{"id":"9512a92e-3773-4da4-a886-a00162cdd8b7","arxiv_id":"2412.16097","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Grouping two opposite-polarization elements in a beyond diagonal RIS achieves the fully connected performance bound in dual-polarized line-of-sight links, and dual-polarized BD-RIS offers gains over conventional RIS even under LoS.","lead":"This paper analyzes dual-polarized smart surfaces called beyond diagonal RIS, and derives formulas for the received power they can deliver relative to conventional surfaces. It finds that pairing one vertical and one horizontal element gives the performance of the most complex design in line-of-sight links, at much lower hardware cost.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Prop. 2 is not proven as written: Eq. (27) has an algebraic error, and the imported [4, Prop. 1] transfer to dual-polarized LoS is not re-derived, so the Pareto-frontier claim is conditional until repaired.","rationale":"I verified the scaling-law derivations in Section III and Prop. 1 in Section IV. The expressions for P_Single, P_Fully, and the gain G in the four scenarios are consistent with the model, and Prop. 1's construction of group size 2 with opposite polarizations does achieve (1+χ)²N²/4. The load-bearing weakness is entirely in Prop. 2, which is needed for the claim that n groups of size 2 form the Pareto frontier. The reader correctly identifies Eq. (27) as erroneous, and I agree this is a real defect in the written proof. However, the error is algebraic and repairable: with the correct (1−χ)² factor, the maximum of the upper bound still occurs at n_h=n, and the equality conditions force each of the n groups to be a balanced size-2 group, so the proposition's statement appears true. The reader also stresses the unproven transfer of [4, Prop. 1]; I regard that as a secondary concern because the relation G=2N−C follows from replacing group-connected subnetworks by tree/forest equivalents and is channel-independent, but the paper does not spell this out. Because the main theorem is plausible and the identified flaw is fixable, I do not move the verdict to reject; the appropriate posture remains conditional acceptance of the scaling laws and Prop. 1, with Prop. 2's proof corrected before the Pareto-frontier claim is treated as fully established. Hence I keep the reader's verdict unchanged, with partial agreement on the weakest assumption.","tokens_in":8949,"tokens_out":28305,"duration_ms":247254,"concrete_test":"Repair Eq. (27) by replacing the middle term with 2n(1−χ)² n_h, then re-derive the derivative: the critical point becomes n_h=n and the maximum is n(1+χ). Independently, for small N (e.g., N=8) and n ∈ {1,2,3}, exhaustively enumerate all partitions of N_v=N_h=N/2 elements into G=N−n groups of integer sizes, computing P = (Σ_g √((v_g χ + h_g)(v_g + h_g χ)))² for χ ∈ {0.01, 0.1, 0.5}; verify that (18) is the maximum and that the n-groups-of-size-2 configuration attains it. This separates the typo from the substantive claim.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central architecture-optimality claim rests on Prop. 2, whose proof contains a concrete algebraic error in Eq. (27). Starting from Q ≤ √((n_v χ + n_h)(n_v + n_h χ)) with n_v+n_h=2n, the radicand expands to 4n²χ + 2n(1−χ)² n_h − (1−χ)² n_h². The printed middle term, 2n(1−χ)n_h, is missing the square on (1−χ). With the printed expression, the derivative vanishes at n_h = n/(1−χ), not at n_h=n, so the asserted maximizer does not follow. With the corrected coefficient, the derivative does vanish at n_h=n and yields Q_max = n(1+χ), so the proposition is likely salvageable, but the proof as published is invalid. In addition, the first line of the proof asserts without re-derivation that an optimal architecture with C=N+n has G=N−n groups, citing [4, Prop. 1] from the uni-polarized setting. That number-of-groups relation is a complexity-counting/forest-equivalence statement and probably transfers, but the paper does not show this; if the import fails, the whole Pareto-frontier characterization lacks support. Prop. 1 and the scaling laws in Table I are correct under the stated model; the weakness is confined to the proof of the Pareto frontier.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies dual-polarized beyond-diagonal RIS (BD-RIS) systems. It derives scaling laws for the received power and the BD-RIS gain over conventional diagonal RIS (D-RIS) in four channel scenarios: Rayleigh or LoS fading, with Tx and Rx having either the same or opposite polarization. It shows that in LoS with opposite polarization, BD-RIS can provide a gain over D-RIS, and that a group-connected RIS with group size 2, in which each group pairs one vertical and one horizontal element, achieves the fully-connected received power upper bound with a low circuit complexity of C = 3N/2 (Prop. 1). The paper further claims a complete Pareto frontier for the performance-complexity trade-off in this LoS scenario, with Prop. 2 interpolating between single-connected and group-2 architectures.","tokens_in":9261,"tokens_out":17832,"duration_ms":142657,"significance":"If the results hold, the paper makes a useful contribution: it identifies a practically relevant scenario in which BD-RIS provides gains over D-RIS under LoS, and it proposes a low-complexity architecture (group size 2 with mixed polarizations) that attains the fully-connected performance bound. The scaling laws in Table I and the gain expressions in Eqs. (5), (7), (10), (11), and (15) are clean, parameter-free, and check out under the stated model. Prop. 1 is a solid result with clear practical implications. However, the Pareto-frontier claim rests on Prop. 2, whose proof as written contains an algebraic error in Eq. (27) and relies on an imported result from [4] without re-derivation; additionally, the proof considers only architectures with exactly N-2n singleton groups, leaving other feasible architectures unaddressed. These are load-bearing gaps, though they appear repairable.","major_comments":[{"comment":"The expansion of the radicand in Eq. (27) is algebraically incorrect: the middle term should be 2n(1-χ)^2 n_h rather than 2n(1-χ)n_h. With the printed expression, the derivative with respect to n_h vanishes at n_h = n/(1-χ), so the asserted maximizer n_h = n does not follow. With the corrected coefficient, the derivative does vanish at n_h = n and yields Q ≤ n(1+χ), so the step is repairable, but the proof as printed is invalid.","section":"Section IV, Prop. 2 proof, Eq. (27)"},{"comment":"The proof assumes without re-derivation that an optimal architecture with C=N+n has G=N-n groups, citing [4, Prop. 1]. That result was established for uni-polarized BD-RIS and may depend on complexity-counting conventions that are not verified here for dual-polarized LoS channels. Unless the transfer is justified or the relation is derived directly in this setting, the Pareto-frontier characterization lacks a load-bearing foundation.","section":"Section IV, Prop. 2 proof, first paragraph"},{"comment":"The proof passes from \"at least N-2n groups of size 1\" to an expression (20) with exactly N-2n singleton groups and n non-singleton groups. This is not without loss of generality: when n≥3, architectures with fewer non-singleton groups are feasible under the complexity budget (e.g., one group of size 3 when the extra budget is 3). For such architectures, the constant term in (21) and the definition of Q over n groups in (22) are not valid. A complete proof must optimize over the number of singleton groups and the sizes of the remaining groups, not only over the polarization assignment of 2n elements partitioned into n pairs.","section":"Section IV, Prop. 2 proof, Eqs. (20)-(22)"}],"minor_comments":[{"comment":"The text reads \"∥^h_R∥ = √(n_v + n_h χ)\" but this is the norm of ^h_T, not ^h_R; the symbol is mislabeled and should be corrected to ∥^h_T∥.","section":"Section IV, proof of Prop. 2, after Eq. (26)"},{"comment":"The statement that \"additional numerical simulations show that the gain under Rician channels...\" provides no simulation setup, parameter choices, or figure; either add details or remove the claim to keep the paper self-contained.","section":"Section III-D, last paragraph"},{"comment":"The abstract states that group-connected RIS with group size 2 provides gains in both Rayleigh and LoS channels; this is true for the opposite-polarization LoS scenario, but in the same-polarization LoS case the gain is G=1. Please add the qualifier \"opposite polarization\" to avoid overgeneralization.","section":"Abstract and Section III-D"}],"recommendation":"major_revision","confidential_remarks":"The core contribution (Prop. 1 and the scaling laws) is sound and publishable, but the proof of Prop. 2 is not complete as written. The algebraic typo in Eq. (27) is easy to fix, but the proof also needs a full treatment of architectures with fewer than n non-singleton groups, and the transfer of [4, Prop. 1] needs to be justified. These are load-bearing for the Pareto-frontier claim, so I recommend major revision rather than rejection, since the issues appear repairable within the manuscript's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe scaling-law work here is solid and worth citing: first dual-polarized BD-RIS analysis, a nice LoS gain result G=(1+chi)^2/(4chi), and Prop. 1 showing group size 2 hits the fully-connected bound at linear complexity. I verified the algebra in Sec. III and Table I; it checks out under the stated models.\n\nThe soft spot is Prop. 2, the Pareto-frontier characterization. The proof as printed has a real algebraic error in Eq. (27). The expansion should be 4n^2 chi + 2n(1-chi)^2 n_h - (1-chi)^2 n_h^2; the printed middle term lacks the square on (1-chi). With the printed expression the derivative vanishes at n_h = n/(1-chi), not n_h=n, so the claimed maximizer doesn't follow. With the corrected coefficient the derivative does vanish at n_h=n and Q_max = n(1+chi), so the result is likely salvageable, but the proof is invalid as-is.\n\nThere's also an unstated import: the first line uses [4, Prop. 1] to set G=N-n groups for complexity C=N+n. That result was for uni-polarized BD-RIS. It's probably a counting/forest-equivalence statement that transfers, but the paper doesn't say why. The same group authored [4], so it's not a citation-pattern problem, just a missing justification.\n\nMinor: in Prop. 2's proof, the line after (25) writes both norms as ||hR_hat||; the second should be ||hT_hat||.\n\nNet: the core claim—that a group-connected RIS with group size 2 is optimal in dual-polarized LoS—is plausible and likely correct, but it rests on a proof that needs repair. The scaling laws and Prop. 1 stand alone. I'd send this to a serious referee, with the expectation that the authors fix Eq. (27) and make the import of [4] explicit. I wouldn't cite Prop. 2 in its current form, but I would cite the scaling laws and the group-2 result.\n\nBring it to reading group if you want to walk through the proof repair; otherwise it's a solid IEEE Comm. Letters type contribution once cleaned up.\n\nBest,","headline":"Solid scaling-law contributions, but the Pareto-frontier proof has a fixable algebraic error and an unstated import; publish after repair.","tokens_in":9811,"tokens_out":3611,"would_cite":true,"duration_ms":28314,"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":"For dual-polarized line-of-sight links, a group-connected RIS whose groups pair one vertical and one horizontal element reaches the fully-connected performance bound while needing only 3N/2 tunable components.","keywords":["beyond diagonal RIS","dual-polarized MIMO","reconfigurable intelligent surface","cross-polar discrimination","line-of-sight channels","performance-complexity trade-off","Pareto frontier","group-connected RIS"],"falsifier":"Enumerate all group partitions of a small surface (say $N=8$) into $n$ pairs and singletons under both polarization assignments, compute the maximum of $(\\sum_g \\lVert h_{R,g}\\rVert\\lVert h_{T,g}\\rVert)^2$ over all orderings, and compare it with $(n(1+\\chi)+(N-2n)\\sqrt{\\chi})^2$; any partition that beats the formula, or any corrected algebra showing a maximum away from $n_h=n$, would disprove Proposition 2.","tokens_in":8741,"feed_emoji":"📡","tokens_out":13603,"duration_ms":99766,"temperature":0.7,"pith_summary":"This paper studies reconfigurable intelligent surfaces (RISs) whose scattering matrices go beyond the diagonal form (BD-RIS), in systems where the surface elements and the transmitter/receiver antennas are dual-polarized. It claims that when the transmitter and receiver have opposite polarizations, BD-RIS beats conventional diagonal RIS even in line-of-sight channels, with a gain that grows as the cross-polar discrimination worsens. The central result is that the fully-connected upper bound on received power is achieved by a much simpler group-connected architecture: split the $N$ elements into $N/2$ pairs, each containing one vertical and one horizontal element. That design needs $3N/2$ tunable impedance components instead of $N(N+1)/2$. The paper also gives the full Pareto frontier of the performance-complexity trade-off, claiming that the optimal BD-RIS with $N+n$ components has $n$ such pairs and $N-2n$ unpaired elements.","feed_headline":"Two-element RIS groups hit the power ceiling","feed_subtitle":"Pairing opposite polarizations reaches the theoretical power limit using only 3N/2 tunable parts.","key_machinery":"The central object is the group-connected BD-RIS scattering matrix, whose tunable impedance network partitions the $N$ elements into groups. The load-bearing identity is that a group holding one vertical and one horizontal element, under opposite Tx/Rx polarization, gives $\\lVert h_{R,g}\\rVert=\\lVert h_{T,g}\\rVert=\\sqrt{1+\\chi}$ regardless of the line-of-sight phases, so each group contributes $1+\\chi$ to the coherent sum and $N/2$ groups produce $((1+\\chi)N/2)^2$, exactly the fully-connected bound $\\lVert h_R\\rVert^2\\lVert h_T\\rVert^2=((1+\\chi)N/2)^2$. For intermediate complexity, the argument rewrites the received power as $(\\sum_g \\lVert h_{R,g}\\rVert\\lVert h_{T,g}\\rVert)^2$ and maximizes the contribution of the $n$ pairs through the Cauchy–Schwarz upper bound $\\lVert \\hat h_R\\rVert\\lVert \\hat h_T\\rVert$, whose optimum at equal numbers of vertical and horizontal elements yields $n(1+\\chi)$.","core_discovery":"The paper establishes that in a dual-polarized RIS-aided link with opposite Tx/Rx polarization, the received-power upper bound for a lossless BD-RIS is $P_R^\\mathrm{fully}=(1+\\chi)^2 N^2/4$, where $\\chi\\in[0,1]$ is the inverse cross-polar discrimination; a diagonal RIS only reaches $\\chi N^2$ in line-of-sight, giving a BD-RIS gain $G=(1+\\chi)^2/(4\\chi)$. Proposition 1 shows that any group-connected RIS with group size 2 whose groups pair opposite polarizations attains the upper bound exactly for every $\\chi$, with circuit complexity $C=3N/2$. Proposition 2 extends this to intermediate complexities: for $C=N+n$, the optimal architecture is $n$ opposite-polarization pairs plus $N-2n$ singletons, delivering $(n(1+\\chi)+(N-2n)\\sqrt{\\chi})^2$. In Rayleigh fading the paper derives separate scaling laws, with gain $4(1+\\chi)^2/(\\pi^2\\chi)$ for opposite polarization and $16/\\pi^2$ for same polarization.","pith_inferences":["Because the group-of-2 gain in line-of-sight does not depend on the LoS phases, the same wiring pattern should remain optimal under phase drift or imperfect channel knowledge, so the architecture could be prototyped with fixed inter-element couplings; the paper does not explore this simplification.","The Pareto formula suggests a graceful deployment strategy: start with $N/2$ opposite-polarization pairs and remove pairs one at a time to lower complexity, with received power decreasing quadratically in the number of remaining pairs; the paper does not present this as a design procedure.","The paper's Rician simulations indicate the gain lies between the LoS and Rayleigh extremes; deriving the explicit Rician scaling law and checking whether group-of-2 remains Pareto-optimal for all Rician factors would be a natural next step that the paper leaves open.","The same polarization-pairing principle may carry over to multi-user BD-RIS architectures, but the paper's analysis is single-user and does not address that setting."],"forward_implications":["In dual-polarized line-of-sight links with opposite transmitter and receiver polarization, a group-connected RIS with group size 2 attains the fully-connected received power $(1+\\chi)^2 N^2/4$ using $3N/2$ tunable impedance components instead of $N(N+1)/2$.","The BD-RIS gain over diagonal RIS in that scenario is $G=(1+\\chi)^2/(4\\chi)$, so the largest relative gains occur at small $\\chi$ (for instance $G=3$ at $\\chi=0.1$) and disappear at $\\chi=1$.","For intermediate complexity $C=N+n$, the claimed Pareto frontier is $(n(1+\\chi)+(N-2n)\\sqrt{\\chi})^2$, realized by $n$ opposite-polarization pairs and $N-2n$ unpaired single elements.","With Rayleigh fading and opposite polarization, the asymptotic gain is $4(1+\\chi)^2/(\\pi^2\\chi)$, while with same polarization it is $16/\\pi^2$ independent of $\\chi$.","Because the group-of-2 design saturates the performance bound, the tree- and fully-connected architectures are not needed for maximum power in this scenario, which can simplify prototype hardware."],"supporting_citations":[{"why":"Supplies the cascaded channel model that represents the received signal as a function of the RIS scattering matrix.","marker":"[1]"},{"why":"Provides the received-power formulas for single- and fully-connected RIS, the group-connected formula used in Proposition 1, and the expectation steps for the Rayleigh scaling laws.","marker":"[2]"},{"why":"Establishes the tree- and forest-connected architectures as least-complex designs reaching maximum performance in single-user systems, the baseline that the group-of-2 result overturns in dual-polarized line-of-sight.","marker":"[3]"},{"why":"Supplies the imported result that fixes the number of groups of an optimal BD-RIS as a function of its circuit complexity, which Proposition 2 relies on.","marker":"[4]"},{"why":"Supplies the dual-polarized channel model with polarization weight vectors parameterized by the inverse cross-polar discrimination.","marker":"[6]"}],"fun_headline_variants":["Opposite polarization pairing hits RIS power ceiling","Group-size-2 RIS matches theoretical power bound","Pairing opposite polarizations reaches full power","Dual-polarized BD-RIS: 3N/2 parts, full power","Low-complexity RIS hits dual-polarized power limit"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof of the performance-complexity frontier takes as given a previously established formula for how many groups an optimal surface has, even though that formula was derived for single-polarization systems, and one algebraic step in the same proof is written with the wrong coefficient, so the claimed optimum at equally many vertical and horizontal elements is not fully verified as printed.","fun_headline_variants_meta":{"raw":{"variants":["Opposite polarization pairing hits RIS power ceiling","Group-size-2 RIS matches theoretical power bound","Pairing opposite polarizations reaches full power","Dual-polarized BD-RIS: 3N/2 parts, full power","Low-complexity RIS hits dual-polarized power limit"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001553,"raw_usage":{"total_tokens":6176,"prompt_tokens":885,"completion_tokens":5291,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":501,"completion_tokens_details":{"reasoning_tokens":5212}},"tokens_in":501,"tokens_out":5291,"duration_ms":37047,"temperature":1.0,"reasoning_tokens":5212,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T10:48:31.342730+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Enumerate all group partitions of a small surface (say $N=8$) into $n$ pairs and singletons under both polarization assignments, compute the maximum of $(\\sum_g \\lVert h_{R,g}\\rVert\\lVert h_{T,g}\\rVert)^2$ over all orderings, and compare it with $(n(1+\\chi)+(N-2n)\\sqrt{\\chi})^2$; any partition that beats the formula, or any corrected algebra showing a maximum away from $n_h=n$, would disprove Proposition 2.","supporting_citations":[{"cited_title":"Beyond diagonal reconfig- urable intelligent surfaces utilizing graph theory: Modeling, architecture design, and optimization,","cited_arxiv_id":null,"evidence_quote":"Establishes the tree- and forest-connected architectures as least-complex designs reaching maximum performance in single-user systems, the baseline that the group-of-2 result overturns in dual-polarized line-of-sight."},{"cited_title":"Pareto frontier for the performance- complexity trade-off in beyond diagonal reconfigurable intelligent sur- faces,","cited_arxiv_id":null,"evidence_quote":"Supplies the imported result that fixes the number of groups of an optimal BD-RIS as a function of its circuit complexity, which Proposition 2 relies on."},{"cited_title":"Limited feedback beamforming systems for dual-polarized MIMO channels,","cited_arxiv_id":null,"evidence_quote":"Supplies the dual-polarized channel model with polarization weight vectors parameterized by the inverse cross-polar discrimination."}],"review_version":1}