Pith. sign in

REVIEW 3 major objections 5 minor 2 cited by

Experiments with an oscillator based Ising machine

T0 review · 3 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read An eight-oscillator analog Ising machine solves small max-cut graphs by phase locking.

desk verdict An honest but under-evidenced report on a small analog oscillator Ising machine; worth refereeing as a proof-of-concept, but the central 'consistently accurate' claim needs quantitative backing. read the letter →

arxiv 2502.03167 v2 pith:MJZNYC4H submitted 2025-02-05 cs.ET nlin.AO

Shrish Roy , Bernd Ulmann This is my paper · ORCID
classification cs.ETnlin.AO
keywords Isingmachineoscillatornetworkmax-cutantiphasesynchronizationSHILanalogcomputingphase-shiftcombinatorialoptimization
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 reports the construction and testing of an analog Ising machine built from eight phase-shift oscillators wired through a configurable matrix of digital potentiometers. The authors' aim is to show that a non-quantum, off-the-shelf oscillator network can solve max-cut problems by letting the oscillators synchronize into binary phases that encode the partition of the graph. They find that when a second-harmonic injection-locking (SHIL) signal is applied at twice the oscillator frequency, the phases reliably binarize and the machine consistently produced accurate solutions for the small max-cut graphs they designed. The result matters because it suggests a cheap, parallel, analog route to combinatorial optimization that avoids both digital simulation cost and quantum hardware. The paper is explicit that the tested problems were author-designed and that coupling strengths had to be tuned per problem, with odd cycles sometimes landing in different partitions on repeated runs.

What carries the argument

The carrying mechanism is a network of eight phase-shift oscillators, each an operational amplifier with three RC stages, coupled through a zero-diagonal 8 by 8 matrix of 10-bit digital potentiometers whose resistances encode the graph's adjacency matrix. The computation is the synchronization dynamics: oscillators connected by an edge tend to settle $\pi$ out of phase, unconnected ones in phase, following Kuramoto-type phase coupling. A second-harmonic injection-locking (SHIL) signal, an external signal at twice the resonance frequency, is added to every oscillator's synchronization input to force the continuous phases to binarize into $0$ or $\pi$, making the final state readable as a graph partition. The phase detectors multiply each oscillator's output against the reference and integrate, giving a clean binary readout of the solution.

What would settle it

Run the machine on an 8-node max-cut graph with a known unique optimum and an odd cycle, using a fixed coupling strength and SHIL setting, for many repeated runs; if the runs give different partitions or a cut below the known optimum, the convergence-to-ground-state claim is refuted. A benchmark version of the same test is to compare the machine's outputs against the known optimal values of the G-set max-cut instances, which the paper has not yet done.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that an oscillator-based Ising machine can be assembled from standard analog computer modules and purpose-built phase-shift oscillators, and that this machine solves small max-cut instances by antiphase synchronization. Each node of the graph is an oscillator; a cut edge is represented by two oscillators settling into opposite phases, $0$ and $\pi$, an uncut edge by in-phase synchrony, and the assignment of phases to the two sides of the partition is read out directly. The mechanism that makes this reliable is an external SHIL signal at twice the oscillator resonance frequency, which forces the phases to binarize; with such a signal at 2.5 V and 6.8 kHz, using sinusoidal, rectangular, or pulse waveforms, the machine 'consistently produced accurate solutions' for the problems tested. The authors also report that coupling strength must be chosen per problem, that odd closed loops can produce different partitions on repeated runs, and that isomorphic graphs give identical results under SHIL.

Load-bearing premise

The load-bearing assumption is that the network settles into the true ground state of the Ising Hamiltonian, the actual max-cut, whenever coupling weights and SHIL parameters are set, rather than into some local minimum; this is not proven, and the paper itself notes that odd closed loops often produce different partitions on repeated runs.

Editorial extensions

If this is right

  • A max-cut graph maps directly to the hardware: the adjacency matrix sets the resistive couplings, and the converged binary phases read out as the partition.
  • With a SHIL signal at twice the oscillator frequency, the machine consistently produced accurate solutions for the author-designed max-cut graphs tested, with no digital search over partitions.
  • Different graphs require different coupling strengths, with values such as 1/10 or 1/5 of full scale working for most, so the hardware is not yet parameter-free.
  • Graphs with odd closed loops can synchronize into different partitions on repeated runs, so the dynamics do not always settle into a unique ground state.
  • Driving the SHIL at three times the oscillator frequency makes the phases ternarize, which the paper suggests could allow states \{-1,0,1\} and greater expressive power.

Reading between the lines

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

  • I infer that the reported accuracy is conditional on per-instance parameter selection, since the authors say each problem needed its own coupling strength and SHIL choice; without a fixed setting that works across a benchmark family, the machine is not yet a drop-in solver.
  • I infer that the odd-loop instability is the main scaling risk: any graph with an odd cycle has competing phase assignments, and if repeated runs sample among them, a single run's answer needs independent verification.
  • The paper has not yet run the G-set benchmarks it names; I infer that a passing result on those graphs without per-graph tuning would be the decisive test of the central claim.
  • Because the 10-bit weights were only lightly exercised, I infer that much coarser couplings, possibly ternary or single-bit weights, may suffice, which would simplify the interconnect for larger machines.
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.

Referee Report

3 major / 5 minor

Summary. The paper describes an experimental oscillator-based Ising machine built from off-the-shelf analog components, starting with a four-oscillator prototype and expanding to an eight-oscillator system. The hardware uses phase-shift oscillators coupled through a matrix of digital potentiometers, with second-harmonic injection locking (SHIL) to binarize oscillator phases into 0 or π. The authors map max-cut problems onto the coupling network, with antiphase synchronization representing the partition, and report that the eight-oscillator machine 'consistently produced accurate solutions' for the small max-cut instances tested. The paper also discusses synchronization theory, the Kuramoto model, and a list of open research questions.

Significance. If the central experimental claim were properly supported, this work would provide a low-cost, accessible hardware platform for exploring oscillator-based Ising machines, with useful implementation details such as oscillator schematics, phase-detector circuits, and a description of the hybrid analog-digital control interface. The paper is honest about several limitations, including the need for per-problem coupling-strength tuning and the observation of run-to-run variability on odd-loop graphs. However, as submitted, the experimental evidence is qualitative and underreported: no quantitative results are given, no repeated-run statistics are provided, and the raw data are only available upon request. The claimed demonstration of solving max-cut problems is therefore not verifiable from the manuscript alone, which is a serious shortcoming for a paper whose contribution is an experimental demonstration.

major comments (3)
  1. [Section 10] The central experimental claim is not supported by quantitative evidence. The section states that the configuration 'consistently produced accurate solutions' but provides no list of test graphs, no cut sizes or energies, no number of runs, no success rates, and no error bars. The term 'consistently' is undefined and cannot be checked by a reader. The statement that 'raw data and results ... will be provided by the authors upon request' does not meet the standard of a reproducible experimental report, and it makes the main finding unfalsifiable from the paper as written.
  2. [Section 10] The coupling strength c is chosen individually per problem, as stated: 'the actual value c has to be chosen individually for every problem to achieve best results.' Since the correct max-cut is evidently known when this choice is made, the procedure is vulnerable to circularity: the experimenter can tune c until the readout matches the known optimum. The paper gives no independent protocol for setting c (for example, a fixed rule based on graph statistics) and does not quantify how sensitive the outcome is to c. This undermines the claim that the machine solves max-cut problems rather than being fitted to the known answers.
  3. [Section 10] The paper acknowledges run-to-run variability, stating that closed odd loops lead to systems 'partitioning into different configurations upon repeated runs,' yet it reports no statistics over repeated measurements. It is therefore unclear whether the 'accurate solutions' represent typical behavior, best-case behavior, or manually selected runs. Without repeated-run statistics, the claim of 'consistently produced accurate solutions' is not falsifiable. This is a load-bearing issue because the existence of incorrect configurations shows that the dynamics do not always reach the ground state, and the paper does not quantify the frequency or conditions of such failures.
minor comments (5)
  1. [Section 9] The oscillator resonance frequency is stated as about 3.8 kHz, but the SHIL signal in Section 10 is reported at 6.8 kHz. Since the second harmonic of a 3.8 kHz signal would be 7.6 kHz, the relationship between the oscillator frequency and the SHIL frequency should be clarified or reconciled.
  2. [Section 8] The mapping between edge weights and coupling coefficients is given as J = -μij, but the sign convention and its connection to the Ising Hamiltonian in Eq. (1) are not explained. A brief derivation or reference would help the reader understand how the physical coupling realizes the max-cut objective.
  3. [Section 9] The number of coupling weights is described as 'n × n − n,' which is mathematically correct but would be clearer as n(n-1). Additionally, the text should clarify whether the potentiometer matrix includes the diagonal (with zero values) as implied by Figure 6.
  4. [Section 3] The discussion of complexity classes (P, NP, NP-hard) is textbook material that is not specific to the present contribution and could be condensed substantially to keep the paper focused on the experimental results.
  5. [General] The references include non-archival or informal sources (for example, Marandi [2019] is a YouTube video and the Stanford G-set is cited with an access date). For a journal submission, please ensure that all key claims about prior Ising-machine work are supported by peer-reviewed references.

Circularity Check

2 steps flagged · score 6.0 of 10

The headline success claim reduces in part to per-problem tuning of coupling strength and SHIL parameters, so the machine's 'accurate solutions' are not demonstrated with fixed, pre-specified inputs.

  1. fitted input called prediction [Section 10, paragraph on coupling strengths]
    "The analysis of results obtained by our machine show that different problems require varying coupling strengths to enhance the likelihood of producing the correct output. Even if all coupling weights are equal values c (which is the case for all problems considered until now), the actual value c has to be chosen individually for every problem to achieve best results, although values like c = 1/10 or c = 1/5 tend to be good candidates for most problem graphs."

    The paper's central claim is that the eight-oscillator machine 'consistently produced accurate solutions' for max-cut. But the accuracy is judged against the known max-cut partition, and the coupling strength c is chosen per problem 'to achieve best results.' Since c is selected after seeing which outputs are correct, the reported solutions are the product of a per-instance parameter search rather than a fixed mapping from graph to solution. The paper gives no pre-registered protocol for choosing c and no raw data or success rates, so the demonstration does not establish that the machine solves max-cut independently of the known answer; it shows only that with a fitted c the hardware can reproduce the desired partitions.

  2. fitted input called prediction [Section 10, 'Effect of SHIL' bullet]
    "The experiments were carried out with and without the presence of a second harmonic injection line (SHIL) at 2.5 V and 6.8 kHz, using sinusiodal, rectangular, and pulse signals. This configuration consistently produced accurate solutions and represents the 'Goldilocks zone' for most of the problems tested."

    The SHIL waveform, amplitude, and frequency are reported as a 'Goldilocks zone' after the fact, and the immediately preceding text states that without SHIL the house graph 'repeatedly produces incorrect solutions.' Thus the successful configuration is selected by trying alternatives until the readout matches the known optimum. Because the success criterion is agreement with the correct max-cut and the configuration is not fixed in advance, the statement 'This configuration consistently produced accurate solutions' is a fitted characterization of the demonstration runs, not a prediction from a fixed input setting.

full rationale

No self-citation chain or imported uniqueness theorem is involved; the cited prior work (Wang, Pikovsky, Belykh, etc.) is external and provides background, not a load-bearing premise authored by the present authors. The circularity is narrower but real: the paper's headline result is an experimental validation, and the validation criterion is agreement with known max-cut solutions. The paper openly admits that the global coupling strength c must be 'chosen individually for every problem to achieve best results,' and that the SHIL configuration was found by testing with and without SHIL at various waveforms and levels. Since the experimenter knows the correct answer when making these choices, and since no raw data, cut sizes, run counts, or success rates are reported, the claim that the machine 'consistently produced accurate solutions' reduces in part to fitting parameters to the target output. The closed-odd-loop caveat that the machine often 'partitioning into different configurations upon repeated runs' further weakens the reliability claim, but that is a correctness and robustness concern rather than a circularity one. Overall, the central demonstration is partially circular because the reported successes are not obtained with fixed, pre-specified inputs; a score of 6 reflects that partial circularity without alleging intent.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central demonstration depends on a per-problem coupling strength c and SHIL parameters that are tuned by the experimenter; the mapping to max-cut and the binarization by SHIL are standard assumptions from prior work. No new physical entities are introduced.

free parameters (4)
  • Coupling weight c = 1/10 or 1/5 typically, but tuned per graph
    Section 10: 'the actual value c has to be chosen individually for every problem to achieve best results'. This is a per-problem fitted parameter.
  • SHIL amplitude = 2.5 V
    Section 10: experiments used SHIL at 2.5 V and 6.8 kHz; described as the 'Goldilocks zone' for most problems.
  • SHIL frequency = 6.8 kHz (second harmonic of ~3.8 kHz oscillators)
    Section 10: SHIL frequency chosen as second harmonic of the oscillator resonance frequency; exact detuning not reported.
  • SHIL signal waveform = rectangular or pulse preferred over sinusoidal
    Section 10: 'a sinusoidal SHIL signal did not work as well as a rectangular one or a pulse signal on most setups'.
assumptions (3)
  • domain assumption The phase relationship of mutually coupled oscillators encodes the Ising spin state, with antiphase (phase difference pi) corresponding to opposite spins.
    Sections 7-8: the mapping from max-cut to oscillator phases assumes that synchronized antiphase states directly represent spin configurations.
  • domain assumption The SHIL signal forces the oscillator phases to binarize into 0 or pi.
    Section 9: 'To ensure that the oscillators can only lock into one of two possible phases, 0 and pi, an external second harmonic injection locking (SHIL) signal is fed into all oscillators.' This is assumed to hold for the tested parameter range.
  • domain assumption The oscillator network converges to the low-energy (ground) state of the coupled system, thereby solving the max-cut problem.
    Section 2 and Section 10: the machine 'progresses towards the Hamiltonian's lowest-energy state by letting the oscillators synchronize'; however, the paper itself notes that odd loops can lead to different configurations on repeated runs, so convergence to the ground state is not guaranteed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Experiments with an oscillator based Ising machine." pith.science (2026). https://pith.science/paper/MJZNYC4H

@misc{pith2026250203167,
  author       = {Pith},
  title        = {Pith review of: Experiments with an oscillator based Ising machine},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MJZNYC4H}},
  note         = {Machine review of arXiv:2502.03167}
}
read the original abstract

Interest in non-algorithmic, unconventional computing is rising in recent years due to more and more apparent short comings of classic stored-program digital computers, such as energy efficiency, degree of parallelism in computations, clock frequency limitations, integration density, silicon utilization, etc. One notable such unconventional approach are oscillator based Ising machines, i.e., systems consisting of a number of oscillators which can be coupled in order to create an analogue for some problem to be solved, while the actual information is encoded in the phase relationships of these oscillators with respect to some reference (typically one of these oscillators). It has been shown that machines of this type are capable of solving NP-hard problems such as max-cut, etc. In the following an experimental Ising machine is presented together with experimental results obtained from this machine.

Figures

Figures reproduced from arXiv: 2502.03167 by the authors.

Figure 1
Figure 1. Typical max-cut graph between max-cut problems and anti-phase synchronization models facilitates an efficient hardware implementation of oscillator based ISING machines. 8 Problem Mapping The mapping is therefore straightforward and corresponds to the adjacency matrix of the given graph. Initial computational tests were conducted using graphs with uniform weights, requiring adjustment of the resistive coupling betwe… view at source ↗
Figure 2
Figure 2. Schematic of the phase-shift-oscillator used be ring oscillators, LC oscillators, or phase-shift oscillators. A discrete implementation as shown below makes ring-oscillators more involved than LC- or phase-shift-oscillators. Eventually it was decided to implement phase-shift-oscillators due to their simplicity and the fact that they do not require physically large inductances. Another design decision is with respect… view at source ↗
Figure 3
Figure 3. figure 3. The actual I [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (5 more)
Figure 3
Figure 3. Figure 3: First incarnation of the oscillator based ISING machine while the oscilloscope on the left is used for debugging purposes. On top of this oscilloscope is an analog phase meter that was initially used to determine the phase relationships of the four oscillators. This sm…
Figure 4
Figure 4. Figure 4: Typical solution on the four oscillator machine 4. Finally, the phases and thus the solution are read out by means of some phase detector circuits [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Eight oscillator ISING machine [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: Schematic of the eight oscillator ISING machine 11 [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Schematic of one out of seven phase detectors used in the eight oscillator ISING machine signal cannot be negative3 , the 4.7 V ZENER diode at the top clamps the output signal at +4.7 V maximum. If the phase of the oscillator is in phase with the reference oscillator, …

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Geometric Theory of Ising Machines

    cs.ET 2025-07 conditional novelty 6.0 of 10

    Ising circuits can express affine nearest-neighbor classifiers with parallelepiped centroids, and removing their spurious local minima is a linear programming problem.

  2. How to Train an Oscillator Ising Machine using Equilibrium Propagation

    cond-mat.dis-nn 2025-05 conditional novelty 5.0 of 10

    Equilibrium Propagation trains a simulated oscillator Ising machine to classify MNIST at 97.2% and Fashion-MNIST at 88.0%, with robustness to quantized parameters and phase noise.

Reference graph

Works this paper leans on

15 extracted references · 15 canonical work pages · cited by 2 Pith papers

  1. [1]

    28 largesymbols

    \@lbibitem[??(????)] \@undefined @set@cites @set@cites \@citex @citex \@undefined begindocument/end [copernicus/natbib] \@citex @citex thebibliography [1] @secline * - @ \@biblabel NAT@ctr \@bibsetup ##1 NAT@ctr 0 @openbib .11em \@plus.33em \@minus.07em `\.=1000 \@noitemerr natbib Empty `thebibliography' environment - @stage@final @manuscript [1] biblio *...

    work page 2020

  2. [2]

    Belykh , M

    I. Belykh , M. Hasler , M. Lauret , H. Nijmeijer , ``Synchronization and graph topology'', in International Journal of Bifurcation and Chaos, Vol. 15, No. 11, 2005, pp. 4323--3433

    work page 2005

  3. [3]

    Bian , F

    Z. Bian , F. Chudak , G. Macready , G. Rose , ``The ising model: Teaching an old problem new tricks'', in D-wave systems, Vol. 2, 2010, pp. 1--32

    work page 2010

  4. [4]

    Bolotow , M

    D. Bolotow , M. Bolotov , L. Smirnov , G. Osipov , A. Pikovsky , ``Twisted states in a system of nonlinearly coupled phase oscillators'', in Regular and Chaotic Dynamics, Vol. 24, 2019, pp. 717--724

    work page 2019

  5. [5]

    Erementhouk , A

    M. Erementhouk , A. Shukla , P. Mazumder , ``On computational capabilities of Ising machines based on nonlinear oscillators'', in Physica D: Nonlinear Phenomena, Vol. 437, 2022, Article No. 133334

    work page 2022

  6. [6]

    Ising , ``Beitrag zur Theorie des Ferromagnetismus'', in Zeitschrift f\"ur Physik, Vol

    E. Ising , ``Beitrag zur Theorie des Ferromagnetismus'', in Zeitschrift f\"ur Physik, Vol. 31, No. 1, Feb. 1925, pp. 253--258

    work page 1925

  7. [7]

    Kawamura , ``Phase synchronization between collective rhythms of fully locked oscillator groups'', in Scientific Reports, Vol

    Y. Kawamura , ``Phase synchronization between collective rhythms of fully locked oscillator groups'', in Scientific Reports, Vol. 4, No. 1, 2014, Article No. 4832

    work page 2014

  8. [8]

    Marandi , Ising machines: Non-von Neumann computing with nonlinear optics, https://www.youtube.com/watch?v=V7BxJsLyubk&t=209s, 2019

    A. Marandi , Ising machines: Non-von Neumann computing with nonlinear optics, https://www.youtube.com/watch?v=V7BxJsLyubk&t=209s, 2019

    work page 2019

Show all 15 references
  1. [9]

    Mohseni , P

    N. Mohseni , P. L. McMahon , T. Byrnes , ``Ising machines as hardware solvers of combinatorial optimization problems'', in Nature Reviews Physics, Vol. 4, No. 6, 2022, pp. 363--379

    2022

  2. [10]

    Pikovsky , M

    A. Pikovsky , M. G. Rosenblum , J. Kurths , Synchronization -- A Universal Concept in Nonlinear Sciences, Cambridge University Press, 2001

    2001

  3. [11]

    Stanford, http://web.stanford.edu/ yyye/yyye/Gset/, accessed 27.01.2025

    2025

  4. [12]

    S. H. Strogatz , Nonlinear dynamics and chaos with applications to physics, biology, chemistry, and engineering, CRC press, 2018

    2018

  5. [13]

    Vaidya , M

    J. Vaidya , M. K. Bashar , N. Shikla , ``Using noise to augment synchronization among oscillators'', in Scientific Reports, Vol. 11, No. 1, 2021, Article No. 4462

    2021

  6. [14]

    Wang , J

    T. Wang , J. Roychowdhury , ``OIM: Oscillator-based Ising machines for solving combinatorial optimisation problems'', in Unconventional Computation and Natural Computation: 18th International Conference, UCNC 2019, Tokyo, Japan, June 3--7, Proceedings 18, 2019, pp. 232--256

    2019

  7. [15]

    Yuan , F

    D. Yuan , F. Lin , L. Wang , D. Liu , J. Yang , Y. Xiao , ``Multistable states in a system of coupled phase oscillators with inertia'', in Scientific Reports, Vol. 7, No. 1, 2017, Article No. 42178

    2017

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.