{"id":"16ba0635-1e95-4be5-9c25-dec4ce5fcf41","arxiv_id":"1909.02342","paper_version":3,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"In butterfly-based networks, achievable classical multicast rates exceed upper bounds on quantum multicast rates, with the gap growing up to one bit per receiver when blocks are added in parallel.","lead":"This paper studies butterfly-shaped communication networks and shows that classical information can be transmitted at higher rates than quantum information through the same topology. The authors find that the gap grows when butterfly blocks are added in parallel, reaching one extra bit per receiver for perfect channels.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Quantum per-receiver rate RQ counts a collective GHZ logical qubit, not independently decodable qubits, so the comparison with per-receiver classical bits conflates different multicast tasks.","rationale":"The paper's identity and erasure rate formulas are explicit and internally consistent, and the limiting one-bit-per-receiver gap follows from elementary cut counting once the authors' definition of a quantum multicast is granted. The reader's conditional verdict is well motivated by the depolarizing typo and by the opacity of the numerical depolarizing rate calculation; those issues are real but not the most load-bearing. The deeper issue is definitional: Eq. (10) bounds physical qubits delivered to a super-receiver, and dividing by r assumes a logical qubit encoded as a GHZ state is equivalent to r individually decodable qubits. It is not; the GHZ encoding is a collective/shared state. Therefore the quantity RQ is not a true per-receiver quantum multicast capacity in the usual sense, and the comparison to RC, where each receiver obtains the bit, mixes tasks. This does not necessarily invalidate all conclusions, because standard perfect quantum multicast to multiple receivers is impossible by no-cloning, but it does mean the quantitative bounds and the 'one extra bit per receiver' statement need substantial clarification or redefinition before the central claim is meaningful as a practical networking statement. The reader's weakest_assumption mentions the division by r and the REE-bound validity, so there is partial agreement, but the present concern is sharper: even granting the REE bound, the conversion to a per-receiver rate is problematic.","tokens_in":10590,"tokens_out":20905,"duration_ms":239663,"concrete_test":"Analyze the identity single-block butterfly with one sender A1 and two receivers B1, B2. Determine whether any protocol using the two available transmission paths can deliver an unknown qubit to B1 and B2 such that each receiver can apply a local decoder and recover the input with fidelity 1. By no-cloning this is impossible; the GHZ encoding α|00>+β|11> yields per-receiver reduced states |α|^2|0><0|+|β|^2|1><1|. Compute the maximum achievable worst-case fidelity (e.g., via a semidefinite program over encoding and local decoding maps) and compare it with the value implied by Eq. (10). If the per-receiver quantum multicast capacity is below RQ, Eq. (10) must be reinterpreted or the comparison to RC revised.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The load-bearing step is the conversion in Eq. (10) of the super-user physical-qubit bound into a per-receiver quantum multicast rate by dividing by r. The paper's quantum message is defined as a GHZ-like logical qubit α|0...0> + β|1...1>, with one physical qubit sent to each receiver. Such a state is a single logical qubit held jointly by the destination set; no individual receiver can locally decode α|0>+β|1> since its reduced state is |α|^2|0><0| + |β|^2|1><1|, which has lost the coherence between |0> and |1>. Classical multicast, by contrast, delivers the message to every receiver individually, and each classical bit counted in RC is locally accessible. Thus RQ and RC count different tasks. The claimed 'one extra bit per receiver' for parallel identity blocks compares a per-receiver classical rate with a per-receiver share of a collective quantum state. If 'multicast' is taken in the standard sense that every receiver can decode the message, perfect single-qubit multicast to r>=2 receivers is forbidden by no-cloning, so the REE-based bound is not an upper bound for that task and the quantitative gap expressed in Eqs. (19)-(21) is not the gap stated in the abstract.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript studies butterfly networks and networks assembled from butterfly blocks, comparing achievable classical multicast rates with upper bounds on quantum multicast rates. For identity, depolarizing, and erasure channels, the authors derive explicit per-receiver rates: the classical rate RC is computed from counting arguments or numerical optimization, while the quantum upper bound RQ comes from the relative entropy of entanglement (REE) network bound of Refs. [32,35,36], divided by the number of receivers. The central claim is that RC exceeds RQ in all cases considered, with the gap growing with the number of parallel butterfly blocks and reaching up to one extra bit per receiver in the erasure/identity limit. The paper also treats series and grid arrangements, showing that the parameter region where classical communication wins remains nonempty.","tokens_in":10828,"tokens_out":18996,"duration_ms":194937,"significance":"If the comparison is meaningful, the paper provides a quantitative demonstration that standard classical network-coding architectures can outperform quantum communication, and it gives explicit, parameter-free counting formulas for the erasure case. The erasure formulas (17)-(18) and (21)-(22) are transparent and checkable, and the use of closed-form REE bounds avoids fitted parameters. However, the significance is undermined by a conceptual mismatch between the tasks counted by RC and RQ: the classical rate counts locally decodable bits per receiver, while the quantum rate counts shares of a GHZ-like logical qubit that no individual receiver can decode. This issue, together with a sign error in Eq. (12), means the quantitative gap claimed in the abstract is not yet established for the stated comparison.","major_comments":[{"comment":"The comparison between RC and RQ conflates two different tasks. RQ is defined for a single-message multicast in which each sender transmits a GHZ-like logical qubit α|0...0>+β|1...1> encoded in r physical qubits. No individual receiver can locally decode this logical qubit: the reduced state of any single receiver is diagonal and carries no coherence between |0> and |1>. In contrast, each classical bit counted in RC is delivered to and locally decodable by every receiver. The claimed gap of up to one extra bit per receiver (Eqs. (19)-(21)) therefore compares locally accessible classical bits with per-receiver shares of a globally distributed quantum state. If 'multicast' is understood in the standard sense that every receiver obtains the message, then a perfect single-qubit multicast to r>=2 receivers is forbidden by no-cloning, so RQ is not an upper bound for that task. The manuscript should either define the quantum task as entanglement distribution and compare with an appropriate classical analogue, or restrict the claims accordingly.","section":"Section II, Eq. (10); Section IV A, Eqs. (19)-(21); Abstract"},{"comment":"The argument of the binary entropy in Eq. (12) is incorrect. For the qubit depolarizing channel defined in Eq. (11), the Choi state has fidelity F = 1 - 3p/4, so the REE bound is 1 - H2(1 - 3p/4), equivalently 1 - H2(3p/4) by symmetry of H2. Equation (12) instead writes 1 - H2((1 - 3p)/4), which gives 1 - H2(1/4) ≈ 0.189 at p=0 instead of 1, and becomes ill-defined for p > 1/3. This error propagates into Eq. (13) and the depolarizing curves in Fig. 2, and the numerical results in those panels should be recalculated after the fix.","section":"Section III, Eq. (12) and Eq. (13); Fig. 2"},{"comment":"The achievable classical rate RC for the depolarizing butterfly is not specified. The text states that the network is deconstructed into equivalent channels and that the overall rate is 'found numerically', but no transition matrix, optimization problem, or code is provided. Since the depolarizing panels of Fig. 2 and the associated claim of a gap over the whole range of p rely on this rate, the authors should give the explicit channel decomposition and the optimization procedure (or a derivation) so that the lower bound is verifiable.","section":"Section III, depolarizing case; Fig. 2"}],"minor_comments":[{"comment":"Equation (19) writes H2(3p/4) while Eq. (12) writes H2((1-3p)/4); after correcting Eq. (12), use the same expression in both places to avoid apparent inconsistency.","section":"Eq. (19)"},{"comment":"The sentence 'the total number of logical qubits correctly received by the destination set is equal to the total number of physical qubits correctly received by each individual receiver' is confusing; since each logical qubit uses r physical qubits, the division in Eq. (10) deserves a clearer one-sentence justification.","section":"Section II, text before Eq. (10)"},{"comment":"The exponent 5 in the network-coding term is not derived in the text; a short explanation of which five edges must succeed for a network-coded bit (e.g., both senders' edges to R1, R1->R2, R2->Bi, and the direct edge Ai->Bi for decoding) would make the counting transparent.","section":"Section III, Eqs. (17)-(18)"},{"comment":"The crossing points eta and eta' are quoted to three decimal places, but the method used to compute them (root finding of which curve difference?) is not described; please state how these values are obtained.","section":"Fig. 3 and Fig. 6"}],"recommendation":"major_revision","confidential_remarks":"The REE upper bound is taken from Refs. [32,35,36], which are predominantly the authors' own prior work; the present contribution is an application of that bound rather than an independent derivation. The conceptual concern in major comment 1 should be resolved by the authors before acceptance; if the task is reframed as GHZ-state distribution, the comparison with classical multicast should be justified or the claims weakened. The paper has been publicly available since September 2019; the editor may wish to consider the timeliness of the submission."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper gives explicit rate formulas for classical and quantum communication over butterfly networks and shows that the classical rate exceeds an upper bound on the quantum rate, with the gap growing when you add parallel blocks. The erasure-channel derivations are clean counting arguments, and the identity-channel limit is a neat observation. The paper is worth reading for those formulas alone.\n\nThe soft spots are real. First, Eq. (12) has a sign error: the REE for the depolarizing channel should be 1 - H2(3p/4), not 1 - H2((1-3p)/4). The p=0 limit gives H2(1/4) ≈ 0.811, which would imply an entanglement of about 0.189 for a maximally entangled state. That's clearly wrong. The correct expression appears later in Eq. (19), so this is likely a typo, but it's in a load-bearing position.\n\nSecond, the depolarizing classical rate is said to come from a numerical optimization, but no algorithm or code is given. A reader cannot verify the curve in Fig. 2. That's a reproducibility gap.\n\nThe bigger issue is conceptual. The quantum rate RQ in Eq. (10) is obtained by dividing a physical-qubit bound by the number of receivers r. The quantum message is a GHZ-like logical qubit shared by the receivers. Each receiver does not locally hold the qubit; they hold a share. Classical multicast gives each receiver an independent bit. So RC and RQ are rates for different tasks. In the identity case, saying '2 bits per receiver vs 1.5 qubits per receiver' compares a locally accessible bit with a collective quantum state. The abstract's 'one extra bit per receiver' is not a gap in transmission of the same kind of information. If the quantum task were to send a qubit that each receiver can decode, no-cloning makes that impossible for two receivers, so the REE bound would not be the relevant upper bound.\n\nThis does not mean the erasure formulas are wrong; they are correct for the task the authors define. But the paper's framing oversells the comparison. A referee should ask the authors to either redefine the quantum task to match the classical one or to state clearly that they are comparing a shared quantum state with independent classical messages.\n\nThe citation pattern is fine; the REE bound is taken from prior work and used as a theorem.\n\nMy verdict: the paper deserves a serious referee but needs major revision. If the authors fix the typo, provide the depolarizing calculation, and reframe the quantum task honestly, it could be a useful contribution to the network-coding literature. As it stands, the central claim is not as strong as advertised.","headline":"Concrete rate formulas for butterfly networks, but the quantum 'multicast' rate counts a shared GHZ qubit, not a qubit per receiver, so the headline gap compares different tasks.","tokens_in":11367,"tokens_out":9069,"would_cite":false,"duration_ms":91361,"reading_group":"maybe","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 shows that classical multicast communication can outpace quantum communication in butterfly-based networks, with the per-receiver gap reaching one bit when many butterfly blocks are wired in parallel.","keywords":["butterfly network","quantum network coding","multicast communication","relative entropy of entanglement","depolarizing channel","erasure channel","teleportation-covariant channel","quantum capacity bounds"],"falsifier":"Demonstrate a single-message multicast over a parallel identity-channel butterfly network (with r receivers) that delivers more than (2r - 1)/r logical qubits per use per receiver; this would refute the REE-based ceiling that the paper's gap relies on.","tokens_in":10373,"feed_emoji":"🦋","tokens_out":8209,"duration_ms":77973,"temperature":0.7,"pith_summary":"The paper asks whether classical network topologies built from butterfly blocks are inherently worse at carrying quantum information than classical information. It compares achievable classical single-message multicast rates with upper bounds on quantum multicast rates for networks whose links are identity, depolarizing, or erasure channels. The central finding is that the classical rate per receiver exceeds the quantum upper bound over wide noise ranges, and for perfect identity channels the gap grows to one full bit per receiver as butterfly blocks are added in parallel. The paper's purpose is to identify these 'negative case' configurations so that quantum networks can be designed away from them.","feed_headline":"Classical beats quantum in butterfly networks by one bit per receiver","feed_subtitle":"Achievable classical multicast rates exceed quantum upper bounds for identity, depolarizing, and erasure channels.","key_machinery":"The load-bearing object is the butterfly network block and its rectangular grids. The quantum side rests on Eq. (10): the quantum multicast rate is bounded by (1/r) times the minimum-cut sum of relative entropies of entanglement of the Choi matrices of the channels crossing the cut, where senders and receivers are treated as super-users and r is the number of receivers. The classical side uses network coding: XOR at the bottleneck node lets each sender's message reach both receivers through a single channel use, and the achievable rates are computed by decomposing the network into effective point-to-point channels and optimizing input distributions. The comparison of these two quantities is what produces the gap.","core_discovery":"For a single butterfly network with two senders and two receivers using identical teleportation-covariant channels, the achievable classical multicast rate per receiver can be strictly larger than the relative-entropy-of-entanglement upper bound on the quantum multicast rate. In the perfect-channel limit, one butterfly block transmits 2 classical bits per use per receiver but is bounded by 1.5 qubits. When N_x butterfly blocks are connected in parallel (with r = N_x + 1 receivers), the quantum bound becomes (2r - 1)/r qubits per receiver, approaching 2 qubits, while the achievable classical rate approaches 3 bits, so the gap tends to one bit per receiver. For depolarizing channels the classical rate exceeds the quantum bound over essentially the whole noise range, and for erasure channels it exceeds the bound below a critical erasure probability (about 0.159 without inter-node communication and 0.244 with it); adding blocks in parallel or allowing classical communication between nodes widens the range.","pith_inferences":["The same REE-cut technique could be applied to continuous-variable teleportation-covariant channels (e.g., Gaussian channels) to check whether the classical-over-quantum gap persists in that regime.","Because the quantum bound uses the super-user relaxation, a tighter multi-sender upper bound might reduce the gap; finding such a bound would be a direct test of how much of the reported gap is real versus an artifact of the relaxation.","The classical rates assume XOR network coding and simple routing; more sophisticated classical coding could push the classical rate higher, making the gap even larger.","One could experimentally realize a small erasure-butterfly network and measure the multicast throughput to see whether the predicted crossing at erasure probability about 0.159 is observed."],"forward_implications":["A quantum network built by simply replacing the wires of a butterfly-based classical network with quantum channels will carry fewer logical qubits than classical bits, by up to one bit per receiver in the parallel-block limit.","Inter-node classical communication helps classical erasure networks widen their advantage: the critical erasure probability at which classical beats quantum rises from about 0.159 to about 0.244 for a single block.","Adding butterfly blocks in parallel increases the classical-versus-quantum gap monotonically; adding blocks in series decreases both rates but the classical rate still beats the quantum bound for a range of erasure probabilities.","The results give a concrete design rule: to avoid quantum performance penalties, do not duplicate butterfly-block topologies that rely on network coding for classical gains."],"supporting_citations":[{"why":"Establishes the XOR network-coding scheme that gives the classical butterfly its 2-bit per receiver rate.","marker":"[1]"},{"why":"Shows that perfect quantum network coding is impossible in the butterfly, underpinning why the quantum rate is expected to be lower.","marker":"[20]"},{"why":"Provides the single-letter relative-entropy-of-entanglement bound for the two-way capacity of teleportation-covariant channels, the basis of the quantum upper bounds.","marker":"[32]"},{"why":"Extends the REE bound to single- and multi-edge flows over quantum networks, used to bound multicast rates.","marker":"[35]"},{"why":"Generalizes the REE bound to multi-sender/multi-receiver networks with super-users, yielding the multicast bound in Eq. (9).","marker":"[36]"},{"why":"Gives the capacity of the binary symmetric channel used to compute achievable classical rates for depolarizing links.","marker":"[39]"},{"why":"Provides the classical and quantum capacities of the erasure channel used for the erasure-network rates.","marker":"[43]"}],"fun_headline_variants":["Butterfly networks show classical beats quantum by one bit per receiver","Quantum networking limits: butterfly networks favor classical","Classical outperforms quantum in butterfly network multicasts","Butterfly networks: classical rate exceeds quantum bound per receiver","Quantum networking gap: butterfly blocks give classical one-bit edge"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire quantum-rate gap rests on the validity of the relative-entropy-of-entanglement network bound, including the super-user relaxation and the division by the number of receivers; if that bound does not actually limit single-message multiple multicasts, the claimed quantum rates are not ceilings and the gap could shrink or vanish.","fun_headline_variants_meta":{"raw":{"variants":["Butterfly networks show classical beats quantum by one bit per receiver","Quantum networking limits: butterfly networks favor classical","Classical outperforms quantum in butterfly network multicasts","Butterfly networks: classical rate exceeds quantum bound per receiver","Quantum networking gap: butterfly blocks give classical one-bit edge"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000816,"raw_usage":{"total_tokens":3564,"prompt_tokens":925,"completion_tokens":2639,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":541,"completion_tokens_details":{"reasoning_tokens":2559}},"tokens_in":541,"tokens_out":2639,"duration_ms":21253,"temperature":1.0,"reasoning_tokens":2559,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T04:52:58.598076+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Demonstrate a single-message multicast over a parallel identity-channel butterfly network (with r receivers) that delivers more than (2r - 1)/r logical qubits per use per receiver; this would refute the REE-based ceiling that the paper's gap relies on.","supporting_citations":[{"cited_title":"We encounter a bottleneck at node R1 as data is waiting to be sent from both senders through the channel (R1, R2)","cited_arxiv_id":null,"evidence_quote":"Establishes the XOR network-coding scheme that gives the classical butterfly its 2-bit per receiver rate."},{"cited_title":"Pirandola, J","cited_arxiv_id":null,"evidence_quote":"Shows that perfect quantum network coding is impossible in the butterfly, underpinning why the quantum rate is expected to be lower."},{"cited_title":"Soeda, Y","cited_arxiv_id":null,"evidence_quote":"Provides the single-letter relative-entropy-of-entanglement bound for the two-way capacity of teleportation-covariant channels, the basis of the quantum upper bounds."},{"cited_title":"Leung, J","cited_arxiv_id":null,"evidence_quote":"Extends the REE bound to single- and multi-edge flows over quantum networks, used to bound multicast rates."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Generalizes the REE bound to multi-sender/multi-receiver networks with super-users, yielding the multicast bound in Eq. (9)."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the classical and quantum capacities of the erasure channel used for the erasure-network rates."}],"review_version":1}