{"id":"1f4aa6f6-8b34-4a70-a6f9-80e663a47499","arxiv_id":"2411.18298","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The capacity-maximizing BD-RIS reflection matrix is Θ = V_F V_G^H, pairing the singular-value directions of the two channel matrices in order of strength.","lead":"This paper derives a closed-form formula for the optimal reflection matrix of a beyond-diagonal reconfigurable intelligent surface (BD-RIS) in a MIMO link with no direct transmitter-receiver path. The optimal configuration pairs the strongest propagation paths in order of strength, and the paper gives the resulting capacity in closed form.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The closed-form Θ is provably optimal only under the explicit no-direct-path assumption; the paper offers no robustness analysis for a direct link, so the practical MIMO claim remains conditional on exact blockage.","rationale":"The proof of Theorem 1 is algebraically sound; my check of the padded SVD construction and the majorization step found no gap. The reader's weakest_assumption correctly identifies the no-direct-path condition as the most sensitive modeling choice for the central claim. The missing error bars and the relation to [14], [15] are presentation issues, not threats to the theorem's internal correctness. I therefore keep the reader's CONDITIONAL verdict (reflected here as no change to the reader's verdict) and recommend a direct-path robustness check to settle whether the practical scope concern actually lands.","tokens_in":7830,"tokens_out":10509,"duration_ms":99162,"concrete_test":"Simulate a MIMO setup with F and G as in the paper and add a Rician direct path H_d = sqrt(κ/(1+κ)) H_LOS + sqrt(1/(1+κ)) H_NLOS for κ ∈ {-∞, 0, 3, 10} dB. Compare the capacity achieved by (8)–(9) with waterfilling against a numerically optimized Θ (e.g., manifold gradient ascent on the unitary group). If the gap is negligible for κ ≥ 0 dB, the no-direct-path concern is not load-bearing; if the gap is significant, the paper's practical claim should be explicitly restricted to H_d = 0.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 1 and its proof are internally sound: under H = FΘG^H, the Lemma 1 majorization bound and the achievability calculation in (16) are correct. The load-bearing point is the Section II assumption that no transmitter–receiver path exists except via the BD-RIS. With a direct path H_d, the channel becomes H_d + FΘG^H; the factorization in (14) and the SVD alignment argument in (16) no longer hold, and Θ = V_F V_G^H is not in general optimal. Since practical MIMO deployments generally have at least some direct-path component, the title-level claim of a capacity-maximizing BD-RIS reflection matrix for a MIMO channel overstates the demonstrated scope. This is a modeling limitation rather than a proof error; the authors should either state the conditional nature more prominently or bound the suboptimality when H_d ≠ 0.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies a narrowband MIMO channel assisted by a fully connected beyond-diagonal RIS, under the assumption that there are no transmitter-receiver paths except through the RIS. The end-to-end channel is H=FΘG^H with Θ constrained to be unitary. The main result, Theorem 1, provides a closed-form global solution: Θ=V_F V_G^H and Q=U_G diag(q_1,...,q_Nt)U_G^H, where the q_i are waterfilling powers, and the capacity is the sum of K=min(Nt,Nr,M) terms involving the products of the ordered singular values of F and G^H. The proof uses a majorization lemma for singular values of matrix products and an explicit achievability calculation. Corollaries characterize the high-SNR permutation invariance and the semi-unitary channel case. Numerical results compare the closed-form solution with the iterative reciprocal BD-RIS algorithm of [11] and with a conventional RIS baseline.","tokens_in":7920,"tokens_out":18741,"duration_ms":170286,"significance":"If the result is taken within its stated no-direct-path model, it is a significant advance: it gives the first closed-form globally optimal BD-RIS reflection matrix for a MIMO channel, replacing iterative algorithms with a parameter-free analytical solution. The proof via majorization is elegant and self-contained, the achievability step is explicit, and the path-pairing interpretation provides genuine geometric insight. The numerical experiments corroborate the analytical claims and also quantify the gap to reciprocal BD-RIS and conventional RIS. These strengths make the paper a useful contribution to the BD-RIS literature, provided the scope is communicated precisely.","major_comments":[{"comment":"The optimality of Theorem 1 is entirely conditional on the Section II assumption that there are no transmitter-receiver paths except via the BD-RIS. If a direct path H_d is present, the channel becomes H=FΘG^H+H_d, the factorization used in equation (14) no longer holds, the singular-value alignment argument in equation (16) fails, and Θ=V_F V_G^H is not generally optimal. Since the title and abstract claim a capacity-maximizing BD-RIS reflection matrix 'for a MIMO channel' without this qualification, the demonstrated scope is narrower than the stated claim. Please qualify the title, abstract, and conclusions, and either add a discussion of the direct-link case or provide a bound on the suboptimality of the proposed Θ when H_d≠0.","section":"Section II, Abstract, and Title"}],"minor_comments":[{"comment":"The second determinant in equation (20) should be with respect to I_{N_t}, not I_{N_r}; as written, it would scale with N_r and contradict equation (12) when N_r≠N_t.","section":"Appendix C, equation (20)"},{"comment":"The proof drops the '+1' when passing from the exact capacity expression to equation (17), so Corollary 1 should be stated as an asymptotic high-SNR result; for finite SNR, a permuted pairing can be strictly suboptimal.","section":"Appendix B, Corollary 1"},{"comment":"The statement that F and G each have K equally large singular values is ambiguous in cases such as N_t≤M≤N_r, where a semi-unitary F with orthonormal columns has M nonzero singular values while K=N_t; please clarify the convention used in the proof, where F^H F=σ_F^2 I_M and G G^H=σ_G^2 I_{N_t}.","section":"Appendix C, Corollary 2"},{"comment":"The plotted curves are averages over many random channel realizations, but no error bars, confidence intervals, or number of realizations are reported; please add this information so the variability of the comparisons can be assessed.","section":"Section IV"},{"comment":"The initialization expression 'FHH*G' is not typeset clearly and the matrix H* is defined only in prose; please rewrite the expression with explicit definitions and dimensions.","section":"Footnote 3"}],"recommendation":"major_revision","confidential_remarks":"The technical core of the paper is sound and the closed-form result is a clear contribution. The requested revision is about scope framing: the no-direct-path assumption must be reflected in the title, abstract, and conclusions, and the authors should either add a robustness discussion for a direct link or explicitly bound the suboptimality when H_d≠0. The proof issues are otherwise local and fixable. No concerns about novelty or citation practice."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know: this paper gives the first closed-form globally optimal BD-RIS reflection matrix for a MIMO channel, and the proof holds up. The main caveat is the no-direct-path assumption, which is explicit in Section II but never revisited for robustness.\n\nThe result itself is clean. Theorem 1 says the optimal Θ is V_F V_G^H, pairing the ordered singular-vector directions of the two channel matrices, with Q a waterfilling covariance in the U_G basis. The proof via Lemma 1's majorization bound is correct and self-contained. The corollaries add real value: at high SNR any permutation among the K strongest singular directions works, and for semi-unitary channels with M ≤ max(Nt,Nr) any unitary Θ is optimal, which tells you when BD-RIS buys nothing over a diagonal RIS. The numerical results are consistent with the theorems, including the semi-unitary behavior in Fig. 4.\n\nWhat's genuinely new is the explicit construction and the geometric interpretation; the capacity expression itself aligns with the already-published results [14] and [15], and the paper says so. That honesty is to its credit. The SVD-pairing technique is standard, but applying it to close the BD-RIS MIMO problem in one shot is a solid contribution to the subfield.\n\nThe soft spots are real but not damaging. The no-direct-path assumption is load-bearing: if H_d ≠ 0, the channel is FΘG^H + H_d, and the Theorem 1 Θ is not generally optimal. The paper states the assumption clearly, so there is no deception, but the title's \"MIMO channels assisted by BD-RIS\" overstates the demonstrated scope. A short paragraph on how a direct path breaks the solution, or a bound on suboptimality, would fix this. Minor issues: a typo in Appendix C where I_Nr appears instead of I_Nt, and the simulation curves have no error bars, though the gaps between curves are large enough that the qualitative conclusions are not in doubt.\n\nThe paper is for RIS/MIMO researchers, especially those working on BD-RIS architectures. It deserves serious refereeing. I would accept it with minor revisions rather than desk-reject.","headline":"Rigorous closed-form optimal BD-RIS solution for no-direct-path MIMO, with a clear geometrical story; the direct-path caveat is real but explicitly scoped.","tokens_in":8486,"tokens_out":2452,"would_cite":true,"duration_ms":22487,"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":"The paper derives the first closed-form capacity-maximizing reflection matrix for a MIMO channel assisted by a beyond-diagonal RIS, showing that the surface should pair the singular-value directions of the two channel matrices in order of…","keywords":["beyond-diagonal RIS","MIMO capacity","closed-form optimization","singular value decomposition","reflection matrix","waterfilling power allocation","semi-unitary channels","reconfigurable intelligent surfaces"],"falsifier":"Search for a counterexample: for randomly generated $F$ and $G$, compute the capacity from Theorem 1 and compare it with the largest value of $\\log_2\\det(I_{N_r} + F\\Theta G^H Q G\\Theta^H F^H/N_0)$ found by numerical optimization over unitary $\\Theta$ and trace-constrained $Q$. The theorem claims equality for every $F,G$, so one instance where the numerical search exceeds the closed form would refute it; under the model, the search should never win.","tokens_in":7596,"feed_emoji":"📡","tokens_out":9728,"duration_ms":81342,"temperature":0.7,"pith_summary":"This paper establishes a closed-form solution for the capacity-maximizing reflection matrix of a beyond-diagonal reconfigurable intelligent surface (BD-RIS) in a narrowband MIMO channel where the transmitter and receiver are connected only through the surface. The optimal surface is $\\Theta = V_F V_G^H$, formed from the right singular vectors of the two channel matrices, and the optimal transmit covariance is waterfilling over the squared singular values $\\sigma_i^2(F)\\sigma_i^2(G^H)$. The result matters because it replaces iterative algorithms with an analytic formula and reveals the geometry: the surface pairs the strongest transmitter-to-surface path with the strongest surface-to-receiver path, then continues in descending order of strength. It also identifies special cases, such as semi-unitary channels with few surface elements, where a conventional diagonal RIS already achieves the same capacity.","feed_headline":"Closed-form optimal BD-RIS reflection matrix found for MIMO","feed_subtitle":"The new matrix formula reveals how a BD-RIS should pair paths and when a simpler RIS is enough.","key_machinery":"The machinery is the singular value decomposition combined with a majorization-type inequality on singular values of matrix products (Lemma 1). The feasible set for the BD-RIS is the unitary group $\\{\\Theta:\\Theta^H\\Theta = I_M\\}$, which lets $\\Theta$ act as a passive rotation between the right singular-vector bases of $G$ and $F$. The proof bounds the capacity for any $\\Theta$ and $Q$ by $\\sum_i \\log_2(1 + q_i\\sigma_i^2(F)\\sigma_i^2(G^H)/N_0)$ using the inequality, then shows that $\\Theta = V_F V_G^H$ and the waterfilling covariance meet the bound term by term. $U_G$ diagonalizes the transmit covariance, while $V_F$ and $V_G$ realize the direction pairing.","core_discovery":"The central claim, stated as Theorem 1, is that for the channel $H = F\\Theta G^H$ with no direct path, the maximum of $\\log_2\\det(I_{N_r} + HQH^H/N_0)$ over unitary $\\Theta$ and positive semidefinite $Q$ with trace bound is attained by $\\Theta = V_F V_G^H$ and $Q = U_G\\operatorname{diag}(q_1,\\dots,q_{N_t})U_G^H$, where $V_F$, $V_G$, and $U_G$ come from the singular value decompositions of $F$ and $G$, and $q_i$ are waterfilling powers. The resulting capacity is $\\sum_{i=1}^K \\log_2(1 + q_i\\sigma_i^2(F)\\sigma_i^2(G^H)/N_0)$ with $K=\\min(N_t,N_r,M)$. Geometrically, the optimal BD-RIS maps the $i$th strongest singular direction of $G$ to the $i$th strongest singular direction of $F$, and the waterfilling level decides how many of those paired paths carry power. Corollary 1 shows that at high SNR the pairing order among the $K$ strongest directions can be any permutation, and Corollary 2 shows that for semi-unitary $F$ and $G$ with $M\\le\\max(N_r,N_t)$, every unitary $\\Theta$ is optimal, so a conventional RIS matches the BD-RIS performance.","pith_inferences":["Beyond the paper, if a direct transmitter–receiver path is added, the effective channel becomes $F\\Theta G^H + H_d$, and the closed form is no longer claimed optimal; a natural test is to quantify how far $\\Theta = V_F V_G^H$ falls from the numerical optimum as the direct path grows.","A design heuristic the paper does not explore is to project the SVD-pairing solution $\\Theta = V_F P V_G^H$ onto the symmetric-unitary set required by reciprocal BD-RIS circuits, then measure the capacity loss; this would connect the closed form to the reciprocal architectures used in earlier work.","A testable consequence of Corollary 2 is a design rule for when BD-RIS hardware is worthwhile: use BD-RIS only when $M > \\max(N_t,N_r)$ or when the singular values of $F$ and $G$ are markedly unequal, since otherwise a conventional RIS already reaches the same capacity."],"forward_implications":["A BD-RIS-assisted MIMO link with no direct path can be configured for maximum capacity in closed form, without iterative search.","The optimal reflection matrix does not depend on the transmit power or noise level; waterfilling only selects how many of the paired singular directions are active.","At high SNR, the $K$ strongest directions of $F$ and $G$ can be paired in any order, so the surface only needs to match the two subspaces, not the exact ordering.","When $F$ and $G$ are semi-unitary with equal singular values and $M\\le\\max(N_t,N_r)$, any unitary reflection matrix achieves the same capacity, and a conventional diagonal RIS is sufficient.","The spatial multiplexing gain is $\\min(N_t,N_r,M)$, so the number of BD-RIS elements caps the number of parallel data streams when it is smaller than both antenna counts."],"supporting_citations":[{"why":"Introduces the BD-RIS architecture and the scattering-parameter model whose general feasible set is the unitary reflection matrices used in the problem formulation.","marker":"[6]"},{"why":"Provides the previous iterative MIMO-capacity algorithm for BD-RIS that the closed-form solution in Theorem 1 replaces and is benchmarked against.","marker":"[11]"},{"why":"Supplies the waterfilling characterization of MIMO capacity and the conventional RIS optimization baseline used in the numerical comparisons.","marker":"[13]"},{"why":"Gives an earlier singular-value-alignment result for line-of-sight MIMO via an intelligent reflecting surface that Theorem 1 is aligned with but proves differently.","marker":"[14]"},{"why":"Provides a related spatial-multiplexing result for RISs that Theorem 1 also aligns with, supporting the singular-value pairing interpretation.","marker":"[15]"},{"why":"Supplies the product singular-value inequality used as the main step of Lemma 1 to upper-bound the capacity of any configuration.","marker":"[16]"},{"why":"Gives the monotonicity theorem used to convert the singular-value product inequality into the logarithmic capacity bound in Lemma 1.","marker":"[17]"}],"fun_headline_variants":["Closed-form BD-RIS matrix solves MIMO capacity","Optimal BD-RIS reflection matrix: closed form","BD-RIS capacity max: closed-form path pairing","First closed-form BD-RIS solution for MIMO","MIMO BD-RIS: closed-form optimal pairing"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the assumption that there are no transmitter-receiver paths except through the BD-RIS; if a direct path exists, the channel becomes $F\\Theta G^H + H_d$ and the closed-form $\\Theta$ from Theorem 1 is generally no longer optimal.","fun_headline_variants_meta":{"raw":{"variants":["Closed-form BD-RIS matrix solves MIMO capacity","Optimal BD-RIS reflection matrix: closed form","BD-RIS capacity max: closed-form path pairing","First closed-form BD-RIS solution for MIMO","MIMO BD-RIS: closed-form optimal pairing"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00032,"raw_usage":{"total_tokens":1856,"prompt_tokens":1047,"completion_tokens":809,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":663,"completion_tokens_details":{"reasoning_tokens":733}},"tokens_in":663,"tokens_out":809,"duration_ms":7569,"temperature":1.0,"reasoning_tokens":733,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:20:10.522333+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Search for a counterexample: for randomly generated $F$ and $G$, compute the capacity from Theorem 1 and compare it with the largest value of $\\log_2\\det(I_{N_r} + F\\Theta G^H Q G\\Theta^H F^H/N_0)$ found by numerical optimization over unitary $\\Theta$ and trace-constrained $Q$. The theorem claims equality for every $F,G$, so one instance where the numerical search exceeds the closed form would refute it; under the model, the search should never win.","supporting_citations":[{"cited_title":"Modeling and architecture design of reconfigurable intelligent surfaces using scattering parameter network analysis,","cited_arxiv_id":null,"evidence_quote":"Introduces the BD-RIS architecture and the scattering-parameter model whose general feasible set is the unitary reflection matrices used in the problem formulation."},{"cited_title":"MIMO capacity maximization with beyond-diagonal RIS,","cited_arxiv_id":null,"evidence_quote":"Provides the previous iterative MIMO-capacity algorithm for BD-RIS that the closed-form solution in Theorem 1 replaces and is benchmarked against."},{"cited_title":"Line-of-sight MIMO via intelligent reflecting surface,","cited_arxiv_id":null,"evidence_quote":"Gives an earlier singular-value-alignment result for line-of-sight MIMO via an intelligent reflecting surface that Theorem 1 is aligned with but proves differently."},{"cited_title":"Spatial multiplexing in near field MIMO channels with reconfigurable intelligent surfaces,","cited_arxiv_id":null,"evidence_quote":"Provides a related spatial-multiplexing result for RISs that Theorem 1 also aligns with, supporting the singular-value pairing interpretation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the product singular-value inequality used as the main step of Lemma 1 to upper-bound the capacity of any configuration."}],"review_version":1}