PHYSICAL REVIEW LETTERS 124, 133901 (2020)

Editors' Suggestion

Improved Phase Locking of Laser Arrays with Nonlinear Coupling

Simon Mahler ,1,* Matthew L. Goh ,1,2 Chene Tradonsky ,1 Asher A. Friesem ,1 and Nir Davidson1
1
Department of Physics of Complex Systems, Weizmann Institute of Science, Rehovot 761001, Israel
2
Department of Quantum Science, RSPhys, The Australian National University, Canberra ACT 0200, Australia

(Received 11 October 2019; accepted 24 February 2020; published 1 April 2020)

An arrangement based on a degenerate cavity laser for forming an array of nonlinearly coupled lasers
with an intracavity saturable absorber is presented. More than 30 lasers were spatially phase locked and
temporally Q switched. The arrangement with nonlinear coupling was found to be 25 times more sensitive
to loss differences and converged five times faster to the lowest loss phase locked state than with linear
coupling, thus providing a unique solution to problems that have several near-degenerate solutions.

DOI: 10.1103/PhysRevLett.124.133901

Phase locking of lasers corresponds to a state where all Nonlinear coupling with SA in the spatial domain was
the lasers have the same frequency and the same constant used to phase lock two lasers [3] and for selecting spatial
relative phase, leading to a coherent superposition of their modes [24,25]. Here we analyze and demonstrate that a SA
fields [1,2]. Accordingly, the total brightness of the lasers is can robustly and efficiently phase lock many lasers.
high and allows focusing of all the lasers to a sharp spot Specifically, inserting a SA at the far-field plane of a laser
[3–5]. Phase locking of lasers has been incorporated in array ensures that the phase locked state (that has sharp and
many investigations, including simulating spin systems strong intensity peaks there [3–6,19]) corresponds to the
[6–9], finding the ground-state solution of complex land- minimal loss state, to be selected by optical feedback (mode
scapes [6,9,10], observing dissipative topological defects competition) [18].
[11,12], and solving the phase retrieval problem [13]. We show experimentally and numerically that nonlinear
Phase locking of laser arrays can be achieved with coupling provides stable phase locking and is inherently
dissipative coupling that leads to a stable state of minimal more robust to alignment errors, aberrations and noise than
loss, which is the phase locked state [6,11,12]. Dissipative linear coupling. Moreover, nonlinear mode coupling pro-
coupling involves mode competition whereby modes of vided by the SA both in the spatial and temporal domains
different losses compete for the same gain [2,6,9,14]. Only can yield multiple coupled copies of the laser array
modes with the lowest loss survive and are amplified by the (corresponding to different longitudinal modes) that all
gain medium. Accordingly, by inserting amplitude and converge to the same minimal loss state. Hence, signifi-
phase linear optical elements into a laser cavity that cantly improving the ability of the coupled lasers to
minimize the loss of the phase locked states, it is possible distinguish between near-degenerate states.
to achieve phase locking with mode competition [2–4,15]. Our experimental arrangement with nonlinear coupling
While phase locking with such linear optical elements of laser arrays is based on a degenerate cavity laser (DCL)
has yielded many exciting results [4–6,9,11,13–18], it [4,5,26], schematically presented in Fig. 1(a). It was
suffers from inherent limitations. It is very sensitive to comprised of two flat mirrors where one served as a back
imperfections, such as positioning errors, mechanical mirror with high 99.5% reflectivity and the other as an
vibrations, thermal effects and other types of aberrations output coupler with 80% reflectivity. A mask of holes for
associated with these intracavity elements. Moreover, in forming the array of lasers was placed at the near-field
many cases, especially for spin simulations and computa- plane, adjacent to the output coupler. In our investigations,
tional problem solving [6,10,13,19], there are two or more the mask was a square array of holes of diameter 200 μm
states with nearly degenerate minimal loss that cannot be and period a ¼ 300 μm. The gain medium was Nd:YAG
distinguished from each other. crystal rod of 0.95 cm diameter and 10.9 cm length placed
In this Letter, we resort to nonlinear coupling between adjacent to the back mirror and optically pumped by quasi-
lasers by means of a saturable absorber (SA). A SA is a CW 100 μs pulsed flash lamps operating at 1 Hz flashing
nonlinear optical element that block light until it saturates, pulse rate, so operating wavelength is λ ¼ 1064 nm.
where its optical loss decreases sharply [20]. It can thus Between the mirrors, two spherical (Fourier) lenses of
affect the temporal modes within the laser so as to obtain focal lengths f ¼ 20 cm and diameters 5.08 cm formed a
passive Q switching and (longitudinal) mode locking, for 4f telescope configuration. Due to the 4f telescope, each
generating short pulses and high output peak powers [1] hole in the mask was precisely imaged onto itself after a
and studying nonlinear laser dynamics [1,3,21–23]. cavity round-trip, to obtain an independent laser.

0031-9007=20=124(13)=133901(5) 133901-1 © 2020 American Physical Society

 PHYSICAL REVIEW LETTERS 124, 133901 (2020)

The nonlinear coupling between the lasers was achieved mechanism of the lasers by the SA. Even more lasers can be
with a Cr:YAG saturable absorber, having initial trans- phase locked with the SA, but with a somewhat lower
mission of 45%, which was inserted in the far-field quality [27]. We also performed numerical simulations to
(Fourier) plane midway between the two lenses of the support our experimental results. The simulations were
4f telescope [4,5]. For comparison, we also performed performed by combining the Fox-Li algorithm [30] and the
experiments with linear coupling by removing the SA and Gerchberg-Saxton algorithm [31] to obtain a combined
displacing the output coupler by half of the Talbot length algorithm [5]. As evident in Fig. 2, the simulated far-field
from the mask, such that the round-trip distance between intensity distributions are in good agreement with the
them (Talbot distance) is equal to the Talbot length ZT ¼ experimental ones, indicating that the SA phase locked
ð2a2 =λÞ and the cavity length becomes 4f þ ZT =2 [5] [27]. the lasers in the in phase state.
First, we detected the time evolution of the laser For some pump pulse realizations, other phase locked
array output power without and with the SA. Figure 1(b) states such as the out of phase state [3–5] can occur. The
shows the results without the SA where the lasing other phase locked states also have sharp peaks in their far-
pulse duration was 200 μs with complicated strong oscil- field intensity distribution and similarly minimize loss.
lations. Figure 1(c) shows the results with the SA where the Some typical results are presented in Fig. 3, showing
lasing pulse duration was reduced to 100 ns indicating experimental far-field intensity distributions with a SA
temporal Q switching [1]. The total energy in both cases for different pump pulse realizations. Figures 3(a) and 3(b)
was similar.
show the in phase and out of phase states, and Fig. 3(c)
Next, we characterized the spatial coherence between the
shows a coexistence state of the in phase and out of phase
lasers by measuring the near-field and far-field intensity
states. To explain such a coexistence state, we note that
distributions without and with the SA. The lasers were
operated close to their lasing threshold. As evident in each laser in the array contains several hundreds of
Fig. 2, the near-field intensity distributions of the square temporal (longitudinal) modes [5,6]. Each of these longi-
array of lasers are essentially the same, while the far-field tudinal modes corresponds to a different realization of the
intensity distributions differ dramatically, indicating differ- spatially coupled lasers so phase locking could be either in
ent spatial coherences. Specifically, the broad Gaussian in the in phase state or in the out of phase state. In the
the far-field intensity distribution without the SA in coexistence state, part of the longitudinal modes phase
Fig. 2(a) indicates no phase relation between the different locked in phase and part out of phase [5,6].
lasers in the array [3–5].
On the other hand, the sharp peaks in the far-field
intensity distribution with the SA in Fig. 2(b) indicate in
phase locking of most, if not all, of the 30 lasers in the array
[3–5]. These high intensity peaks increase the saturation of
the SA and minimize loss. This minimal nonlinear loss
combined with mode competition explain the phase locking

FIG. 2. Experimental and simulated near-field and far-field

FIG. 1. Degenerate cavity laser (DCL) arrangement and ex- intensity distributions of the DCL arrangement. (a) Without and
perimental temporal evolution of the laser array intensity. (b) with a SA in the far-field plane. The broad Gaussian far-field
(a) DCL arrangement with a mask of holes in the near-field intensity distribution without the SA indicates that there is no
plane so as to form an array of lasers and a saturable absorber phase relation between the lasers whereas the sharp peaks with
(SA) in the far-field plane so as to non-linearly couple them. the SA indicate near-perfect phase locking of the entire array. As
(b) Time evolution of the laser array intensity without the SA. evident, there is good agreement between the experimental results
(c) Time evolution with the SA indicating temporal Q switching. and the numerical simulations.

133901-2
 PHYSICAL REVIEW LETTERS 124, 133901 (2020)

FIG. 3. Typical experimental far-field intensity distributions for

different pump pulse realizations. (a) In phase state with SA,
(b) out of phase state with SA and (c) coexistence state of in phase
and out of phase states with SA. (d) Distribution of the number of
states for 50 different pump pulse realizations, top row: linear
coupling (Talbot diffraction), bottom row: nonlinear coupling
(SA) near lasing threshold.

We determined the likelihood of the coexistence state

when using either linear or nonlinear coupling. The results
are presented in Fig. 3(d), for 50 different pump pulse
realizations where all the near-field intensity distributions
were the same. For the linear coupling (by Talbot diffrac-
tion with Talbot distance ZT for which the two phase locked
states are exactly degenerated), we always observed coex-
istence states only (top row). Such behavior can be easily
understood: for linear coupling, the different longitudinal
modes act as an ensemble of independent realizations, FIG. 4. The effect of the SA on the convergence to the lowest
where each can have a different phase locked state. The loss phase locked state. (a) Calculated losses of the in phase and
probability that all of them have the same state is expo- out of phase states as a function of Talbot distance in the range
nentially small. For linear coupling, the loss of the ½1to1.2ZT . The SA increases the loss of the out of phase state by
coexistence state is minimal, similar to that of in phase ∼0.5% as compared to the in phase state thereby shifting the
degeneracy distance between them from 1ZT . to 1.05ZT . (b) Mea-
or out of phase state.
sured relative occurrence of the in phase, out of phase, and
For nonlinear coupling (with SA), we found that the
coexistence states as a function of Talbot distance. (c) Measured
likelihood of the coexistence state is completely suppressed
expectation value of the in phase and out of phase states as a
near the lasing threshold and all the longitudinal modes
function of Talbot distance. (d) Typical far-field intensity dis-
choose the same phase locked state (bottom row). Such
tribution for single realizations and averaged distribution over the
suppression can be explained by noting that in the
100 realizations at 1.05ZT .
coexistence state, there are many far-field peaks whose
intensity is relatively low and saturate less the SA, thereby
increasing the loss. In addition, the nonlinear coupling
between longitudinal modes provided by the SA forces the when an additional state (e.g., in phase state) with nearly
longitudinal modes to have the same phase locked state. As identical but slightly higher loss is present. The effects of
a result, near lasing threshold where mode competition is the SA on the convergence to the lowest loss phase locked
the strongest, a single phase locked state is enforced state are presented in Fig. 4. The linear losses of the in
and the coexistence state is suppressed [32]. We also phase and out of phase states were controlled by varying
found that high above the lasing threshold, the likelihood the Talbot distance (round-trip distance between the near-
of the coexistence state was not completely suppressed field mask and the output coupler) [5].
although it is lower with nonlinear coupling than with Using an eigenvalues modal analysis [33], we calculated
linear. the losses, without and with a SA, of the in phase and out of
Next, we investigated whether adding nonlinear coupling phase states for different Talbot distances in the range
to linearly coupled lasers can improve the convergence to ½1to1.2ZT . The results are presented in Fig. 4(a). For linear
the lowest loss phase locked state (e.g., out of phase state), coupling, there is a degeneracy between the two phase

133901-3
 PHYSICAL REVIEW LETTERS 124, 133901 (2020)

locked states at 1ZT . The difference in loss between the two of phase to in phase state, which is ≃5 times sharper with
states increases as the Talbot distance increases, where the nonlinear coupling.
loss of the out of phase state becomes significantly lower Finally, Fig. 4(d) shows experimental far-field intensity
[34]. When adding nonlinear coupling, the calculated distributions at 1.05ZT for two typical realizations and the
losses of the in phase state are slightly reduced (by averaged distribution over the 100 realizations, for both
∼0.5%) compared with those of the out of phase state linear and nonlinear coupling. For linear coupling, the
for all Talbot distances, shifting the degeneracy distance distributions for all the different single realizations are
from 1ZT to about 1.04ZT [Fig. 4(a) right] [35]. identical (and hence also is the averaged distribution) and
We measured the relative occurrence of each phase correspond to the coexistence state, indicating that the
locked state (in phase, out of phase, or coexistence state) longitudinal modes are uncoupled. For nonlinear coupling,
as a function of the Talbot distance for 100 different the distribution for the single realizations correspond either
pump pulse realizations [36]. The results are presented in to the in phase or out of phase state (no coexistence state),
Fig. 4(b). For linear coupling, the minimal loss phase indicating that the longitudinal modes are coupled and the
locked state (i.e., out of phase state) with ≥ 90%
averaged distribution reveals both states. We also repeated
occurrence rate is achieved only at distances larger than
the experiments for a larger range of Talbot distances and
1.15ZT where the loss difference between in phase and
numerically simulated the far-field intensity distribu-
out of phase states is large [≃5% in Fig. 4(a) left]. For
tions [27].
smaller loss difference, the coexistence state is dominant,
We calculated the loss functional that is minimized by
indicating that some of the longitudinal modes select the
the lasers at the vicinity of its global minimum [27]. Loss
wrong phase locked state.
functionals for nonlinear coupling were theoretically inves-
For the nonlinear coupling, the occurrence of the
tigated also for optical parametric oscillators and non-
coexistence state is negligible (occurs only around the
equilibrium condensates (polariton arrays) [9,19]. In
degeneracy distance 1.04ZT with a very small probabil-
polariton arrays, the nonlinear coupling introduced a phase
ity), confirming that with nonlinear coupling all longi-
lag in the system that shifted or destabilized the stationary
tudinal modes select the same phase locked state (as
point, with improved convergence to the ground-state
already noted in Fig. 3 without Talbot diffraction). As the
solution [19].
Talbot distance is varied across the degeneracy distance
We also showed analogy with strongly coupled Ising
1.04ZT , a sharp transition occurs between the in phase
spins that produce significant magnetization even for
and the out of phase states. The minimal loss state with
magnetic energy smaller than their thermal energy [27].
≥ 90% occurrence rate is achieved at Talbot distances
To conclude, we showed that an array of 30 lasers can be
1.03 and 1.06ZT when the difference in loss is ≃0.2%
phase locked efficiently and robustly, by resorting to
(≃25 times smaller than for linear coupling). We also
nonlinear coupling with a saturable absorber in the far-
found that the results for the nonlinear coupling are more
field. The nonlinear coupling was found to significantly
robust and repeatable [27].
The method used to obtain the results in Fig. 4(b) only improve the ability of the lasers to converge to the correct
minimal loss phase locked state, that is mapped to the
provides the relative occurrence of each phase locked state
ground state of the classical XY spin Hamiltonian [6]. It is
and does not provide the ratio of intensities between the in
25 times more sensitive to differences in loss with five
phase and out of phase states. Figure 4(c) shows the
times faster convergence to the lowest loss state than with
expectation value of the in phase and out of phase states
linear coupling. The nonlinear coupling forces all longi-
as a function of the Talbot distance (obtained by summing
tudinal modes of the lasers to have the same phase
the far-field intensity distributions of all 100 realizations
locked state.
and calculating the expectation intensity value of each
We also showed that nonlinear coupling inherently
phase locked state).
improves the phase locking of many lasers. Specifically,
For nonlinear coupling, the results in Figs. 4(b) and 4(c)
reducing the sensitivity to misalignment errors, mechanical
have the same behavior (simply because there are only in
vibrations, thermal effects, and various aberrations.
phase or out of phase states with no coexistence state). For
Moreover, nonlinear coupling also improves the phase
linear coupling, the minimal loss state with ≥ 90% expect-
locking by converging to a single (minimal loss) phase
ation value is achieved again only for Talbot distances
locking state. We plan to study the effect of nonlinear
larger than 1.15ZT , where the difference in loss is large
coupling on the time dynamics of the degenerate cavity
(∼5%). Even “majority selection” of the longitudinal
laser after replacing the flash lamps pumping by diode
modes (expectation value above 50%) fails for Talbot
pumping.
distances smaller than 1.08ZT , where the difference in loss
is ≃1.3%. The improvement of nonlinear coupling over The authors wish to acknowledge the Israel Science
linear coupling for finding the lowest loss state can also be Foundation and the Israeli Planning and Budgeting
quantified by using the slope of the transition from the out Committee Fellowship Program for their support.

133901-4
 PHYSICAL REVIEW LETTERS 124, 133901 (2020)

* [19] N. Stroev and N. G. Berloff, arXiv:1910.00842.

[email protected]
[20] M. Hercher, Appl. Opt. 6, 947 (1967).
[1] A. E. Siegman, Lasers (University Science Books, Mill
[21] A. K. Wojcik, N. Yu, L. Diehl, F. Capasso, and A. A.
Valey, California, 1986), 1st ed., Vol. 1, p. 1.
Belyanin, Opt. Eng. 49, 1 (2010).
[2] A. F. Glova, Quantum Electron. 33, 283 (2003).
[22] H. G. Winful and D. T. Walton, Opt. Lett. 17, 1688 (1992).
[3] F. Kong, L. Liu, C. Sanders, Y. C. Chen, and K. K. Lee,
[23] E. Lacot and F. Stoeckel, J. Opt. Soc. Am. B 13, 2034
Appl. Phys. Lett. 90, 151110 (2007).
(1996).
[4] S. Mahler, C. Tradonsky, R. Chriki, A. A. Friesem, and N.
[24] X. Chen, Y. Bai, M. Jiang, L. Li, Y. Zhou, H. Wang, Z. Ren,
Davidson, OSA Continuum 2, 2077 (2019).
and J. Bai, Appl. Phys. Express 5, 122701 (2012).
[5] C. Tradonsky, V. Pal, R. Chriki, N. Davidson, and A. A.
[25] A. A. Ishaaya, N. Davidson, and A. A. Friesem, Opt.
Friesem, Appl. Opt. 56, A126 (2017).
Express 13, 4952 (2005).
[6] M. Nixon, E. Ronen, A. A. Friesem, and N. Davidson, Phys.
[26] J. A. Arnaud, Appl. Opt. 8, 189 (1969).
Rev. Lett. 110, 184102 (2013).
[27] See the Supplemental Material at http://link.aps.org/
[7] S. Tamate, Y. Yamamoto, A. Marandi, P. McMahon, and S.
supplemental/10.1103/PhysRevLett.124.133901 for further
Utsunomiya, arXiv:1608.00358.
[8] N. G. Berloff, M. Silva, K. Kalinin, A. Askitopoulos, J. D. calculated and experimental results, and technical details,
Töpfer, P. Cilibrizzi, W. Langbein, and P. G. Lagoudakis, which includes Refs. [1,5,6,28,29].
Nat. Mater. 16, 1120 (2017). [28] H. E. Stanley, Introduction to Phase Transitions and Criti-
[9] Y. Yamamoto, K. Aihara, T. Leleu, K. Kawarabayashi, S. cal Phenomena (Oxford University Press, New York, 1971).
Kako, M. Fejer, K. Inoue, and H. Takesue, npj Quantum Inf. [29] J. Strecka and M. Jascur, Acta Phys. Slovaca 65, 235 (2015).
3, 49 (2017). [30] A. G. Fox and T. Li, Bell Syst. Tech. J. 40, 453 (1961).
[10] A. Marandi, Z. Wang, K. Takata, R. L. Byer, and Y. [31] R. W. Gerchberg and W. O. Saxton, Optik (Stuttgart) 35,
Yamamoto, Nat. Photonics 8, 937 (2014). 237 (1972).
[11] V. Pal, C. Tradonsky, R. Chriki, A. A. Friesem, and N. [32] The in phase state is the more dominant because it loss is
Davidson, Phys. Rev. Lett. 119, 013902 (2017). slightly smaller than the out of phase state [as in Fig. 4(a)].
[12] S. Mahler, V. Pal, C. Tradonsky, R. Chriki, A. A. Friesem, [33] D. Mehuys, W. Streifer, R. G. Waarts, and D. F. Welch, Opt.
and N. Davidson, J. Phys. B 52, 205401 (2019). Lett. 16, 823 (1991).
[13] C. Tradonsky, I. Gershenzon, V. Pal, R. Chriki, A. A. [34] To keep our system simple with only two low loss phase
Friesem, O. Raz, and N. Davidson, Sci. Adv. 5, eaax4530 locked states, we suppressed two additional low loss phase
(2019). locked states, see the Supplemental Material [27].
[14] V. Eckhouse, M. Fridman, N. Davidson, and A. A. Friesem, [35] The eigenvalues calculation uses saturation function for the
Phys. Rev. Lett. 100, 024102 (2008). SA that contains four parameters (SA’s intensities and
[15] Y. Zhou, L. Liu, C. Etson, Y. Abranyos, A. Padilla, and Y. C. transmissions thresholds) that are adjusted to set the
Chen, Appl. Phys. Lett. 84, 3025 (2004). degeneracy distance to agree with the Talbot distance where
[16] J. Guillot, A. Desfarges-Berthelemot, V. Kermène, and A. the out of phase state intersects the in phase state in
Barthélémy, Opt. Lett. 36, 2907 (2011). Figs. 4(b) and 4(c) that is about 1.04ZT .
[17] S. Sivaramakrishnan, W. Z. Chang, A. Galvanauskas, and [36] We analyzed the far-field intensity distribution of each
H. G. Winful, IEEE J. Quantum Electron. 51, 1 (2015). realization and set an intensity threshold value (≈15% of
[18] M. Nixon, O. Katz, E. Small, Y. Bromberg, A. A. Friesem, the maximal intensity) for determining whether the in phase
Y. Silberberg, and N. Davidson, Nat. Photonics 7, 919 or out of phase or coexistence state is present.
(2013).

133901-5