pith. sign in

arxiv: 2602.20701 · v2 · submitted 2026-02-24 · ⚛️ physics.optics

Native QR Factorization on Programmable Photonic Meshes

S.A. Fldzhyan , S.S. Straupe , M.Yu. Saygin This is my paper

Pith reviewed 2026-05-15 20:04 UTC · model grok-4.3

classification ⚛️ physics.optics
keywords QR factorizationphotonic meshinterferometeroptical computingmatrix decompositionHessenberg reductionbidiagonalizationunitary transformation
0
0 comments X

The pith

A programmable photonic mesh performs QR factorization through local power routing in O(N log N) operations.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proposes configuring a mesh of tunable interferometers to compute the QR factorization of a matrix by routing optical power locally in successive steps. The upper triangular factor emerges directly at the optical outputs, with the number of operations scaling as O(N log N) rather than the cubic scaling of digital methods. This native optical approach also allows reuse of the configured mesh for eigenvalue algorithms and provides procedures for related decompositions like Hessenberg reduction.

Core claim

The paper establishes that a programmable unitary interferometer mesh can be configured via sequences of local power routing steps within tunable two-mode interferometric elements to implement QR factorization, allowing direct readout of the upper triangular factor from the optical outputs, with physical operations scaling as O(N log2 N).

What carries the argument

Programmable unitary interferometer mesh configured through local power routing steps in tunable two-mode elements to realize the QR decomposition.

If this is right

  • The same architecture supports iterative spectral computations by reusing the configured interferometer in a mirrored arrangement for the QR eigenvalue algorithm.
  • Related optical procedures enable Hessenberg reduction and bidiagonalization as preprocessors for QR and SVD workflows.
  • The approach exhibits comparable asymptotic complexity to systolic array architectures for blocked QR decomposition.
  • It is more efficient than digital methods for Hessenberg reduction and bidiagonalization.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If losses remain low, this could integrate into larger photonic circuits for real-time linear algebra in sensing or communications.
  • Scaling to bigger matrices would require confirming that imperfections do not accumulate to spoil the factorization.
  • Hybrid optical-digital systems might use this mesh as a fast core for specific decomposition tasks.

Load-bearing premise

A programmable unitary interferometer mesh can be exactly configured through local power routing steps to implement the QR factorization without optical losses, crosstalk, or deviations that corrupt the output R factor.

What would settle it

An experiment inputting a known test matrix into the configured mesh and checking whether the measured optical outputs match the expected R factor within error tolerances.

Figures

Figures reproduced from arXiv: 2602.20701 by M.Yu. Saygin, S.A. Fldzhyan, S.S. Straupe.

Figure 2
Figure 2. Figure 2: (c). By sequentially applying the power concentra￾tion property within the constituent T blocks, the total optical power of any input incident on Ys can always be routed into the uppermost channel. A notable fea￾ture of this scheme is its capacity for self-configuration [36]: the system can align itself using only local feedback loops that maximize power at designated nodes, poten￾tially eliminating the ne… view at source ↗
Figure 1
Figure 1. Figure 1: FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 4
Figure 4. Figure 4: , the resulting R has the form R =   r11 r12 r13 r14 r15 r16 0 r22 r23 r24 r25 r26 0 0 r33 r34 r35 r36 0 0 0 r44 r45 r46 0 0 0 0 r55 r56 0 0 0 0 0 r66   . (3) The number of physical operations is O(N log2 N): there are N column injections and ∼ O(log2 N) local configu￾ration steps per column. E. Recovering Q Defining Q := U † yields the standard QR factorization A = QR. (4) The entries of Q a… view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6 [PITH_FULL_IMAGE:figures/full_fig_p004_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7 [PITH_FULL_IMAGE:figures/full_fig_p005_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8 [PITH_FULL_IMAGE:figures/full_fig_p006_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9 [PITH_FULL_IMAGE:figures/full_fig_p007_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10 [PITH_FULL_IMAGE:figures/full_fig_p007_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: FIG. 11 [PITH_FULL_IMAGE:figures/full_fig_p008_11.png] view at source ↗
read the original abstract

We propose a photonic native procedure for computing the QR factorization of a matrix using a programmable unitary interferometer mesh. The method configures the mesh through a sequence of local power routing steps within tunable two mode interferometric elements, while reading out the resulting upper triangular factor directly from the optical outputs. The number of physical operations grows as $ O(N\log_2N)$ with matrix size $N$, reducing the runtime relative to standard digital QR routines, which scale cubically ($O(N^3)$). Beyond single factorizations, the same architecture supports iterative spectral computations by reusing the configured interferometer in a mirrored arrangement that implements the core update step of the QR eigenvalue algorithm. We also describe related optical procedures for Hessenberg reduction and bidiagonalization, serving as compatible preprocessors for QR and SVD workflows. A comparison with the systolic array computational architecture is provided. Our approach exhibits comparable asymptotic complexity for blocked QR decomposition and is more efficient for Hessenberg reduction and bidiagonalization.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes a native photonic procedure for QR factorization on a programmable unitary interferometer mesh. The mesh is configured via a sequence of local power routing steps in tunable two-mode interferometric elements, allowing the upper-triangular R factor to be read directly from the optical outputs. The claimed complexity is O(N log₂ N) physical operations, in contrast to the O(N³) scaling of standard digital QR routines. The architecture is further shown to support iterative spectral computations via a mirrored reuse of the configured mesh for the QR eigenvalue algorithm, along with related procedures for Hessenberg reduction and bidiagonalization that serve as preprocessors for QR and SVD workflows. A comparison to systolic-array architectures is included.

Significance. If the local-routing procedure can be shown to produce an exact unitary Q (or a sufficiently accurate approximation) for general input matrices, the work would constitute a meaningful advance in optical linear-algebra accelerators. The O(N log₂ N) scaling, together with the ability to reuse the same mesh for iterative eigenvalue steps, could enable substantial runtime and energy advantages over digital implementations for large-scale matrix factorizations and spectral computations, particularly in hybrid photonic-electronic pipelines.

major comments (2)
  1. [Abstract and §3] Abstract and §3 (configuration procedure): the central O(N log₂ N) scaling claim is stated without an explicit derivation, step-by-step algorithm, or operation count that demonstrates how the sequence of local power-routing adjustments produces the global QR decomposition for an arbitrary matrix.
  2. [§4 and abstract] §4 (error model) and abstract: the procedure assumes that the configured mesh implements an exact unitary Q so that the optical outputs equal the true R factor, yet no insertion-loss, crosstalk, or phase-drift model is supplied, nor is a tolerance analysis given showing that the output R remains accurate for general inputs.
minor comments (2)
  1. [Abstract] The abstract states that the architecture supports Hessenberg reduction and bidiagonalization but does not indicate whether these procedures inherit the same O(N log₂ N) scaling or require additional mesh reconfigurations.
  2. [§5] Figure captions and §5 (comparison): the systolic-array comparison would benefit from an explicit table listing latency, number of tunable elements, and energy per operation for both approaches at the same matrix size.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive feedback. The two major comments identify important gaps in the presentation of the algorithm and its robustness. We will revise the manuscript to address both points directly.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3 (configuration procedure): the central O(N log₂ N) scaling claim is stated without an explicit derivation, step-by-step algorithm, or operation count that demonstrates how the sequence of local power-routing adjustments produces the global QR decomposition for an arbitrary matrix.

    Authors: We agree that the current manuscript lacks a self-contained derivation and explicit algorithm. In the revised version we will expand §3 with (i) a step-by-step description of the local power-routing procedure, (ii) pseudocode that maps each routing step to the corresponding Givens-like rotation on the mesh, and (iii) a detailed operation count showing that the parallel depth is O(log₂ N) while the total number of tunable-element adjustments remains O(N log₂ N) for an arbitrary dense matrix. This will make the claimed complexity transparent and demonstrate how the sequence yields the exact QR factorization under the ideal unitary model. revision: yes

  2. Referee: [§4 and abstract] §4 (error model) and abstract: the procedure assumes that the configured mesh implements an exact unitary Q so that the optical outputs equal the true R factor, yet no insertion-loss, crosstalk, or phase-drift model is supplied, nor is a tolerance analysis given showing that the output R remains accurate for general inputs.

    Authors: We acknowledge that the present manuscript treats the mesh as ideal and does not quantify non-idealities. In the revision we will augment §4 with a realistic error model that includes insertion loss, crosstalk, and phase drift. We will add numerical tolerance analysis (Monte-Carlo simulations over random matrices) that reports the relative error in the recovered R factor as a function of mesh size and noise level, thereby demonstrating the regime in which the optical outputs remain sufficiently accurate for practical use. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper presents a direct configuration procedure for implementing QR factorization on a programmable photonic mesh via a sequence of local power routing steps in tunable two-mode elements, with the upper-triangular factor R read from optical outputs. The claimed O(N log2 N) scaling follows from counting the physical operations required by the mesh reconfiguration sequence for an N-dimensional input, without any fitted parameters, self-referential definitions, or load-bearing self-citations that reduce the central claim to its own inputs. The derivation remains self-contained against the stated assumptions of ideal unitary behavior and does not rename known results or smuggle ansatzes via citation.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The proposal rests on standard domain assumptions about unitary transformations in photonic meshes and the ability to perform exact local routing without loss; no free parameters or invented entities are introduced in the abstract.

axioms (1)
  • domain assumption Programmable unitary interferometer meshes can implement arbitrary unitary transformations via local tuning of two-mode elements
    Invoked implicitly when stating that the mesh can be configured to produce the QR output directly.

pith-pipeline@v0.9.0 · 5476 in / 1217 out tokens · 47167 ms · 2026-05-15T20:04:29.567478+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

53 extracted references · 53 canonical work pages

  1. [1]

    Bogaerts, D

    W. Bogaerts, D. P´ erez, J. Capmany, D. A. B. Miller, J. Poon, D. Englund, F. Morichetti, and A. Melloni, Na- ture586, 207 (2020)

    work page 2020

  2. [2]

    B. Wu, H. Zhou, J. Dong, and X. Zhang, Applied Physics Reviews11, 011309 (2024)

    work page 2024

  3. [3]

    T. G. de Brugi` ere, R. Mezher, S. Currie, and S. Mans- field, New designs of linear optical interferometers with minimal depth and component count (2025), arXiv:2504.06059 [quant-ph]

    work page arXiv 2025

  4. [4]

    Marchesin, M

    F. Marchesin, M. Hejda, T. M. Carmona, S. D. Carlo, A. Savino, F. Pavanello, T. V. Vaerenbergh, and P. Bi- enstman, Optics Express33, 2227 (2025)

    work page 2025

  5. [5]

    Girouard and N

    V. Girouard and N. Quesada, Journal of the Optical So- ciety of America B43, A66 (2026)

    work page 2026

  6. [6]

    Kondratyev, V

    I. Kondratyev, V. Ivanova, S. Fldzhyan, A. Ar- genchiev, N. Kostyuchenko, S. Zhuravitskii, N. Skryabin, I. Dyakonov, M. Saygin, S. Straupe, A. Korneev, and S. Kulik, Photonics Research12, A28 (2024)

    work page 2024

  7. [7]

    Kondratyev, K

    I. Kondratyev, K. Urusova, A. Argenchiev, N. Klush- nikov, S. Kuzmin, N. Skryabin, A. Golikov, V. Kova- lyuk, G. Goltsman, I. Dyakonov, S. Straupe, and S. Ku- lik, Physical Review Applied25, 034072 (2026)

    work page 2026

  8. [8]

    Kuzmin, I

    S. Kuzmin, I. Dyakonov, and S. Straupe, Physical Review A112, 053515 (2025)

    work page 2025

  9. [9]

    Y. Xiao, Y. Zhao, W. Wang, Z. Cheng, X. Peng, H. Tang, S. Liu, and Y. Tang, Optics Express33, 32190 (2025)

    work page 2025

  10. [10]

    N. A. Nemkov and S. S. Straupe, Complexity- energy trade-off in programmable unitary interferometers (2025), arXiv:2507.22972 [physics]

    work page arXiv 2025

  11. [11]

    Taguchi, Journal of the Optical Society of America B 42, 2207 (2025)

    Y. Taguchi, Journal of the Optical Society of America B 42, 2207 (2025)

    work page 2025

  12. [12]

    Hamerly, J

    R. Hamerly, J. R. Basani, A. Sludds, S. K. Vadlamani, and D. Englund, APL Photonics10, 110803 (2025)

    work page 2025

  13. [13]

    Talib, P

    H. Talib, P. D. Sewell, A. Vukovic, and S. Phang, Optical and Quantum Electronics57, 590 (2025)

    work page 2025

  14. [14]

    J. Lin, K. Yang, Q. Fu, P. Wang, S. Dai, W. Chen, D. Kong, J. Li, T. Dai, and J. Yang, Journal of Lightwave Technology43, 1024 (2025)

    work page 2025

  15. [15]

    R. Tang, M. Okano, K. Toprasertpong, S. Takagi, D. En- glund, and M. Takenaka, Optics Express30, 33940 (2022)

    work page 2022

  16. [16]

    R. Tang, M. Okano, C. Zhang, K. Toprasertpong, S. Tak- agi, and M. Takenaka, Optica12, 812 (2025)

    work page 2025

  17. [17]

    Milanizadeh, E

    M. Milanizadeh, E. Damiani, T. Jonuzi, M. J. Mencagli, B. Edwards, D. A. Miller, N. Engheta, A. Melloni, and F. Morichetti, inEuropean Conference on Integrated Op- tics 2020 (ECIO)(2020)

    work page 2020

  18. [18]

    Peserico, B

    N. Peserico, B. J. Shastri, and V. J. Sorger, Journal of Lightwave Technology41, 3704 (2023)

    work page 2023

  19. [19]

    M. Chen, Q. Cheng, M. Ayata, M. Holm, and R. Penty, Photonics Research10, 2488 (2022)

    work page 2022

  20. [20]

    M. Chen, Y. Wang, C. Yao, A. Wonfor, S. Yang, R. Penty, and Q. Cheng, Nature Communications15, 5926 (2024)

    work page 2024

  21. [21]

    T. M. Carmona, F. Marchesin, M. P. Abrate, P. Bienst- man, S. D. Carlo, and A. S. Senior, LuxIA: A Lightweight Unitary matriX-based Framework Built on an Iterative Algorithm for Photonic Neural Network Training (2025), arXiv:2512.22264 [cs]

    work page arXiv 2025

  22. [22]

    Cavicchioli, D

    G. Cavicchioli, D. A. B. Miller, N. Engheta, A. Melloni, and F. Morichetti, inOptical Fiber Communication Con- ference(Optica Publishing Group, 2024) pp. Th1A–2

    work page 2024

  23. [23]

    E. E. Tyrtyshnikov,A Brief Introduction to Numerical Analysis(Birkh¨ auser Boston, Boston, MA, 1997)

    work page 1997

  24. [24]

    G. H. Golub and C. F. Van Loan,Matrix Computations, 4th ed., Johns Hopkins Studies in the Mathematical Sci- ences (The Johns Hopkins University Press, Baltimore, 2013)

    work page 2013

  25. [25]

    D. P. Arbenz,Lecture Notes on Solving Large Scale Eigenvalue Problems(Computer Science Department, ETH Z¨ urich, 2016)

    work page 2016

  26. [26]

    Khachaturian, R

    A. Khachaturian, R. Fatemi, and A. Hajimiri, IEEE Open Journal of the Solid-State Circuits Society1, 263 (2021)

    work page 2021

  27. [27]

    Bantysh, K

    B. Bantysh, K. Katamadze, A. Chernyavskiy, and Y. Bogdanov, Optics Express31, 16729 (2023)

    work page 2023

  28. [28]

    B. I. Bantysh, A. Y. Chernyavskiy, S. A. Fldzhyan, and Y. I. Bogdanov, Laser Physics Letters21, 015203 (2024)

    work page 2024

  29. [29]

    M. Reck, A. Zeilinger, H. J. Bernstein, and P. Bertani, Physical Review Letters73, 58 (1994)

    work page 1994

  30. [30]

    R. J. Potton, Reports on Progress in Physics67, 717 (2004)

    work page 2004

  31. [31]

    H. Kung, C. Leiserson, C.-M. U. P. P. D. of COMPUTER SCIENCE., and C. M. U. C. S. Department,Systolic Ar- rays for (VLSI), CMU-CS (Carnegie-Mellon University, Department of Computer Science, 1978)

    work page 1978

  32. [32]

    W. R. Clements, P. C. Humphreys, B. J. Metcalf, W. S. Kolthammer, and I. A. Walsmley, Optica3, 1460 (2016)

    work page 2016

  33. [33]

    Hamerly, S

    R. Hamerly, S. Bandyopadhyay, and D. Englund, Nature Communications13, 6831 (2022)

    work page 2022

  34. [34]

    Di Tria, G

    A. Di Tria, G. Cavicchioli, P. Giannoccaro, F. Morichetti, A. Melloni, G. Ferrari, M. Sampietro, and F. Zanetto, Laser & Photonics Reviews , e00610 (2025)

    work page 2025

  35. [35]

    N. J. Russell, L. Chakhmakhchyan, J. L. O’Brien, and A. Laing, New Journal of Physics19, 033007 (2017)

    work page 2017

  36. [36]

    D. A. B. Miller, Photonics Research1, 1 (2013)

    work page 2013

  37. [37]

    Chiles, S

    J. Chiles, S. M. Buckley, S. W. Nam, R. P. Mirin, and J. M. Shainline, APL Photonics3, 106101 (2018)

    work page 2018

  38. [38]

    S. Xu, J. Wang, S. Yi, X. Zhao, B. Liu, J. Shao, and W. Zou, Optics Express30, 42057 (2022)

    work page 2022

  39. [39]

    Giamougiannis, A

    G. Giamougiannis, A. Tsakyridis, M. Moralis-Pegios, G. Mourgias-Alexandris, A. R. Totovic, G. Dabos, M. Kirtas, N. Passalis, A. Tefas, D. Kalavrouziotis, D. Syrivelis, P. Bakopoulos, E. Mentovich, D. Lazovsky, and N. Pleros, Advanced Photonics5, 016004 (2023)

    work page 2023

  40. [40]

    R. Tang, R. Tanomura, T. Tanemura, and Y. Nakano, Physical Review Applied21, 014054 (2024)

    work page 2024

  41. [41]

    S. A. Fldzhyan, M. Yu. Saygin, and S. S. Straupe, Phys- ical Review Research8, 013021 (2026)

    work page 2026

  42. [42]

    Kung, IEEE ASSP Magazine2, 4 (1985)

    S. Kung, IEEE ASSP Magazine2, 4 (1985)

    work page 1985

  43. [43]

    Kurzak, P

    J. Kurzak, P. Luszczek, I. Yamazaki, Y. Robert, and J. Dongarra, Supercomputing Frontiers and Innovations 4, 10.14529/jsfi170101 (2017)

    work page doi:10.14529/jsfi170101 2017

  44. [44]

    N. P. Jouppi, C. Young, N. Patil, D. Patterson, G. Agrawal, R. Bajwa, S. Bates, S. Bhatia, N. Boden, A. Borchers, and et. al., ACM SIGARCH Computer Ar- chitecture News45, 1 (2017)

    work page 2017

  45. [45]

    J. Lu, D. Qu, J. Qu, R. Fong, G. H. Ahn, and J. Vuck- ovic, A systolic update scheme to overcome memory bandwidth limitations in gpu-accelerated fdtd simula- tions (2025), arXiv:2502.20610 [physics.optics]

    work page arXiv 2025

  46. [46]

    W. M. Gentleman and H. T. Kung, in25th Annual Tech- 10 nical Symposium, edited by T. F. Tao (San Diego, 1982) pp. 19–26

    work page 1982

  47. [47]

    Bojanczyk, R

    A. Bojanczyk, R. P. Brent, and H. T. Kung, SIAM Jour- nal on Scientific and Statistical Computing5, 95 (1984)

    work page 1984

  48. [48]

    Tiskin, Future Generation Computer Systems23, 179 (2007)

    A. Tiskin, Future Generation Computer Systems23, 179 (2007)

    work page 2007

  49. [49]

    G. H. Y. Li, M. Parto, J. Ge, Q.-X. Ji, M. Gao, Y. Yu, J. Williams, R. M. Gray, C. R. Leefmans, N. Englebert, K. J. Vahala, and A. Marandi, All-optical computing with beyond 100-ghz clock rates (2025), arXiv:2501.05756 [physics.optics]

    work page arXiv 2025

  50. [50]

    Feldmann, N

    J. Feldmann, N. Youngblood, M. Karpov, H. Gehring, X. Li, M. Stappers, M. Le Gallo, X. Fu, A. Lukashchuk, A. S. Raja, J. Liu, C. D. Wright, A. Sebastian, T. J. Kippenberg, W. H. P. Pernice, and H. Bhaskaran, Nature 589, 52–58 (2021)

    work page 2021

  51. [51]

    Virtanen, R

    P. Virtanen, R. Gommers, T. E. Oliphant, M. Haber- land, T. Reddy, D. Cournapeau, E. Burovski, P. Peter- son, W. Weckesser, J. Bright, and et. al., Nature Methods 17, 261 (2020)

    work page 2020

  52. [52]

    Hamerly, S

    R. Hamerly, S. Bandyopadhyay, and D. Englund, Physi- cal Review Applied18, 024019 (2022)

    work page 2022

  53. [53]

    D. M. Pozar,Microwave Engineering, fourth edition ed. (John Wiley & Sons, Inc, Hoboken, NJ, 2012)

    work page 2012