independent edge pairs on the same four nodes. If the move cannot be made, a new random edge pair of G is picked. exploited in computational geneti… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 16 canonical work pages

[17]

Graph structure discovery with QAOA

Mario Szegedy. Graph structure discovery with QAOA. in preparation, 2019

work page 2019
[1]

Cirq: A python framework for creating, editing, and invoking Noisy Intermediate Scale Quantum (NISQ) circuits, 2019

work page 2019
[2]

Qiskit: An open-source framework for quantum computing, 2019

Héctor Abraham et al. Qiskit: An open-source framework for quantum computing, 2019

work page 2019
[3]

Graph isomorphism in quasipolynomial time [extended abstract]

László Babai. Graph isomorphism in quasipolynomial time [extended abstract]. In Proceed- ings of the 48th Annual ACM SIGACT Symposium on Theory of Computing, STOC 2016, Cam- bridge, MA, USA, June 18-21, 2016 , pages 684–697, 2016

work page 2016
[4]

László Babai, Paul Erdös, and Stanley M. Selkow. Random graph isomorphism. SIAM J. Comput., 9(3):628–635, 1980

work page 1980
[5]

Local approximation of the Maximum Cut in regular graphs

Étienne Bamas and Louis Esperet. Local approximation of the maximum cut in regular graphs. arXiv preprint arXiv:1902.04899, 2019. 9

work page Pith review arXiv 1902
[6]

Boixo, S

S. Boixo, S. V . Isakov, V . N. Smelyanskiy, and H. Neven. Simulation of low-depth quantum circuits as complex undirected graphical models. ArXiv e-prints, December 2017

work page 2017
[7]

Graph iso- morphism and gaussian boson sampling, 2018

Kamil Bradler, Shmuel Friedland, Josh Izaac, Nathan Killoran, and Daiqin Su. Graph iso- morphism and gaussian boson sampling, 2018

work page 2018

Show all 21 references

[8]

Classical simulation of intermediate-size quantum circuits

Jianxin Chen, Fang Zhang, Cupjin Huang, Michael Newman, and Yaoyun Shi. Classical simulation of intermediate-size quantum circuits. arXiv preprint arXiv:1805.01450, 2018

2018 arXiv
[9]

Wilson, and Edwin R

David Emms, Simone Severini, Richard C. Wilson, and Edwin R. Hancock. Coined quantum walks lift the cospectrality of graphs and trees. Pattern Recognition, 42(9):1988–2002, 2009

1988
[10]

A quantum approximate optimization algorithm

Edward Farhi, Jeffrey Goldstone, and Sam Gutmann. A quantum approximate optimization algorithm. arXiv preprint arXiv:1411.4028, 2014

2014 arXiv
[11]

Classical and quantum bounded depth approximation algorithms

Matthew B Hastings. Classical and quantum bounded depth approximation algorithms. arXiv preprint arXiv:1905.07047, 2019

1905 arXiv
[12]

Eugene M. Luks. Isomorphism of graphs of bounded valence can be tested in polynomial time. J. Comput. Syst. Sci. , 25(1):42–65, 1982

1982
[13]

Phase-modiﬁed CTQW unable to distinguish strongly regular graphs efﬁciently

A Mahasinghe, J A Izaac, J B Wang, and J K Wijerathna. Phase-modiﬁed CTQW unable to distinguish strongly regular graphs efﬁciently. Journal of Physics A: Mathematical and Theoret- ical, 48(26):265301, jun 2015

2015
[14]

Markov and Yaoyun Shi

Igor L. Markov and Yaoyun Shi. Simulating quantum computation by contracting tensor networks. SIAM Journal on Computing , 38(3):963–981, 2008

2008
[15]

Schuyler Fried, N

E. Schuyler Fried, N. P . D. Sawaya, Y. Cao, I. D. Kivlichan, J. Romero, and A. Aspuru-Guzik. qTorch: The Quantum Tensor Contraction Handler. ArXiv e-prints, September 2017

2017
[16]

Training the Quantum Approximate Optimization Algo- rithm without access to a quantum processing unit

Michael Streif and Martin Leib. Training the Quantum Approximate Optimization Algo- rithm without access to a quantum processing unit. arXiv preprint arXiv:1908.08862, 2019

1908 arXiv
[18]

H. Wang, J. Wu, X. Yang, and X. Yi. A graph isomorphism algorithm using signatures computed via quantum walk search model. Journal of Physics A Mathematical General , 48(11):115302, March 2015

2015
[19]

Quantum approximate optimization algorithm for MaxCut: A fermionic view

Zhihui Wang, Stuart Hadﬁeld, Zhang Jiang, and Eleanor G Rieffel. Quantum approximate optimization algorithm for MaxCut: A fermionic view. Physical Review A, 97(2):022304, 2018

2018
[20]

Alibaba Cloud Quantum Development Kit: Large-Scale Classical Simulation of Quantum Circuits

Fang Zhang, Cupjin Huang, Michael Newman, Junjie Cai, Huanjun Yu, Zhengxiong Tian, Bo Yuan, Haihong Xu, Junyin Wu, Xun Gao, et al. Alibaba Cloud Quantum Development Kit: Large-Scale Classical Simulation of Quantum Circuits. arXiv preprint arXiv:1907.11217 , 2019

1907 arXiv
[21]

Quantum Approximate Optimization Algorithm: performance, mechanism, and implementation on near-term devices

Leo Zhou, Sheng-Tao Wang, Soonwon Choi, Hannes Pichler, and Mikhail D Lukin. Quantum Approximate Optimization Algorithm: performance, mechanism, and implementation on near-term devices. arXiv preprint arXiv:1812.01041, 2018. 10

2018 arXiv

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [17]

Graph structure discovery with QAOA

Mario Szegedy. Graph structure discovery with QAOA. in preparation, 2019

work page 2019

[2] [1]

Cirq: A python framework for creating, editing, and invoking Noisy Intermediate Scale Quantum (NISQ) circuits, 2019

work page 2019

[3] [2]

Qiskit: An open-source framework for quantum computing, 2019

Héctor Abraham et al. Qiskit: An open-source framework for quantum computing, 2019

work page 2019

[4] [3]

Graph isomorphism in quasipolynomial time [extended abstract]

László Babai. Graph isomorphism in quasipolynomial time [extended abstract]. In Proceed- ings of the 48th Annual ACM SIGACT Symposium on Theory of Computing, STOC 2016, Cam- bridge, MA, USA, June 18-21, 2016 , pages 684–697, 2016

work page 2016

[5] [4]

László Babai, Paul Erdös, and Stanley M. Selkow. Random graph isomorphism. SIAM J. Comput., 9(3):628–635, 1980

work page 1980

[6] [5]

Local approximation of the Maximum Cut in regular graphs

Étienne Bamas and Louis Esperet. Local approximation of the maximum cut in regular graphs. arXiv preprint arXiv:1902.04899, 2019. 9

work page Pith review arXiv 1902

[7] [6]

Boixo, S

S. Boixo, S. V . Isakov, V . N. Smelyanskiy, and H. Neven. Simulation of low-depth quantum circuits as complex undirected graphical models. ArXiv e-prints, December 2017

work page 2017

[8] [7]

Graph iso- morphism and gaussian boson sampling, 2018

Kamil Bradler, Shmuel Friedland, Josh Izaac, Nathan Killoran, and Daiqin Su. Graph iso- morphism and gaussian boson sampling, 2018

work page 2018