Pith. sign in

REVIEW 2 major objections 5 minor 1 cited by

Realified tensor networks: quantum circuit simulation on real-valued matrix accelerators

T0 review · 2 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that any complex tensor network can be converted by a local graph rewrite into an equivalent real tensor network whose contraction costs at most three times the arithmetic of the same network over the reals, making…

desk verdict The realification cost law is proven and the benchmarks are solid; the order-transfer flatness claim is real but scoped, and the authors are transparent about it. read the letter →

arxiv 2608.03987 v2 pith:WG6Q5SDQ submitted 2026-08-04 quant-ph

Yusheng Zhao , Xiwei Pan , Enji Xiong , Chengkai Zhu , Jinguo Liu This is my paper · ORCID
classification quant-ph
keywords tensor-networkcontractionquantumcircuitsimulationrealificationreal-valuedmatrixacceleratorsthree-multiplicationformulacostlawcontraction-ordertransferAscend910
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 targets the mismatch between quantum circuit simulation, which naturally works with complex tensor networks, and modern matrix accelerators that expose only real matrix multiplication. It proposes a local graph rewrite called realification: every complex tensor gains one extra dimension-2 index holding its real and imaginary parts, and wherever two complex pieces merge, a fixed rank-3 structure tensor realizes the three-multiplication (3M) formula inside the network. The central result is a proven cost law: the converted real network costs $1+2m+r$ times as many real multiplications as the same network contracted over the reals, where $m$ and $r$ are the volume shares of two-complex and one-complex steps, so the overhead always lies between $1\times$ and $3\times$, and every intermediate tensor at most doubles in size. On 67 circuits the law is verified across the full range, complex-optimal contraction orders transfer to the realified network to within $5\times10^{-4}$ relative arithmetic cost on 66 of them, and on an Ascend 910 the rewrite beat both per-GEMM complex lowerings on all twelve random circuits and on 52 of 55 structured cells. If correct, the result means real-only tensor engines can natively simulate complex quantum circuits, with arithmetic overhead bounded by a known constant and measurable wall-clock speedups.

What carries the argument

The load-bearing object is the structure tensor $C\in\mathbb{R}^{2\times2\times2}$ given by $C_{abc}=\mathrm{Re}(i^{a+b+c-3})$, with matrix slices $\begin{pmatrix}1&0\\0&-1\end{pmatrix}$ and $\begin{pmatrix}0&-1\\-1&0\end{pmatrix}$. It is the multiplication tensor of $\mathbb{C}$ as a real $*$-algebra, permutation-symmetric in all three legs, and its rank-3 decomposition is the three-multiplication (3M) identity. Contracting two realified tensors' added dimension-2 ('green') legs through $C$, followed by the sign flip $Z$, reproduces the complex product; conjugation and phase rotations become local linear maps on the green leg. The spider-fusion rules (permutation invariance, conjugate covariance, unit rule, and the cascade rule) make every tree-aligned wiring equivalent, which is what keeps the contraction order free and supports the per-step cost accounting.

What would settle it

Take a mostly-real tensor network with only two complex tensors among many real ones and rerun optimization directly on the realified network with a much larger search budget than the paper used; a discovered contraction order that beats the converted order by more than $5\times10^{-4}$ in relative arithmetic cost would falsify the transfer-flatness claim for that family.

Watch

Extended reading notes

Core claim

The central discovery is that complex tensor networks need not be reconstructed as complex GEMMs on real-only hardware; the network itself can be rewritten over the reals. Each complex tensor $A=A_R+iA_I$ becomes a real tensor with one extra dimension-2 index, conjugation becomes a sign flip $Z=\mathrm{diag}(1,-1)$, and complex multiplication is reproduced by contracting the two green legs through the permutation-symmetric structure tensor $C_{abc}=\mathrm{Re}(i^{a+b+c-3})$, whose rank-3 decomposition is exactly the three-multiplication (3M) formula. The rewrite is aligned with the chosen contraction tree, so the contraction order remains a free choice rather than being pinned to a fixed worldline. The paper proves the tight cost law $1+2m+r$ for arithmetic overhead, with every intermediate at most doubled in size, and verifies it on 67 circuits; on an Ascend 910 the realified executor beat the four-real-GEMM baseline on all twelve random circuits and on 52 of 55 structured cells, with median speedups of roughly $1.7\times$ on random circuits and $1.4\times$ on structured ones, while applying the same 3M identity per GEMM, without the network rewrite, showed no systematic wall-clock gain.

Load-bearing premise

The load-bearing assumption is that the order in which the network's multiplications are grouped, when optimized on the original complex network, remains almost as cheap after the real rewrite; the benchmarks verify this with a finite computer search rather than a proof, and the paper itself constructs a family of networks where the converted order is up to nearly twice as expensive as reoptimizing.

Editorial extensions

If this is right

  • Any complex tensor network, not just a quantum circuit, can be executed on real-only matrix engines with arithmetic overhead between $1\times$ and $3\times$.
  • Mostly-real circuits automatically cost less than the flat 3M ceiling: at a fixed fraction of T gates, spatially clustered placements in the paper's Clifford+T cells dropped the overhead from $2.93\times$ to $1.98\times$.
  • Per-GEMM complex lowering is never better than network-level realification when any contraction step is a pass or a ride, because the rewrite charges the $3\times$ factor only where two complex operands merge.
  • Because every tensor and contraction is real after the rewrite, reverse-mode differentiation of the realified network uses ordinary real-valued automatic differentiation; only structure-preserving factorizations such as SVD remain open.
  • Contraction orders optimized on the complex network transfer to the realified network within $5\times10^{-4}$ relative arithmetic cost on 66 of 67 benchmark circuits, so the pipeline avoids a full reoptimization pass on those circuit classes.

Reading between the lines

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

  • Editorial inference: the same structure-tensor construction should carry over to any real $*$-algebra with a low-rank multiplication tensor, so quaternion- or Clifford-valued tensor networks could inherit the bounded-overhead realification argument.
  • Editorial inference: because the paper constructs a family where a skeleton-optimal tree is up to nearly $2\times$ worse than the realified optimum, real deployments on mostly-real circuits may still need a green-aware reoptimization pass even though the 67-circuit suite did not; an automated detector for that gap region would complement the method.
  • Editorial inference: since the rewrite preserves byte-level peak memory and produces an ordinary real tensor-network contraction, it should compose with distributed slicing and bfloat16 precision splitting; the paper states these compose with the representation but leaves slicing interactions uncharacterized, so measuring sliced runs is a natural next experiment.
  • Editorial inference: the method treats a hardware restriction as a graph rewrite, so real-only engines remain viable for complex simulation; the paper's hardware results suggest the wall-clock gain is partly stack-specific, so porting the realified representation to other real-only accelerators would test how much of the gain is intrinsic.
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

2 major / 5 minor

Summary. The paper presents a graph-rewrite compilation ('network-3M') that converts any complex tensor network into a real-valued tensor network for execution on real-only GEMM accelerators. Each complex tensor acquires a dimension-2 auxiliary (green) index; pairwise contraction of two green operands is mediated by a permutation-symmetric rank-3 structure tensor C realizing Gauss's 3M formula, while rides and passes cost 2x and 1x respectively. The central theoretical result, Theorem 4, states that for any fixed binary tree the realified arithmetic overhead is 1+2m+r in real scalar multiplications, bounded between 1 and 3, with every intermediate at most doubled in element count. The paper validates the accounting on 67 circuits, reports that complex-gate placement rather than count governs overhead, claims that contraction orders optimized on the complex network transfer to the realified network with relative gap below 5e-4 on 66 of 67 circuits, and reports Ascend 910 wall-clock speedups of the rewritten executor over GEMM-4M and GEMM-3M baselines. Appendix A analyzes the landscape and proves that worst-case conversion gaps up to factor 2 are possible, so the transfer flatness is an empirical property of the benchmark suite rather than a universal theorem.

Significance. Theorem 4 is clean and, as far as I can tell, correct: the 3x merge cost follows from the explicit rank-3 decomposition of Eq. (7) together with Winograd's minimality, the 2x ride cost follows from the spectator green leg, and the memory bound follows from tree-aligned wiring. The paper is admirably explicit about the scope of its evidence: it distinguishes the theorem from the search-budget-verified density hypothesis, states that annealing certifies only what it samples, and ships reproducible contraction plans and a pinned toolchain. If the empirical claims survive scrutiny, this gives a practical recipe for running complex tensor-network contraction natively on real-only accelerators with bounded arithmetic overhead, demonstrated wall-clock wins, and a clean real-valued differentiation rule. The main caveat is that the headline 'orders transfer' claim is empirical and not guaranteed by the theory; the paper's own Theorem 8 exhibits a 2x worst-case conversion gap.

major comments (2)
  1. [Abstract, §3.1, Appendix A] The abstract's unconditional phrasing that 'contraction orders transfer from the complex network with a relative arithmetic-cost gap below 5e-4 on 66 of 67 circuits' goes beyond what is established. Theorem 6 holds only under the density hypothesis (9), which Appendix A states is 'verified within search budget rather than proved', and 'annealing certifies only what it samples'. Moreover, Theorem 8 explicitly constructs green-sparse families for which the converted skeleton-optimal tree is within a factor of two of the realified optimum, and the 5-qubit outlier is presented as a miniature instance of that mechanism. Please add to the abstract and Section 4 the same qualification already present in §3.1 ('within the annealer's search budget'), and rephrase the abstract so it does not read as a universal transfer guarantee.
  2. [§3.1, Appendix B] The cross-check statement that 'independently re-optimizing each circuit's realified einsum ... produces trees that beat the archived schedule on all twelve circuits' appears to contradict the preceding flatness claim unless 'beat' refers to the optimizer's loop-volume convention tc_R (total loop iterations), which includes the factor-vector projection costs quantified in Eq. (11), and not to the Theorem 4 multiplication count. The main text does not state this, and Table 5's tc_R values are on a different scale from the multiplication counts in Table 1. Please state explicitly which metric is used for 'beat', and discuss whether the loop-volume improvement would affect the wall-clock comparison; as written, a reader cannot tell whether the no-reoptimization pipeline is optimal in the model actually executed on the accelerator.
minor comments (5)
  1. [Fig. 9(b), §3.1] Figure 9(b) defines the pipeline gap using 'overhead (Eq. (6))', while the abstract and text speak of a 'relative arithmetic-cost gap'. If o is only the overhead multiplier, the two quantities differ when the compared trees have different skeleton volumes, as in the 5-qubit exception. Please use one definition, preferably the total realified cost, and state it consistently in the figure caption and Section 3.1.
  2. [Abstract, §3.1] The phrase 'the law holds across the real-to-complex range' is a bookkeeping identity once (m,r) are read from the tree; the substantive empirical content is the location of (m,r) as a function of circuit structure. Consider phrasing such as 'the audited (m,r) place every circuit on the band predicted by Eq. (6)'.
  3. [Fig. 9(b)] Row labels such as 'random (12) (9)' are cryptic; please explain what the parenthesized numbers denote and what the left-edge annotation counts.
  4. [Table 1] Please clarify in the table caption that the 'overhead' column is computed relative to the best skeleton found across all optimization passes, not the tree's own skeleton; this is needed to understand the 5-qubit row's 2.01x value versus the law value 1.94x.
  5. [§3.2] The term 'device-clean' in the structured benchmark set is defined only implicitly through fallback flags; please state in the main text how many cells were excluded and why, since the exclusion bears on the 52-of-55 win record.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the cost law is an exact pricing rule for an explicit construction, and the transfer claim is an empirically caveated benchmark finding rather than a fitted prediction.

full rationale

The derivation chain is self-contained and does not reduce to its inputs. Theorem 4 is an exact pricing rule for the proposed realified construction: passes cost 1x, rides cost 2x because the dimension-2 green leg is a spectator, and merges cost 3x because Eq. (7) gives an explicit rank-3 tensor decomposition of C and Winograd's independent minimality result bounds the real rank of complex multiplication from below. The parameters m and r are defined from the real-skeleton contraction-tree volumes, not fitted from the realified costs, so the law is a theorem plus an implementation audit rather than a fitted model renamed as a prediction. The benchmark tables compare the audited realified cost against the same-tree (m, r) values, which is bookkeeping verification of the construction, not a circular statistical fit. The contraction-order-transfer claim is empirical and is explicitly qualified in Appendix A: the density hypothesis of Theorem 6 is 'verified within search budget rather than proved' and 'annealing certifies only what it samples.' That is a strength-of-evidence limitation, not a circular reduction, and Theorem 8 even constructs a green-sparse family where flatness fails, so the paper does not posit the transfer claim as a logical consequence of a self-citation. The self-references that occur (Bauer for the *-algebra tensor, Yao and omeco as tooling) are provenance or implementation artifacts; the algebraic identities are proved in Appendix D from the closed form of C, and no uniqueness theorem from the authors' prior work is imported to force the choice. No parameter is fitted to a subset and then used to predict a closely related quantity, and no known result is merely renamed. The only soft spot is the search-budget dependence of the flatness claim, which belongs to correctness risk rather than circularity, and the paper discloses it.

Assumptions & free parameters 0 free parameters · 4 assumptions · 1 invented entities

The central cost law rests on standard field axioms for complex numbers, the known minimal rank-3 decomposition of complex multiplication, and the domain assumption that the target accelerators are real-only. No free parameters are fitted to data. The only assumption that is not fully proved for the benchmark suite is the density hypothesis used for order-transfer flatness, and the paper discloses that it is verified within the annealer's search budget.

assumptions (4)
  • standard math Complex multiplication has real tensor rank 3, with the Gauss 3M decomposition attaining the minimum (Winograd).
    Used in Theorem 4 to price merges at 3x the real-skeleton cost. The decomposition is exhibited in Eq. (7) and minimality is cited to Winograd [11].
  • domain assumption Modern matrix accelerators such as TPUs, GPU tensor cores, and NPUs expose only real GEMM, so complex arithmetic must be emulated.
    Stated in Sec. 1 and used throughout the motivation and the hardware benchmark. The authors themselves show the wall-clock advantage disappears on an A800 with native complex arithmetic (44 of 55 rows fastest in native complex).
  • domain assumption Contraction trees are binary and acyclic; spider fusion applies to any tree of C nodes with Z-dressed internal edges, and cyclic wirings are excluded because they introduce scalar factors of 2.
    Sec. 2.2 and Appendix D. The construction is tree-aligned, so cycles never arise in the method, but the paper relies on this to transfer wirings between trees.
  • ad hoc to paper The density hypothesis m(T) >= 1 - epsilon for every tree T with V(T) <= 3V* is assumed in Theorem 6.
    Appendix A uses this hypothesis to prove flatness of the reoptimization landscape. For the benchmark suite it is verified within search budget, not proved, and the authors state this limitation explicitly.
invented entities (1)
  • Structure tensor C with Z-dressed multiplication M = CZ independent evidence
    purpose: Encodes complex multiplication as a symmetric real rank-3 tensor so that realified contractions reproduce complex products exactly.
    This is a mathematical construction, not a new physical entity. Its closed form Re(i^(a+b+c-3)) and the rank-3 factorization (7) are directly checkable by substitution, and the spider fusion identities are verified in Appendix D.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Realified tensor networks: quantum circuit simulation on real-valued matrix accelerators." pith.science (2026). https://pith.science/paper/WG6Q5SDQ

@misc{pith2026260803987,
  author       = {Pith},
  title        = {Pith review of: Realified tensor networks: quantum circuit simulation on real-valued matrix accelerators},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WG6Q5SDQ}},
  note         = {Machine review of arXiv:2608.03987}
}
abstract

Tensor-network contraction simulates quantum circuits, but modern matrix accelerators (NPUs, TPUs) expose only real GEMM pipelines, so the complex networks of quantum simulation must be reconstructed in software. We resolve the mismatch by a realification rewrite that maps any complex tensor network to a real one. At each merge of two complex tensors, a rank-3 structure tensor realizes Gauss's three-multiplication (3M) formula; contractions with one or no complex operand need only two or one real products. We prove a tight cost law: overhead $1 + 2m + r$ in real multiplications, where $m$ and $r$ are the volume fractions of two- and one-complex-operand contractions, never exceeding $3\times$ relative to real contraction, with every intermediate at most doubled in size. On 67 circuits (random, Clifford+$T$, QAOA, VQE), the law holds across the real-to-complex range and complex-gate placement, not count, governs cost. Contraction orders transfer from the complex network with a relative arithmetic-cost gap below $5\times 10^{-4}$ on 66 of 67 circuits; the exception closes under a few steps of low-temperature simulated annealing. On an Ascend 910 NPU the rewrite beat both the four-real-GEMM baseline and a per-GEMM Gauss lowering on all twelve random circuits and on 52 of 55 structured cells (three cells slower by at most 12\%); the four-GEMM baseline was slower by a median $1.7\times$ (random) and $1.4\times$ (structured). Realification makes complex tensor-network contraction native to real-only matrix engines.

Figures

Figures reproduced from arXiv: 2608.03987 by the authors.

Figure 1
Figure 1. From quantum circuit to real-valued contraction. (a) A quantum circuit [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. (a) M distinguishes input legs (solid) from output (dashed). (b) Absorbing Z yields C, invariant under all permutations of its legs. 𝑇𝐶 = 𝑇𝐴 𝑇𝐵 𝒞︀ 𝑍 [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Matrix product C = AB in the realified picture: the green legs of TA and TB merge through C, followed by Z on the output. Conjugation and phases. Quantum circuit networks also require conjugation (bra vectors, ad￾joints) and global phases. Both are local operations on the green leg, and neither adds contraction￾stage multiplications: conjugation is a sign flip, and a global phase, known when the network is built, is… view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: (a) Conjugation: Z on the green leg gives TA∗ . (b) Global phase: the rotation Rφ on the green leg gives Te iφA. Neither adds contraction-stage multiplications: the sign flip is free, and static phases are absorbed into leaf data at construction. 2.2 Tree freedom and t…
Figure 4
Figure 4. Figure 4: (a) Conjugation: Z on the green leg gives TA∗ . (b) Global phase: the rotation Rφ on the green leg gives Te iφA. Neither adds contraction-stage multiplications: the sign flip is free, and static phases are absorbed into leaf data at construction. 𝑇𝐴 𝑇𝐵 𝑇𝐶 𝒞︀ 𝑍 𝒞︀ 𝑍 (𝐴𝐵…
Figure 5
Figure 5. Figure 5: One complex network, two realifications. Contracting the same three-tensor chain along [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: The algebra of the realified representation. (a) Matrix product (see also Fig. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: The three pairwise-contraction cases and their arithmetic cost relative to the same step [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Measured overhead versus merge fraction m for all 67 benchmark instances: twelve random circuits (orange circles), 24 Clifford+T (purple triangles), 10 QAOA (blue diamonds), and 21 VQE (green squares). The law (5) confines every point to the band between the solid line…
Figure 9
Figure 9. Figure 9: Pipeline comparison: reoptimization after realification is unnecessary on the tested [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]
Figure 10
Figure 10. Figure 10: Forward and backward contractions with green-index labels. (a) Forward product: [PITH_FULL_IMAGE:figures/full_fig_p022_10.png]

Review history (3 revisions) →

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Aicir: A Full-Stack Quantum Circuit Simulator with AscendNPU Support

    quant-ph 2026-08 conditional novelty 6.0 of 10

    Aicir is a full-stack quantum circuit simulator with a native Huawei Ascend NPU backend, using paired real tensors and fixed-rank gate views, validated with CPU fallback disabled.

Reference graph

Works this paper leans on

35 extracted references · 3 canonical work pages · cited by 1 Pith paper

  1. [1]

    Markov and Yaoyun Shi

    Igor L. Markov and Yaoyun Shi. Simulating quantum computation by contracting tensor networks.SIAM Journal on Computing, 38(3):963–981, 2008

    2008

  2. [2]

    A flexible high-performance simulator for verifying and benchmarking quantum circuits implemented on real hardware.npj Quantum Information, 5:86, 2019

    Benjamin Villalonga, Sergio Boixo, Bron Nelson, Christopher Henze, Eleanor Rieffel, Rupak Biswas, and Salvatore Mandr` a. A flexible high-performance simulator for verifying and benchmarking quantum circuits implemented on real hardware.npj Quantum Information, 5:86, 2019

    2019

  3. [3]

    Hyper-optimized tensor network contraction.Quantum, 5:410, 2021

    Johnnie Gray and Stefanos Kourtis. Hyper-optimized tensor network contraction.Quantum, 5:410, 2021

    2021

  4. [4]

    Solving the sampling problem of the Sycamore quantum circuits.Physical Review Letters, 129:090502, 2022

    Feng Pan, Keyang Chen, and Pan Zhang. Solving the sampling problem of the Sycamore quantum circuits.Physical Review Letters, 129:090502, 2022

    2022

  5. [5]

    Quantum supremacy using a programmable superconducting processor

    Frank Arute et al. Quantum supremacy using a programmable superconducting processor. Nature, 574:505–510, 2019

    2019

  6. [6]

    Adam G. M. Lewis et al. Large-scale distributed linear algebra with tensor processing units. Proceedings of the National Academy of Sciences, 119(33):e2122762119, 2022

    2022

  7. [7]

    Quantum circuit simu- lation by SGEMM emulation on tensor cores and automatic precision selection

    Hiroyuki Ootomo, Hidetaka Manabe, Kenji Harada, and Rio Yokota. Quantum circuit simu- lation by SGEMM emulation on tensor cores and automatic precision selection. InHigh Per- formance Computing (ISC High Performance 2023), volume 13948 ofLecture Notes in Com- puter Science, pages 259–276, 2023. doi: 10.1007/978-3-031-32041-5 14. arXiv:2303.08989. 22

    arXiv 2023

  8. [8]

    Simulation of quantum physics with tensor processing units: brute-force computation of ground states and time evolution.arXiv preprint arXiv:2111.10466, 2021

    Markus Hauru et al. Simulation of quantum physics with tensor processing units: brute-force computation of ground states and time evolution.arXiv preprint arXiv:2111.10466, 2021

    arXiv 2021

Show all 35 references
  1. [9]

    Simulation of quantum many-body dynamics with tensor processing units: Floquet prethermalization.PRX Quantum, 3:020331, 2022

    Alan Morningstar et al. Simulation of quantum many-body dynamics with tensor processing units: Floquet prethermalization.PRX Quantum, 3:020331, 2022

    2022

  2. [10]

    Density matrix renormalization group with tensor processing units

    Martin Ganahl et al. Density matrix renormalization group with tensor processing units. PRX Quantum, 4:010317, 2023

    2023

  3. [11]

    On multiplication of 2×2 matrices.Linear Algebra and its Applications, 4:381–388, 1971

    Shmuel Winograd. On multiplication of 2×2 matrices.Linear Algebra and its Applications, 4:381–388, 1971. doi: 10.1016/0024-3795(71)90009-7

    1971 doi

  4. [12]

    Nicholas J. Higham. Stability of a method for multiplying complex matrices with three real matrix multiplications.SIAM Journal on Matrix Analysis and Applications, 13(3):681–687,

  5. [13]

    Van Zee and Tyler M

    Field G. Van Zee and Tyler M. Smith. Implementing high-performance complex matrix multiplication via the 3M and 4M methods.ACM Transactions on Mathematical Software, 44(1):1–36, 2018. doi: 10.1145/3086466

    2018 doi

  6. [14]

    A 2 rebit gate universal for quantum computing.arXiv preprint quant-ph/0210187, 2002

    Terry Rudolph and Lov Grover. A 2 rebit gate universal for quantum computing.arXiv preprint quant-ph/0210187, 2002

    2002 arXiv

  7. [15]

    Both Toffoli and controlled-NOT need little help to do universal quantum computing.Quantum Information and Computation, 3(1):84–92, 2003

    Yaoyun Shi. Both Toffoli and controlled-NOT need little help to do universal quantum computing.Quantum Information and Computation, 3(1):84–92, 2003. doi: 10.26421/QIC3. 1-7

    2003 doi

  8. [16]

    A simple proof that Toffoli and Hadamard are quantum universal.arXiv preprint quant-ph/0301040, 2003

    Dorit Aharonov. A simple proof that Toffoli and Hadamard are quantum universal.arXiv preprint quant-ph/0301040, 2003

    2003 arXiv

  9. [17]

    Simulating quantum systems using real Hilbert spaces.Physical Review Letters, 102(2):020505, 2009

    Matthew McKague, Michele Mosca, and Nicolas Gisin. Simulating quantum systems using real Hilbert spaces.Physical Review Letters, 102(2):020505, 2009

    2009

  10. [18]

    Quantum mechanics is *-algebras and tensor networks, 2020

    Andreas Bauer. Quantum mechanics is *-algebras and tensor networks, 2020

    2020

  11. [19]

    Tensor types and their use in physics

    Andreas Bauer and Alexander Nietner. Tensor types and their use in physics. 2022

    2022

  12. [20]

    XLA: Compiling machine learning for peak performance, 2020

    Amit Sabne. XLA: Compiling machine learning for peak performance, 2020. URLhttps: //research.google/pubs/xla-compiling-machine-learning-for-peak-performance/. Google Research

    2020

  13. [21]

    TVM: An automated end-to-end optimizing compiler for deep learning

    Tianqi Chen, Thierry Moreau, Ziheng Jiang, Lianmin Zheng, Eddie Yan, Meghan Cowan, Haichen Shen, Leyuan Wang, Yuwei Hu, Luis Ceze, Carlos Guestrin, and Arvind Krishna- murthy. TVM: An automated end-to-end optimizing compiler for deep learning. In13th USENIX Symposium on Operat...

    2018

  14. [22]

    PyTorch 2: Faster machine learning through dynamic Python bytecode transformation and graph compilation

    Jason Ansel et al. PyTorch 2: Faster machine learning through dynamic Python bytecode transformation and graph compilation. In29th ACM International Conference on Architec- tural Support for Programming Languages and Operating Systems (ASPLOS), pages 929–947, 2024

    2024

  15. [23]

    Interacting quantum observables: categorical algebra and diagrammatics.New Journal of Physics, 13:043016, 2011

    Bob Coecke and Ross Duncan. Interacting quantum observables: categorical algebra and diagrammatics.New Journal of Physics, 13:043016, 2011. doi: 10.1088/1367-2630/13/4/ 043016

    2011 doi

  16. [24]

    Cambridge University Press, 2017

    Bob Coecke and Aleks Kissinger.Picturing Quantum Processes: A First Course in Quantum Theory and Diagrammatic Reasoning. Cambridge University Press, 2017. 23

    2017

  17. [25]

    The complex gradient operator and the CR-calculus, 2009

    Ken Kreutz-Delgado. The complex gradient operator and the CR-calculus, 2009

    2009

  18. [26]

    Differentiable programming tensor networks.Physical Review X, 9:031041, 2019

    Hai-Jun Liao, Jin-Guo Liu, Lei Wang, and Tao Xiang. Differentiable programming tensor networks.Physical Review X, 9:031041, 2019. doi: 10.1103/PhysRevX.9.031041

    2019 doi

  19. [27]

    Automatic differentiation for complex valued SVD, 2019

    Zhou-Quan Wan and Shi-Xin Zhang. Automatic differentiation for complex valued SVD, 2019

    2019

  20. [28]

    Use and implementation of autodifferentiation in tensor network methods with complex scalars, 2019

    Claudius Hubig. Use and implementation of autodifferentiation in tensor network methods with complex scalars, 2019

    2019

  21. [29]

    Multi-tensor contraction for XEB verification of quantum circuits.arXiv preprint arXiv:2108.05665, 2021

    Gleb Kalachev, Pavel Panteleev, and Man-Hong Yung. Multi-tensor contraction for XEB verification of quantum circuits.arXiv preprint arXiv:2108.05665, 2021

    2021 arXiv

  22. [30]

    OMECO: One more einsum contraction order

    Jinguo Liu. OMECO: One more einsum contraction order. GitHub repository, 2026. URL https://github.com/GiggleLiu/omeco. Rust library for tensor-network contraction-order optimization

    2026

  23. [31]

    Harun Bayraktar, Ali Charara, David Clark, Saul Cohen, Timothy Costa, Yao-Lung L. Fang, Yang Gao, Jack Guan, John Gunnels, Azzam Haidar, Andreas Hehn, Markus Hohnerbach, Matthew Jones, Tom Lubowe, Dmitry Lyakh, Shinya Morino, Paul Springer, Sam Stan- wyck, Igor Terentyev, Saty...

    2023 arXiv

  24. [32]

    D. F. Robinson. Comparison of labeled trees with valency three.Journal of Combinatorial Theory, Series B, 11(2):105–119, 1971. doi: 10.1016/0095-8956(71)90020-7

    1971 doi

  25. [33]

    Yao.jl: Extensible, efficient framework for quantum algorithm design.Quantum, 4:341, 2020

    Xiu-Zhe Luo, Jin-Guo Liu, Pan Zhang, and Lei Wang. Yao.jl: Extensible, efficient framework for quantum algorithm design.Quantum, 4:341, 2020

    2020

  26. [34]

    Tensorcircuit: a quantum software framework for the NISQ era.Quantum, 7:912, 2023

    Shi-Xin Zhang, Jonathan Allcock, Zhou-Quan Wan, Shuo Liu, Jiace Sun, Hao Yu, Xing-Han Yang, Jiezhong Qiu, Zhaofeng Ye, Yu-Qin Chen, Chee-Kong Lee, Yi-Cong Zheng, Shao-Kai Jian, Hong Yao, Chang-Yu Hsieh, and Shengyu Zhang. Tensorcircuit: a quantum software framework for the NIS...

    2023

  27. [1992]

    doi: 10.1137/0613043

    doi

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.