{"id":"dd352471-ed59-4ba2-804e-be5b2922f2f7","arxiv_id":"2509.02668","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Existing NISQ optimization passes can reduce estimated FTQC resource requirements, with rotation-gate reduction providing the largest gains.","lead":"This paper tests whether existing NISQ quantum circuit optimization passes can reduce the physical resources needed for fault-tolerant quantum circuits, using Microsoft's Azure resource estimator. It finds that optimizing rotation gates helps a lot, while optimizing Clifford gates barely matters, and it offers guidance for FTQC compiler development.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Concern: the \"no need to start from scratch\" claim is tested only through Azure QRE's default PSSPC cost model; if Clifford gates are not effectively free or rotation-layer scheduling differs under another FTQC scheme, Findings 1 and 3 may not generalize.","rationale":"The reader's weakest assumption identifies the Azure QRE cost model as the key unvalidated bridge between the NISQ optimization results and the FTQC conclusion. My reading agrees: the paper's central claim is an empirical generalization about FTQC compiler development, but the only FTQC metric measured is produced by one estimator with default parameters. The paper does acknowledge many caveats internally, and the qualitative result that rotation-gate reduction helps is likely robust because non-Clifford rotations are expensive under essentially all known FTQC schemes. However, the more surprising findings - Clifford optimization being useless and the qubit/runtime trade-off - are directly tied to the PSSPC assumption of free Clifford gates. A sensitivity or cross-estimator check would settle whether these findings are properties of FTQC itself or artifacts of the chosen cost model. Since the reader already issued a CONDITIONAL verdict asking for sensitivity analysis, my recommendation is to keep that verdict rather than change it.","tokens_in":11384,"tokens_out":5174,"duration_ms":52437,"concrete_test":"Re-run the complete Qiskit/TKET pass benchmark on at least qft, qpeexact, qpeinexact, and ae with (i) Azure QRE configurations other than the defaults, including error rates of 1e-3 and 1e-4 and a Clifford-costed accounting that assigns a nonzero cost to Clifford gates, and (ii) an independent estimator (pyLIQTR or Qualtran) on the same circuits. The concern is settled if Finding 2 (rotation reduction drives FTQC benefit) and the sign of the Finding 3 qubit-runtime trade-off are preserved; if Clifford-only passes become materially beneficial or the qft TemplateOptimization qubit/runtime signs reverse, the headline conclusion must be restricted to the PSSPC cost model.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central argument equates Azure QRE's default PSSPC-based estimates with FTQC resource requirements. In that model Clifford gates and pi/2-multiple rotations are treated as free, so Finding 1 (Clifford optimization does not help FTQC) and the qubit-runtime trade-off in Finding 3 are statements about that specific accounting. Section V.E then generalizes to 'existing NISQ optimization routines' without restricting the conclusion to PSSPC. Under other credible FTQC implementations, e.g. schemes without resource-free multi-qubit Pauli measurements or with finite-cost Clifford operations, reducing Clifford count can reduce lattice-surgery overhead, and changes in rotation-gate layers may not map to the same qubit-versus-runtime trade. The observed FTQC improvements for qft and qpe (Tables I-II: e.g. CommutativeCancellation -14% #G, -40% #Q, -23% t) are therefore not demonstrated to be stable across plausible cost models. The small 5-10 qubit circuit set and the use of a single tool with default parameters make this a single-point estimate rather than a broadly established result. This is a caveat rather than an invalidation: rotation-gate reduction is likely expensive under most architectures, so Finding 2 is probably robust, but the unqualified 'do not start from scratch' answer depends on cost-model portability.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper investigates whether quantum circuit optimization passes developed for the NISQ era can be reused to reduce FTQC resource requirements. The authors take 11 small benchmark circuits from MQT Bench, apply five Qiskit and five TKET optimization passes, and compare NISQ gate counts with Azure Quantum Resource Estimator outputs (physical qubits and runtime) before and after optimization. They report three findings: Clifford-gate optimization helps NISQ gate counts but not FTQC estimates; reducing non-Clifford rotation gates produces strong FTQC improvements; circuit rearrangement trades off physical qubits against runtime. Based on these results, the paper concludes that FTQC optimization does not have to be developed from scratch, but that NISQ routines should be adapted with attention to gate types and the qubit-runtime trade-off.","tokens_in":11722,"tokens_out":5095,"duration_ms":48331,"significance":"If the conclusions hold, the paper provides a useful empirical starting point for FTQC compiler design and gives NISQ tool developers concrete guidance on where to invest effort. The study is honest about its proxy nature and includes an open-source implementation, an external resource estimator (which avoids circularity), multiple optimization passes, and a range of circuits. The main risk is that the conclusions are derived from a single cost model (Azure QRE with default PSSPC parameters) and a small circuit set, so the breadth of the 'do not start from scratch' answer is not yet fully established. The paper is nevertheless a credible and reproducible contribution to an emerging discussion.","major_comments":[{"comment":"The central conclusion is stated without restricting its scope to the resource estimator's PSSPC model. In the PSSPC layout used by Azure QRE, Clifford gates are free by construction, so Finding 1 is a property of that accounting, and Finding 3 reflects rotation-layer scheduling under the same model. The paper should either explicitly scope the answer to PSSPC-like schemes or add a portability check (e.g., using Qualtran or pyLIQTR, or varying the QRE protocol and parameters) before saying 'we can build upon existing NISQ optimization routines.' Without this, the load-bearing generalization from Section V.E is not fully supported.","section":"Section IV.A, V.B, V.E"},{"comment":"The quantitative evidence is limited to relative percentage changes from a single run with default parameters and no absolute resource numbers or error estimates. For example, the qwalk-v-chain row shows -41% in #Q and -1% in t under HoareOptimizer, but without absolute values and sensitivity analysis it is difficult to tell whether the effect is robust. Please report the underlying QRE outputs and vary at least the logical error budget, qubit model, and T-gate approximation settings to confirm that the qualitative findings and the ranking of passes are stable.","section":"Tables I and II"},{"comment":"The mechanistic claim that rotation-gate reduction drives the FTQC benefit is inferred from the QRE model rather than demonstrated directly on the circuits. The paper should report the actual numbers of rotation gates and rotation-gate layers before and after optimization for representative cases such as qft and qpe. This would make the causal claim testable and would strengthen the practical guideline concerning which NISQ passes to adapt.","section":"Section V.C"}],"minor_comments":[{"comment":"Capitalize 'Software Development Kit' in the phrase 'IBM's quantum Software Development Kit (SDK) Qiskit'.","section":"Section I"},{"comment":"The text refers to green and no color in the tables; ensure the printed version is self-contained when colors are not visible, for example by using symbols or boldface.","section":"Section V.B"},{"comment":"The phrase 'rotation gates (with angles ≠ π/2 or multiples thereof)' is ambiguous; write 'rotation gates with angles that are not multiples of π/2'.","section":"Section V.E"},{"comment":"Specify whether 'TK gates' refers to TK1 gates only, and state how the custom TK1 decomposition affects the final gate-set translation for TKET circuits.","section":"Section IV.B"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper to know about: Forster et al. run ten NISQ optimization passes from Qiskit and TKET over eleven MQT Bench circuits and ask whether the same optimizations still help after the circuits are translated into FTQC resource estimates via Azure Quantum Resource Estimator. The new thing is the systematic comparison matrix: which passes actually move the FTQC qubit and runtime numbers, not just gate count. Their headline answer—no, we don't have to start from scratch—is supported for the tested configurations, with the important caveat that the conclusion is only as strong as the PSSPC cost model.\n\nWhat the paper does well: it uses an external tool for the FTQC side, so the findings are empirical properties of Azure QRE rather than fitted guesses; it ships code; the circuit set is a standard benchmark. The three findings are sensible: Clifford-gate removal helps NISQ but not FTQC under this cost model; rotation-gate reduction gives large FTQC benefits; and passes that rearrange gates can trade off qubits and runtime. The discussion in Section V.E is appropriately cautious—they explicitly warn against blind application.\n\nWhere it is soft: the whole evaluation is single-point. One estimator, default parameters, eleven small circuits, no error bars, no absolute resource numbers, no sensitivity analysis. The stress-test concern is legitimate: if a different FTQC scheme doesn't make Clifford gates effectively free, Finding 1 and the qubit-runtime trade-off in Finding 3 may look different. The paper actually acknowledges the PSSPC dependence in Section IV.A, but the \"No, we don't have to start from scratch\" framing in Section V.E could have been restricted to that scheme more explicitly. I'd also like to see at least one alternative cost model or a discussion of how results might shift. That said, the rotation-gate finding is probably robust across schemes, so the core practical advice holds.\n\nIs it worth a serious referee? Yes. It is a clean empirical study that gives compiler developers a useful starting map, and it is honest about its own limitations. It doesn't overclaim—it presents guidelines, not laws. I would recommend accepting it for peer review, with major revision only in the sense of asking for a broader benchmark and sensitivity analysis. It belongs in a quantum software venue.\n\nMy verdict: if I were an editor, I'd send it to review. It's not a breakthrough, but it is exactly the kind of systematic evaluation the compiler community needs.\n\nBest.","headline":"A solid empirical benchmark mapping NISQ optimization passes to FTQC resource estimates, with the caveat that all conclusions rest on Azure QRE's PSSPC cost model.","tokens_in":12158,"tokens_out":2308,"would_cite":true,"duration_ms":20889,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper asks whether NISQ optimization routines can be reused for fault-tolerant quantum computing, and answers that they can, provided the focus shifts from total gate count to rotation-gate reduction.","keywords":["NISQ","FTQC","quantum circuit optimization","resource estimation","rotation gates","Clifford gates","qubit-runtime trade-off","quantum circuit compilation"],"falsifier":"Run the same before/after optimization comparison using a resource-estimation tool or error-correction model that assigns non-negligible cost to Clifford gates; if Clifford-optimization passes then produce FTQC savings comparable to rotation-gate reductions, Finding 1 would not generalize, and the paper's central guidance would need to be limited to PSSPC-style schemes.","tokens_in":11138,"feed_emoji":"⚛️","tokens_out":7428,"duration_ms":62055,"temperature":0.7,"pith_summary":"Fault-tolerant quantum computing (FTQC) will require error-corrected circuits whose qubit and gate overheads are far larger than today's NISQ devices, so circuit optimization for FTQC is a pressing problem. This paper asks whether the optimization passes already developed for NISQ circuits can be used as a starting point for FTQC optimization, rather than requiring entirely new tools. It reports that existing Qiskit and TKET optimization passes do reduce the estimated FTQC resource requirements on a set of benchmark circuits. The benefit is not uniform: reductions in rotation gates with non-Clifford angles drive the FTQC improvements, while Clifford-gate elimination helps NISQ metrics but not estimated FTQC resources. The authors conclude that NISQ routines can serve as a basis for FTQC optimization, but must be fine-tuned toward rotation-gate reduction and the qubit-runtime trade-off.","feed_headline":"NISQ optimizers carry over to fault-tolerant quantum computing","feed_subtitle":"Existing Qiskit and TKET passes reduce estimated FTQC qubit and runtime needs, largely by cutting rotation gates.","key_machinery":"The central mechanism is the cost asymmetry between Clifford gates and rotation gates in the resource-estimation model. The Azure Quantum Resource Estimator, built on the parallel synthesis sequential Pauli computation (PSSPC) scheme, treats Clifford gates as effectively free and therefore only accounts for logical qubits, T gates, single-qubit rotations, rotation-gate layers, CCZ gates, CCiX gates, and measurements. Because arbitrary-angle rotations are implemented by approximating them with T gates, reducing their count or rearranging their layers directly changes the estimated number of physical qubits and the runtime. This asymmetric cost structure is what makes rotation-gate reduction the dominant lever, and it also explains why Clifford-focused optimization passes show no FTQC benefit.","core_discovery":"On the paper's own terms, the central discovery is that current NISQ optimization techniques transfer to the FTQC setting: they reduce the estimated physical-qubit count and runtime of fault-tolerant circuits, so FTQC-oriented optimizers do not have to be written from scratch. This is established by running ten optimization passes from Qiskit and TKET on benchmark circuits, estimating FTQC resources with the Azure Quantum Resource Estimator before and after each pass, and comparing relative changes. Three findings emerge: Clifford-gate optimization improves NISQ gate counts but leaves estimated FTQC resources unchanged; reducing the number of rotation gates with angles that are not multiples of $\\pi/2$ produces FTQC benefits that are often much larger than the NISQ gate-count reduction; and passes that rearrange gates can leave gate counts untouched while shifting the balance between physical qubits and runtime. These findings support the paper's conclusion that existing NISQ routines should be adapted rather than replaced.","pith_inferences":["This suggests that NISQ methods that specialize in reducing T-count or synthesizing rotations, rather than generic peephole passes, are the most promising candidates to port to FTQC compilers.","A plausible two-stage compiler strategy would be to first run generic NISQ optimization for rotation reduction, then apply an FTQC-specific layout that eliminates Clifford gates, rather than adapting a single pass end-to-end.","A direct extension would be to use the same estimator-based methodology to search over optimization pass orderings for a Pareto front of qubit-runtime trade-offs, which the paper does not do.","Because the estimator ignores Clifford gates, circuits dominated by Clifford subroutines may behave differently; rerunning the evaluation on arithmetic or stabilizer-heavy benchmarks would test whether the rotation-gate finding is universal."],"forward_implications":["FTQC optimizer development can begin from existing NISQ passes rather than from a blank slate.","Reducing rotation gates with non-Clifford angles should be the first optimization target for FTQC, since it is the main source of estimated resource savings.","Clifford-gate optimization should be deprioritized for FTQC unless the underlying error-correction scheme charges for Clifford gates.","Optimization passes that rearrange gates should be applied with the specific use-case in mind, because they trade physical qubits against runtime.","Benchmarking for FTQC should track T and rotation counts and rotation-gate layers, not only total gate count."],"supporting_citations":[{"why":"Provides the resource estimator used to convert each circuit into estimated FTQC physical-qubit counts and runtimes, the central measurement instrument of the study.","marker":"[32]"},{"why":"Defines the PSSPC scheme and cost model under which Clifford gates are treated as free, which underpins Finding 1 and the estimator's input categories.","marker":"[33]"},{"why":"Supplies the Qiskit SDK whose five optimization passes are evaluated.","marker":"[29]"},{"why":"Supplies the TKET SDK whose five optimization passes are evaluated.","marker":"[30]"},{"why":"Supplies the MQT Bench circuits used as the representative benchmark set.","marker":"[31]"},{"why":"Supports the use of resource estimation as a development-time tool for quantum applications, motivating the methodology.","marker":"[34]"}],"fun_headline_variants":["NISQ optimizers worth keeping for fault-tolerant era","Rotation reduction, not Clifford passes, benefits FTQC","NISQ gate reduction doesn't guarantee FTQC savings","Adapt NISQ optimizers, don't rewrite for FTQC"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the Azure Quantum Resource Estimator's cost model, with its PSSPC assumption that Clifford gates are free, accurately captures real FTQC resource requirements; if that model is unrepresentative of other error-correction schemes, the findings on which optimizations help may not generalize.","fun_headline_variants_meta":{"raw":{"variants":["NISQ optimizers worth keeping for fault-tolerant era","Rotation reduction, not Clifford passes, benefits FTQC","NISQ gate reduction doesn't guarantee FTQC savings","Adapt NISQ optimizers, don't rewrite for FTQC"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000558,"raw_usage":{"total_tokens":2699,"prompt_tokens":1036,"completion_tokens":1663,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":652,"completion_tokens_details":{"reasoning_tokens":1595}},"tokens_in":652,"tokens_out":1663,"duration_ms":11435,"temperature":1.0,"reasoning_tokens":1595,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T16:35:40.115762+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same before/after optimization comparison using a resource-estimation tool or error-correction model that assigns non-negligible cost to Clifford gates; if Clifford-optimization passes then produce FTQC savings comparable to rotation-gate reductions, Finding 1 would not generalize, and the paper's central guidance would need to be limited to PSSPC-style schemes.","supporting_citations":[{"cited_title":"Using azure quantum resource estimator for assessing performance of fault tolerant quantum computation,","cited_arxiv_id":null,"evidence_quote":"Provides the resource estimator used to convert each circuit into estimated FTQC physical-qubit counts and runtimes, the central measurement instrument of the study."},{"cited_title":"Quantum computing with Qiskit,","cited_arxiv_id":null,"evidence_quote":"Supplies the Qiskit SDK whose five optimization passes are evaluated."},{"cited_title":"t|ket〉: A retargetable compiler for NISQ devices,","cited_arxiv_id":null,"evidence_quote":"Supplies the TKET SDK whose five optimization passes are evaluated."},{"cited_title":"MQT Bench: Benchmarking software and design automation tools for quan- tum computing,","cited_arxiv_id":null,"evidence_quote":"Supplies the MQT Bench circuits used as the representative benchmark set."},{"cited_title":"Utilizing resource estimation for the development of quantum comput- ing applications,","cited_arxiv_id":null,"evidence_quote":"Supports the use of resource estimation as a development-time tool for quantum applications, motivating the methodology."}],"review_version":2}