{"id":"c3513d6d-fbb1-4504-8789-025c5a699752","arxiv_id":"2607.15651","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"FSIM jointly optimizes beamforming, phase shifts, and in-layer element positions to outperform fixed-element stacked intelligent metasurfaces in simulated multiuser downlink sum rate.","lead":"Researchers propose putting movable 'fluid' antenna elements inside stacked intelligent metasurfaces, letting each layer reconfigure its geometry to focus signals on users. Simulations at 28 GHz show the design raises multiuser sum-rate above fixed-element stacked metasurfaces and other flexible-surface baselines.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Phase-shift subproblem (III-B) uses an invalid identity: q_{k,k} as defined is not the coefficient of ψ_l in g_k p_k, so the SCA/SDP update optimizes a different objective.","rationale":"The reader's concern about unvalidated channel modeling is reasonable but requires external measurements or full-wave simulation. A more immediately decisive problem is internal: the phase-shift optimization derivation in §III-B does not follow from the paper's own system model. For non-diagonal A_l, the paper's q_{k,k} misrepresents the dependence of g_k p_k on ψ_l. Consequently the SCA/rank-one phase update in (12)-(15) and Algorithm 1 is not maximizing the stated sum-rate, so the Section IV gains are not reproducible from the text. This does not necessarily sink the underlying FSIM idea—the algebra may be a typo and the authors' code may be correct—but as written the central algorithmic claim is unsupported. The verdict should remain CONDITIONAL: require correction of the q_{k,k} definition (and any related phase-shift expressions) or release of code/implementation details that verify the intended optimization. The reader's weakest_assumption did not identify this concern, hence disagreement.","tokens_in":10715,"tokens_out":18354,"duration_ms":164230,"concrete_test":"Implement a two-layer, two-element numerical check: L=2, N=2, random non-diagonal W_2, random phases, random ¯g and p. Compute the true objective by directly multiplying the chain in (3): |¯g^H Φ_2 W_2 Φ_1 W_1 p|^2. Compare it with the paper's phase-shift expression using q_{k,k} = diag(¯g^H) W_2 Φ_1 W_1 p (with A_2 = I, C_2 = W_2Φ_1 under their definitions). If the two values differ—as the L=1 counterexample shows they do—the phase-shift subproblem is invalid. More generally, re-derive §III-B using x^H A diag(ψ) B v = ψ^T[(Bv)⊙(A^T x^*)] and verify the q vector against that identity.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Section III-B defines q_{k,k} = diag(¯g_k^H) A_l C_l W_1 p_k and claims |g_k p_k|^2 = |ψ_l^H q_{k,k}|^2 with A_l = Φ_L W_L ... Φ_{l+1}W_{l+1}, C_l = W_lΦ_{l-1}...Φ_1. Directly from (3), g_k p_k = ¯g_k^H A_l diag(ψ_l) C_l W_1 p_k. For non-diagonal A_l this equals ψ_l^T[(C_l W_1 p_k) ⊙ (A_l^T ¯g_k^*)], so the correct conjugate-form vector is q'_k = (C_l W_1 p_k)^* ⊙ (A_l^H ¯g_k), not the paper's vector. The difference is not a constant phase: with L=1, N=2, take ¯g = [1,j]^T, W_1 p = [1,1]^T, ψ = [1,j]^T; the true |g p|^2 is 4, while the paper's |ψ^H diag(¯g^*) W_1 p|^2 is 0. Hence equations (12)-(15) are not an SCA of the stated sum-rate in the phase-shift variable, and Algorithm 1's phase update is solving a different problem. This is an internal algebraic inconsistency, independent of whether the physical channel model in (2)-(3) is accurate. The simulations in Section IV therefore do not support the headline comparison unless either q_{k,k} is corrected or the code is made available and shown to implement the true objective.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a fluid-element stacked intelligent metasurface (FSIM) architecture for multiuser downlink MISO networks. Each metasurface layer's meta-atoms can move within a two-dimensional fluidic region, adding positional degrees of freedom to the conventional phase-only SIM. The authors formulate a sum-rate maximization problem jointly over transmit beamforming, phase shifts, and meta-atom positions. They develop an alternating optimization (AO) algorithm: the beamforming and phase-shift subproblems are handled via semidefinite programming with successive convex approximation and penalty-based rank-one relaxation, while the position subproblem is addressed by a projected gradient method. Simulation results claim the proposed FSIM outperforms fixed-position SIMs, partially fluid structures, patch-based designs, and benchmark algorithms including GA, PSO, ZF, and MMSE.","tokens_in":11202,"tokens_out":7479,"duration_ms":74083,"significance":"If the claimed results were correct, the paper would make a useful contribution by showing that moving meta-atoms within SIM layers yields meaningful sum-rate gains beyond phase-only reconfiguration. The system model, with a layered near-field channel and Rician user links, is clearly presented, and the comparisons against external benchmarks (GA/PSO/ZF/MMSE) are appropriate. The paper ships no code or proofs, but the algorithmic structure is standard. However, the central phase-shift reformulation contains a serious algebraic error: the identity used to derive the phase-shift subproblem is wrong for non-diagonal A_l. This means the reported phase updates optimize a different objective than the stated sum-rate, so the simulation results do not, as they stand, support the headline claims. The position projection in (23) also does not demonstrably enforce the physical spacing constraints. These are load-bearing issues that require correction and re-evaluation of the numerical results.","major_comments":[{"comment":"The expansion |g_k p_k|^2 = |ψ_l^H q_{k,k}|^2 is invalid when A_l is non-diagonal. From (3), g_k p_k = \\bar g_k^H A_l diag(ψ_l) C_l W_1 p_k. Setting a=\\bar g_k, b=W_1 p_k, we have a^H A_l diag(ψ_l) C_l b = ψ_l^T[(C_l b) \\odot (A_l^T a^*)], not ψ_l^H diag(a^*) A_l C_l b. For L=1, N=2, a=[1,j]^T, W_1 p=[1,1]^T, ψ=[1,j]^T, the true |g p|^2 is 4 while |ψ^H q|^2 is 0. Therefore (12)-(15) and Algorithm 1, step 4, optimize a different objective from the sum-rate in (5). This error must be corrected and the simulations rerun before the performance claims can be assessed.","section":"III-B, Eq. (12) and the definition of q_{k,k}"},{"comment":"The projection Π_XFE in (23) is not a valid projection onto the feasible set RX defined by (1a)-(1e). The formula clamps each coordinate using a single neighboring coordinate, but the minimum-distance constraints (1a)-(1b) involve pairwise spacings across the entire layer. No argument is given that iterates satisfy all spacing constraints simultaneously, and no numerical check is reported. Since the optimized positions are used in every simulation, the physical feasibility of the reported configurations is not established.","section":"III-C, Eq. (23)"},{"comment":"Because of the phase-shift reformulation error, all simulated sum-rate values attributed to the proposed algorithm are not maximizing the stated objective. Consequently, the comparisons in Figs. 1-4 and the abstract's conclusion that the proposed FSIM 'significantly outperforms' prior designs are unsupported by the present manuscript. Additionally, the penalty-SCA loop has no convergence proof; the paper should report the final penalty values or rank metrics to show that the rank-one relaxation is tight for the reported results.","section":"Section IV and Algorithm 1"}],"minor_comments":[{"comment":"The distance expression contains a typo: it should be sqrt( d_layer^2 + (x^l_{n_h,n_v}-x^{l-1}_{n_h,n_v})^2 + (z^l_{n_h,n_v}-z^{l-1}_{n_h,n_v})^2 ). The minus sign is missing in the z-term.","section":"Eq. (2)"},{"comment":"The gradient ∇z_k^T(F^{(t)}) is written with a factor '· ln 2' in the numerator; it should be divided by ln 2, since z_k is a log2 function. This is likely a typographical error but should be corrected.","section":"III-A, Eq. (11) vicinity"},{"comment":"The definitions of A_l and C_l are ambiguous. The text should explicitly state the product indices and that S = A_l diag(ψ_l) C_l, otherwise the decomposition cannot be followed.","section":"III-B definitions"},{"comment":"The observation that ZF performs worse than the random baseline is surprising and should be explained; without justification, it weakens confidence in the benchmark setup.","section":"Fig. 1"},{"comment":"The label 'FSIM as TX (5λ)' is not defined in the text. Clarify what this benchmark represents.","section":"Fig. 3"},{"comment":"Constraint (1e) forces all meta-atoms on a layer to share the same y-coordinate, which is inconsistent with the preceding statement that they move within a two-dimensional panel. Clarify the geometry.","section":"II, Eq. (1e)"}],"recommendation":"major_revision","confidential_remarks":"The phase-shift identity error is the core issue: it invalidates the numerical results as reported. The paper could become publishable if the authors correct the algebra, rerun all simulations, and verify that the position projection satisfies the spacing constraints. The heavy self-citation pattern is not a reason for rejection but should be trimmed during revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe FSIM idea is a fresh combination: letting the meta-atoms of a stacked intelligent metasurface slide in-plane, and using a position-gradient update inside an AO loop. That is genuinely new relative to the SIM and fluid-element literature, and the paper does useful simulation exploration of layer count, thickness, and fluid region shape. The position-gradient equations (19)-(22) look plausible, and the convergence plots show the algorithm's behavior even if the comparisons are all self-benchmarked.\n\nThe problem is in Section III-B. The claimed identity |g_k p_k|^2 = |ψ_l^H q_{k,k}|^2 with q_{k,k} = diag(¯g^H) A_l C_l W_1 p_k is wrong. For non-diagonal A_l, the correct conjugate-form vector is q'_k = (C_l W_1 p_k)^* ⊙ (A_l^H ¯g_k), not that. The stress-test's counterexample (L=1, N=2, ¯g=[1,j]^T, W_1 p=[1,1]^T, ψ=[1,j]^T) gives true |g p|^2=4 while the paper's expression is 0. So equations (12)-(15) are not an SCA of the sum-rate; Algorithm 1's phase update solves a different problem. Unless the implementation in code used the correct expression, the simulation results in Section IV don't support the conclusions. No code is provided, so this cannot be checked.\n\nThere's also a smaller issue: the projection step (23) uses a max/min that doesn't obviously enforce the (1a)-(1b) minimum-distance constraints between all element pairs, so the position update may leave the feasible set.\n\nThe paper also suffers from missing hyperparameters, no error bars, and blanket \"significantly outperforms\" claims. The physical channel model (2)-(3) is unvalidated, but that's a typical assumption for this line of work.\n\nWho gets value: researchers considering movable-element SIMs will find the system model and the idea worth a look, but they should not trust the stated algorithm without a fix. The paper deserves peer review because the concept is interesting and the algebra is likely fixable; a good referee could push the authors to correct the derivation, release code, and re-run the simulations. If the corrected algorithm still shows gains, it would be a solid incremental contribution. As is, it's not publishable.\n\nRecommendation: send to peer review, but with a clear message that the phase-shift derivation needs to be corrected before any acceptance.","headline":"FSIM concept is fresh, but the phase-shift subproblem rests on an invalid identity that undermines the simulation results.","tokens_in":11676,"tokens_out":10240,"would_cite":false,"duration_ms":81692,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A stacked intelligent metasurface whose meta-atoms can slide within each layer outperforms fixed-element SIMs in multiuser downlink sum-rate.","keywords":["Stacked intelligent metasurface","Fluid element","Movable meta-atoms","Sum-rate maximization","Alternating optimization","Successive convex approximation","Multiuser MISO","Near-field beamforming"],"falsifier":"Measure or full-wave-simulate the transmission between two FSIM layers as meta-atoms are physically displaced; if the observed inter-layer coupling deviates from the free-space spherical-wave formula in equation (2), the claimed sum-rate advantage of fluid positioning cannot be realized.","tokens_in":10615,"feed_emoji":"📡","tokens_out":2342,"duration_ms":33018,"temperature":0.7,"pith_summary":"This paper tries to establish that adding positional freedom to a stacked intelligent metasurface (SIM) meaningfully improves multiuser downlink performance. The authors propose FSIM, where the meta-atoms in each layer move within a two-dimensional fluidic region, and jointly optimize element positions, base-station beamforming, and phase-shifts. They show through simulation that this joint position-plus-phase control yields higher sum-rate than conventional fixed-position SIMs, patch-based structures, partially movable designs, and existing flexible intelligent metasurfaces. The practical payoff, if true, is that a compact multi-layer surface with movable elements can better shape cascaded channels toward users and away from interference.","feed_headline":"Moving meta-atoms lift multiuser rates past fixed SIMs","feed_subtitle":"Fluid-positioned stacked metasurfaces beat fixed, patch, and partial-fluid designs in sum-rate simulations.","key_machinery":"The load-bearing object is the cascaded near-field channel model S = Φ_L W_L ... Φ_2 W_2 Φ_1, where each W_l is a matrix of free-space spherical-wave coupling coefficients between adjacent layers. The key trick is that moving a meta-atom changes only the geometric distance and direction in W_l, so position becomes an explicit optimization variable with closed-form derivatives. The solution machinery is alternating optimization: beamforming and phase-shift subproblems are relaxed to semidefinite programs solved by successive convex approximation plus rank-one penalty, while the position subproblem uses projected gradient ascent.","core_discovery":"The paper's central claim is that letting meta-atoms move within each layer of a stacked intelligent metasurface adds a new degree of freedom that phase-only control cannot match. By reformulating the cascaded channel S = Φ_L W_L ... Φ_2 W_2 Φ_1, where each W_l depends on element geometry, the authors show that element positions directly reshape the inter-layer coupling and therefore the end-to-end channel. An alternating optimization algorithm updates beamforming, phase-shifts, and positions, with position updates using the gradient of sum-rate with respect to element coordinates. Simulation results report that full-boundary fluidity achieves the highest rates among all compared fluid archi","pith_inferences":["Editorial inference: The result depends on the assumption that moving a meta-atom changes only the geometric distance in W_l while leaving element patterns and mutual coupling unchanged; if real fluid elements alter their electromagnetic response when displaced, the simulated gains could shrink.","Editorial inference: The position-gradient update could be extended to dynamic user locations, allowing the FSIM to track users by continuously adjusting element positions rather than only phases.","Editorial inference: The same position-optimization idea might apply to other layered reconfigurable surfaces where element placement, not just phase, can be tuned, potentially improving interference suppression in dense networks."],"forward_implications":["If FSIM works as modeled, full-boundary element mobility across all layers is the most valuable configuration, beating partial, patch-limited, single-axis, and discrete-port fluid designs.","A compact FSIM with small inter-layer thickness is preferable, because thicker layers weaken inter-layer coupling and reduce the achievable sum-rate.","The update order matters: optimizing positions before beamforming and phase-shifts (P-B-T) yields the highest sum-rate in simulations.","The proposed alternating optimization algorithm converges and outperforms metaheuristic methods, sub-schemes with disabled variables, and conventional beamforming benchmarks.","More layers and more elements increase sum-rate, especially in the near-field region, because they add finer-grained spatial degrees of freedom."],"fun_headline_variants":["Moving meta-atoms boost multiuser rates in stacked metasurfaces","Fluid meta-atom positions add new flexibility to stacked intelligent metasurfaces","Stacked metasurfaces with movable elements improve downlink sum-rate","Position-tunable meta-atoms outperform fixed designs in SIM networks","Letting meta-atoms drift in stacked metasurfaces raises multiuser rate"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The inter-layer coupling model treats each displaced meta-atom as a point-like scatterer whose only effect is a changed propagation distance; if actual movement alters element patterns, mutual coupling, or parasitic responses, the computed sum-rates overstate the real benefit.","fun_headline_variants_meta":{"raw":{"variants":["Moving meta-atoms boost multiuser rates in stacked metasurfaces","Fluid meta-atom positions add new flexibility to stacked intelligent metasurfaces","Stacked metasurfaces with movable elements improve downlink sum-rate","Position-tunable meta-atoms outperform fixed designs in SIM networks","Letting meta-atoms drift in stacked metasurfaces raises multiuser rate"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000728,"raw_usage":{"total_tokens":3149,"prompt_tokens":846,"completion_tokens":2303,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":590,"completion_tokens_details":{"reasoning_tokens":2220}},"tokens_in":590,"tokens_out":2303,"duration_ms":16180,"temperature":1.0,"reasoning_tokens":2220,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T22:40:27.693328+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure or full-wave-simulate the transmission between two FSIM layers as meta-atoms are physically displaced; if the observed inter-layer coupling deviates from the free-space spherical-wave formula in equation (2), the claimed sum-rate advantage of fluid positioning cannot be realized.","supporting_citations":[],"review_version":1}