Pith. sign in

REVIEW 5 major objections 5 minor 24 references

Clifford and Non-Clifford Splitting in Quantum Circuits: Applications and ZX-Calculus Detection Procedure

T0 review · 5 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A ZX-Calculus procedure, the Clifford border split, separates Clifford+T circuits into Clifford and non-Clifford parts and makes their classical simulation faster for up to about ten qubits.

desk verdict The paper has one clean idea (the projector-circuit distribution identity) and an under-specified, unproven border-splitting algorithm that everything else leans on; the split needs a proof or a counterexample before the speedups can be trusted. read the letter →

arxiv 2504.16004 v1 pith:CKGCWYOU submitted 2025-04-22 quant-ph

Fernando Lima , Arcesio Castañeda Medina This is my paper · ORCID
classification quant-ph
keywords CliffordborderdetectionZX-CalculusClifford+TcircuitsstatevectorsimulationstabilizertableauVQEcircuitsplittingGottesman-Knilltheorem
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 claims that many Clifford+T quantum circuits can be written as a product of a Clifford unitary and a non-Clifford unitary, separated by a detectable cut it calls the Clifford border, and that this split has practical payoff. The payoff is demonstrated with a composite simulation strategy: the Clifford segment is handled by a fast stabilizer-tableau method, converted into a statevector, and the remaining non-Clifford segment is evolved normally. On the paper's random 2-to-10-qubit Clifford+T circuits, this composite simulation is systematically faster than running the whole circuit with a statevector simulator, with the advantage disappearing around 11 qubits. The same split is used to speed up VQE for a small molecule and to reduce the quantum workload in expectation-value estimation by absorbing the Clifford segment into the measured observable. A sympathetic reader would take the paper to establish a new ZX-Calculus-based splitting algorithm and a set of concrete small-circuit speedups.

What carries the argument

The load-bearing mechanism is the Clifford border itself, defined as the cut in a circuit-like ZX diagram that separates a Clifford unitary on one side from a non-Clifford unitary on the other, found by iterative spider fusion/unfusion and recursive handling of two-qubit gates that straddle the cut. The other half of the machinery is the composite simulation pattern built on that cut: a stabilizer tableau for the Clifford segment, a tableau-to-statevector conversion, and statevector evolution of the non-Clifford segment. In the $U_C U_{NC}$ direction, the same cut turns a Clifford segment into a classical observable transformation, because Clifford conjugation maps Pauli observables to Pauli observables.

What would settle it

Run the border detection on a family of random 2-to-10-qubit Clifford+T circuits and check two things: whether the extracted left section passes a Clifford tableau stabilizer test, and whether the composite simulation is faster than full statevector simulation on each circuit. A single circuit in which the left section is not Clifford, or where the composite method loses on any tested sample in that range, would contradict the paper's central claim.

Watch

Extended reading notes

Core claim

The central discovery is the Clifford Border detection procedure: a circuit-level ZX diagram is rewritten so that every non-Clifford spider, meaning a spider whose phase is not an integer multiple of $\pi/2$, is pushed as far right as possible using the spider fusion and unfusion rules. The parsed vertices left of the pushed spiders mark the border, and any two-qubit gate crossing the border is recursively pulled onto the non-Clifford side, giving the circuit the form $U_C U_{NC}$ (or $U_{NC} U_C$ after running the same procedure on the adjoint). This split enables a composite simulator that applies a Clifford tableau to $U_C$, converts the resulting stabilizer generators into a statevector, and evolves that statevector through $U_{NC}$. On randomly generated Clifford+T circuits with 2 to 11 qubits, T-gate probability 20%, and depth 40, the composite method is systematically faster than full statevector simulation for 2 through 10 qubits and crosses over around 11 qubits; the same pattern carries into the paper's H2 VQE experiments.

Load-bearing premise

The load-bearing premise is that the iterative spider fusion/unfusion pushing always comes to a stop with a diagram whose left section is a genuine Clifford unitary; no proof of termination, uniqueness, or correctness is given in the paper.

Editorial extensions

If this is right

  • Random Clifford+T circuits with a non-trivial left Clifford section can be classically simulated by tableau-plus-statevector faster than by full statevector simulation for up to about 10 qubits; the paper reports execution times as low as 5% of the baseline at 2-3 qubits.
  • The splitting procedure shortens the circuit that must run on quantum hardware for a meaningful share of small benchmark circuits, with depth reductions spread across the 10%-90% range.
  • For VQE with Clifford+T ansatzes and statevector simulation, replacing the Clifford part by a precomputed stabilizer state and computing expectation values directly from the statevector reduces wall-clock time, with the gain shrinking as ansatz depth grows.
  • For the $U_C U_{NC}$ ordering, the Clifford segment can be moved out of the quantum execution path by classically transforming the Pauli observable at $O(n^2)$ cost.

Reading between the lines

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

  • The procedure's termination and uniqueness are not established; if two different orders of spider rewrites can produce different borders or leave a non-Clifford spider on the left, circuits would need a canonical rewrite order before the split can be treated as well-defined.
  • Because the crossover at about 11 qubits is set by the cost of the tableau-to-statevector conversion, faster conversion algorithms would push the size range where composite simulation wins beyond 10 qubits without changing the splitting procedure.
  • The same border detection could be applied recursively to isolate multiple Clifford segments, leading to alternating $U_C U_{NC} U_C U_{NC}\cdots$ forms; the paper notes the two-sided $U_C U_{NC} U_C$ possibility as future work but does not implement it.
  • The paper's argument suggests a hybrid execution mode, keeping the Clifford segment classical and sending only the non-Clifford segment to a quantum device, although no such experiment is reported.
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

5 major / 5 minor

Summary. The manuscript proposes a procedure called Clifford Border Detection that, starting from an extracted circuit-like ZX diagram of a Clifford+T circuit, attempts to push non-Clifford spiders as far right as possible and then defines a cut separating a left Clifford unitary U_C from a right non-Clifford unitary U_NC, so that the full circuit equals U_C·U_NC or, after adjointing, U_NC·U_C. It then presents several applications of such a split: classical statevector simulation by initializing the stabilizer state through Stim's tableau-to-statevector conversion and evolving through the non-Clifford section; VQE speedups on small Clifford+T ansatzes; replacement of right Clifford sections by projector circuits that reproduce the output distribution; and classical transformation of observables for expectation-value estimation. Benchmarks on QASMBench, MQT Bench, and randomly generated 2- to 10-qubit Clifford+T circuits are used to support claims of depth reduction and simulation speedup up to about 10 qubits.

Significance. If the border split were proven correct and efficiently implementable, the paper would contribute a practical heuristic for splitting Clifford+T circuits and several plausible use cases. The composite simulation method is a sensible way to leverage Stim's fast tableau representation and to_statevector conversion on the Clifford part, and the projector-distribution argument in Section 3.2.2 is a neat observation. The paper is also honest about the naive pushing procedure's dependence on circuit shape and about the narrow qubit range of the claimed speedups. However, the central splitting algorithm is currently specified only informally and without a correctness proof, and no code or data are provided; these omissions make the empirical claims difficult to evaluate. The paper does not rely on fitted parameters, which is a point in its favor, but the absence of a machine-checked or even fully formal description of the main procedure is a serious gap for an algorithmic claim of this kind.

major comments (5)
  1. [Section 2, Clifford Border Detection Algorithm] The pushing step is not a well-defined rewriting procedure. The text says to 'iteratively use the spider fusion and unfusion rule' on non-Clifford spiders 'when possible, commuting them with adjacent green spiders until it is no longer possible'. Fusion and unfusion only merge or split same-colour spiders connected by a plain wire; they are not a commutation rule, and moving a non-Clifford phase through a multi-qubit Clifford spider generally changes the phase into a multi-qubit Pauli rotation. The manuscript gives no termination argument, no confluence argument, and no proof that the final left section contains only phases that are integer multiples of pi/2. The recursive border update ('until no such situation occurs') also needs a termination proof. Since every downstream use case in Sections 3.1 and 3.2 assumes an exact split U = U_C U_NC, this is a load-bearing gap.
  2. [Section 3.1.3, Figures 9 and 10] The reported composite simulation time does not state whether the time spent by the border detection and pushing procedure is included. If it is not included, the end-to-end comparison is not a fair comparison; if it is included, the timing methodology should say so explicitly and report the splitting overhead separately. The text should also specify the random circuit generation procedure, including random seeds, the gate set, the sampling distribution, and the versions of Qiskit and Stim used, since otherwise the claim that the composite procedure is 'systematically better' cannot be reproduced.
  3. [Section 3.1.3, statistical support] The claim that the composite method is systematically better for 2 to 10 qubits is supported only by averages over 7 runs with no confidence intervals or statistical tests. Given the visible variance in the plots and the authors' own explanation of local spikes by random variation in the extracted Clifford depth, the evidence is not sufficient for a strong universal claim. More runs or a per-circuit paired comparison would be needed to separate the effect of the splitting from random circuit-to-circuit variation.
  4. [Section 3.1.2, Figure 7] The statement that points above the upper boundary are 'guaranteed' to be improved by projector initialization is not justified. The projector-circuit depth bound of 3 + n^2 assumes that every controlled generator can be implemented with a single controlled operation at unit depth, and it ignores the cost of synthesizing controlled Pauli rotations and the effect of failed post-selection: measurement outcomes other than |0...0> require discarding or correction, so the expected cost of a successful projection can be larger than the indicated bound.
  5. [General, reproducibility] The manuscript does not include code or data for the border detection algorithm, the random circuit generation, or the timing benchmarks. For an algorithmic paper whose central claims are empirical, this makes it impossible to check whether the pushing procedure terminates and yields a valid split on the claimed circuit classes, and it prevents independent verification of Figures 4 through 11. A public implementation of the full pipeline and benchmark scripts should be provided.
minor comments (5)
  1. [Section 1.1] The definition of spiders says 'n-input and m-input green objects' and 'n-input and m-input red objects'; the second occurrence should read 'n-input and m-output', and the input/output convention should be stated explicitly.
  2. [Throughout] The phrases 'n-dimensional Hilbert space' and 'n-dimensional Clifford group' should be '2^n-dimensional Hilbert space' and 'n-qubit Clifford group', respectively, to avoid confusion between Hilbert-space dimension and qubit count.
  3. [Figure 3] The caption refers to vertices 61 and 62 without explaining how the reader can identify them in the diagram; a clearer annotation of the border, the pushed spiders, and the moved CZ gate would be helpful.
  4. [Section 3.1.3] There is a typo: 'sistematically' should be 'systematically'. The sentence 'as can be seen for a qubit count of 11' could also be rephrased for clarity.
  5. [Section 3.2.2] The notation X_n is used both for a Pauli string that toggles qubits according to bitstring n and for the number of qubits n; this makes the derivation harder to follow and should be clarified.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the Clifford-border split is an under-specified heuristic but not an input-output tautology, and all downstream benchmarks are external.

full rationale

The derivation chain is not circular. Section 2's border detection is a constructive heuristic: it pushes non-Clifford ZX spiders to the right and defines a cut by the positions reached; the assertion that the left section is a Clifford unitary is an unproven correctness claim, not an identification of the border with the Clifford section by definition. The downstream uses in Sections 3.1.3, 3.1.4, 3.2.1, and 3.2.2 assume the split and then compare against external tools (Qiskit AerSimulator, Stim, Qiskit Nature); no fitted parameter is later relabeled as a prediction. The projector-state derivation follows from standard Pauli commutation and the Gottesman-Knill tableau formalism, and the expectation-value transformation A' = U_C^dag A U_C is a mathematical identity, not a fitted input. The external algorithms cited for the statevector conversion (Stim) and for Clifford absorption (ref. [15], QuCLEAR) are not prior work by the present authors, so there is no load-bearing self-citation chain. The paper's genuine weakness is that the pushing procedure is under-specified and its termination and exactness are not proved; that is a correctness and reproducibility risk, not circular reasoning. The depth-reduction percentages are descriptive outputs of the algorithm rather than predictions obtained from the algorithm's own outputs, and even the VQE speedup is attributed to an external implementation detail (Statevector.expectation_value) rather than to a quantity fitted from the split. Therefore no circular step can be exhibited from the manuscript.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The central mathematical content relies on standard theorems: the soundness of ZX-calculus rewrite rules, graph-like diagram extraction, the Gottesman-Knill theorem, and Clifford group properties. The paper assumes the circuits are in the Clifford+T gate set and that the border detection pushes non-Clifford spiders successfully, which is an unproven domain assumption. No free parameters are fitted; the benchmark generation settings, such as qubit number, depth, and T-gate probability, are experimental inputs, not parameters of the derivation.

assumptions (6)
  • standard math ZX-calculus rewrite rules are sound and complete for the diagrams considered (rules cited from [13]).
    The border detection algorithm in Section 2 relies on rewriting diagrams with the f and h rules of Figure 1, which are standard sound ZX-calculus rules.
  • standard math Every graph-like ZX diagram can be extracted into a circuit-like form via the procedure of Backens et al. [2].
    The Clifford Border detection starts from a circuit-like ZX diagram obtained by extraction, as stated at the beginning of Section 2.
  • standard math Stabilizer circuits can be simulated classically in polynomial time via the Gottesman-Knill theorem [1,11].
    The use cases in Section 3 explicitly rely on the Gottesman-Knill theorem, cited in the introduction and Section 1.2.
  • standard math The Clifford group normalizes the Pauli group, so conjugating a Pauli observable by a Clifford yields another Pauli [16].
    This is used in Section 3.2.1 for observable transformation and in Section 3.2.2 for the commutation relations of transformed Pauli strings.
  • domain assumption The circuits under consideration are Clifford+T circuits, meaning gates are drawn from Clifford gates and T-like phase gates, and non-Clifford spiders are those with phases not multiples of pi/2.
    The algorithm in Section 2 is defined for 'Clifford+T ZX diagrams', and the benchmarks are generated as Clifford+T circuits with a given T-gate probability.
  • domain assumption For the projector circuit distribution result in Section 3.2.2, the non-Clifford section is applied to |0> and the full circuit is in UC UNC form, with decoupled generators for the stabilizer state U_C*|0>.
    The derivation assumes the target circuit is split in this exact order and that the stabilizer generators of the Clifford-transformed computational basis states can be tracked via commutation relations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Clifford and Non-Clifford Splitting in Quantum Circuits: Applications and ZX-Calculus Detection Procedure." pith.science (2026). https://pith.science/paper/CKGCWYOU

@misc{pith2026250416004,
  author       = {Pith},
  title        = {Pith review of: Clifford and Non-Clifford Splitting in Quantum Circuits: Applications and ZX-Calculus Detection Procedure},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CKGCWYOU}},
  note         = {Machine review of arXiv:2504.16004}
}
read the original abstract

Classical simulation of quantum circuits is a pivotal part of the quantum computing landscape, specially within the NISQ era, where the constraints imposed by available hardware are unavoidable. The Gottesman-Knill theorem further motivates this argument by accentuating the importance of Clifford circuits and their role on this topic of simulation. In this work, we propose and analyze use cases that come from quantum circuits that can be written as product between a Clifford and a Non-Clifford unitary, these ranging from fully classical emulation, hybrid quantum-classical execution or even quantum algorithm simplification. To further complement this analysis, we make use of ZX-Calculus and its assets to detect a limiting border of these circuits that would allow for a separation between a Clifford section and a Non-Clifford section. To achieve this, we present a novel procedure for parsing ZX diagrams, that not only allows for the detection of this border but also simplifies the circuit extraction process.

Figures

Figures reproduced from arXiv: 2504.16004 by the authors.

Figure 1
Figure 1. Set of ZX-Calculus rewrite rules One particular set of diagrams that is greatly important to this work is the set of Graph-like ZX-Diagrams, whose form will allow for application of a circuit extraction procedure that will be introduced shortly. In accordance to its definition 7 , a ZX-Diagram is graph-like when: • It only contains Z-spiders • Its Z-spiders are exclusively connected via Hadamard wires • It has no pa… view at source ↗
Figure 2
Figure 2. Example of post-selection generator projector circuit for a a 3- [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Example of the Clifford Border on a 4-qubit Clifford+T circuit [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Counts for each percentile corresponding to the relative depth [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Counts for each percentile corresponding to the relative depth [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Number of gates of each randomly generated Clifford+T circuit [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 8
Figure 8. Figure 8: Execution time of Qiskit’s Statevector simulator for randomly [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: Execution time of Qiskit’s Statevector simulator for randomly [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Execution time of Qiskit’s Statevector simulator for ran [PITH_FULL_IMAGE:figures/full_fig_p008_10.png]
Figure 11
Figure 11. Figure 11: Execution times of VQE procedure for the [PITH_FULL_IMAGE:figures/full_fig_p009_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

24 extracted references · 13 canonical work pages

  1. [1]

    Aaronson S., Gottesman D., 2004, Physical Review A, 70

    work page 2004

  2. [2]

    Backens M., Miller-Bakewell H., de Felice G., Lobski L., van de We- tering J., 2021, Quantum, 5, 421

    work page 2021

  3. [3]

    E., et al., 2022, Assessing requirements to scale to practical quantum advantage ( arXiv:2211.07629), https://arxiv

    Beverland M. E., et al., 2022, Assessing requirements to scale to practical quantum advantage ( arXiv:2211.07629), https://arxiv. org/abs/2211.07629

    arXiv 2022

  4. [4]

    Bravyi S., Kitaev A., 2005, Physical Review A, 71

    work page 2005

  5. [5]

    pp 104–113, doi:10.1109/QCE53715.2022.00029

    Camps D., Van Beeumen R., 2022, in 2022 IEEE International Con- ference on Quantum Computing and Engineering (QCE). pp 104–113, doi:10.1109/QCE53715.2022.00029

    arXiv 2022

  6. [6]

    Chen S., Cotler J., Huang H.-Y., Li J., 2022, The Complexity of NISQ (arXiv:2210.07234), https://arxiv.org/abs/2210.07234

    arXiv 2022

  7. [7]

    Duncan R., Kissinger A., Perdrix S., van de Wetering J., 2020, Quan- tum, 4, 279

    work page 2020

  8. [8]

    Ekert A., Hosgood T., Kay A., Macchiavello C., , Introduction to Quantum Information Science, https://qubit.guide

    work page

Show all 24 references
  1. [9]

    Gidney C., 2021, Quantum, 5, 497

    2021

  2. [10]

    Gottesman D., 1997, Stabilizer Codes and Quantum Error Correc- tion (arXiv:quant-ph/9705052), https://arxiv.org/abs/quant-ph/ 9705052

    1997 arXiv

  3. [11]

    Gottesman D., 1998, The Heisenberg Representation of Quantum Computers ( arXiv:quant-ph/9807006), https://arxiv.org/abs/qu ant-ph/9807006

    1998 arXiv

  4. [12]

    Gottesman D., 2009, An Introduction to Quantum Error Correction and Fault-Tolerant Quantum Computation (arXiv:0904.2557), http s://arxiv.org/abs/0904.2557

    2009 arXiv

  5. [13]

    Kissinger A., van de Wetering J., 2020, Electronic Proceedings in Theoretical Computer Science, 318, 229–241

    2020

  6. [14]

    Li A., Stein S., Krishnamoorthy S., Ang J., 2023, ACM Transac- tions on Quantum Computing, 4

    2023

  7. [15]

    H., Hovland P., 2025, QuCLEAR: Clifford Extraction and Absorption for Quantum Circuit Optimization ( arXiv:2408.13316), https://arxiv.org/abs/2408.1 3316

    Liu J., Gonzales A., Huang B., Saleem Z. H., Hovland P., 2025, QuCLEAR: Clifford Extraction and Absorption for Quantum Circuit Optimization ( arXiv:2408.13316), https://arxiv.org/abs/2408.1 3316

    2025 arXiv

  8. [16]

    Mastel K., 2023, The Clifford theory of the n-qubit Clifford group (arXiv:2307.05810), https://arxiv.org/abs/2307.05810

    2023

  9. [17]

    Mih´ alikov´ a I., Pivoluska M., Plesch M., Fri´ ak M., Nagaj D.,ˇSob M., 2022, Nanomaterials, 12, 243

    2022

  10. [18]

    Quetschlich N., Burgholzer L., Wille R., 2023, Quantum, 7, 1062

    2023

  11. [19]

    R., Wallman J

    Sanders Y. R., Wallman J. J., Sanders B. C., 2015, New Journal of Physics, 18, 012002

    2015

  12. [20]

    J., 2021, Nature Communications, 12

    Wang S., Fontana E., Cerezo M., Sharma K., Sone A., Cincio L., Coles P. J., 2021, Nature Communications, 12

    2021

  13. [21]

    Xu X., Benjamin S., Sun J., Yuan X., Zhang P., 2023, A Herculean task: Classical simulation of quantum computers (arXiv:2302.08880), https://arxiv.org/abs/2302.08880

    2023 arXiv

  14. [22]

    G., Martiel S., 2024, Faster and shorter synthesis of Hamiltonian simulation circuits ( arXiv:2404.03280), https://arxi v.org/abs/2404.03280

    de Brugi` ere T. G., Martiel S., 2024, Faster and shorter synthesis of Hamiltonian simulation circuits ( arXiv:2404.03280), https://arxi v.org/abs/2404.03280

    2024 arXiv

  15. [23]

    de Silva N., Salmon W., Yin M., 2024, Fast algorithms for classical specifications of stabiliser states and Clifford gates (arXiv:2311.10357), https://arxiv.org/abs/2311.10357

    2024 arXiv

  16. [24]

    van de Wetering J., 2020, ZX-calculus for the working quantum computer scientist ( arXiv:2012.13966), https://arxiv.org/abs/20 12.13966 13

    2020 arXiv

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.