Pith. sign in

REVIEW 1 major objections 4 minor 50 references

In-situ Adjoint Wave Control in Reconfigurable Non-Hermitian Nonlinear Systems

T0 review · 1 major / 4 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 In a wave-chaotic network with one nonlinear defect, the hardware itself can measure the full gradient of an objective from one forward and one adjoint pass, enabling optimization without a digital twin.

desk verdict Genuinely new phase-engineered adjoint protocol for nonlinear wave systems; the model-free claim needs a gradient sanity check and a self-calibration path for f'(I). read the letter →

arxiv 2608.13503 v1 pith:4TOLCGYJ submitted 2026-08-13 physics.optics cond-mat.dis-nneess.SP

John Guillamon , William Tuxbury , Cheng-Zhen Wang , Owen Miller , Zin Lin , Tsampikos Kottos This is my paper · ORCID
classification physics.opticscond-mat.dis-nneess.SP
keywords in-situadjointoptimizationnonlinearwavecontrolnon-Hermitianscatteringwave-chaoticnetworkscoherentperfectabsorptionasymmetrictransportphysicalmultipath
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 establishes an in-situ adjoint protocol—called iNAP—for optimizing nonlinear, non-reciprocal, multipath wave systems directly in hardware. The claim is that a single forward measurement plus one carefully phase-engineered adjoint measurement gives the complete gradient of any objective with respect to all tunable parameters, without solving the nonlinear equation numerically or building a digital twin. The physical system performs both propagations, so losses, detunings, and scattering complexity are automatically included. Demonstrations on a diode-loaded coaxial-cable loop achieve on-demand power splitting, near-perfect coherent absorption, and asymmetric transport; a 21-vertex in-silico network shows the same protocol scaling to hundreds of bonds. A sympathetic reader would care because this points toward self-optimizing wave devices in complex, partially unknown environments.

What carries the argument

The load-bearing mechanism is the pump-probe linearization of the nonlinear steady-state equation. A strong forward excitation $b$ sets the operating point $\boldsymbol{\Phi}$; injecting the phase-engineered source $b_{\rm adj}=e^{-i\theta^*}b-\epsilon e^{i\theta^*}(\partial g/\partial\boldsymbol{\Phi})^T$ produces a weak perturbation whose first-order equation matches the adjoint equation, provided the phase offset is set to $\theta^*=\frac{1}{2}\arg[f'(I_{n_0})\Phi_{n_0}^2]$. The adjoint field is extracted as the normalized field difference $(\boldsymbol{\Psi}-e^{-i\theta^*}\boldsymbol{\Phi})/\epsilon$, and the gradient follows from $dg/dp=\partial g/\partial p+2\,\mathrm{Re}\{\boldsymbol{\Lambda}^T(\partial F/\partial p)\}$. The same nonlinear defect serves both as the source of non-reciprocity and as the local element whose phase-conjugation invariance makes the adjoint emulation exact; the multipath network repeatedly returns the wave to the defect, enriching the input-output map without extra hardware.

What would settle it

Run iNAP on a loop whose nonlinear defect has an unknown, uncharacterized response, set $\theta^*$ from the measured steady-state field alone, and compare the resulting gradient with a brute-force finite-difference gradient obtained by perturbing each parameter; if the two disagree systematically, the phase-condition step is carrying unacknowledged model information. A second test is to increase the probe amplitude $\epsilon$ until the measured $\boldsymbol{\Lambda}$ deviates from the true Jacobian-vector product, locating the linear-response range the protocol implicitly assumes.

Watch

Extended reading notes

Core claim

The central discovery is that the adjoint equation of a nonlinear scattering problem can be physically emulated by the same hardware that runs the forward problem, even when the nonlinear defect breaks reciprocity. The trick is to linearize the nonlinear response around the operating point set by a strong forward pump, then inject a weak adjoint probe whose global phase is rotated so that the field at the defect is invariant under complex conjugation. With the phase offset $\theta^*=\frac{1}{2}\arg[f'(I_{n_0})\Phi_{n_0}^2]$ chosen from the measured steady-state field, the perturbation equation coincides with the adjoint equation; measuring the field difference $\boldsymbol{\Lambda}=(\boldsymbol{\Psi}-e^{-i\theta^*}\boldsymbol{\Phi})/\epsilon$ then yields the adjoint field, and Eq. (3) assembles $dg/dp$ from that single adjoint field plus the forward field. No Jacobian entries need be computed one by one, and no time-reversal or reciprocity assumption is needed.

Load-bearing premise

The protocol requires knowing the complex derivative $f'(I_{n_0})$ of the nonlinear defect's response at the operating point, because the phase offset $\theta^*$ is built from it; in the experiment that derivative comes from a fitted temporal-coupled-mode model of the defect, so in a genuinely unknown environment the protocol is not yet fully model-free unless $f'$ can be measured locally.

Editorial extensions

If this is right

  • Any number of tunable parameters can be optimized with just two physical measurements—one forward and one adjoint—so the cost of gradient evaluation no longer grows with the dimension of the control space.
  • Non-reciprocal nonlinear platforms, previously ruled out for in-situ adjoint optimization, become trainable in hardware without time-reversal symmetry or reciprocity.
  • The same experimental loop, with no hardware changes, can be reprogrammed on the fly for different objectives—power splitting, coherent perfect absorption, asymmetric transport—by changing only the objective used to build the adjoint source.
  • In the scale-up network, objective values reach $10^{-7}$ for power splitting, absorption $0.999$, an asymmetric transport ratio near $10^{27}$, and invisibility error $9\times10^{-3}$, supporting scalability to complex multipath geometries.
  • Because the full optimization loop runs in hardware, the simulation–reality gap from parasitic reflections, calibration errors, and losses is bypassed rather than corrected.

Reading between the lines

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

  • If the phase offset can be calibrated with a short local probe rather than a global fitted model, iNAP may work in fully unknown media, turning wave-chaotic hardware into a self-tuning component that needs no prior modeling at all.
  • The two-pass gradient readout is structurally the same operation as backpropagation in a physical neural network; a natural extension is to train networks with several nonlinear elements by applying the phase condition at each defect sequentially.
  • Multipath returns act like a random feature expansion of the single defect's nonlinearity, so the effective number of controllable degrees of freedom may grow faster than the number of physical parameters; this is testable by comparing how quickly objective error falls with iteration count as network size grows.
  • Applying iNAP when the nonlinearity is distributed, or when gain is present, would require the richer adjoint structure the paper names as future work; the same phase-conjugation invariance argument might then be recast as a generalized reciprocity condition.
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

1 major / 4 minor

Summary. The manuscript proposes an in-situ adjoint optimization protocol, iNAP, for nonlinear multipath wave systems with a single localized nonlinear defect. The idea is to use the physical system itself for both the forward and adjoint propagations: an adjoint source is formed by phase-rotating the forward excitation and superposing a weak perturbation proportional to the objective gradient, and the adjoint field is extracted from the field difference between two measurements. The authors derive the adjoint equations for a scalar nonlinear defect, implement the protocol on a microwave cable network with a diode-loaded resonator, and demonstrate convergence for power splitting, coherent perfect absorption, and asymmetric transport. They also report in-silico scale-up on a 21-vertex network. The central claim is that the gradient can be evaluated from measurements alone, without a digital twin or numerical solution of the nonlinear equation.

Significance. If correct, the protocol would be a valuable tool for gradient-based optimization of nonlinear complex wave systems, reducing gradient evaluation to two physical measurements per iteration and avoiding the simulation-reality gap. The paper is clearly written and includes a concrete hardware demonstration with careful component characterization. The derivation of the adjoint equations for the single-defect case is a useful contribution. However, the advertised model-free operation is not fully established: the enabling phase condition requires the complex derivative f'(I) of the local nonlinear response, which in the only implementation is taken from a fitted TCMT model, not from a model-free measurement. Moreover, the field difference defined in Eq. (6) appears to carry an uncompensated phase factor relative to the adjoint field of Eq. (2), so the gradient formula Eq. (3) may be biased as written. The lack of any finite-difference validation means the experimental convergence curves cannot rule out this bias.

major comments (1)
  1. [Physical Platform, Methods (TCMT fit)] The enabling phase condition theta* = (1/2)arg[f'(I_{n0})Phi_{n0}^2] requires knowledge of the complex derivative f'(I) of the nonlinear defect at the operating point. In the experiment, f'(I) is not measured locally; it is computed from the fitted TCMT model of Eq. (16) with parameters z0, z1, chi_s, and gamma_i^2 extracted from S-parameter fits (Methods). The Abstract and Results claim that the protocol operates 'without a digital twin' and is 'agnostic to the specific functional form of the nonlinearity', but the only implementation depends on a fitted model for exactly the quantity that sets the adjoint mapping. No model-free procedure for obtaining f' or self-calibrating theta* is described. The in-silico scale-up in Fig. 3 uses the same fitted nonlinearity, so it does not independently test the model-free premise. The authors should either demonstrate a local measurement of f' (for example, a small-signal pump-probe calibration at the defect) or substantially soften the claims about partially unknown environments.
minor comments (4)
  1. [General] The title on the first page, 'In-situ Nonlinear Adjoint Propagation for Multi-Path Wave Control', differs from the metadata title; please unify the title across all versions.
  2. [Discussion] The statement that iNAP 'self-calibrates to losses and detunings' is stronger than what is demonstrated, given the dependence of theta* on the fitted f'(I).
  3. [Eq. (10)] In the objective for power splitting, the denominator runs over alpha=1 to M (all output ports) while the sum in the numerator runs over alpha'=1 to M_tar; the notation should be clarified so the normalization is unambiguous.
  4. [Fig. 2] The axis labels in the panels of Fig. 2 are small and difficult to read; consider increasing the font size. The caption should also state that the in-silico curves use the same fitted TCMT parameters as the hardware model.

Circularity Check

1 steps flagged · score 3.0 of 10

The enabling phase offset θ★ requires the fitted derivative f′(I) of the diode model, so the claimed model-free, no-digital-twin gradient is not demonstrated; the core hardware measurement itself is nevertheless not constructed from the optimization target.

  1. fitted input called prediction [Results, paragraph after Eq. (4); Methods, 'Experimental Implementation and Characterization of the Nonlinear Vertex']
    "In practice, we measure the phase of the steady-state field Φ_{n0} at the nonlinear element and set θ★ = 1/2 arg[f′(I_{n0})Φ^2_{n0}]. ... Rather than assuming a particular nonlinear response, as in Ref. (40,41), our protocol remains agnostic to the specific functional form of the nonlinearity, treating f(|Φ_{n0}|^2) in full generality. ... The complex parameters z0, z1, χ together with the coupling strengths γ1, γ2, γ3 were extracted from direct fits to the measured scattering parameters of the nonlinear vertex (see Methods)."

    The central claim is that one forward and one adjoint measurement give the full gradient dg/dp 'without a digital twin' and while remaining 'agnostic' to the nonlinearity. But the measured difference Lambda in Eq. (6) is the true adjoint field only if the linearized forward equation (4) coincides with the adjoint equation (2), and that coincidence is enforced by the phase choice theta★ = (1/2)arg[f′(I_{n0})Phi^2_{n0}]. This requires the complex derivative f′(I) of the nonlinear defect at the operating point; the paper's only implementation obtains f′ from the fitted saturable TCMT model with parameters z0, z1, chi extracted from measured S-parameters. Thus the 'model-free' component of the gradient prediction is not actually model-free: it is conditioned on a fitted derivative.

full rationale

The derivation of Eqs. (1)-(3) and (9) is a standard adjoint calculation, and the experimental loop is genuinely in-situ in the sense that Phi and Psi are measured fields and Lambda is constructed from Eq. (6) rather than from a simulation of the full nonlinear equation. The optimization objectives are not fitted targets used to produce the gradient; the hardware provides the field responses, so the main loop is not circular by construction. The limitation is more specific: the phase condition theta★ = (1/2)arg[f′(I_{n0})Phi^2_{n0}] is load-bearing for the equivalence between the measured linearized perturbation and the adjoint field, and it requires f′(I). The paper obtains f′(I) from the fitted TCMT diode model (Eq. (16) and the fitted parameters in Methods), not from a model-free local measurement or self-calibration. Hence the claimed 'without a digital twin' and 'agnostic to the specific functional form' statements are overstated: the demonstration depends on a fitted derivative at exactly the enabling step. A biased f′ would bias the gradient through the phase offset, and the reported convergence curves alone cannot exclude that possibility. The in-silico scale-up validation also uses the same fitted nonlinearity, so it is a self-consistency check rather than an independent test of model freedom. No direct finite-difference comparison of the iNAP gradient is reported. The self-citations (Refs. 29, 32, 46) are prior independent publications describing the nonlinear vertex, the fitting procedure, and earlier linear in-situ adjoint work; they do not by themselves make the central derivation circular. Overall, the paper has real independent content, but its flagship model-free claim is partly underwritten by a fitted input, giving a moderate circularity score of 3.

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

The protocol depends on hardware parameters fitted to measured S-parameters and on the assumption that the nonlinear defect's derivative is known for setting the phase condition; the remaining assumptions are standard wave-scattering and TCMT modeling choices.

free parameters (5)
  • z0 = (-86.4 - i 59.2) MHz
    Complex linear response of the diode-loaded resonator, obtained by fitting measured S-parameters (Methods).
  • z1 = (-86.4 - i 50.0) MHz
    Saturable nonlinear response coefficient of the diode-loaded resonator from the same fit; its real part matches z0 in the text.
  • chi_s = (1.5 + i) x 10^9 (mW s)^-1
    Saturation parameter of the diode nonlinearity, extracted from S-parameter fits.
  • gamma_i^2 for i=1,2,3 = 62.5 MHz each
    Coupling strengths of the resonator to three kink antennas, extracted from fits.
  • cable complex refractive index = n_r=1.212, n_i ~ 2e-3
    Fitted to measured frequency-dependent transmission through a cable; enters k and all bond-length gradients.
assumptions (5)
  • domain assumption The nonlinear defect response depends only on the local intensity |a|^2 at the fundamental frequency; higher harmonics are negligible.
    Invoked in Methods (TCMT, higher harmonics are negligible) and in the global phase invariance used to rotate the adjoint source.
  • domain assumption The linear background operator H_L and projector P_n0 are symmetric, so the adjoint and forward linearized operators coincide once the anomalous term is real.
    Used to write Eq. (2) and to equate Eq. (4) with the adjoint; relies on a reciprocal linear network without magnetic bias.
  • domain assumption The weak probe remains in the linear-response regime around the pump operating point (epsilon to 0), so first-order perturbation theory suffices.
    Required by Eqs. (4)-(6); not quantified with a linearity check.
  • ad hoc to paper The complex derivative f'(I_n0) is known or measurable, and the global phase rotation theta* can be set accordingly.
    The agnostic-to-nonlinearity claim is not met; theta* uses f' from a fitted TCMT model (Methods).
  • domain assumption The steady-state field is single-valued at the operating point, with no bistability in the operating regime.
    Supplementary S1 notes multistability branches are not relevant for the fitted platform; otherwise the branch must be tracked by continuity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of In-situ Adjoint Wave Control in Reconfigurable Non-Hermitian Nonlinear Systems." pith.science (2026). https://pith.science/paper/4TOLCGYJ

@misc{pith2026260813503,
  author       = {Pith},
  title        = {Pith review of: In-situ Adjoint Wave Control in Reconfigurable Non-Hermitian Nonlinear Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4TOLCGYJ}},
  note         = {Machine review of arXiv:2608.13503}
}
read the original abstract

Complex multipath environments are usually avoided in wave-based information processing because repeated scattering creates many interfering propagation paths, obscuring controllability and generating extreme sensitivity to perturbations. The addition of nonlinear mechanisms fundamentally alters the wave-control landscape by breaking the superposition principle that underpins most wave-management strategies. Here, we show that these two apparent impediments -- multipath complexity and nonlinearity -- can instead be harnessed as key resources for physical optimization. We demonstrate an in-situ adjoint optimization protocol in a wave-chaotic platform incorporating a single localized nonlinear defect, in which the system itself performs both the forward and the adjoint propagations required for gradient evaluation. Recurrent multipath returns repeatedly expose the wave to the defect, producing from a minimal hardware a rich nonlinear input-output map with many pathway-mediated degrees of freedom. At the same time, a suitable adjoint excitation enables direct extraction of the sensitivities from measurements alone, without a digital twin or conventional numerical backpropagation. We experimentally validate the protocol on a minimal nonlinear multipath platform composed of incommensurate coaxial cables connected via T-junctions, one of which hosts a diode-loaded cavity. Our approach opens a route to adaptive wireless communications, imaging and analog intelligence in complex, partially unknown environments where conventional modeling is impractical.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 35 canonical work pages

  1. [1]

    Lagendijk, B

    A. Lagendijk, B. van Tiggelen, D. S. Wiersma, Fifty years of Anderson localization.Physics To- day62(8), 24–29 (2009), doi:10.1063/1.3206091,https://doi.org/10.1063/1.3206091. 22

    doi:10.1063/1.3206091 2009

  2. [2]

    H. Cao, A. P. Mosk, S. Rotter, Shaping the propagation of light in complex media.Nat. Phys. 18(9), 994–1007 (2022)

    work page 2022

  3. [3]

    Gigan, O

    S. Gigan, O. Katz, H. De Aguiar,et al., Roadmap on wavefront shaping and deep imaging in complex media.Journal of Physics: Photonics4(4), 042501 (2022)

    work page 2022

  4. [4]

    A. P. Mosk, A. Lagendijk, G. Lerosey, M. Fink, Controlling waves in space and time for imaging and focusing in complex media.Nature Photonics6, 283–292 (2012),https://doi.org/ 10.1038/nphoton.2012.88

    work page doi:10.1038/nphoton.2012.88 2012

  5. [5]

    Lerosey, J

    G. Lerosey, J. de Rosny, A. Tourin, M. Fink, Focusing Beyond the Diffraction Limit with Far- Field Time Reversal.Science315(5815), 1120–1122 (2007), doi:10.1126/science.1134824, https://www.science.org/doi/abs/10.1126/science.1134824

    work page doi:10.1126/science.1134824 2007

  6. [6]

    Bender,et al., Depth-targeted energy delivery deep inside scattering media.Nature Physics 18(3), 309–315 (2022)

    N. Bender,et al., Depth-targeted energy delivery deep inside scattering media.Nature Physics 18(3), 309–315 (2022)

    work page 2022

  7. [7]

    McIntosh,et al., Delivering broadband light deep inside diffusive media.Nat

    R. McIntosh,et al., Delivering broadband light deep inside diffusive media.Nat. Photon.pp. 1–7 (2024)

    work page 2024

  8. [8]

    C. W. Hsu, S. F. Liew, A. Goetschy, H. Cao, A. Douglas Stone, Correlation-enhanced control of wave focusing in disordered media.Nat. Phys.13(5), 497–502 (2017)

    work page 2017

Show all 50 references
  1. [9]

    W. Wan,et al., Time-Reversed Lasing and Interferometric Control of Absorption.Science 331(6019), 889–892 (2011), doi:10.1126/science.1200735,https://www.science.org/ doi/abs/10.1126/science.1200735

    2011 doi

  2. [10]

    L. Chen, T. Kottos, S. M. Anlage, Perfect absorption in complex scattering systems with or without hidden symmetries.Nat. Comm.11(1), 5826 (2020)

    2020

  3. [11]

    Y. V. Fyodorov, S. Suwunnarat, T. Kottos, Distribution of zeros of the S-matrix of chaotic cavities with localized losses and coherent perfect absorption: non-perturbative results.Journal of Physics A: Mathematical and Theoretical50(30), 30LT01 (2017)

    2017

  4. [12]

    Baranov, A

    D. Baranov, A. Krasnok, T. e. a. Shegai, Coherent perfect absorbers: linear control of light with light.Nat Rev Mater2, 17064 (2017), doi:10.1038/natrevmats.2017.64. 23

    2017 doi

  5. [13]

    Bender, A

    N. Bender, A. Goetschy, C. W. Hsu,et al., Coherent enhancement of optical remission in diffusive media.Proc. Natl. Acad. Sci. U.S.A.119(2022)

    2022

  6. [14]

    S. M. Popoff,et al., Measuring the Transmission Matrix in Optics: An Approach to the Study and Control of Light Propagation in Disordered Media.Phys. Rev. Lett.104, 100601 (2010), doi:10.1103/PhysRevLett.104.100601,https://link.aps.org/doi/10. 1103/PhysRevLett.104.100601

    2010 doi

  7. [15]

    Pl ¨oschner, T

    M. Pl ¨oschner, T. Tyc, T. Cizm´ar, Seeing through chaos in multimode fibres.Nature Photonics 9, 529 (2015)

    2015

  8. [16]

    O. Katz, E. Small, Y. Bromberg, Y. Silberberg, Focusing and compression of ultrashort pulses through scattering media.Nature Photonics5(6), 372–377 (2011), doi:10.1038/nphoton.2011. 72,https://www.nature.com/articles/nphoton.2011.72

    2011 doi

  9. [17]

    Wang,et al., Guiding waves through chaos: Universal bounds for targeted mode transport

    C.-Z. Wang,et al., Guiding waves through chaos: Universal bounds for targeted mode transport. Science Advances12(2026)

    2026

  10. [18]

    Jiang,et al., Coherent control of chaotic optical microcavity with reflectionless scattering modes.Nat

    X. Jiang,et al., Coherent control of chaotic optical microcavity with reflectionless scattering modes.Nat. Phys.20(1), 109–115 (2024)

    2024

  11. [19]

    del Hougne, F

    P. del Hougne, F. Lemoult, M. Fink, G. Lerosey, Spatiotemporal Wave Front Shaping in a Microwave Cavity.Phys. Rev. Lett.117, 134302 (2016), doi:10.1103/PhysRevLett.117.134302, https://link.aps.org/doi/10.1103/PhysRevLett.117.134302

    2016 doi

  12. [20]

    H. Q. Yang,et al., Adaptively programmable metasurface for intelligent wireless communi- cations in complex environments.Nature Communications16(1), 6070 (2025), doi:10.1038/ s41467-025-61409-6,https://www.nature.com/articles/s41467-025-61409-6

    2025

  13. [21]

    Wetzstein, A

    G. Wetzstein, A. Ozcan, S. Gigan,et al., Inference in artificial intelligence with deep optics and photonics.Nature588(7836), 39–47 (2020)

    2020

  14. [22]

    Silva,et al., Performing Mathematical Operations with Metamaterials.Science343(6167), 160–163 (2014), doi:10.1126/science.1242818

    A. Silva,et al., Performing Mathematical Operations with Metamaterials.Science343(6167), 160–163 (2014), doi:10.1126/science.1242818. 24

    2014 doi

  15. [23]

    Zangeneh-Nejad, D

    F. Zangeneh-Nejad, D. L. Sounas, A. Al` u, R. Fleury, Analogue Computing with Metamaterials. Nature Reviews Materials6, 207–225 (2021), doi:10.1038/s41578-020-00243-2

    2021 doi

  16. [24]

    D. R. Solli, B. Jalali, Analog Optical Computing.Nature Photonics9, 704–706 (2015), doi: 10.1038/nphoton.2015.208

    2015 doi

  17. [25]

    S. E. Skipetrov, R. Maynard, Diffuse Waves in Nonlinear Disordered Media, inWave Scat- tering in Complex Media: From Theory to Applications, B. A. van Tiggelen, S. E. Skipetrov, Eds. (Springer, Dordrecht), vol. 107 ofNATO Science Series II: Mathematics, Physics and Chemistry, ...

    2003 doi

  18. [26]

    Suntharalingam, L

    A. Suntharalingam, L. Fern ´andez-Alc´azar, R. Kononchuk, T. Kottos, Noise resilient exceptional-point voltmeters enabled by oscillation quenching phenomena.Nature Commu- nications14, 5515 (2023), doi:10.1038/s41467-023-41189-7,https://doi.org/10.1038/ s41467-023-41189-7

    2023 doi

  19. [27]

    D.-Y. Chen, L. Dong, Q.-A. Huang, Inductor-capacitor passive wireless sensors using nonlinear parity-time symmetric configurations.Nature Communications15, 9312 (2024), doi:10.1038/ s41467-024-53655-x

    2024

  20. [28]

    Suntharalingam,et al., Symmetry-violation-driven hysteresis loops as measurands for noise- resilient sensors.Physical Review Applied23, 064043 (2025), doi:10.1103/zm1g-xnn5

    A. Suntharalingam,et al., Symmetry-violation-driven hysteresis loops as measurands for noise- resilient sensors.Physical Review Applied23, 064043 (2025), doi:10.1103/zm1g-xnn5

    2025 doi

  21. [29]

    C.-Z. Wang, J. Guillamon, W. Tuxbury, U. Kuhl, T. Kottos, Nonlinearity-induced scattering zero degeneracies for spectral management of coherent perfect absorption in complex systems.Phys. Rev. Appl.22, 064093 (2024), doi:10.1103/PhysRevApplied.22.064093,https://link.aps. org/d...

    2024 doi

  22. [30]

    Galiffi, A

    E. Galiffi, A. Harwood, S. Vezzoli, et al., Optical coherent perfect absorption and amplification in a time-varying medium.Nat. Photon.20, 163–169 (2026),https://doi.org/10.1038/ s41566-025-01833-8

    2026

  23. [31]

    Go ¨ıcoechea, J

    A. Go ¨ıcoechea, J. H¨ upfl, S. Rotter, F. m. c. Sarrazin, M. Davy, Detecting and Focusing on a Nonlinear Target in a Complex Medium.Phys. Rev. Lett.134, 25 183802 (2025), doi:10.1103/PhysRevLett.134.183802,https://link.aps.org/doi/10. 1103/PhysRevLett.134.183802

    2025 doi

  24. [32]

    C.-Z. Wang, R. Kononchuk, U. Kuhl, T. Kottos, Loss-Induced Violation of the Fundamen- tal Transmittance-Asymmetry Bound in Nonlinear Complex Wave Systems.Phys. Rev. Lett. 131, 123801 (2023), doi:10.1103/PhysRevLett.131.123801,https://link.aps.org/doi/ 10.1103/PhysRevLett.131.123801

    2023 doi

  25. [33]

    T. W. Hughes, M. Minkov, I. A. D. Williamson, S. Fan, Adjoint Method and Inverse Design for Nonlinear Nanophotonic Devices.ACS Photonics5(12), 4781–4787 (2018), doi:10.1021/ acsphotonics.8b01522,https://doi.org/10.1021/acsphotonics.8b01522

    2018 doi

  26. [34]

    L. G. Wright, T. Onodera, M. M. Stein,et al., Deep physical neural networks trained with backpropagation.Nature601(7894), 549–555 (2022)

    2022

  27. [35]

    Y. Shen, N. C. Harris, S. Skirlo,et al., Deep learning with coherent nanophotonic circuits. Nature photonics11(7), 441–446 (2017)

    2017

  28. [36]

    Momeni, B

    A. Momeni, B. Rahmani, B. Sceller, G. Wright, Logan,et al., Training of physical neural networks.Nature645(2025),https://doi.org/10.1038/s41586-025-09384-2

    2025 doi

  29. [37]

    Momeni, B

    A. Momeni, B. Rahmani, M. Mall ´ejac, P. del Hougne, R. Fleury, Backpropagation-Free Training of Deep Physical Neural Networks.Science382(6676), 1297–1303 (2023), doi: 10.1126/science.adi8474

    2023 doi

  30. [38]

    Wang,et al., Image Sensing with Multilayer Nonlinear Optical Neural Networks.Nature Photonics17, 408–415 (2023), doi:10.1038/s41566-023-01170-8

    T. Wang,et al., Image Sensing with Multilayer Nonlinear Optical Neural Networks.Nature Photonics17, 408–415 (2023), doi:10.1038/s41566-023-01170-8

    2023 doi

  31. [39]

    ´A. Papp, W. Porod, G. Csaba, Nanoscale Neural Network Using Non-Linear Spin-Wave Inter- ference.Nature Communications12, 6422 (2021), doi:10.1038/s41467-021-26711-z

    2021 doi

  32. [40]

    X. Guo, T. D. Barrett, W. Z. M., A. I. Lvovsky, Backpropagation through nonlinear units for the all-optical training of neural networks.Photonics Research9(2021)

    2021

  33. [41]

    Spall, X

    J. Spall, X. Guo, A. I. Lvovsky, Training neural networks with end-to-end optical backpropa- gation.Advanced Photonics7(2025),https://doi.org/10.1117/1.AP.7.1.016004. 26

    2025 doi

  34. [42]

    T. W. Hughes, M. Minkov, Y. Shi, S. Fan, Training of photonic neural networks through in situ backpropagation and gradient measurement.Optica5(7), 864–871 (2018)

    2018

  35. [43]

    S. Pai,et al., Experimentally realized in situ backpropagation for deep learning in photonic neural networks.Science380(6643), 398–404 (2023), doi:10.1126/science.ade8450,https: //www.science.org/doi/10.1126/science.ade8450

    2023 doi

  36. [44]

    L ´opez-Pastor, F

    V. L ´opez-Pastor, F. Marquardt, Self-Learning Machines Based on Hamiltonian Echo Back- propagation.Phys. Rev. X13, 031020 (2023), doi:10.1103/PhysRevX.13.031020,https: //link.aps.org/doi/10.1103/PhysRevX.13.031020

    2023 doi

  37. [45]

    Dal Cin, F

    N. Dal Cin, F. Marquardt, C. C. Wanjura, Training nonlinear optical neural networks with Scattering Backpropagation.arXiv preprint arXiv:2508.11750(2025), doi:10.48550/arXiv. 2508.11750,https://arxiv.org/abs/2508.11750

    2025 doi arXiv

  38. [46]

    Guillamon, C.-Z

    J. Guillamon, C.-Z. Wang, Z. Lin, T. Kottos, In-situ physical adjoint computing in multiple- scattering electromagnetic environments for wave control.Nature Communications16(2025), https://doi.org/10.1038/s41467-025-66385-5

    2025 doi

  39. [47]

    Kottos, U

    T. Kottos, U. Smilansky, Chaotic Scattering on Graphs.Phys. Rev. Lett.85, 968–971 (2000), doi:10.1103/PhysRevLett.85.968,https://link.aps.org/doi/10.1103/PhysRevLett. 85.968

    2000 doi

  40. [48]

    Kottos, H

    T. Kottos, H. Schanz, Quantum graphs: a model for quantum chaos.Physica E9(3), 523–530 (2001)

    2001

  41. [49]

    Kuchment, Quantum graphs and their applications.Special issue of Waves in Random Media 14(1) (2004)

    P. Kuchment, Quantum graphs and their applications.Special issue of Waves in Random Media 14(1) (2004)

    2004

  42. [50]

    C. E. Baum,Electromagnetic Topology for the Analysis and Design of Complex Elec- tromagnetic Systems(Springer Netherlands, Dordrecht), pp. 467–547 (1986), doi:10.1007/ 978-94-017-0445-8-17,https://doi.org/10.1007/978-94-017-0445-8-17. 27 Supplementary Materials for In-situ Non...

    1986 doi

Pith tools

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