{"id":"dbb6031b-4de7-41de-b08f-3f4b8b880f8b","arxiv_id":"2507.16767","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"The sum mutual information of a multi-RIS MIMO multiple-access channel is asymptotically Gaussian with closed-form mean and variance, enabling statistical phase optimization that works down to small antenna counts.","lead":"This chapter derives closed-form approximations for the average and the spread of the information rate of a multi-antenna uplink served by several programmable reflecting surfaces, and shows the rate distribution is nearly Gaussian. The practical payoff is that surface phase patterns can be tuned from channel statistics alone, without fast channel tracking, with the biggest gains when signals near the surfaces are already correlated.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (14) is not a real positive variance in the M=1, K=0 limit: det Λ < 0, so −log det Λ is undefined or negative; the Gaussian outage claim rests on an invalid variance formula.","rationale":"The chapter's central claim is the Gaussian approximation of sum-MI, which requires both a correct mean (10) and a correct variance (14). The reader correctly notes that the variance is black-boxed and the vanishing of higher cumulants is unproved. But there is a sharper, checkable problem: Eqs. (14)-(24), as displayed, cannot be a variance. In the simplest limit, a single TX with no RIS and uncorrelated identity covariance, the 2×2 matrix Λ has diagonal entries −r_d^2 and −t_d^2 with off-diagonal −1; its determinant is r_d^2 t_d^2 − 1 < 0 for all ρ>0. Thus −log det Λ has an imaginary part and a negative real part, while the variance of mutual information is real and positive. This is independent of any large-N convergence or higher-cumulant issue. I am not claiming the underlying statistical-physics method is wrong or that the authors are being careless; the displayed formula may contain a sign or identity error that a revision could repair. But as printed, the strongest claim fails a necessary sanity check, so the reader's conditional acceptance should be moved to rejection unless the variance formula is corrected and re-derived. The mean-based optimization sections may survive, but the headline outage and Gaussianity result depends on Eq. (14).","tokens_in":20093,"tokens_out":15073,"duration_ms":174542,"concrete_test":"Run the one-line sanity check: set M=1, K=0, N_t=N_r=N with N large, R_d=T_d=I_N, Q=I_N (ρ=1). Solve the fixed-point equations to get r_d=t_d=(√5−1)/2≈0.618. Assemble Λ from (15)-(24), which is [[−r_d^2, −1], [−1, −t_d^2]], and evaluate −log det(Λ). If the expression is non-real or negative while a Monte Carlo estimate of Var[log det(I+HH†/N)] is positive, then Eq. (14) is not a valid variance formula. This check distinguishes a typographical sign error (fixable) from a genuine statistical claim.","verdict_should_be":"REJECT","load_bearing_attack":"The load-bearing piece of the central claim is the variance formula (14), because the Gaussian outage approximation in Fig. 5 needs a valid positive variance. As printed, (14) fails a basic definiteness check. The diagonal blocks defined in (16)-(24) are all non-positive, and the off-diagonal blocks include −I. For the natural specialization M=1, K=0, N_t=N_r=N, R_d=T_d=I, Q=ρI, the fixed-point equations give r_d = (−1+√(1+4ρ))/2 and t_d = 1/(1+r_d), so Λ in (15) reduces to the 2×2 matrix [[−r_d^2, −1], [−1, −t_d^2]]. For any ρ>0, det Λ = r_d^2 t_d^2 − 1 < 0; for ρ=1, det Λ ≈ −0.854. Hence −log det Λ is not a real number, and its real part is negative, so it cannot equal Var(I), which is positive. The same sign pattern persists for K≥1. Thus Eq. (14) is invalid as written, independently of the unproved vanishing of higher cumulants. The mean formula (10) and the optimization results are not directly affected, but the chapter's headline statistical claim is built on a quantity that is not a variance.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The chapter studies an uplink MIMO multiple-access channel assisted by multiple reconfigurable intelligent surfaces (RISs). It imports from the authors' prior work [69] a large-system closed-form expression for the ergodic sum mutual information, states an asymptotic variance formula and an asymptotic Gaussianity result for the sum-MI, and uses these to approximate outage probabilities and to motivate two RIS phase-optimization schemes, one semi-optimal and one gradient-based. Numerical sections compare the analytic ergodic rates and the Gaussian outage approximation with Monte Carlo simulations, reporting good agreement for moderate antenna and RIS sizes. The central advertised contributions are the variance formula (14), the Gaussian approximation for outage, and the statistically driven RIS optimization methodology.","tokens_in":20337,"tokens_out":7652,"duration_ms":84061,"significance":"If the statistical results are correct, the chapter would provide a computationally efficient performance characterization of multi-RIS MIMO-MAC systems, including outage probabilities and capacity-region boundaries, together with a decoupled RIS optimization procedure that uses only channel covariance information. The ergodic-rate comparisons and the demonstrated optimization gains for small angle spread are valuable engineering results, and the Monte Carlo validation in Fig. 5 is a strength in principle. However, the validity of the central statistical claim is not established as written because the variance formula (14) fails an elementary positive-definiteness check. The optimization sections are less affected, but the outage and Gaussianity claims, which are the chapter's distinctive contribution, rest on an expression that cannot be correct in its present form.","major_comments":[{"comment":"The variance formula Var(I) = -log det(Lambda) is not a valid positive real variance. In the specialization K=0, M=1, N_t=N_r=N, R_d=T_d=I, and Q=rho I, the fixed-point equations above (14) give r_d=(-1+sqrt(1+4rho))/2 and t_d=1/(1+r_d). With the definitions in (16)-(24), Lambda in (15) reduces to [[-r_d^2, -1],[-1, -t_d^2]], whose determinant is r_d^2 t_d^2 - 1 < 0 for every rho>0. Hence -log det(Lambda) is not real-valued, and on the principal branch its real part is negative, so it cannot equal the variance of the mutual information. The diagonal blocks defined in (16)-(24) are non-positive and the off-diagonal -I blocks persist for K>=1, so this is not an artifact of the K=0 limit. This invalidates the variance input to the Gaussian outage approximation in Fig. 5 unless a corrected expression, with a derivation, is supplied.","section":"Section 3.2, Eq. (14)"},{"comment":"The statement that all higher cumulants of the sum-MI vanish in the large-N_t limit, and the subsequent claim that the sum-MI is asymptotically jointly Gaussian, are asserted without proof or a precise theorem statement. This Gaussianity is the load-bearing step that converts the two moments into the outage approximation used in Fig. 5 and into the outage capacity-region discussion at the end of Section 3.2. The authors should either provide the argument or state exactly which theorem from [65] or [70], under which conditions, implies the result.","section":"Section 3.2, paragraph after Eq. (14)"},{"comment":"The main asymptotic mean expression (10) and the fixed-point equations (10)-(13) are imported from the authors' own [69] without derivation or a precise statement of the asymptotic regime. Because the variance and optimization results inherit any error in these equations, the chapter should state the exact assumptions under which (10) is the correct limit, and ideally include enough of the derivation that the variance calculation can be checked independently. This is a self-containedness and correctness-risk concern, not a claim that the prior result is wrong.","section":"Section 3.2, Eq. (10)"},{"comment":"The Monte Carlo agreement shown in Fig. 5 cannot, as it stands, corroborate Eq. (14), because the printed formula does not produce a real positive variance in the no-RIS limit. If the simulations or the analytic curves used a corrected version of (14), the authors need to state that corrected expression explicitly; otherwise the reader cannot reproduce the figure from the text.","section":"Section 5, Fig. 5"}],"minor_comments":[{"comment":"The caption states that the crosses are obtained with the fully-optimal methodology of Section 5, but Section 5 is the numerical-results section; the fully-optimal gradient method is described in Section 4.2.","section":"Section 5, Fig. 2 caption"},{"comment":"The channel-model paragraph cites \"[60, 60, 61]\", duplicating reference [60]; one of the two citations should be removed or replaced.","section":"Section 2.3"},{"comment":"Equation (32) has a typographical error in the exponent: the expression e(q_r - q_t,km) x_n is missing the imaginary unit and should read e^{i(q_r - q_t,km) x_n}.","section":"Section 4.1, Eq. (32)"},{"comment":"The convergence condition in Algorithm 1 contains two identically zero terms, |t_dm^(i)-t_dm^(i)| and |t1,km^(i-1)-t1,km^(i-1)|; these are likely intended to be |t_dm^(i)-t_dm^(i-1)| and |t1,km^(i)-t1,km^(i-1)|, respectively.","section":"Algorithm 1, line 11"}],"recommendation":"major_revision","confidential_remarks":"The chapter overlaps substantially with the authors' own published work [54] and [69], and its incremental contribution is the variance/Gaussianity/outage analysis and the optimization studies. That incremental contribution is currently undermined by the apparent sign/structural error in Eq. (14). If a corrected, derived variance expression cannot be supplied, the chapter's distinctive statistical claim is not established. The editor may also wish to confirm that the amount of material repeated from [69] is appropriate for the intended venue."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a serious extension of the authors' earlier large-system work on RIS-assisted MIMO, with a genuinely useful decoupling insight for phase optimization and a numerically well-supported Gaussian outage approximation down to 10^-3. But I checked the variance formula and it is not a variance. In the M=1, K=0 limit (single TX, no RIS) with N_t=N_r and Q=ρI, the fixed-point equations give r_d = (-1+√(1+4ρ))/2 and t_d = 1/(1+r_d), and Λ from (15) reduces to [[-r_d^2, -1], [-1, -t_d^2]]. det Λ = r_d^2 t_d^2 - 1 < 0 for every ρ>0, so -log det Λ is complex. A variance can't be negative or complex. This is not a cosmetic issue: the Gaussian-outage claim in Fig. 5 rests directly on Eq. (14). The paper also asserts, without proof, that higher cumulants vanish, and the mean expression (10) is imported from self-cited [69]. None of this is hidden; it's just not addressed.\n\nWhat the paper does well: it pushes the MAC analysis to multiple RISs, gives a closed-form variance and Gaussian approximation that match Monte Carlo in the regimes actually simulated, and — the strongest part — shows that the RIS phase optimization separates per RIS, so the phases can be optimized independently. That is a real practical takeaway. The comparison of semi-optimal versus full-gradient optimization, including the honest disclosure that the full-gradient method fails at large angular separations, is also useful.\n\nThe soft spots are proportionate to the state of the manuscript. The variance derivation is absent. The vanishing of higher cumulants is asserted, not proved. The K=0 collapse is a concrete red flag that should force a careful revisit. If the correct variance is recoverable by fixing a sign or an ordering issue in Λ, there is a valuable paper here. But as printed, the load-bearing statistic is invalid.\n\nWho is this for? Researchers in statistical-physics-based capacity analysis of RIS/MIMO and 6G system designers who want fast statistical optimization. It deserves a serious referee, but the referee should send it back with the request to correct and verify the variance formula. I would not cite Eq. (14) in its current form.","headline":"Useful decoupling and outage approximation, but the variance formula (14) is invalid as printed—it gives a complex value in the no-RIS limit—so the paper needs correction before its central statistical claim can be trusted.","tokens_in":697,"tokens_out":1581,"would_cite":false,"duration_ms":81546,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["94A15","94A40","60B20"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper establishes that the sum mutual information of a multi-user MIMO link with multiple RISs is asymptotically Gaussian, with closed-form mean and variance, accurate down to $10^{-3}$ outage for moderate system sizes.","keywords":["reconfigurable intelligent surfaces","MIMO multiple access channel","capacity region","outage probability","large-system analysis","statistical channel state information","RIS phase optimization","Gaussian approximation"],"falsifier":"Run Monte Carlo for the sum-MI at, for example, $N_t=4$, $N_r=8$, $N_s=400$ with correlated RIS-side channels and compute the empirical fourth cumulant: if it does not shrink relative to the square of the variance as $N_t$ increases, or if the Gaussian fit to the outage curve breaks down above $10^{-3}$, the central Gaussianity claim is wrong. A second check is whether the variance from Eq. (14) matches the Monte Carlo variance when the RIS phases are not the identity, since the optimization results inherit any error in the fixed-point equations.","tokens_in":19806,"feed_emoji":"📶","tokens_out":8511,"duration_ms":84083,"temperature":0.7,"pith_summary":"This paper tries to give a closed-form statistical description of a wireless uplink in which several multi-antenna users talk to one multi-antenna receiver through several reconfigurable intelligent surfaces, programmable reflectors that shape the radio environment. Its central claim is that the total mutual information of this channel is asymptotically Gaussian, so the outage probability can be computed from just its mean and variance, which the paper derives in closed form. The authors also show numerically that this Gaussian approximation stays accurate down to outage probabilities of $10^{-3}$ even for modest antenna counts and metasurface sizes. If the claim holds, system designers can dimension such links and optimize the surfaces from channel covariance statistics alone, without instantaneous channel estimates or brute-force simulation.","feed_headline":"A Gaussian law predicts multi-RIS MIMO outage rates","feed_subtitle":"Closed-form mean and variance match simulations down to one-in-a-thousand outage, even for small arrays.","key_machinery":"The load-bearing object is the asymptotic sum-MI expression (10), together with the fixed-point scalar parameters $t_{dm}, r_{dm}, t_{1k}, r_{1km}, t_{2km}, r_{2km}$ defined by the trace equations below it, and the variance formula (14), which is the negative log-determinant of the matrix $\\boldsymbol{\\Lambda}$ in (15). The key structural fact is that each RIS phase matrix $\\boldsymbol{\\Phi}_k$ appears only inside $\\boldsymbol{\\Sigma}_{km} = \\mathbf{S}_{t,km}^{1/2} \\boldsymbol{\\Phi}_k^\\dagger \\mathbf{S}_{r,k} \\boldsymbol{\\Phi}_k \\mathbf{S}_{t,km}^{1/2}$, within its own log-determinant, which decouples the optimization into one problem per RIS and makes the gradient updates in (29) and (31) depend only on local covariance knowledge. The vanishing-angle-spread reduction to rank-one matrices $\\mathbf{S}_{t,m}=N_s\\mathbf{u}_m\\mathbf{u}_m^\\dagger$ and $\\mathbf{S}_r=N_s\\mathbf{v}\\mathbf{v}^\\dagger$ produces the scalar figure of merit $\\kappa_m(\\boldsymbol{\\Phi}) = \\mathbf{v}^\\dagger\\boldsymbol{\\Phi}\\mathbf{u}_m$, which turns the phase design into aligning each RIS element's phase with the wave-vector differences $\\Delta \\mathbf{q}_m$.","core_discovery":"On the paper's own terms, the discovery is that in the limit where antenna counts and RIS element counts grow at fixed ratios, the ergodic sum mutual information of a Kronecker-correlated MIMO multiple-access channel with multiple RISs has an asymptotically Gaussian distribution whose mean is given by Eq. (10) and whose variance is given by Eq. (14), a negative log-determinant of a $(2M+4MK)$-dimensional matrix. Because the higher cumulants are asserted to vanish, outage probabilities follow from the two moments, and the Gaussian law is confirmed by Monte Carlo simulations down to $10^{-3}$ for systems as small as $N_t=4$, $N_r=8$ or $12$, and $N_s=400$ or $900$. The same asymptotic expressions reveal that the RIS phase matrices enter only through the sandwich combinations $\\boldsymbol{\\Sigma}_{km} = \\mathbf{S}_{t,km}^{1/2} \\boldsymbol{\\Phi}_k^\\dagger \\mathbf{S}_{r,k} \\boldsymbol{\\Phi}_k \\mathbf{S}_{t,km}^{1/2}$ in separate log-determinant terms, so each RIS can be optimized independently using only the covariance matrices of the channels impinging on and leaving it. The optimization gains are largest when those covariance matrices are highly correlated, namely at small angle spread, and the capacity region of the multi-user system is obtained by maximizing the weighted sum of these asymptotic expressions.","pith_inferences":["A natural test that goes beyond the paper is to compute the fourth cumulant of the sum-MI directly for small $N_t$ and check whether it decays relative to the variance as $N_t$ grows; that would separate a genuine large-system theorem from a numerical coincidence.","The per-RIS decoupling suggests a distributed deployment rule: each surface can be configured from its own local impinging and outgoing covariance estimates, so the result could extend to networks where surfaces are managed by different nodes without sharing full channel state.","Because the Gaussian approximation gives the full outage curve from two moments, a further extension would be to use it for finite-blocklength or delay-constrained metrics, where tail probabilities rather than ergodic averages are what matter.","If the covariance matrices are imperfectly estimated, the semi-optimal method's reliance on the leading eigenvectors of $\\mathbf{S}_{r,k}$ and $\\mathbf{S}_{t,km}$ may be more robust than the full gradient method, and this is testable by adding estimation noise to the covariances."],"forward_implications":["Outage probabilities for block-fading multi-RIS MIMO links can be read off from the Gaussian law with the closed-form mean and variance, matching Monte Carlo down to $10^{-3}$ for moderate sizes.","RIS phase profiles can be optimized per surface, in parallel, from statistical covariance matrices only, and the semi-optimal largest-eigenvalue method performs as well as full gradient ascent in the tested regimes.","The benefit of statistical RIS optimization grows as the angle spread shrinks, so the technique is most valuable for the correlated channels expected at higher carrier frequencies.","Coarse phase quantization retains most of the gain: 1-bit quantization still gives significant improvement and 2-bit is nearly optimal.","The ergodic capacity region of the MIMO-MAC-RIS system is obtained by sweeping the priority vector $\\boldsymbol{\\mu}$ in the weighted asymptotic sum-MI, with optimized RIS phases changing the usual pentagon into a curved boundary."],"supporting_citations":[{"why":"It supplies the asymptotic sum-MI expression and the fixed-point equations (10)-(13) that the chapter extends to variance, Gaussianity, and optimization.","marker":"[69]"},{"why":"It provides the single-RIS large-system analysis, the vanishing-angle-spread phase solution, and the correlation matrix model this chapter generalizes to multiple RISs and TXs.","marker":"[54]"},{"why":"It establishes the polymatroid structure and outage capacity region framework for fading multiple-access channels used to define the capacity region and outage.","marker":"[65]"},{"why":"It gives the weighted sum-MI maximization functional for tracing the MAC capacity boundary, adapted here with RIS phase matrices.","marker":"[68]"},{"why":"It supplies the Kronecker-product correlation model used for all channel covariance matrices.","marker":"[60]"},{"why":"It provides the angular-spread integral formula for the RIS correlation matrices appearing in (4).","marker":"[62]"},{"why":"It supplies the statistical-physics tools invoked to argue joint Gaussianity of the mutual information for different active TX sets.","marker":"[70]"}],"fun_headline_variants":["Multi-RIS MIMO outage rates follow Gaussian law","Closed-form stats for multi-RIS MIMO outages","Small arrays obey Gaussian law for multi-RIS systems","RIS optimization gains rise with channel correlation","Statistical physics simplifies multi-RIS capacity analysis"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that all fluctuations of the sum mutual information beyond its mean and variance disappear as the antenna and RIS element counts grow, so the Gaussian approximation becomes exact; the paper asserts this without proof, and the fixed-point equations that feed the mean and variance are themselves taken as given from earlier work.","fun_headline_variants_meta":{"raw":{"variants":["Multi-RIS MIMO outage rates follow Gaussian law","Closed-form stats for multi-RIS MIMO outages","Small arrays obey Gaussian law for multi-RIS systems","RIS optimization gains rise with channel correlation","Statistical physics simplifies multi-RIS capacity analysis"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000331,"raw_usage":{"total_tokens":1899,"prompt_tokens":1058,"completion_tokens":841,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":674,"completion_tokens_details":{"reasoning_tokens":770}},"tokens_in":674,"tokens_out":841,"duration_ms":9563,"temperature":1.0,"reasoning_tokens":770,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:02:22.964245+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run Monte Carlo for the sum-MI at, for example, $N_t=4$, $N_r=8$, $N_s=400$ with correlated RIS-side channels and compute the empirical fourth cumulant: if it does not shrink relative to the square of the variance as $N_t$ increases, or if the Gaussian fit to the outage curve breaks down above $10^{-3}$, the central Gaussianity claim is wrong. A second check is whether the variance from Eq. (14) matches the Monte Carlo variance when the RIS phases are not the identity, since the optimization results inherit any error in the fixed-point equations.","supporting_citations":[{"cited_title":"MIMO MAC empowered by reconfigurable intelli- gent surfaces: Capacity region and large system analysis,","cited_arxiv_id":null,"evidence_quote":"It supplies the asymptotic sum-MI expression and the fixed-point equations (10)-(13) that the chapter extends to variance, Gaussianity, and optimization."},{"cited_title":"Reconfigurable intelligent surfaces and capacity optimization: A large system analysis,","cited_arxiv_id":null,"evidence_quote":"It provides the single-RIS large-system analysis, the vanishing-angle-spread phase solution, and the correlation matrix model this chapter generalizes to multiple RISs and TXs."},{"cited_title":"Multiaccess fading channels Part I: Polymatroid structure, optimal resource allocation and throughput capacities,","cited_arxiv_id":null,"evidence_quote":"It establishes the polymatroid structure and outage capacity region framework for fading multiple-access channels used to define the capacity region and outage."},{"cited_title":"Iterative water-filling for Gaussian vector multiple- access channels,","cited_arxiv_id":null,"evidence_quote":"It gives the weighted sum-MI maximization functional for tracing the MAC capacity boundary, adapted here with RIS phase matrices."},{"cited_title":"New results for the multivariate Nakagami-𝑚 fading model with arbitrary correlation matrix and applications,","cited_arxiv_id":null,"evidence_quote":"It supplies the Kronecker-product correlation model used for all channel covariance matrices."},{"cited_title":"Communication through a diffusive medium: Coherence and capacity,","cited_arxiv_id":null,"evidence_quote":"It provides the angular-spread integral formula for the RIS correlation matrices appearing in (4)."},{"cited_title":"MIMO capacity through correlated channels in the presence of correlated interferers and noise: A (not so) large N analysis,","cited_arxiv_id":null,"evidence_quote":"It supplies the statistical-physics tools invoked to argue joint Gaussianity of the mutual information for different active TX sets."}],"review_version":1}