Abstract
Recently, there has been growing interest in the utilization of physical systems as heuristic optimizers for classical spin Hamiltonians. A prominent approach employs gain-dissipative optical oscillator networks for this purpose. Unfortunately, these systems inherently suffer from an inexact mapping between the oscillator network loss rate and the spin Hamiltonian due to additional degrees of freedom present in the system such as oscillation amplitude. In this work, we theoretically analyze and experimentally demonstrate a scheme for the alleviation of this difficulty. The scheme involves control over the laser oscillator amplitude through modification of individual laser oscillator loss. We demonstrate this approach in a laser network classical XY model simulator based on a digital degenerate cavity laser. We prove that for each XY model energy minimum there corresponds a unique set of laser loss values that leads to a network state with identical oscillation amplitudes and to phase values that coincide with the XY model minimum. We experimentally demonstrate an eight fold improvement in the deviation from the minimal XY energy by employing our proposed solution scheme.
1 Introduction
Optimization problems are at the heart of numerous fields of science and industry from drug discovery [1] to artificial intelligence [2]. Unfortunately, many of the problems found in these applications are proven to be in the NP-hard complexity class rendering their solution impractical even at modest input size [3]. Due to the significant applicability of these problems, various computational approaches have been developed to find practically useful approximations of their solutions in polynomial time [4], [5]. These approximation algorithms include non-linear programming [6], semidefinite programming [7], [8], local search algorithms [9], evolution inspired algorithms [10], and physically inspired heuristic algorithms [11] among others.
Physically inspired algorithms are typically heuristic algorithms based on a mapping between the cost function and a physical energy landscape. Optimization is then carried out by mimicking the dynamics of physical systems towards low energy states [12]. In physical terms, the optimization problem is converted to a ground state search problem. For example, the simulated annealing (SA) algorithm mimics the cooling of metals by stochastic dynamics [13], the quantum annealing algorithm mimics quantum dynamic of the ground state evolution [14], the particle swarm algorithm mimics the manner in which avian flocks find food sources through distributed non-linear dynamics [15]. Extensive literature exists on the use and benchmarking of these algorithms for various applications [16], [17] as well as on the success of such an approach in comparison with state-of-the-art methods [18], [19], [20].
An alternative to ground-state search algorithms implemented on digital computers is the realization of specialized hardware setups. Prominent examples of systems for ground-state search include dedicated hardware for neural network training [21], [22], [23], photonic Ising machines and XY simulators [24], [25], [26], [27], [28], superconducting qubit annealing machines [14], [29], polariton based XY simulators [30, 31], electronic Ising machines [32], [33], [34], memristor network systems [35], [36], [37] and others. Many of these systems are aimed at finding the ground state of classical spin models. This is of special interest since many NP-complete problems can be mapped to such models [38]. Recent results on the universality of these models provide additional flexibility in mapping a given optimization problem to a given spin model [39].
Generally, two main ingredients are required of a physical ground state finder: (i) A correspondence between the physical system’s stable states and the optimization function i. e. model energy landscape minima, (ii) A mechanism that ensures the evolution of the system to a stable state corresponding to an optimum of the optimization problem i. e. the model ground state.
An exact correspondence between a system’s stable state and a model energy minimum would require the elimination of all physical degrees of freedom (DOF) absent from the model. For example, a continuous scalar DOF might be mapped to a discrete spin DOF. Achieving an exact correspondence is thus challenging both from the theoretical and experimental aspects [40], [41], [42]. However, additional DOFs can present an opportunity to improve the optimization success rate by embedding the model dynamics in the higher dimensional dynamics of the physical system. For example, such embedding could help to avoid trapping in local minima [8], [43].
Ensuring the evolution of a physical system to its global ground state poses a significant challenge for complex non-convex energy landscapes [44]. It is highly unlikely that physical systems can find the global ground state of such Hamiltonians in sub-expnonential time [45, 46] Thus the main research question is whether physical optimization machines can outperform digital computers at these hard tasks by harnessing additional resources. Such resources might include short iteration times [47], quantum tunneling and coherence [14], inherent parallelism [47], favorable scaling [48], or other resources [49].
In gain-dissipative optical oscillator networks one aims to find the ground state of classical spin models such as the Ising model and the XY model. In such systems, the phase of each optical oscillator (either optical parametric oscillator - OPO or laser) is mapped to a spin DOF. The idea put forward in [50], is to utilize the mode-competition property of optical oscillators to select the network state with the lowest loss rate. In principle, there exists an external driving rate (pump) range for which only the mode associated with the lowest loss is a stable fixed point. Tuning the pump to this range could in principle result in finding the ground state of the spin model. Several types of optical oscillator networks based on this operation principle have been implemented and their efficacy at finding low energy states of various instances of spin models was studied [24], [25], [50].
It was shown that an approximate mapping can be established between the spin model energy and the oscillator network loss rate when the inter-oscillator coupling is low [40]. In this regime, the oscillation amplitudes are almost the same for all oscillators. This reduces the loss rate of the network to an equivalent spin model energy [40]. On the other hand, low coupling strength leads to small energy gaps and thus long relaxation times to the ground state. Additionally, this leads to increased sensitivity to imperfections and noise, dictating the use of finite-size coupling in practical applications. This, in turn, leads to unequal (heterogeneous) oscillation amplitudes which preclude the exact mapping between the loss rate and the spin model energy. Since this limitation stems from a finite size coupling between oscillators, it is inherent to any coupled oscillator network. Several theoretical approaches for mediating this effect have been suggested [41], [42], [51]. Both methods rely on equalizing the amplitude of the oscillators by imposing additional dynamical equations on each oscillator. To the author’s knowledge, such techniques have not been experimentally demonstrated and studied to date.
In this work, we theoretically analyze and experimentally demonstrate the problem of unequal amplitudes. This is carried out on a simple laser network acting as an XY model ground state finder. We find that an inherent contradiction exists between finding the ground state in the vicinity of the oscillation threshold and accurately mapping the network state to an XY state. To solve this, we devise and experimentally demonstrate a scheme for the solution of the amplitude heterogeneity problem in a laser network. The implemented solution is based on controlling the individual oscillator loss rates to achieve equal oscillation amplitudes. The phase in such states is found to correspond to XY model minima. We prove that for each XY model minimum, the set of laser network parameters for which the laser network phase values coincide with the XY model is unique.
2 Problem presentation
We illustrate the amplitude heterogeneity problem and our solution to it with a simple laser network - the “house graph” [52] with negative (anti-ferromagnetic) couplings, shown in Figure 1(a). The classical XY Hamiltonian for the house graph is given by
where
We propose to overcome this effect by tuning the laser oscillator parameters. As shown in Section 4, if the loss of each laser is adjusted such that the minimal loss state has equal amplitudes for all lasers, the phases corresponding to the XY ground state are exactly recovered. This verifies the exact mapping between the minimal loss state of equal amplitude lasers to the XY ground state.
3 Experiment
The laser network used to simulate the XY model is implemented in a digital degenerate ring cavity laser [47], schematically depicted in Figure 2. The cavity includes two 4f imaging telescopes, an Nd:YAG gain medium, a spatial light modulator (SLM), an optical isolator and an output coupler. The gain medium is pumped by a Xenon flash lamp, resulting in 200 μs quasi-CW pulses. The pumping rate is controllable by changing the voltage of the flash-lamp activation pulse. The 8f telescopes image the field distribution from the SLM onto itself after a cavity roundtrip [54]. For more details, see Supplementary Information.
The SLM is utilized as a digital mirror whose complex-valued reflectivity at each pixel is controlled [55]. This control is used to generate a network of single mode lasers with any desired geometry (“house” network here) by imposing lasing only on specific spots where a non-zero controllable reflectivity is defined. This adjustable reflectively is then used to control the loss of each laser independently via a closed loop control scheme. We first use this scheme to compensate for loss and gain imperfections in our system and to form a network of identical laser oscillators and later we use it to form a network of lasers with equal amplitude (see Section 6). We adjust the phase of the reflection coefficient for each laser in the network independently with an additional closed-loop control scheme to compensate for aberrations in our laser cavity and ensure identical frequency for all lasers.
In the degenerate cavity used in this work, each laser spot corresponds to an independent laser oscillator [56]. Two methods are applied in this work to introduce diffractive coupling between lasers, (i) a circular aperture placed at the Fourier plane of the second 4f telescope as depicted in Figure 2(a) generates weak coupling [57] and (ii) a lens placed instead of the aperture, generates strong (Talbot) coupling [58], see Supplementary Material for more details.
The measurement of the laser network phase and amplitude state is carried out by using an interferometric apparatus schematically depicted in Figure 2(b). On one arm of the interferometer, a pinhole and a lens serve to select and expand a single laser. In the second arm, a 4f telescope is used to directly image the laser field distribution on the SLM. The light from both arms is recombined on a CCD detector resulting in interference fringes on each laser spot (see Figure 2(c)).
Figure 1(d) depicts the measured state of a network of identical lasers with anti-ferromagnetic coupling. The coupling strength (where a bond exists) is
Next, we adjust the loss of each laser such that all lasers have the same amplitude (while maintaining identical frequencies and pump values) by applying the amplitude control scheme detailed in Section 6. The loss modification of laser oscillators was experimentally found to be 0.204 for the uppermost laser (in units of inverse cavity round-trip time) and −0.041, 0.032, 0.062, 0.0014 for the lower four (starting from the upper left laser and going counter-clockwise). We denote it as the modified loss vector
Next, to investigate the network state as a function of the amplitude heterogeneity, we scan the laser loss vector by interpolating between the identical oscillator and the equal amplitude states in the following manner
where x is the interpolation parameter in loss space and
P−Pth | x = 0 | x = 1 | |||||
---|---|---|---|---|---|---|---|
ΔI/I[%] | ΔI/I[%] | ||||||
i | 5% | κ | E | 146 ± 1 | 13.4 ± 0.4 | 135 ± 1 | 6.2 ± 0.1 |
T | 148.3 | 22.4 | 136.8 | 0 | |||
ii | 5% | E | 150 ± 1 | 20.9 ± 0.3 | 141 ± 1 | 2.7 ± 0.1 | |
T | 143.9 | 14.2 | 136.8 | 0 | |||
iii | 5% | E | 133 ± 1 | 4 ± 0.04 | 140 ± 2 | 3 ± 0.04 | |
T | 139.0 | 4.2 | 136.8 | 0 | |||
iv | 40% | E | 174 ± 4 | 55.9 ± 2 | – | – | |
T | 180 | 56 | 136.8 | 0 |
The experimental and theoretical values for the phase and amplitude heterogeneity for all measured cases are summarized in Table 1 at the identical oscillator and equal amplitude states. Qualitatively it is seen in Figure 3(b) that the deviation from the XY model ground state phase changes from negative values (corresponding to a phase
Figure 3(c) directly plots the deviation from the XY ground state phase versus the amplitude heterogeneity. It is seen that both numerically and experimentally, the data collapses to a single curve which indicates that in this example, the phase deviation depends only on the amplitude heterogeneity. Focusing on the unmodified laser network fixed point (x = 0), it is seen that higher coupling strength leads to significantly larger amplitude heterogeneity and consequently extreme phase deviations from the XY model phase exceeding 40°. Unfortunately, the correction of this extreme heterogeneity by amplitude feedback was experimentally unavailable due to high parasitic loss values present in our implementation of strong coupling [57]. The theoretical plot for strong coupling values and pump close to threshold reveals a very sharp transition from extreme amplitude heterogeneity to equal amplitudes as a function of x.
The final test of the effect of amplitude heterogeneity on the performance of a laser network as an XY model optimizer is the effect on the XY energy (Eq. (1)) calculated from the measured laser network phase values. The energy calculated from measured phase values and from simulation results is plotted in Figure 4 versus x for two of the cases in Table 1. Figure 4 reveals that the theoretical minimum of the XY model energy is obtained to the best approximation at the equal amplitude point x = 1. A good general agreement is found between the measured values and the simulation across the whole x range. In addition, it is evident again that at high coupling strength the deviation from the XY energy minimum is more severe at the identical oscillator network state and that generally the dependence on x is steeper around the equal amplitudes point.
4 Analysis
In the following we show that the proposed method of oscillator loss tuning indeed leads to equal amplitude solutions of the laser rate equations. Moreover, we show that the loss values for each XY minimum are unique. The analysis starts from the coupled laser rate equations for M class-B, identical frequency, laser oscillators [60]
where Em(t) is the slowly-varying electric field of the mth laser, Gm(t) is the mth active medium gain,
First, we establish the connection between the laser network and the classical XY model. To this end we assume that the oscillation amplitude is an identical constant i. e. Am(t) = |Em(t)| = A, ∀m. By setting
where HXY is the classical XY model Hamiltonian defined in Eq. (1). Thus, the phase values at equal amplitude fixed points correspond to the phase values of the XY model extremal points i. e. an exact correspondence is reached between the laser network loss and the XY energy.
To achieve this exact correspondence, the inherent amplitude heterogeneity has to be addressed. In the following, we outline our proposed scheme for achieving this by tailoring the loss of each oscillator
Now we self-consistently assume that we can modify the loss such that all lasers now have the same amplitude Am = A, ∀m and consequently check if such a solution can indeed exist. It immediately follows that there exist a set of (unknown) loss values that allow for such a solution i. e.
where δ is defined through Eq. (8). Denoting the loss as
Let us now demonstrate our approach for the house graph example. Using Eq. (7), the added loss which corresponds to the XY ground state is uniquely defined by
To directly show the correspondence, note that summing up Eq. (7), we obtain:
Thus, the added loss is directly related to the XY model energy at the fixed point. Moreover, the XY energy of the state can be inferred solely from the loss modification values
To understand the extreme intensity heterogeneity found near the oscillation threshold, we observe Eq. (3) slightly above the oscillation threshold. Since the lasing transition is a super-critical pitchfork bifurcation [62], we can assume |Em|≪1 ∀m and arrive at the following eigenvalue equation for the fixed points:
where
5 Discussion and conclusions
As a first step, our approach has been demonstrated for a small oscillator network, with nearest-neighbour coupling, and a single stable fixed point. We believe that by using the closed-loop amplitude control scheme this approach can be directly extended for larger arrays with arbitrary connectivity [64]. However, this approach finds some fixed point of the XY problem, without a guarantee for finding the ground state. Thus, future research should be aimed to explore the joint dynamics of the closed loop loss control and the oscillators as well as to devise schemes guaranteeing convergence of the joint dynamics to the global optimum.
We have demonstrated and studied the effects of amplitude heterogeneity on the performance of a laser oscillator network XY simulator. The effect was studied on a small and simple laser network – the house graph with anti-ferromagnetic coupling. It was found that amplitude heterogeneity is most severe at the minimum loss state (infinitesimally above oscillation threshold) and that the phase values at this state deviate the most from the XY model ground state phases. We proposed and demonstrated a scheme for the solution of the amplitude heterogeneity effect. The scheme involves changing the laser oscillator loss rate to achieve equal amplitudes. We found that upon equalization of the amplitudes the laser phases recover the XY ground state phases. We also proved that the set of loss values equalizing the amplitudes is unique for each XY model minimum.
6 Methods
6.1 Phase measurement
As shown in a recent publication [65], multiple longitudinal modes that coexist in our system are effectively uncoupled. Thus the measured interference fringes (Figure 2(b) and (c)) amplitude and phase i. e. the complex valued coherence [66] corresponds to an ensemble average over the coherence [67]. To interpret these ensemble averaged results, note that for time-reversal symmetric systems (identical frequency oscillators with symmetric real valued coupling matrix), complex conjugation of each field solution also yields a valid solution. Hence, a state with phases
6.2 Amplitude control
Control over the laser intensities, to suppress intensity heterogeneity, is carried out by a laser loss closed loop control scheme as follows. Operating at quasi-CW mode, the pulse averaged intensity of each laser is measured using the imaging apparatus. This intensity is averaged over a pre-defined number of pulses after which the SLM reflection coefficient of each laser is modified according to a proportional control feedback [68] scheme with a constant target intensity for all lasers. This protocol is carried out until all intensities converge to the target value (up to a predefined tolerance).
6.3 Frequency control
To achieve the required phase measurement resolution, the laser frequency heterogeneity i. e. detuning imperfection was also addressed. This imperfection is caused by phase aberrations and cavity misalignment which lead to systematic phase deviations in the laser network fixed points. To compensate for this imperfection, the oscillator phase is measured per pulse and controlled via the SLM reflection phase.
6.4 Experimental imperfection compensation
Due to experimental imperfections, making the oscillators identical entails the nullification of parasitic loss variation. To that end, we applied the intensity feedback scheme separately to two sub-graphs of the house graph: a square graph and a triangle graph. Since both subgraphs have equal amplitude fixed points when the oscillators are identical, finding the loss values that result in equal amplitudes compensates for parasitic loss heterogeneity. The loss values resulting in equal amplitudes are found by using the closed loop intensity feedback as described the previous paragraph.
Funding source: Abramson Family Center for Young Scientists
Funding source: Israel Science Foundation
Award Identifier / Grant number: 950/19
Award Identifier / Grant number: 1881/17
Acknowledgments
O.R. is the incumbent of the Shlomo and Michla Tomarin career development chair, and is supported by the Abramson Family Center for Young Scientists and by the Israel Science Foundation, Grant No. 950/19 and 1881/17.
Author contribution: All the authors have accepted responsibility for the entire content of this submitted manuscript and approved submission.
Research funding: This research was funded by Abramson Family Center for Young Scientists and Israel Science Foundation.
Employment or leadership: None declared.
Honorarium: None declared.
Conflict of interest statement: The authors declare no conflicts of interest regarding this article.
References
[1] I. D. Kuntz, “Structure-based strategies for drug design and discovery,” Science, vol. 257, no. 5073, pp. 1078–1082, Sep. 1992, https://doi.org/10.1126/science.257.5073.1078.Search in Google Scholar
[2] J. J. Hopfield, “Neural networks and physical systems with emergent collective computational abilities,” Proc. Natl. Acad. Sci. U.S.A., vol. 79, no. 8, pp. 2554–2558, 1982, https://doi.org/10.1073/pnas.79.8.2554.Search in Google Scholar
[3] M. R. Garey and D. S. Johnson, Computers and Intractability, vol. 174, San Francisco. W. H. Freeman, 1979.Search in Google Scholar
[4] V. V. Vazirani, Approximation Algorithms, Berlin, Springer Science & Business Media, 2013.Search in Google Scholar
[5] E.-G. Talbi, Metaheuristics: From Design to Implementation, vol. 74, New Jersy, US, John Wiley & Sons, 2009.10.1002/9780470496916Search in Google Scholar
[6] D. G. Luenberger and Y. Ye, and Others, Linear And Nonlinear Programming, vol. 2, Berlin, Springer, 1984.Search in Google Scholar
[7] M. X. Goemans and D. P. Williamson, “Improved approximation algorithms for maximum cut and satisflability problems using semidefinite programming,” J. ACM, vol. 42, no. 6, pp. 1115–1145, Nov. 1995, https://doi.org/10.1145/227683.227684.Search in Google Scholar
[8] L. Vandenberghe and S. Boyd, “Semidefinite programming,” SIAM Rev., vol. 38, no. 1, pp. 49–95, 1996, https://doi.org/10.1137/1038003.Search in Google Scholar
[9] E. Aarts, E. H. L. Aarts, and J. K. Lenstra, Local Search in Combinatorial Optimization, New Jersy, US, Princeton University Press, 2003.10.1515/9780691187563Search in Google Scholar
[10] G. Zames, N. M. Ajlouni, N. M. Ajlouni et al., “Genetic algorithms in search, optimization and machine learning.” Inf. Technol. J., vol. 3, no. 1, pp. 301–302, 1981.Search in Google Scholar
[11] X.-S. Yang, Nature-Inspired Optimization Algorithms, Amsterdam, Netherlands, Elsevier, 2014.10.1016/B978-0-12-416743-8.00005-1Search in Google Scholar
[12] E. Aarts and J. Korst, Simulated annealing and Boltzmann machines: A Stochastic Approach to Combinatorial Optimization and Neural Computing. Chichester: Wiley Interscience Series in Discrete Mathematics and Optimization, 1989.Search in Google Scholar
[13] S. Kirkpatrick, C. D. Gelatt, and M. P. Vecchi, “Optimization by simulated annealing,” Science, vol. 220, no. 4598, pp. 671–680, 1983, https://doi.org/10.1126/science.220.4598.671.Search in Google Scholar
[14] A. Das and B. K. Chakrabarti, Quantum Annealing and Related Optimization Methods, vol. 679, Berlin, Springer Science & Business Media, 2005.10.1007/11526216Search in Google Scholar
[15] J. Kennedy and R. Eberhart, “Particle swarm optimization,” in Proceedings of ICNN’95-International Conference on Neural Networks, Perth, vol. 4, IEEE, 1995, pp. 1942–1948.Search in Google Scholar
[16] J. Vesterstrøm and R. Thomsen, “A comparative study of differential evolution, particle swarm optimization, and evolutionary algorithms on numerical benchmark problems,” in Proceedings of the 2004 Congress on Evolutionary Computation, CEC2004, vol. 2, 2004, pp. 1980–1987.10.1109/CEC.2004.1331139Search in Google Scholar
[17] J. A. Parejo, A. Ruiz-Cortés, S. Lozano, and P. Fernandez, “Metaheuristic optimization frameworks: a survey and benchmarking,” Soft Comput., vol. 16, no. 3, pp. 527–561, Mar. 2012, https://doi.org/10.1007/s00500-011-0754-8.Search in Google Scholar
[18] X. Yin, B. Sedighi, M. Varga, M. Ercsey-Ravasz, Z. Toroczkai, and X. S. Hu, “Efficient analog circuits for boolean satisfiability,” IEEE Trans. Very Large Scale Integr. Syst., vol. 26, no. 1, pp. 155–167, Jan. 2018, https://doi.org/10.1109/tvlsi.2017.2754192.Search in Google Scholar
[19] F. L. Traversa, P. Cicotti, F. Sheldon, and M. Di Ventra, “Evidence of exponential speed-up in the solution of hard optimization problems,” Complexity, vol. 2018, 2018, Art no. 7982851, [Online]. Available at: https://doi.org/10.1155/2018/7982851.Search in Google Scholar
[20] D. Molnár, F. Molnár, M. Varga, Z. Toroczkai, and M. Ercsey-Ravasz, “A continuous-time MaxSAT solver with high analog performance,” Nat. Commun., vol. 9, no. 1, pp. 1–12, Dec. 2018, https://doi.org/10.1038/s41467-018-07327-2.Search in Google Scholar
[21] D. Steinkraus, I. Buck, and P. Y. Simard, “Using GPUs for machine learning algorithms,” in Proceedings of the International Conference on Document Analysis and Recognition, ICDAR, vol. 2005, 2005, pp. 1115–1120.10.1109/ICDAR.2005.251Search in Google Scholar
[22] J. D. Owens, M. Houston, D. Luebke, S. Green, J. E. Stone, and J. C. Phillips, “GPU computing,” Proc. IEEE, vol. 96, no. 5, pp. 879–899, May. 2008, https://doi.org/10.1109/jproc.2008.917757.Search in Google Scholar
[23] L. Larger, M. C. Soriano, D. Brunner et al., “Photonic information processing beyond Turing: an optoelectronic implementation of reservoir computing,” Optics Express, vol. 20, no. 3, p. 3241, Jan. 2012, https://doi.org/10.1364/oe.20.003241.Search in Google Scholar
[24] P. L. McMahon, A. Marandi, Y. Haribara et al., “A fully programmable 100-spin coherent Ising machine with all-to-all connections,” Science, vol. 354, no. 6312, pp. 614–617, Nov. 2016, https://doi.org/10.1126/science.aah5178.Search in Google Scholar
[25] T. Inagaki, Y. Haribara, K. Igarashi et al., “A coherent Ising machine for 2000-node optimization problems,” Science, vol. 354, no. 6312, pp. 603–606, Nov. 2016, https://doi.org/10.1126/science.aah4243.Search in Google Scholar
[26] Y. Takeda, S. Tamate, Y. Yamamoto, H. Takesue, T. Inagaki, S. Utsunomiya Boltzmann sampling for an XY model using a non-degenerate optical parametric oscillator network. Quantum Sci Technol, vol. 3 no. 1, 2017, Art no. 014004.10.1364/FIO.2017.FM4E.3Search in Google Scholar
[27] M. Nixon, E. Ronen, A. A. Friesem, and N. Davidson, “Observing geometric frustration with thousands of coupled lasers,” Phys. Rev. Lett., vol. 110, no. 18, May. 2013, https://doi.org/10.1103/physrevlett.110.184102.Search in Google Scholar
[28] Y. Okawachi, M. Yu, X. Ji, J. K. Jang, M. Lipson, and A. L. Gaeta, “Coupled degenerate parametric oscillators towards photonic coherent ising machine,” in Conference on Lasers and Electro-OpticsSan Jose, California United States. The Optical Society, May. 2019, p. FM1D.6.10.1364/CLEO_QELS.2019.FM1D.6Search in Google Scholar
[29] S. Boixo, T. F. Rønnow, S. V. Isakov, et al., “Evidence for quantum annealing with more than one hundred qubits,” Nat. Phys., vol. 10, no. 3, pp. 218–224, 2014, https://doi.org/10.1038/nphys2900.Search in Google Scholar
[30] N. G. Berloff, M. Silva, K. Kalinin, et al., “Realizing the classical XY Hamiltonian in polariton simulators,” Nat. Mater., vol. 16, no. 11, pp. 1120–1126, Aug. 2017, https://doi.org/10.1038/nmat4971.Search in Google Scholar
[31] P. G. Lagoudakis and N. G. Berloff, “A polariton graph simulator,” New J. Phys., vol. 19, no. 12, Aug. 2017, Art no. 125008, https://doi.org/10.1088/1367-2630/aa924b.Search in Google Scholar
[32] M. Yamaoka, C. Yoshimura, M. Hayashi, T. Okuyama, H. Aoki, H. Mizuno “20k-spin Ising chip for combinational optimization problem with CMOS annealing.” Digest of Technical Papers - IEEE International Solid-State Circuits Conference, vol. 58. San Francisco, CA, USA: Institute of Electrical and Electronics Engineers Inc, 2015. p. 432–433.10.1109/ISSCC.2015.7063111Search in Google Scholar
[33] I. Mahboob, H. Okamoto, and H. Yamaguchi, “An electromechanical Ising Hamiltonian,” Science Advances, vol. 2, no. 6, 2016, https://doi.org/10.1126/sciadv.1600236.Search in Google Scholar
[34] T. Wang, J. Roychowdhury OIM: oscillator-based ising machines for solving combinatorial optimisation problems. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 11493. Tokyo, Japan: LNCS, Springer-Verlag, 2019. p. 232–256.10.1007/978-3-030-19311-9_19Search in Google Scholar
[35] A. Thomas, Memristor-based neural networks. J Phys D Appl Phys., vol. 46, no. 9, 2013, Art no. 093001.10.1088/0022-3727/46/9/093001Search in Google Scholar
[36] A. Adamatzky and L. Chua, Memristor Networks, Berlin, Springer Science & Business Media, 2013.10.1007/978-3-319-02630-5Search in Google Scholar
[37] D. Stathis, I. Vourkas, and G. C. Sirakoulis, “Shortest path computing using memristor-based circuits and cellular automata,” in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 8751, pp. 398–407, 2014, https://doi.org/10.1007/978-3-319-11520-7_41.Search in Google Scholar
[38] A. Lucas, “Ising formulations of many NP problems,” Front. Phys., vol. 2, pp. 1–14, 2014, https://doi.org/10.3389/fphy.2014.00005.Search in Google Scholar
[39] G. De Las Cuevas and T. S. Cubitt, “Simple universal models capture all classical spin physics,” Science, vol. 351, no. 6278, pp. 1180–1183, Mar. 2016, https://doi.org/10.1126/science.aab3326.Search in Google Scholar
[40] Z. Wang, A. Marandi, K. Wen, R. L. Byer, and Y. Yamamoto, “Coherent Ising machine based on degenerate optical parametric oscillators,” Phys. Rev. A: At., Mol., Opt. Phys., vol. 88, no. 6, Dec. 2013, https://doi.org/10.1103/physreva.88.063853.Search in Google Scholar
[41] T. Leleu, Y. Yamamoto, P. L. McMahon, and K. Aihara, “Destabilization of local minima in analog spin systems by correction of amplitude heterogeneity,” Phys. Rev. Lett., vol. 122, no. 4, Feb. 2019, https://doi.org/10.1103/PhysRevLett.122.040607.Search in Google Scholar
[42] K. P. Kalinin and N. G. Berloff, “Global optimization of spin Hamiltonians with gain-dissipative systems,” Sci. Rep., vol. 8, no. 1, Dec. 2018, https://doi.org/10.1038/s41598-018-35416-1.Search in Google Scholar
[43] Z. Q. Luo, W. K. Ma, A. So, Y. Ye, and S. Zhang, “Semidefinite relaxation of quadratic optimization problems,” in IEEE Signal Processing Magazine, vol. 27, no. 3, Piscataway, NJ, USA: Institute of Electrical and Electronics Engineers Inc., 2010, pp. 20–34.10.1109/MSP.2010.936019Search in Google Scholar
[44] M. Mézard, G. Parisi, and M. Virasoro, Spin glass theory and beyond: An Introduction to the Replica Method and Its Applications, vol. 9, Singapore, World Scientific Publishing Company, 1987.10.1142/0271Search in Google Scholar
[45] S. Aaronson, “The limits of quantum computers,” Sci. Am., vol. 298, no. 3, pp. 62–69, 2008, https://doi.org/10.1038/scientificamerican0308-62.Search in Google Scholar
[46] Guest Column: NP-complete problems and physical reality,” ACM SIGACT - SIGACT News, vol. 36, no. 1, p. 30, 2005. [Online].10.1145/1052796.1052804Search in Google Scholar
[47] C. Tradonsky, I. Gershenzon, V. Pal et al., “Rapid laser solver for the phase retrieval problem,” Sci. Adv., vol. 5, no. 10, 2019, Art no. eaax4530, https://doi.org/10.1126/sciadv.aax4530.Search in Google Scholar
[48] A. J. MacFaden, G. S. Gordon, and T. D. Wilkinson, “An optical Fourier transform coprocessor with direct phase determination,” Sci. Rep., vol. 7, no. 1, pp. 1–8, 2017, https://doi.org/10.1038/s41598-017-13733-1.Search in Google Scholar
[49] Y. Yamamoto, K. Aihara, T. Leleu, K. I. Kawarabayashi, S. Kako, M. Fejer, et al. Coherent Ising machines—Optical neural networks operating at the quantum limit. NPJ Quantum Info., vol. 3, no. 1, pp. 1–15, 2017.10.1364/CLEOPR.2018.W1D.1Search in Google Scholar
[50] A. Marandi, Z. Wang, K. Takata, R. L. Byer, and Y. Yamamoto, “Network of time-multiplexed optical parametric oscillators as a coherent Ising machine,” Nat. Photon., vol. 8, no. 12, pp. 937–942, Nov. 2014, https://doi.org/10.1038/nphoton.2014.249.Search in Google Scholar
[51] K. Kalinin and N. Berloff, Networks of non-equilibrium condensates for global optimization. New J Phys., vol. 20, no. 11, 2018, Art no. 113023.10.1088/1367-2630/aae8aeSearch in Google Scholar
[52] D. B. West and Others, Introduction to Graph Theory, vol. 2, Upper Saddle River, Prentice-Hall, 2001.Search in Google Scholar
[53] H. T. Diep, Frustrated Spin Systems, 2nd ed. Singapore, World Scientific Publishing Co., 2013.10.1142/8676Search in Google Scholar
[54] J. A. Arnaud, “Degenerate optical cavities II: effect of misalignments,” Appl. Optic., vol. 8, no. 9, p. 1909, 1969, https://doi.org/10.1364/AO.8.001909.Search in Google Scholar
[55] S. Ngcobo, I. Litvin, L. Burger, and A. Forbes, “A digital laser for on-demand laser modes,” Nat. Commun., vol. 4, no. 1, pp. 1–6, Aug. 2013, https://doi.org/10.1038/ncomms3289.Search in Google Scholar
[56] M. Nixon, M. Friedman, E. Ronen, A. A. Friesem, N. Davidson, and I. Kanter, “Synchronized cluster formation in coupled laser networks,” Phys. Rev. Lett., vol. 106, no. 22, Jun. 2011, Art no. 223901, https://doi.org/10.1103/physrevlett.106.223901.Search in Google Scholar
[57] C. Tradonsky, V. Pal, R. Chriki, N. Davidson, and A. A. Friesem, “Talbot diffraction and Fourier filtering for phase locking an array of lasers,” Appl. Optic., vol. 56, no. 1, p. A126, Jan. 2017, https://doi.org/10.1364/ao.56.00a126.Search in Google Scholar
[58] S. Mahler, C. Tradonsky, R. Chriki, A. A. Friesem, and N. Davidson, “Coupling of laser arrays with intracavity elements in the far-field,” OSA Continuum, vol. 2, no. 6, p. 2077, Jun. 2019, https://doi.org/10.1364/osac.2.002077.Search in Google Scholar
[59] A. E. Siegmann, Lasers, Mill Valley, CA, University Scinece Books, 1986.Search in Google Scholar
[60] F Rogister, K. S. Thornburg, L. Fabiny, M. Möller, and R. Roy, “Power-law spatial correlations in arrays of locally coupled lasers,” Phys. Rev. Lett., vol. 92, no. 9, 2004, Art no. 093905, https://doi.org/10.1103/physrevlett.92.093905.Search in Google Scholar
[61] S. Tamate, Y. Yamamoto, A. Marandi, P. McMahon, and S. Utsunomiya, “Simulating the classical XY model with a laser network,” Aug. 2016 [Online]. Available at: http://arxiv.org/abs/1608.00358.Search in Google Scholar
[62] P. Mandel, Theoretical Problems in Cavity Nonlinear Optics, Cambridge, UK, Cambridge University Press, 2005, no. 21.Search in Google Scholar
[63] R. A. Horn and C. R. Johnson, Matrix Analysis, Cambridge, UK, Cambridge University Press, 2012.10.1017/CBO9781139020411Search in Google Scholar
[64] J. W. Goodman, Linear Space-Variant Optical Data Processing, Berlin, Heidelberg, Springer, 1981, pp. 235–260.10.1007/3540105220_13Search in Google Scholar
[65] S. Mahler, M. L. Goh, C. Tradonsky, A. A. Friesem, and N. Davidson, “Improved phase locking of laser arrays with nonlinear coupling,” Phys. Rev. Lett., vol. 124, no. 13, Apr. 2020, Art no. 133901, https://doi.org/10.1103/PhysRevLett.124.13390.Search in Google Scholar
[66] J. W. Goodman, Statistical Optics, New Jersy, US, John Wiley & Sons, 2015.Search in Google Scholar
[67] V. Pal, S. Mahler, C. Tradonsky, A. A. Friesem, and N. Davidson, “Rapid fair sampling of XY spin Hamiltonian with a laser simulator,” Dec. 2019 [Online]. Available at: http://arxiv.org/abs/1912.10689.10.1103/PhysRevResearch.2.033008Search in Google Scholar
[68] B. W. Bequette, Process Control: Modeling, Design, and Simulation. Upper Saddle River, NJ, USA: Prentice Hall Professional, 2003.Search in Google Scholar
Supplementary material
The online version of this article offers supplementary material (https://doi.org/10.1515/nanoph-2020-0137).
© 2020 Igor Gershenzon et al., published by De Gruyter, Berlin/Boston
This work is licensed under the Creative Commons Attribution 4.0 International License.