{"id":"ab074bdd-0ac2-4713-8342-287668abbbeb","arxiv_id":"2412.20736","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The thesis proves that fault-tolerant encoder and decoder circuits can achieve entanglement-assisted communication rates close to the ideal capacity, and reports trapped-ion error-detection experiments plus a new bound on the entanglement advantage.","lead":"This PhD thesis shows that quantum communication can be protected against noisy encoding and decoding circuits without losing much of the theoretical maximum rate, and that entanglement's advantage over unassisted communication can be bounded for many channels. It combines information-theoretic proofs with small-scale trapped ion experiments.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The δ chosen in Theorem 2.4.4's proof is too small: Eq. (2.4) does not vanish (the prefactor d_B^{(p+δ)n} outruns e^{-nδ²/(2log²λ_min)}), so the AVP rate bound and the explicit f(p) in Theorem 2.6.3 are unsupported as written.","rationale":"Good-faith reading: the thesis's central claim is that fault-tolerant entanglement-assisted communication achieves rates approaching the noiseless entanglement-assisted capacity as the gate error p → 0. For that qualitative claim it suffices to have an AVP coding theorem with any vanishing penalty; the exact O(p log p) rate is a strengthening. I therefore checked the proof of Theorem 2.4.4. The decisive step is Eq. (2.4), which balances the postselection prefactor d_B^{(p+δ)n} against the non-typicality error e^{-nδ²/(2(log λ_min)²)}. With the δ chosen in the text, the balance fails: the prefactor grows like e^{n p log d_B}, while the decay is only e^{-n p log d_B / |log p|}. This is an internal, checkable gap, not a disagreement with consensus. The reader's flagged interface theorem from [27] is a legitimate dependency, but it is a referenced theorem rather than an internal step; the AVP issue is more immediately load-bearing because it is the exact inequality Theorem 2.6.3 reduces to. The issue is probably repairable with δ of order |log p|√p, at the cost of a weaker O(√p log²p) penalty, so I would not reject the thesis outright; it should be accepted conditionally only if this calculation is corrected and the stated bounds updated. The qualitative limit C_F^ea(p)(T) → C^ea(T) may survive, but the proof as written does not establish even the stated quantitative bound without modification.","tokens_in":56790,"tokens_out":19710,"duration_ms":204867,"concrete_test":"Re-run the asymptotic check in Eq. (2.4) for d_A = d_B = 2, p = 10^{-6}, λ_min = p²/(d_A d_B), and the stated δ = sqrt(2 log(d_B) p |log λ_min|): compute d_B^{(p+δ)n} e^{-nδ²/(2(log λ_min)²)} symbolically and confirm whether it grows with n. Then repeat with δ = c |log p|√p for a suitable constant c and verify that Eq. (2.4) holds while η(δ) → 0. If the second choice works, update Theorem 2.4.4's penalty and Theorem 2.6.3's f2(p) accordingly; if neither works, the AVP theorem requires a different coding argument.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Load-bearing concern: the proof of the AVP coding theorem (Theorem 2.4.4) contains an asymptotic rate-balance error that propagates into the central Chapter 2 result. After invoking [27, Lem. IV.10], the proof requires d_B^{(p+δ)n} e^{-nδ²/(2(log λ_min)²)} → 0 (Eq. 2.4), but with the stated δ = sqrt(2 log(d_B) · p · |log λ_min|) and λ_min lower-bounded by p²/(d_A d_B), the prefactor has exponent Θ(p log d_B) per channel use (even taking the auxiliary δ in the lemma to zero), while the exponential decay exponent is δ²/(2 log²λ_min) = Θ(p log d_B / |log p|). For small p the decay is slower than the prefactor growth, so the product diverges. The subsequent rate condition R < I − p log d_B − η(δ) therefore does not guarantee Eq. (2.5). A larger δ of order |log p|√p would restore decay with a penalty O(√p log²p), which still vanishes, but not the O(p log p) stated in Theorem 2.4.4 and used in §2.8.2. Since Theorem 2.6.3 explicitly invokes this AVP bound at perturbation 2(j1+j2)cp, the quantitative lower bound on C_F^ea(p)(T) is not proved as written, although the qualitative limit theorem might survive a repair.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This PhD thesis studies the limits of quantum communication under noise, with and without entanglement assistance. Chapter 2, reproducing arXiv:2210.02939, introduces the fault-tolerant entanglement-assisted capacity C^ea_F(p)(T) for the i.i.d. Pauli gate-error model and develops the technical machinery: effective interface lemmas, an entanglement-assisted capacity under arbitrarily varying perturbation (AVP), fault-tolerant entanglement distillation, and finally a lower bound C^ea_F(p)(T) >= C^ea(T) - ..., with a threshold theorem recovering C^ea(T) as p tends to zero. Chapter 3 reports on experimental investigations of the [[4,2,2]] error-detection code on trapped-ion hardware and states a threshold theorem for demonstrations of fault-tolerance. Chapter 4, explicitly in preparation, proposes bounding the ratio C^ea(T)/C(T) for a large class of finite-dimensional channels via channel-divergence techniques. The thesis is honest about which parts are published, ongoing, or in preparation.","tokens_in":57173,"tokens_out":13208,"duration_ms":129446,"significance":"If the main result is established, Chapter 2 provides a meaningful fault-tolerant analogue of the entanglement-assisted capacity, extending Christandl-Muller-Hermes' unassisted results to the entanglement-assisted setting and showing that noisy encoder/decoder circuits need not reduce the asymptotic rate. The modular proof structure--separating circuit-level fault-tolerance, the AVP information-theoretic model, and distillation--is a genuine strength, and the explicit interface lemmas and theorem statements make the dependence on prior work clear. The quantitative claims in Theorem 2.4.4 and Theorem 2.6.3 are, however, not proved as written because of the rate-balance error described below; the qualitative threshold theorem appears likely repairable. Chapters 3 and 4 are ongoing/in preparation and therefore carry less weight in the evaluation, though they are clearly labeled as such.","major_comments":[{"comment":"The proof's rate balance is not valid as written. After applying [27, Lem. IV.10], Eq. (2.4) requires d_B^{(p+\\tildeδ)n} e^{-n δ^2/(2 log^2 λ_min)} -> 0. With the stated δ = sqrt(2 log(d_B) p |log λ_min|) and λ_min >= p^2/(d_A d_B), the decay rate is δ^2/(2 log^2 λ_min) ≈ (log d_B) p / (2 log(1/p)) per channel use, while the prefactor contributes at least (p + \\tildeδ) n log d_B; even for \\tildeδ -> 0 the prefactor outruns the exponential, so Eq. (2.4) does not vanish and the rate condition (2.6) does not imply Eq. (2.5). A choice δ = Θ(√p |log λ_min|) would restore the convergence with a penalty Θ(√p log^2(1/p)), but then the stated bound g(p) = O(p log p) in Theorem 2.4.4 is no longer what is proved. Independently of the balance issue, the displayed g(p) is Θ(√p log^(3/2)(1/p)) rather than O(p log p), so the asymptotic claim in Theorem 2.4.4 and the scaling discussion in §2.8.2 need correction. The qualitative continuity in Theorem 2.4.5 and Corollary 2.4.6 may survive such a repair.","section":"§2.4.2, Theorem 2.4.4 and Eq. (2.4)"},{"comment":"The statement and proof of Theorem 2.6.3 disagree on the definition of f1(p): the displayed theorem has f1(p) = (h(4cp) + 4cp log(3)) j2 / (1 - h(4cp) - 4cp log(3)), while the proof derives f1(p) = (h(4cp) + 4cp log(3)) log(2) / (1 - h(4cp) - 4cp log(3)). Since the final lower bound uses 4 f1(p) C^ea(T)/C(T), the stated bound is not the bound actually proved when j2 ≠ 1. Moreover, because the proof invokes Theorem 2.4.4 at perturbation strength 2(j1+j2)cp, the quantitative bound in Theorem 2.6.3 inherits the unresolved AVP gap from the first comment. The threshold formulation in Theorem 2.6.1 is qualitative and may be recoverable, but Theorem 2.6.3 as printed is not established.","section":"§2.6, Theorem 2.6.3"}],"minor_comments":[{"comment":"There are several typographical issues, including 'comparitive', 'occuring', and 'architechtures' (e.g., Chapter 3 and §2.8.3), and expressions like '2j1+j2' in Theorem 2.6.3 should be typeset as 2^{j1+j2} to avoid ambiguity.","section":"Throughout"},{"comment":"The proof of Theorem 2.5.4 is only a sketch; since this theorem is a load-bearing subroutine in Section 2.6, the final version should expand the proof or give a precise citation/statement of the underlying threshold argument.","section":"§2.5, Theorem 2.5.4"},{"comment":"The class of channels to which the claimed capacity-ratio bound applies is introduced only through a far/near-replacer dichotomy; the boundary of this class should be stated formally in the main text, not only in the chapter's opening discussion.","section":"Chapter 4"}],"recommendation":"major_revision","confidential_remarks":"The gap in Theorem 2.4.4 is the main obstacle. It is localized and appears fixable, so I do not recommend rejection. The thesis's Chapter 2 is the core; Chapters 3 and 4 are not yet at a publishable standard but are labeled accordingly. I would ask the authors to redo the AVP calculation, update the scaling statements, and reconcile the f1(p) inconsistency before acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this thesis deserves a serious referee, but the central quantitative claim in Chapter 2 is not proved as written. I checked the stress-test concern against the proof and it lands. In Theorem 2.4.4, the chosen δ ≈ sqrt(2 log(d_B) p |log λ_min|) gives an exponential decay with exponent roughly n p log(d_B)/|log p|, while the prefactor d_B^{(p+δ)n} grows like exp(n(p log d_B + sqrt(p log p) log d_B)). For small p the prefactor outruns the decay, so Eq. (2.4) does not go to zero. The explicit g(p) in Theorem 2.4.4 is therefore unsupported, and Theorem 2.6.3 inherits the problem because it invokes that AVP bound. A larger δ of order |log p| sqrt(p) would restore the qualitative statement C_F^ea(p)(T) → C^ea(T), but with a penalty O(sqrt(p) log^2 p), not O(p log p). So the qualitative threshold theorem looks repairable, but the stated rates should not be relied on.\n\nCredit where it is due: the Chapter 2 machinery is modular and careful—the effective interface lemmas, the reduction to an effective channel, and the fault-tolerant entanglement distillation are genuine extensions of Christandl–Müller-Hermes. Chapter 3's threshold theorem for error-detection demonstrations is clean, and the trapped-ion discussion is honest, especially the observation that Aria did not show the expected improvement. Chapter 4 is explicitly in preparation, so there is not enough detail to evaluate it; the abstract's claim about bounding the entanglement advantage should be treated as a research announcement, not a result.\n\nOther soft spots, in proportion: Theorem 1.3.2, the interface failure bound, is imported from [27] and is load-bearing; for a thesis that is acceptable, but it should be flagged as a premise rather than a proof. Chapter 3 has no data artifacts, so its experimental claims are not independently checkable from the text. Reproducing arXiv:2210.02939 as Chapter 2 is fine for a thesis, but a journal submission would need to be framed as a version of record issue. The thesis is open about all of this, including explicit labels of what is ongoing and what is in preparation.\n\nWho is this for: people working on fault-tolerant communication and on small-scale error-detection demonstrations. It is a useful, honest map of the area. A serious referee should engage with it, but the §2.4.2 gap must be fixed before Theorem 2.4.4 or Theorem 2.6.3 are cited for their quantitative bounds. Send it to peer review, conditionally: I would not desk reject, but I would insist the AVP rate-balance argument be corrected.","headline":"Useful thesis with a real gap in the AVP proof: the stated δ does not make Eq. (2.4) vanish, so the explicit capacity bound in Chapter 2 is unsupported as written, though the qualitative recovery likely survives a repair.","tokens_in":779,"tokens_out":1880,"would_cite":false,"duration_ms":57251,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P45","81P68"],"pacs":[],"model":"deepseek-v4-flash","headline":"This thesis proves that entanglement-assisted communication survives faulty encoder and decoder gates: below a gate-error threshold the achievable rate is within $f(p)$ of the noiseless capacity, with $f(p)\\to0$, and it bounds the rate…","keywords":["quantum communication","entanglement-assisted capacity","fault-tolerance","quantum error correction","threshold theorem","entanglement distillation","arbitrarily varying perturbation","quantum Shannon theory"],"falsifier":"Simulate the interface circuits of the concatenated seven-qubit code under the i.i.d. Pauli noise model and measure the probability that an encoding or decoding interface is incorrect as a function of $p$; if this probability does not scale as $O(p)$ for small $p$, the bound $\\mathrm{Prob}(\\text{interface incorrect})\\le 2cp$ that carries the effective-channel theorem and the capacity theorem is false. Alternatively, on a channel such as the qubit depolarizing channel, compute the explicit lower bound of the main capacity theorem and search for gate errors $p$ below threshold where the bound is violated by a numerical estimate of the fault-tolerant capacity.","tokens_in":56589,"feed_emoji":"⚛️","tokens_out":8128,"duration_ms":76022,"temperature":0.7,"pith_summary":"The thesis's central result is that entanglement-assisted communication over any finite-dimensional quantum channel is robust to noise in the encoding and decoding hardware: when the circuits that implement the encoder and decoder suffer independent gate errors of probability $p$ below a threshold, communication can still be carried out at a rate within $f(p)$ of the ideal entanglement-assisted capacity $C^{\\mathrm{ea}}(T)$, with $f(p)\\to 0$ as $p\\to 0$. The proof compiles a standard noiseless coding scheme into a concatenated quantum error-correcting code, moves qubits between the code space and the raw channel through interface circuits, and accounts for residual interface failures as a weak perturbation of the channel for which a new coding theorem is proved. A fault-tolerant entanglement distillation subroutine restores the maximally entangled resource states that pass through noisy interfaces. The thesis also contains a hardware-oriented study of a four-qubit error-detection code on trapped-ion devices and a separate result bounding how much entanglement assistance can improve rates over unassisted communication for a large class of channels.","feed_headline":"Noisy quantum gates barely cut entanglement-assisted capacity","feed_subtitle":"A threshold theorem shows faulty encoder and decoder circuits still reach the noiseless rate as gate error p tends to zero.","key_machinery":"The load-bearing mechanism is the effective-channel reduction. A noiseless entanglement-assisted coding scheme is compiled into fault-tolerant circuits in a concatenated seven-qubit code; interface circuits, whose failure probability is imported as at most $2cp$ under the i.i.d. Pauli model, move qubits between the code space and the physical channel. The effective-interface lemmas show that the noisy compiled circuit is close, in induced 1-to-1 norm, to the ideal circuit acting on an effective channel $T_{p,\\mathcal N}=(1-q)(T\\otimes \\mathrm{Tr}_S)+q\\mathcal N$ with $q=2(j_1+j_2)cp$, fed with an arbitrary syndrome state $\\sigma_S$. The thesis then proves a coding theorem for entanglement-assisted communication under 'arbitrarily varying perturbation'—an infimum over syndrome states and perturbations $\\mathcal N$—giving $C^{\\mathrm{ea}}_{\\mathrm{AVP}}(p,T)\\ge C^{\\mathrm{ea}}(T)-g(p)$ with $g(p)=O(p\\log p)$. A fault-tolerant entanglement distillation subroutine, based on a one-way distillation protocol and using some channel copies for the classical communication it needs, converts the noisy resource states produced by the interfaces into near-perfect maximally entangled states in the code space.","core_discovery":"On the paper's own terms, the main discovery is a threshold-type coding theorem for fault-tolerant entanglement-assisted communication. For every finite-dimensional quantum channel $T$ and every target accuracy $\\eta>0$, there is a gate-error threshold $p_{\\mathrm{th}}(\\eta,T)>0$ such that for all $0\\le p\\le p_{\\mathrm{th}}$, $C^{\\mathrm{ea}}_{F(p)}(T)\\ge C^{\\mathrm{ea}}(T)-\\eta$, where $F(p)$ is the i.i.d. Pauli noise model on the encoder and decoder circuits; equivalently, $\\lim_{p\\to0}C^{\\mathrm{ea}}_{F(p)}(T)=C^{\\mathrm{ea}}(T)$. The theorem is proved by exhibiting explicit fault-tolerant encoders and decoders: a noiseless entanglement-assisted scheme is first approximated by circuits, then implemented in a concatenated seven-qubit code whose level grows with the block length, with interface circuits converting between the code space and the physical channel. After the fault-tolerant compilation, the whole setup is close to a noiseless scheme for an effective channel of the form $(1-q)(T\\otimes\\mathrm{Tr}_S)+q\\mathcal N$ with a correlated syndrome input, and the thesis proves a coding theorem showing that such 'arbitrarily varying perturbations' reduce the rate by at most a function $g(q)=O(q\\log q)$. Fault-tolerant entanglement distillation, run through a subset of channel uses, supplies the clean entanglement needed by the assisted scheme.","pith_inferences":["The effective-channel reduction is modular in the error-correcting code: any family of concatenated codes whose interface circuits fail with probability $O(p)$ and whose extended rectangles satisfy a threshold theorem would inherit the same capacity-recovery result, so verifying such interface bounds for topological or LDPC constructions is a natural next test.","Because the loss bound depends on the ratio $C^{\\mathrm{ea}}(T)/C(T)$, the thesis's own conjecture that this ratio is bounded by a dimension-dependent factor would upgrade the threshold result from channel-dependent to uniform in dimension; the thesis explicitly notes this dependence.","On real hardware the explicit rate loss is dominated by a $p\\log p$ term and is far too large for practical finite-block-length use; the thesis's plots suggest the proof-of-principle nature and point toward tighter continuity bounds or different codes as the route to practical rates.","The trapped-ion comparison offers a device-level check of the i.i.d. Pauli assumption: a newer device with lower gate error but no observed encoded gain is consistent with correlated errors that the model excludes, so hardware experiments can directly inform which noise models the capacity theorem should be extended to."],"forward_implications":["For any finite-dimensional quantum channel, fault-tolerant encoder and decoder circuits achieve entanglement-assisted communication at rates arbitrarily close to the noiseless capacity once the local gate error is below a threshold.","The communication channel itself does not need to be below the fault-tolerance threshold; only the local gates in the encoder and decoder do, so the result applies to on-chip communication and to links that are noisier than the local hardware.","As $p\\to0$, the fault-tolerant entanglement-assisted capacity equals the standard entanglement-assisted capacity, recovering the noiseless quantum Shannon-theoretic limit.","The same modular construction yields fault-tolerant entanglement distillation: noisy resource states entering the code space through faulty interfaces can be purified at the price of a fraction of channel uses, rather than being unusable.","The resource-inequality version of the coding theorem expresses the trade-off as $\\langle T\\rangle + H(A)(T\\otimes\\mathrm{id})(\\varphi)[qq]\\ge_{\\mathrm{FT}(p)} I(A':B)(T\\otimes\\mathrm{id})(\\varphi)[c\\to c]$ up to $O(p\\log p)$ corrections."],"supporting_citations":[{"why":"Supplies the fault-tolerant communication framework, the interface correctness theorem with failure probability at most $2cp$, and the classical-communication coding theorem used as a subroutine.","marker":"[27]"},{"why":"Provides the threshold theorem and well-behaved extended rectangle analysis for concatenated codes that bounds logical error probabilities in the compiled circuits.","marker":"[6]"},{"why":"Constructs the interface circuits for concatenated codes whose low failure probability is the imported premise of the effective-channel lemmas.","marker":"[80]"},{"why":"Gives the one-way entanglement distillation protocol used to purify noisy resource states in the code space.","marker":"[32]"},{"why":"Provides the noiseless entanglement-assisted coding theorem and explicit error bounds that the fault-tolerant construction starts from.","marker":"[120]"},{"why":"Defines the entanglement-assisted capacity, gives the superdense coding motivation, and states the conjecture on capacity quotients that Chapter 4 addresses.","marker":"[14]"},{"why":"Introduces related fully quantum arbitrarily varying channels, providing background and contrast for the arbitrarily varying perturbation model.","marker":"[17]"}],"fun_headline_variants":["Faulty gates preserve quantum capacity at low error","Threshold theorem: quantum capacity robust to gate noise","Entanglement-assisted coding survives faulty circuits","Small gate errors don't hurt quantum communication rate"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole argument borrows an imported guarantee that the interface circuits moving qubits in and out of the error-correcting code fail with probability at most a constant times the gate error $p$; the thesis does not re-prove that guarantee, and if it fails the effective-channel reduction collapses.","fun_headline_variants_meta":{"raw":{"variants":["Faulty gates preserve quantum capacity at low error","Threshold theorem: quantum capacity robust to gate noise","Entanglement-assisted coding survives faulty circuits","Small gate errors don't hurt quantum communication rate"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000203,"raw_usage":{"total_tokens":1409,"prompt_tokens":990,"completion_tokens":419,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":606,"completion_tokens_details":{"reasoning_tokens":361}},"tokens_in":606,"tokens_out":419,"duration_ms":4471,"temperature":1.0,"reasoning_tokens":361,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T23:12:08.909375+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the interface circuits of the concatenated seven-qubit code under the i.i.d. Pauli noise model and measure the probability that an encoding or decoding interface is incorrect as a function of $p$; if this probability does not scale as $O(p)$ for small $p$, the bound $\\mathrm{Prob}(\\text{interface incorrect})\\le 2cp$ that carries the effective-channel theorem and the capacity theorem is false. Alternatively, on a channel such as the qubit depolarizing channel, compute the explicit lower bound of the main capacity theorem and search for gate errors $p$ below threshold where the bound is violated by a numerical estimate of the fault-tolerant capacity.","supporting_citations":[{"cited_title":"Long-distance quantum communication over noisy networks without long-time quantum memory","cited_arxiv_id":null,"evidence_quote":"Constructs the interface circuits for concatenated codes whose low failure probability is the imported premise of the effective-channel lemmas."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the noiseless entanglement-assisted coding theorem and explicit error bounds that the fault-tolerant construction starts from."},{"cited_title":"Fully Quantum Arbitrarily Varying Channels: Random Coding Capacity and Capacity Dichotomy","cited_arxiv_id":null,"evidence_quote":"Introduces related fully quantum arbitrarily varying channels, providing background and contrast for the arbitrarily varying perturbation model."}],"review_version":1}