{"id":"d3dab33d-d23d-4bbe-929d-2e8072f6a944","arxiv_id":"2411.16263","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"First achievable-rate bounds for fully quantum relay channels, including an exact capacity formula for Hadamard relay channels.","lead":"This paper proves the first information-theoretic bounds for sending classical messages over a fully quantum relay channel, where the relay sends and receives quantum states. It also gives an exact capacity formula for a special class of relay channels and introduces two new coding strategies.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Converse for Theorem 3 does not justify passing from arbitrary code-induced states to the product-state max in Eq. (24); entangled encoding across time may make the A_i state depend on Y^{i-1}.","rationale":"I read the paper in good faith. The achievability proofs are detailed and use standard tools, and the Hadamard capacity formula is plausible. The reader's conditional verdict is appropriate: the weakest point is indeed Appendix C, where the converse moves from mutual informations evaluated on arbitrary code-induced states to the product-state maximum in Theorem 3. My reading sharpens the concern: the missing step is not merely displaying the tensor product, but proving that the conditional reduced state on A_i can be indexed by X0 alone, independent of the relay history, or alternatively that enlarging X0 to include the history is legitimate and that the resulting alphabet growth does not break single-letterization. I found no independent evidence, such as a machine-checked proof or a parameter-free derivation, that would resolve this gap. I therefore see no reason to change the reader's conditional verdict, but also no basis for rejection without first attempting the concrete re-derivation described above.","tokens_in":29618,"tokens_out":24196,"duration_ms":247449,"concrete_test":"Re-derive the last step of Appendix C with the explicit construction X0,i = (M, Y1^{i-1}) and X1,i = Y1^{i-1}, setting theta^{x0}_A to the conditional reduced state on A_i; verify that the state entering I(X0X1;B) is exactly NAD→BE(theta^{x0}_A ⊗ zeta^{x1}_D) under the code-induced ensemble. Then test the same reduction on a two-use code for a Hadamard relay channel whose marginal MAD→Y1 is a Bell-basis measurement and whose Alice encoder uses a Bell state across A1 and A2, comparing the two-letter rate with n times the RHS of (24). The theorem is falsified if the code-induced rate exceeds the product-state maximum; otherwise, the missing lemma is the only obstruction.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is Theorem 3, the exact capacity formula (24) for Hadamard relay channels. The converse in Appendix C first obtains, via Fano and data processing, the bounds (101a)-(101b), which are stated in terms of the actual code-induced states. It then defines X0,i = M and X1,i = Y1^{i-1} in (106), and in (108) asserts that the resulting expression is an instance of the RHS of (24). This step is load-bearing and is not justified. The RHS of (24) maximizes over input ensembles in which, for each pair (x0,x1), the channel input is a tensor product theta^{x0}_A ⊗ zeta^{x1}_D, with theta indexed only by the x0 label. For an arbitrary code, Alice's encoding map F_{M→A^n} is allowed to be a general CPTP map, so A_i can be entangled with earlier input systems A^{i-1}. Since Y1^{i-1} is a function of A^{i-1}, the reduced state of A_i conditioned on (M, Y1^{i-1}) can depend on Y1^{i-1}. Thus a collection theta^{x0}_A indexed by x0 = M alone need not reproduce the conditional state on A_i, and the paper does not prove that the code-induced ensemble can be replaced by a product-state ensemble without changing the value of the min in (105). A repair may exist by enlarging X0 to include Y1^{i-1}, but the proof does not state such a construction, nor does it address the growing alphabet of Y1^{i-1} in the single-letterization step. As written, the exact-capacity conclusion does not follow from the converse.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the fully quantum relay channel N_{AD→BE}, where Alice transmits a classical message through a memoryless channel with a strictly causal quantum relay. It presents three achievable-rate bounds: a partial decode-forward bound (Theorem 1), a measure-forward bound (Theorem 4), and an assist-forward bound for orthogonal-receiver-component channels (Theorem 5). It also claims an exact single-letter capacity formula for Hadamard relay channels (Theorem 3), obtained by showing that full decode-forward is optimal. The paper includes a recovery of the classical-quantum relay result of Savov et al., a wired-network example, and a depolarizing-relay example with an explicit lower bound. The appendices contain the achievability proofs and the Hadamard converse.","tokens_in":29957,"tokens_out":6532,"duration_ms":78714,"significance":"If correct, Theorem 3 would be the first exact single-letter capacity formula for a fully quantum relay channel, and Theorems 1, 4, and 5 would provide the first general achievable-rate bounds for this model. The achievability proofs use standard quantum packing, covering, and gentle-measurement arguments, and the depolarizing example is worked out in explicit algebra. The paper is also commendably candid about the lack of cardinality bounds on auxiliary variables. However, the Hadamard converse as written contains a load-bearing gap: the reduction from an arbitrary code-induced ensemble to the product-state ensemble in Eq. (24) is not justified. The significance of the result therefore depends on whether this gap can be repaired.","major_comments":[{"comment":"The converse for Theorem 3 does not justify the passage from the code-induced states to the product-state maximization in Eq. (24). In (106) the proof defines X0,i = M and X1,i = Y1^{i-1}, and in (108) it asserts that the resulting expression is an instance of the RHS of (24). But the RHS of (24) maximizes over ensembles in which, for each pair (x0, x1), the channel input is θ^{x0}_A ⊗ ζ^{x1}_D with θ indexed only by x0. For a general encoding map F_{M→A^n}, the system A_i can be entangled with A^{i-1}; since Y1^{i-1} depends on A^{i-1}, the reduced state of A_i conditioned on (M, Y1^{i-1}) can depend on Y1^{i-1}. A collection θ^{x0}_A indexed by x0 = M alone need not reproduce that conditional state, and the paper does not prove that the code-induced ensemble can be replaced by a product-state ensemble without changing the value of the min in (105). The proof also does not address the fact that the alphabet of X1,i = Y1^{i-1} grows with i, so the single-letterization step in (108) is not established. A repair may exist by exploiting the entanglement-breaking or degraded structure of the Hadamard channel, or by enlarging X0 to include the past relay outputs, but as written the exact-capacity conclusion does not follow from the given converse.","section":"Appendix C, Eq. (108)"}],"minor_comments":[{"comment":"The phrase \"full dicode-forward strategy\" appears to be a typo for \"full decode-forward strategy.\"","section":"Appendix C, first paragraph"},{"comment":"The word \"Thoerem 3\" should be \"Theorem 3.\"","section":"Section VI-B, paragraph 2"},{"comment":"Equation (95) has a missing closing parenthesis: the exponent should be n(1+2δ)H(B|U X0 X1)_ω, with the closing parenthesis after the entropy expression.","section":"Appendix B, Eq. (95)"},{"comment":"The parameter α in Example 2 is introduced as a free variable in the derivation and later set to α = q/2; the notation p ∗ q/2 in Eq. (44) is ambiguous and should be parenthesized, e.g., p ∗ (q/2), to avoid confusion with the binary operation associativity.","section":"Example 2 and Appendix F"},{"comment":"The acknowledgment that no cardinality bounds are given for U, X0, X1, Y1, Z1, G0, G1 is useful; it would be helpful to state explicitly that the three rate formulas are therefore not known to be computable in finite time.","section":"Section VI-B"}],"recommendation":"major_revision","confidential_remarks":"The paper's main claim is significant and the achievability portions appear internally consistent, but the Hadamard converse in Appendix C is the load-bearing step for the exact-capacity theorem and is not currently justified. This is a fixable gap if a proper converse can be supplied using the special structure of Hadamard (entanglement-breaking) channels, but it requires substantial new argument rather than a local correction. I recommend major revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a serious paper on fully quantum relay channels, and the three lower bounds are likely real contributions. But the claimed exact capacity for Hadamard relay channels (Theorem 3) is not proven as written. The converse in Appendix C has a gap that is load-bearing, not cosmetic.\n\nThe paper's real strengths are the coding schemes. The partial decode-forward bound generalizes Savov et al. to fully quantum inputs, and the recovery of the c-q result as a special case is a nice check. Measure-forward is a sensible quantum analog of compress-forward with the relay performing measurements, and assist-forward—where Alice generates entanglement between relay and receiver while sending the message—is a genuinely new idea. The achievability proofs (Appendices B, D, E) follow standard packing/covering/gentle-measurement arguments and look internally consistent. The depolarizing relay example's algebra checks out.\n\nThe problem is Appendix C. The proof of Theorem 3 runs an arbitrary code, gets Fano bound (101a)-(101b) in terms of the code-induced states, then defines X0,i = M and X1,i = Y1^{i-1} and asserts without further argument that (108) is an instance of the right-hand side of (24). But (24) maximizes over input ensembles where, for each (x0,x1), the channel input is a product state θ_A^{x0} ⊗ ζ_D^{x1}, with θ indexed only by x0. An arbitrary code's Alice encoder is a CPTP map that can entangle A_i with A^{i-1}, which is correlated with Y1^{i-1}; so the conditional state on A_i given (M, Y1^{i-1}) can depend on Y1^{i-1} and need not be a product state with D_i. The proof never shows that the code-induced ensemble can be replaced by such a product-state ensemble without decreasing the mutual information terms. Just enlarging X0 to include the history breaks the single-letterization; that route needs real work. So the exact-capacity conclusion does not follow from the given proof. The lower bounds in Theorems 1, 4, and 5 are unaffected.\n\nAlso, the paper cites [45] but doesn't explain how the claimed novelty stands relative to it; that should be reconciled. The absence of alphabet cardinality bounds is acknowledged and standard for this kind of multi-letter expression.\n\nBottom line: a serious, well-structured paper with solid achievability results, but the headline capacity theorem is currently unproven. A referee should engage with the converse rather than desk-reject; if the gap is fixable, the Hadamard result would be a substantial theorem. I'd cite the lower bounds and would send it to a careful referee.","headline":"Solid new lower bounds for fully quantum relay channels, but the Hadamard capacity theorem has a load-bearing converse gap that is not fixed in this version.","tokens_in":30482,"tokens_out":9397,"would_cite":true,"duration_ms":92395,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P45","94A40"],"pacs":["03.67.Hk","03.67.-a"],"model":"deepseek-v4-flash","headline":"The paper establishes the first exact capacity formula for Hadamard quantum relay channels, plus three lower bounds for general fully quantum relays.","keywords":["quantum relay channel","Hadamard relay channel","partial decode-forward","measure-forward","assist-forward","entanglement assistance","quantum Shannon theory","capacity"],"falsifier":"Exhibit a family of $(2^{nR},n,\\varepsilon_n)$ codes for a Hadamard relay channel with $\\varepsilon_n\\to 0$ and $R$ strictly larger than the claimed maximin formula; the natural route is a code whose conditional input state, given $(M=m,Y_1^{i-1}=y^{i-1})$, has non-product correlation between $A_i$ and $D_i$, so that the Appendix C step $I(MB^{i-1};B_i)\\le I(MY_1^{i-1};B_i)$ no longer dominates. Concretely, for a two-block code with entangled $AD$ inputs, compute whether the rate exceeds $\\max\\min\\{I(X_0X_1;B),I(X_0;Y_1|X_1)\\}$; any such example would refute the tightness of full decode-forward for Hadamard channels.","tokens_in":29402,"feed_emoji":"📡","tokens_out":9835,"duration_ms":84499,"temperature":0.7,"pith_summary":"This paper claims to give the first general rate bounds for a fully quantum relay channel—a three-terminal memoryless network in which the sender's and relay's transmissions are quantum systems and both receivers are quantum. It establishes three achievable-rate formulas: partial decode-forward, in which the relay decodes part of the message and re-encodes it; measure-forward, in which the relay measures its received system and sends a compressed classical description; and assist-forward, in which the sender uses the broadcast part of the channel to distribute entanglement between relay and destination before the relay transmits with that assistance. For the special class of Hadamard relay channels, where the relay's observation is classical and Bob's channel is a degraded version of the relay's, the partial decode-forward rate is shown to be tight, giving an exact single-letter capacity formula. A sympathetic reader should care because exact capacity formulas for fully quantum relay channels have been open until now, and relay limits are the basic building block for understanding multihop quantum networks and repeaters.","feed_headline":"Exact capacity found for Hadamard quantum relay channels","feed_subtitle":"If correct, the first single-letter formula for a fully quantum relay, plus three new rate bounds.","key_machinery":"The load-bearing machinery is block-Markov coding with a strictly-causal quantum relay: at time $i$ the relay encodes $D_i$ from its previously received systems $E^{i-1}$ together with 'leftover' systems $\\bar E^{i-1}$ left by earlier encoding operations. Achievability rests on the quantum packing lemma (for decoding measurements) and the gentle measurement lemma (so that successive measurements do not destroy the state), organized with the quantum method of types. For Hadamard channels the converse exploits the degraded structure $N^H = P_{Y_1\\to BY_1}\\circ M_{AD\\to Y_1}$ to reduce an arbitrary code to single-letter mutual information terms via the data-processing inequality, taking $X_0=M$ and $X_1=Y_1^{i-1}$. For assist-forward, the central formula is the rate-limited entanglement-assisted capacity, applied to the relay-to-destination link after Alice distributes entanglement through the broadcast component.","core_discovery":"The central discovery, stated as Theorem 3, is that a Hadamard relay channel $N^H_{AD\\to BY_1}$—a fully quantum channel whose relay output is a classical letter $Y_1$ and which is degraded so that $N^H = P_{Y_1\\to BY_1}\\circ M_{AD\\to Y_1}$—has capacity\n$$C(N)=\\max_{p_{X_0X_1},\\,\\$theta^{{x_0}}$_A\\otimes\\$zeta^{{x_1}}$_D}\\min\\{ I(X_0X_1;B)_\\omega,\\; I(X_0;Y_1|X_1)_\\omega \\},$$\nwith the maximum over distributions $p_{X_0X_1}$ and product input states, and this is achieved by full decode-forward: the relay decodes the entire message and re-encodes it. For a general fully quantum relay channel, Theorems 1, 4, and 5 respectively establish the partial decode-forward lower bound $R_{\\mathrm{PD-F}}$, the measure-forward lower bound $R_{\\mathrm{M-F}}$, and the assist-forward lower bound $R_{\\mathrm{A-F}}$; the last combines block-Markov coding, constant-composition coding, rate-limited entanglement assistance, and broadcast subspace transmission. The paper also derives the classical-quantum partial decode-forward bound as a special case, and it computes a closed-form measure-forward rate $1-h(p\\ast q/2)$ for a depolarizing relay channel with orthogonal receiver components.","pith_inferences":["One extension implicit in the proof is a stochastic-degraded version of Theorem 3: if a degrading map exists only on the marginals rather than on the full channel, the same data-processing argument may still yield the same single-letter formula, though the paper does not claim this.","The converse's reduction to product states suggests a precise test: if entangled inputs across time blocks can beat the formula, the capacity would need additional coherent-information terms beyond the two mutual informations in the claimed expression.","The assist-forward construction suggests a testable network-level design: even a very noisy relay-to-destination link can carry a positive rate if the sender uses the broadcast phase to pre-distribute entanglement, so comparing the assist-forward rate against a cutset upper bound for the depolarizing example would show how much the block-Markov protocol loses.","For the depolarizing relay channel, computing the cutset upper bound and comparing with $1-h(p\\ast q/2)$ would reveal whether the measure-forward rate is tight or merely a lower bound; the paper leaves that comparison open."],"forward_implications":["For every Hadamard relay channel, the capacity is a single-letter maximization over product input ensembles, and the full decode-forward strategy attains it; the relay's classical observation $Y_1$ is what limits the rate through $I(X_0;Y_1|X_1)$.","The partial decode-forward bound contains direct transmission as the case $U=\\varnothing$, so it recovers the direct-transmission lower bound from classical-quantum channel capacity and yields the anti-degraded classical-quantum capacity $\\max_{x_1}\\max_{p_X} I(X;B|X_1=x_1)$.","The measure-forward bound generalizes classical compress-forward to fully quantum relays and produces a positive achievable rate $1-h(p\\ast q/2)$ for a depolarizing relay where both marginals are completely depolarizing and direct transmission would give zero.","When the relay channel is a Stinespring dilation, the full decode-forward formula reduces to a bound resembling environment-assisted distillation rates, $\\min\\{H(B),H(E)\\}$ over product inputs."],"supporting_citations":[{"why":"Supplies the classical-quantum partial decode-forward bound that this paper extends to fully quantum channels and recovers as a special case (Corollary 2).","marker":"[33]"},{"why":"Provides the classical relay coding strategies (partial decode-forward and compress-forward) and block-Markov structure that Theorems 1 and 4 quantize.","marker":"[49]"},{"why":"Provides the quantum packing lemma and constant-composition coding tools used in all achievability proofs, plus the gentle measurement lemma.","marker":"[54]"},{"why":"Supplies the rate-limited entanglement-assisted communication formula used in the assist-forward scheme.","marker":"[53]"},{"why":"Provides the broadcast subspace transmission (father protocol) used to distribute entanglement from Alice to relay and Bob in the assist-forward scheme.","marker":"[55]"},{"why":"Supplies the classical-quantum channel capacity formula that gives the direct-transmission lower bound and motivates the packing-based decoding.","marker":"[52]"}],"fun_headline_variants":["Hadamard relay channel capacity exactly determined","Three new rate bounds for quantum relay channels","Exact capacity for Hadamard quantum relay channels","Quantum relay: partial decode, measure, assist strategies"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The converse for the Hadamard capacity assumes that, conditioned on the message and the relay's past classical outputs, the sender's and relay's input states factorize as a product $\\theta^{x_0}_A\\otimes\\zeta^{x_1}_D$; if a code uses entanglement between the $A$ and $D$ inputs across time blocks, that factorization can fail and the claimed exact formula would not follow from the proof.","fun_headline_variants_meta":{"raw":{"variants":["Hadamard relay channel capacity exactly determined","Three new rate bounds for quantum relay channels","Exact capacity for Hadamard quantum relay channels","Quantum relay: partial decode, measure, assist strategies"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000609,"raw_usage":{"total_tokens":2886,"prompt_tokens":1044,"completion_tokens":1842,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":660,"completion_tokens_details":{"reasoning_tokens":1792}},"tokens_in":660,"tokens_out":1842,"duration_ms":14888,"temperature":1.0,"reasoning_tokens":1792,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T13:20:56.268002+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Exhibit a family of $(2^{nR},n,\\varepsilon_n)$ codes for a Hadamard relay channel with $\\varepsilon_n\\to 0$ and $R$ strictly larger than the claimed maximin formula; the natural route is a code whose conditional input state, given $(M=m,Y_1^{i-1}=y^{i-1})$, has non-product correlation between $A_i$ and $D_i$, so that the Appendix C step $I(MB^{i-1};B_i)\\le I(MY_1^{i-1};B_i)$ no longer dominates. Concretely, for a two-block code with entangled $AD$ inputs, compute whether the rate exceeds $\\max\\min\\{I(X_0X_1;B),I(X_0;Y_1|X_1)\\}$; any such example would refute the tightness of full decode-forward for Hadamard channels.","supporting_citations":[{"cited_title":"Partial decode-forward for quantum relay channels,","cited_arxiv_id":null,"evidence_quote":"Supplies the classical-quantum partial decode-forward bound that this paper extends to fully quantum channels and recovers as a special case (Corollary 2)."},{"cited_title":"Capacity theorems for the relay channel,","cited_arxiv_id":null,"evidence_quote":"Provides the classical relay coding strategies (partial decode-forward and compress-forward) and block-Markov structure that Theorems 1 and 4 quantize."},{"cited_title":"Coding theorem and strong converse for quantum channels,","cited_arxiv_id":null,"evidence_quote":"Provides the quantum packing lemma and constant-composition coding tools used in all achievability proofs, plus the gentle measurement lemma."},{"cited_title":"The classical capacity achievable by a quantum channel assisted by a limited entanglement,","cited_arxiv_id":null,"evidence_quote":"Supplies the rate-limited entanglement-assisted communication formula used in the assist-forward scheme."},{"cited_title":"A father protocol for quantum broadcast channels,","cited_arxiv_id":null,"evidence_quote":"Provides the broadcast subspace transmission (father protocol) used to distribute entanglement from Alice to relay and Bob in the assist-forward scheme."},{"cited_title":"The capacity of the quantum channel with general signal states,","cited_arxiv_id":null,"evidence_quote":"Supplies the classical-quantum channel capacity formula that gives the direct-transmission lower bound and motivates the packing-based decoding."}],"review_version":1}