Heisenberg picture tensor network formalism for optical circuits
Pith reviewed 2026-05-23 03:26 UTC · model grok-4.3
The pith
A Heisenberg-picture tensor network computes boson-sampling permanents at the complexity of the fastest classical algorithms.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
By working in the Heisenberg picture the evolution of creation operators is represented by a matrix product state whose graphical contraction directly yields the permanent of the interferometer submatrix. This construction matches the complexity of the best known classical permanent algorithms while incorporating loss and distinguishability through local modifications to the same tensor network.
What carries the argument
The graphical representation of the operator-basis matrix product state that encodes the unitary matrix action on photon creation operators.
If this is right
- Boson-sampling probabilities can be obtained with the same asymptotic cost as Ryser-style permanent algorithms.
- Photon loss and partial distinguishability are included by local tensor modifications without changing the overall scaling.
- The framework applies to any linear optical circuit whose unitary matrix is known.
- Output statistics for large interferometers become accessible on classical hardware at optimal permanent complexity.
Where Pith is reading between the lines
- The approach may extend to other sampling tasks whose amplitudes are permanents of submatrices.
- Hybrid verification schemes could use the tensor network to certify small subsystems of larger optical experiments.
- Similar Heisenberg-picture constructions might apply to other linear bosonic systems beyond optics.
Load-bearing premise
The graphical structure of the operator-basis matrix product state encodes all input-output relations so that permanent extraction incurs no extra contraction cost beyond the best classical methods.
What would settle it
Apply the procedure to a known 4-mode, 3-photon interferometer whose permanent is 6 and check whether the reported amplitude equals the exact value with runtime scaling linearly in the number of modes.
Figures
read the original abstract
Tensor network formalisms have emerged as powerful tools for simulating quantum state evolution. While widely applied in the study of optical quantum circuits, such as Boson Sampling, existing tensor network approaches fail to address the complexity mismatch between tensor contractions and the calculation of photon-counting probability amplitudes. Here, we present an alternative tensor network framework that exploits the input-output relations of quantum optical circuits encoded in the unitary interferometer matrix. Our approach bridges the complexity gap by enabling the computation of the permanent -- central to Boson Sampling -- with the same computational complexity as the best known classical algorithm based on a graphical representation of the operator-basis MPS that we introduce. Furthermore, we exploit the flexibility of tensor networks to extend our formalism to incorporate partial distinguishability and photon loss, two key imperfections in practical interferometry experiments. This work offers a significant step forward in the simulation of large-scale quantum optical systems and the understanding of their computational complexity.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript introduces a Heisenberg-picture tensor network formalism for optical quantum circuits, including Boson Sampling. It claims to resolve the complexity mismatch of prior tensor-network methods by introducing a graphical representation of an operator-basis matrix product state (MPS) whose structure encodes the input-output relations of the unitary interferometer matrix, thereby allowing permanent evaluation at the same O(n 2^n) cost as the best classical algorithms; the framework is further extended to incorporate partial distinguishability and photon loss.
Significance. If the central complexity claim is established with explicit contraction analysis, the work would supply a tensor-network route to ideal and imperfect Boson Sampling that matches the best known classical scaling while retaining the flexibility of tensor networks for modeling realistic experimental imperfections.
major comments (2)
- [Abstract] Abstract: the claim that the graphical representation of the operator-basis MPS yields the permanent at exactly the complexity of the best classical algorithm (without extra cost from tensor contractions) is load-bearing yet unsupported by any derivation, bond-dimension bound, or contraction-order argument. The skeptic note correctly identifies that any growth of virtual dimension with mode or photon number would reintroduce the complexity gap the paper asserts to close.
- The manuscript provides no explicit pseudocode, complexity proof, or numerical verification that the MPS contraction order and bond dimensions remain bounded independently of photon number when the unitary is encoded via the graphical operator-basis representation.
Simulated Author's Rebuttal
We thank the referee for their careful reading and constructive feedback. We address each major comment below and will revise the manuscript to provide the requested explicit derivations and supporting material.
read point-by-point responses
-
Referee: [Abstract] Abstract: the claim that the graphical representation of the operator-basis MPS yields the permanent at exactly the complexity of the best classical algorithm (without extra cost from tensor contractions) is load-bearing yet unsupported by any derivation, bond-dimension bound, or contraction-order argument. The skeptic note correctly identifies that any growth of virtual dimension with mode or photon number would reintroduce the complexity gap the paper asserts to close.
Authors: We agree that the abstract claim requires explicit supporting analysis. The graphical operator-basis MPS construction in the manuscript encodes the unitary matrix elements such that the network structure directly implements the inclusion-exclusion summation underlying the Ryser formula. In the revision we will add a dedicated subsection that (i) proves the virtual bond dimension remains bounded by a small constant independent of photon number n (due to the local operator-basis representation of each mode), and (ii) shows that an optimal left-to-right contraction order reproduces exactly the O(n 2^n) arithmetic operations of the best classical permanent algorithm with no additional tensor-contraction overhead. revision: yes
-
Referee: The manuscript provides no explicit pseudocode, complexity proof, or numerical verification that the MPS contraction order and bond dimensions remain bounded independently of photon number when the unitary is encoded via the graphical operator-basis representation.
Authors: We acknowledge the absence of these explicit elements. The current text relies on the graphical construction to imply the complexity result. In the revised version we will insert (a) pseudocode for the contraction algorithm, (b) a formal proof establishing that bond dimensions stay O(1) with respect to both mode number and photon number, and (c) small-n numerical benchmarks confirming that observed runtimes track the expected O(n 2^n) scaling without extra factors. revision: yes
Circularity Check
No circularity: new graphical MPS representation presented as independent construction whose contraction cost is asserted to match Ryser without definitional reduction
full rationale
The paper introduces an operator-basis MPS and its graphical representation as a novel Heisenberg-picture tensor network. The central claim is that this representation computes the permanent at the complexity of the best external classical algorithm (Ryser-style O(n 2^n)). No equation or step is shown to define the permanent output in terms of itself or to rename a fitted parameter as a prediction. No self-citation is invoked as the sole justification for uniqueness or for the contraction cost bound. The derivation chain therefore remains self-contained against external benchmarks; any verification of the claimed contraction overhead would be an independent check rather than a reduction by construction.
Axiom & Free-Parameter Ledger
Reference graph
Works this paper leans on
-
[1]
(C9) Analogously, for the last couple of rows we obtain 1 2 3 4 7, 8 (1 + u3 7u4 8/u4 7u3 8 , u 4 7u3 8) 7, 8 (1 + u2 7u4 8/u4 7u2 8 , u 4 7u2 8) 7, 8 (1 + u2 7u3 8/u3 7u2 8 , u 3 7u2 8) 7, 8 (1 + u1 7u4 8/u4 7u1 8 , u 4 7u1 8) 7, 8 (1 + u1 7u3 8/u3 7u1 8 , u 3 7u1 8) 7, 8 (1 + u1 7u2 8/u2 7u1 8 , u 2 7u1
-
[2]
(C9) there is exactly one row in Eq
(C10) 9 For each row in Eq. (C9) there is exactly one row in Eq. (C10) that can combine with it without resulting in zero. Each non-zero combination contributes a term to the final result, which is the product of the four ˜u values associated with the two rows. Consequently, the final result can be expressed as follows: α¯n = = (u1 1u2 3 + u2 1u1 3)(u3 7u...
-
[3]
+ (u1 1u3 3 + u3 1u1 3)(u2 7u4 8 + u4 7u2 8) + (u1 1u4 3 + u4 1u1 3)(u2 7u3 8 + u3 7u2
-
[4]
+ (u2 1u3 3 + u3 1u2 3)(u1 7u4 8 + u4 7u1 8) + (u2 1u4 3 + u4 1u2 3)(u1 7u3 8 + u3 7u1
-
[5]
+ (u3 1u4 3 + u4 1u3 3)(u1 7u2 8 + u2 7u1
-
[6]
, (C11) which corresponds to the permanent in Eq. (C6). Appendix D: N¨ aive scaling for Permanent For the sake of argument we consider the case ni = 1 for i ≤ n and 0 otherwise. Using the mixed-product property of the Kronecker product over all the matrix products in Eq. (5) yields A[1] n1 ¯i A[2] n2 ¯i . . . A[M] nM ¯i = = X Pn k=1 σ(1) k =n1 . . . X Pn ...
-
[7]
Or´ us, Nature Reviews Physics1, 538 (2019)
R. Or´ us, Nature Reviews Physics1, 538 (2019)
work page 2019
-
[8]
Montangero, Introduction to tensor network methods (Springer Cham, 2018)
S. Montangero, Introduction to tensor network methods (Springer Cham, 2018)
work page 2018
-
[9]
Schollw¨ ock, Annals of Physics326, 96 (2011)
U. Schollw¨ ock, Annals of Physics326, 96 (2011)
work page 2011
-
[10]
K. Audenaert, J. Eisert, M. B. Plenio, and R. F. Werner, Physical Review A 66, 042327 (2002)
work page 2002
- [11]
-
[12]
Vidal, Physical Review Letters 91, 147902 (2003)
G. Vidal, Physical Review Letters 91, 147902 (2003)
work page 2003
-
[13]
Or´ us, Annals of Physics349, 117–158 (2014)
R. Or´ us, Annals of Physics349, 117–158 (2014)
work page 2014
-
[14]
The Computational Complexity of Linear Optics
S. Aaronson and A. Arkhipov, arXiv:1011.3245
work page internal anchor Pith review Pith/arXiv arXiv
-
[15]
R. Garc´ ıa-Patr´ on, J. J. Renema, and V. Shchesnovich, Quantum 3, 169 (2019)
work page 2019
-
[16]
C. Oh, M. Liu, Y. Alexeev, B. Fefferman, and L. Jiang, Nature Physics 20, 1461 (2024)
work page 2024
-
[17]
M. J. Hartmann, J. Prior, S. R. Clark, and M. B. Plenio, Physical Review Letters 102, 057202 (2009)
work page 2009
-
[18]
S. R. Clark, J. Prior, M. J. Hartmann, D. Jaksch, and M. B. Plenio, New Journal of Physics 12, 025005 (2010)
work page 2010
- [19]
-
[20]
C. Oh, K. Noh, B. Fefferman, and L. Jiang, Physical Review A 104, 022407 (2021)
work page 2021
-
[21]
H. J. Ryser, Combinatorial Mathematics, 1st ed., Vol. 14 (Mathematical Association of America, 1963)
work page 1963
-
[22]
On classi- cal simulation algorithms for noisy boson sampling,
C. Oh, L. Jiang, and B. Fefferman, “On classi- cal simulation algorithms for noisy boson sampling,” arXiv:2301.11532
- [23]
- [24]
-
[25]
Note that each A matrix corresponds to a single element of the matrix U, specifically A[k] tj σ(k) j → u tj k . Consequently, if we denote with w(ti) the multiplicity of tj into ¯t, for a fixed ¯n, the set of the A matrices identifies a unique ma- trix U ′ obtained by repeating w(ti) times its tith column and ni times its ith row. These matrices are preci...
-
[26]
Generalizing and Derandomizing Gurvits's Approximation Algorithm for the Permanent
S. Aaronson and T. Hance, “Generalizing and derandom- izing gurvits’s approximation algorithm for the perma- nent,” arXiv:1212.0025 [quant-ph]
work page internal anchor Pith review Pith/arXiv arXiv
-
[27]
J. J. Renema, A. Menssen, W. R. Clements, G. Triginer, W. S. Kolthammer, and I. A. Walmsley, Physical Review Letters 120, 220502 (2018)
work page 2018
-
[28]
P. S. Denis and P. Grim, Journal of Philosophical Logic 26, 181 (1997)
work page 1997
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.