{"id":"346a3d27-8750-4146-a38a-104ac82a5ed7","arxiv_id":"1908.03965","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"The paper extends IRS beamforming optimization to broadcast, multicast, max-min fairness, and multi-IRS/multi-antenna settings using standard SDR alternating optimization, without numerical validation.","lead":"This paper formulates a broad set of optimization problems for wireless networks aided by intelligent reflecting surfaces (IRSs), covering power control, quality-of-service, and max-min fairness across unicast, broadcast, and multicast traffic, with extensions to multi-antenna users and multiple IRSs. It proposes alternating-optimization algorithms based on semidefinite relaxation, but provides no simulations, convergence proofs, or approximation bounds.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The alternating SDR-plus-randomization loop has no feasibility or monotonicity guarantee, and Algorithms 1–2 explicitly break when no randomized phase vector satisfies the SINR constraints; the 'efficient algorithms' claim is therefore unsupported.","rationale":"The reader's weakest assumption and my concern coincide: the algorithms' validity rests entirely on the SDR-plus-randomization alternation, and the paper supplies no evidence for it. I see no more fundamental flaw in the problem formulations or SDR derivations; the transformations (e.g., Eq. (30)–(39)) are standard and mostly correct. The most serious gap is not a mathematical error but an unsupported central claim: the abstract promises efficient algorithms, yet the algorithms can terminate at an infeasible point and no convergence or approximation analysis is given. The paper's own conclusion explicitly defers approximation bounds and experiments, so this is not an external expectation but an acknowledged absence. A possible secondary issue is that the max-min W-subproblem (P4b) is bilinear in t and X and would require bisection over t, which the text does not specify; however, this is fixable and less load-bearing than the feasibility/convergence gap. The formulations and taxonomy have value, and conditional acceptance with a required validation or explicit heuristic labeling is appropriate. I therefore agree with the reader's CONDITIONAL verdict and see no reason to change it.","tokens_in":32488,"tokens_out":6118,"duration_ms":65951,"concrete_test":"Implement Algorithm 1 for a small multicast QoS instance, e.g., M=2 BS antennas, N=8 IRS elements, K=3 single-antenna users in one group, with random i.i.d. Rayleigh channels and unit-modulus IRS phases. Use Z=10^4 Gaussian randomization samples per iteration, run 100 channel realizations with 10 random initializations each, and record (i) how often the loop exits at line 19 with no feasible φ, and (ii) the sequence f^(r) at every iteration. If line 19 is reached for any instance whose SDR relaxation (P1d) is feasible, or if f^(r) fails to be nonincreasing on feasible iterations, then the central claim that Algorithm 1 efficiently solves P1 is disproved; even if not, a convergence or approximation argument is still missing.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that Algorithms 1–4 efficiently solve the formulated nonconvex IRS beamforming/phase-shift problems. What must be true is that each alternating step, after dropping rank constraints and applying Gaussian randomization, returns a feasible point and that the objective progresses monotonically. The paper establishes neither. In Algorithm 1 (lines 19–20) and Algorithm 2 (lines 18–19), the loop simply breaks when no randomized φ satisfies the SINR inequality (34)/(47); there is no proof that such a φ exists even when the SDR (P1d)/(P4e) is feasible, no approximation bound for the randomization step, and no monotonicity or convergence argument for the outer alternation. The same pattern recurs in Algorithms 3 and 4. Moreover, the W-subproblem for multicast is NP-hard in general, so invoking 'methods of [12]' imports a heuristic without a performance guarantee. The paper's own conclusion lists 'proving approximation bounds' and 'conducting extensive experiments' as future work, confirming that the algorithmic claim is not yet supported. Hence the abstract's assertion of 'efficient algorithms' is the load-bearing point that remains unsecured.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript considers an IRS-aided downlink system with a multi-antenna base station and mobile users, and formulates total-power minimization under SINR constraints (power control under QoS) and max-min fair SINR maximization for unicast, broadcast, and multicast traffic. It extends these formulations to multi-antenna users, multiple IRSs, and their combination. The solution approach in Section IV is alternating optimization: for fixed phase shifts, the beamforming subproblem is relaxed by dropping rank-one constraints and post-processed by Gaussian randomization; for fixed beamformers, the phase-shift subproblem is transformed into a semidefinite program in V = vv^H via the auxiliary variable v = t[φ;1], then relaxed and randomized. Algorithms 1-4 instantiate this scheme. The paper claims these are efficient algorithms, with the stated novelty that prior IRS work considered only unicast power control and no IRS work considered max-min fair QoS.","tokens_in":32795,"tokens_out":8052,"duration_ms":82072,"significance":"If the algorithmic claims were supported, the paper would provide a useful unified treatment of IRS beamforming across a broad set of system models; the problem formulations in Tables I-IV are clearly organized and the SDR steps from SINR constraints to trace constraints are mostly correct. The paper also honestly lists approximation bounds and extensive experiments as future work. The central limitation is that the proposed algorithms are heuristics with no convergence, feasibility, or approximation guarantees and no simulations, so the principal claim of 'efficient algorithms' is not yet established. The contribution is therefore best viewed as a formulation-plus-heuristic paper pending substantial additional support.","major_comments":[{"comment":"The alternating algorithm stops without returning a feasible point when none of the Z randomized phase vectors satisfies the SINR condition (34) (or (47), (65), (80)). No argument is given that such a φ exists whenever the SDR (P1d) is feasible, nor that the objective f^(r) is monotone, nor that the iterates converge; the stopping criterion in line 7 is also not quantified. Since feasible output is the central promise of solving Problems (P1)-(P6), this gap undermines the 'efficient algorithms' claim and must be closed by either a proof under stated assumptions or a clearly labeled heuristic with numerical evidence.","section":"Section IV-A, Algorithm 1 lines 19-20 (and Algorithms 2-4)"},{"comment":"The statement that 'appropriate post-processing such as Gaussian randomization ... is applied to convert the candidate solution into a solution which satisfies the rank constraint' is an assertion, not a consequence of the cited references [13]-[14]. For SINR-constrained feasibility and for the multicast W-subproblem, whose hardness the paper itself leaves as future work, Gaussian randomization does not generally produce a feasible rank-one point from a feasible SDR solution. The authors should either prove a feasibility or approximation result for these specific problems or explicitly present the methods as heuristics and support them with simulations.","section":"Section IV-A, after Eq. (39) and after Eq. (48)"},{"comment":"In the max-min fair case, t is obtained from the SDR (P4b) before rank-one recovery. After Gaussian randomization on W, the randomized beamformers are not shown to achieve the same t, yet the same t is used in the phase-shift feasibility check (47)/(80). If the randomized W achieves a smaller achievable t, the phase-shift step can fail and the loop breaks at line 19 even when the original problem is feasible. The algorithm should recompute t from the feasible W (e.g., via bisection) or otherwise ensure that the t used in the φ-update is achievable.","section":"Section IV-B, Algorithm 2 lines 4-19; Algorithm 4"},{"comment":"The abstract and the contribution list say 'efficient algorithms' are proposed, but no convergence rate, per-iteration complexity, or numerical demonstration is provided. The conclusion explicitly defers 'proving approximation bounds' and 'conducting extensive experiments' to future work. At minimum, the claims should be scaled back to 'heuristic algorithms' unless complexity and empirical behavior are added.","section":"Section VI (Conclusion)"}],"minor_comments":[{"comment":"The sentence 'Replacing k by j in Eq. (21)' should yield |h_i^H(Φ)w_j|^2 = trace(X_j H_i(Φ)), not trace(X_k H_j(Φ)); the resulting inequality (23) is correct, so this appears to be a typo.","section":"Section IV-A, after Eq. (21)"},{"comment":"The phrase 'we express the the constraint' contains a duplicated 'the'; similarly, Algorithm 1 line 5 and Algorithm 3 line 5 use 'object function' where 'objective function' is intended.","section":"Section IV-A, text near Eq. (23)"},{"comment":"The stopping rule 'the relative difference ... is small' is not a testable condition; a specific tolerance or a normalized convergence criterion should be given.","section":"Algorithm 1 line 7 and Algorithm 2 line 6"},{"comment":"The displayed SINR constraint contains a stray vertical bar and an unmatched absolute-value bracket; the denominator should be a sum of nonnegative terms, not |w_j^H H_k(Φ)w_j + σ_k^2|.","section":"Table II, row (P2-MA)"},{"comment":"The paper states that the SDPs can be solved efficiently via [15], but it does not give the dimensions or per-iteration complexity of the SDPs solved in Algorithms 1-4; adding these would help substantiate the 'efficient' claim.","section":"Section IV-A and IV-B"}],"recommendation":"major_revision","confidential_remarks":"The paper appears to be an early preprint ('Under submission' on every page). The main risk is the gap between the claimed 'efficient algorithms' and the actual heuristic content; if the authors add a convergence or approximation analysis and numerical experiments, and temper the claims accordingly, the paper could be publishable, but in its current form the algorithmic contribution is not sufficiently supported."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a useful problem-formulation paper built on standard SDR alternating optimization, and it is honest about what is missing. The new material is the problem taxonomy—max-min fair QoS for IRS-aided downlinks, plus broadcast/multicast and multi-antenna/multi-IRS extensions—which I don't think appears in the cited IRS literature. The SDR derivations from SINR constraints to trace constraints check out (modulo a typo after Eq. (21)), and the tables are a convenient reference.\n\nThe soft spot is the algorithm claim. Algorithms 1–4 are alternating optimization: given W solve a relaxed SDP for the phase vector, randomize to get a rank-one candidate, and stop if no random draw satisfies the SINR inequalities. There is no convergence proof, no approximation bound, and no guarantee that the randomization step returns a feasible point even when the relaxation is feasible. The paper's own conclusion lists proving approximation bounds and running experiments as future work, which is the right thing to say but also confirms the 'efficient algorithms' in the abstract is doing more work than the analysis supports. The multicast W-subproblem is NP-hard in general, so 'apply methods of [12]' imports a heuristic without performance guarantees. For a formulation paper that's not disqualifying, but the claims should be scaled to what is shown.\n\nI don't think the reader's skepticism overshoots. The stress-test note lands. That said, the paper is not circular and does not fit parameters to data; it leans on [4] and [12] for the hard subroutines, and the self-citation to the author's survey [27] is harmless. The central formulations are original enough to be worth having in the literature.\n\nWho is this for? Someone working on IRS beamforming who wants the problem definitions and the standard SDR baseline before trying something stronger. It deserves a serious referee, but the referee should push for either simulations, convergence guarantees, or a title/abstract that says 'formulations and SDR heuristic' instead of 'efficient algorithms.' I'd accept for peer review.","headline":"A genuinely useful IRS problem-formulation paper whose 'efficient algorithms' claim outruns the analysis; referee it, but push for validation or softer claims.","tokens_in":33257,"tokens_out":2359,"would_cite":true,"duration_ms":22037,"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":"One alternating recipe solves IRS power and fairness beamforming.","keywords":["intelligent reflecting surfaces","6G wireless networks","power control under quality of service","max-min fair beamforming","semidefinite relaxation","alternating optimization","multicast beamforming","phase shift design"],"falsifier":"Take a small system small enough for brute force—say one base station with two antennas, one IRS with two elements whose phase shifts are discrete, and two single-antenna users—and enumerate every phase configuration to compute the true least transmit power meeting the SINR targets; then run the paper's Algorithm 1 from several random initial phases. If the algorithm's returned power is consistently higher than the brute-force optimum by more than the randomization tolerance, or if it stops with no feasible phase vector when one exists, the central claim of an efficient near-optimal solution is contradicted.","tokens_in":32292,"feed_emoji":"📶","tokens_out":7670,"duration_ms":69510,"temperature":0.7,"pith_summary":"Intelligent reflecting surfaces (IRSs) are thin arrays of reconfigurable elements that reflect base-station signals so that they arrive stronger at mobile users, and they are expected to be a part of 6G networks. This paper claims that a single optimization template—alternate between designing the base station's transmit beamformers and the IRS's phase-shift matrix, relaxing the hard rank-one constraints, then recovering feasible solutions by Gaussian randomization—can handle two core problems: minimizing transmit power while meeting per-user signal-to-interference-plus-noise ratio targets, and maximizing the minimum fair SINR. The template is shown to cover unicast, broadcast, and multicast traffic, and to extend to mobile users with several antennas and to systems with several IRSs. Existing IRS studies had addressed only the unicast power-control case, so the paper's contribution is a unified formulation and a concrete algorithmic recipe for the broader design space. If the algorithms deliver what the formulations promise, network designers can use one toolbox to configure IRS-aided downlinks under power, fairness, and traffic-pattern constraints.","feed_headline":"One alternating recipe solves IRS power and fairness beamforming","feed_subtitle":"The same two convex relaxations handle unicast, broadcast, and multicast downlinks","key_machinery":"The load-bearing object is the lifted phase-shift matrix V := vv^H, built from v = t[φ; 1] where φ collects the IRS reflection coefficients β_n $e^{{jθ_n}}$ and t is a unit-modulus auxiliary variable. This lifting turns the nonconvex phase-shift feasibility constraint—a product of phase coefficients inside SINR terms—into a linear constraint on V plus a rank-one constraint, which is then relaxed by dropping rank-one and solved as a semidefinite program. The same lifting is reused for the beamformer side with X_k := w_k w_k^H, so that every SINR inequality becomes a trace inequality. Alternating optimization cycles between the W-side SDP and the Φ-side SDP, with Gaussian randomization post-processing to produce rank-one candidates.","core_discovery":"The paper's central claim is that joint design of the BS beamforming matrix W and the IRS reflection coefficient matrix Φ can be reduced, for all six problem combinations (power control or max-min fairness, times unicast, broadcast, or multicast), to two alternating semidefinite programs. On the beamforming side, the rank-one beamformers X_k = w_k w_k^H are relaxed to positive semidefinite matrices; on the phase side, the phase vector φ is lifted through v = t[φ; 1] into V = vv^H, whose diagonal entries encode the unit-modulus or amplitude constraints of the IRS. Dropping rank-one constraints makes both subproblems convex, and a final Gaussian-randomization step is used to project the relaxed solutions back to rank-one beamformers and phase shifts. The paper asserts that this recipe works for multi-antenna mobile users, where the channel term becomes a sum over receive antennas, and for multiple IRSs, where all phase vectors are stacked into one lifted variable, and that the max-min fairness problem is solved by introducing the auxiliary fairness level t and using the same alternating machinery. No optimality or convergence proof is given; the contribution is the unified problem formulation and the efficient SDR-based algorithms.","pith_inferences":["A natural, testable extension is to replace the hard stopping rule in Algorithms 1 and 2, which breaks when no randomized phase vector satisfies the SINR constraints, with an adaptive search that scales the number of Gaussian samples Z with the channel dimensions; the paper gives no guidance on how Z should scale, so success probability versus Z and N is an open empirical question.","Because the max-min fair problem contains the power-control problem as a limiting case through bisection on t, any approximation guarantee proven for one would transfer to the other; the paper leaves this transfer implicit.","The same lifted V formulation should carry over to related IRS objectives such as weighted sum-rate or energy efficiency, since those objectives also enter the SDP only through the same trace inequalities; the paper lists weighted sum-rate as future work without developing the connection.","If channel estimation errors are introduced, the alternating SDR template would likely need a robust reformulation; the paper states this as a future direction, so a concrete next step is to replace the exact channel matrices by their estimated versions and measure SINR violation rates."],"forward_implications":["The max-min fair QoS problems are solved by the same alternating SDR machinery as power control: introducing the fairness level t turns them into feasibility checks with scaled SINR targets t·γ_i, so no new algorithmic structure is needed.","Multi-antenna mobile users reduce to replacing each channel outer product |h_i(Φ)^H w_k|^2 with w_k^H H_i(Φ) w_k, where H_i(Φ) sums over the user's antennas, and all subsequent lifting steps carry through unchanged.","Multiple IRSs are handled by stacking the per-surface phase vectors into a single φ and lifting to a larger V, so the number of IRSs changes only the dimension of the semidefinite programs.","Discrete phase-shift constraints are accommodated at the final mapping step after Gaussian randomization, because the SDP relaxation itself only fixes the diagonal magnitudes of V.","If the SDR solutions are near-optimal, the paper's formulations give a common framework for configuring IRS-aided 6G downlinks under power limits, QoS targets, and fairness criteria."],"supporting_citations":[{"why":"Establishes the unicast power-control-under-QoS baseline with continuous phase shifts that this paper generalizes to broadcast, multicast, and fairness.","marker":"[3]"},{"why":"Supplies the phase-shift SDP formulation, including the SINR-residual variant, that the paper reuses for the Φ-side subproblem.","marker":"[4]"},{"why":"Adds discrete phase-shift constraints and the idea of mapping continuous solutions to discrete values, which the algorithms adopt in their final step.","marker":"[5]"},{"why":"Provides the SDR method and max-min fairness scaling for multicast transmit beamforming that the paper applies to the W-side subproblem.","marker":"[12]"},{"why":"Supplies Gaussian randomization as the post-processing that converts rank-relaxed SDP solutions into feasible rank-one beamformers and phase vectors.","marker":"[13]"},{"why":"Provides the multicast beamforming SDR-with-randomization approach whose randA, randB, and randC variants the paper cites for post-processing.","marker":"[14]"},{"why":"Justifies semidefinite relaxation of quadratic optimization problems, the theoretical basis for solving the relaxed subproblems.","marker":"[15]"}],"fun_headline_variants":["Two SDR relaxations unify IRS beamforming across traffic types","Alt SDR solves IRS power and fairness for all downlink modes","One alternating recipe handles six IRS optimization problems","Two convex relaxations power IRS beamforming for all traffic","Unified SDR algorithm: power, fairness, unicast to multicast"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that after dropping the rank-one constraints, the semidefinite relaxation plus Gaussian randomization reliably returns a feasible and near-optimal beamformer and phase shift at each alternating step, even though the paper gives no proof that such a solution exists or that the alternating objective improves monotonically.","fun_headline_variants_meta":{"raw":{"variants":["Two SDR relaxations unify IRS beamforming across traffic types","Alt SDR solves IRS power and fairness for all downlink modes","One alternating recipe handles six IRS optimization problems","Two convex relaxations power IRS beamforming for all traffic","Unified SDR algorithm: power, fairness, unicast to multicast"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000595,"raw_usage":{"total_tokens":2819,"prompt_tokens":1010,"completion_tokens":1809,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":626,"completion_tokens_details":{"reasoning_tokens":1738}},"tokens_in":626,"tokens_out":1809,"duration_ms":13308,"temperature":1.0,"reasoning_tokens":1738,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:56:15.664294+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a small system small enough for brute force—say one base station with two antennas, one IRS with two elements whose phase shifts are discrete, and two single-antenna users—and enumerate every phase configuration to compute the true least transmit power meeting the SINR targets; then run the paper's Algorithm 1 from several random initial phases. If the algorithm's returned power is consistently higher than the brute-force optimum by more than the randomization tolerance, or if it stops with no feasible phase vector when one exists, the central claim of an efficient near-optimal solution is contradicted.","supporting_citations":[{"cited_title":"Intelligent reflecting surface enhan ced wireless network: Joint active and passive beamforming design,","cited_arxiv_id":null,"evidence_quote":"Establishes the unicast power-control-under-QoS baseline with continuous phase shifts that this paper generalizes to broadcast, multicast, and fairness."},{"cited_title":"Beamforming Optimization for Wireless Network Aided by Intelligent Reflecting Surface with Discrete Phase Shifts","cited_arxiv_id":"1906.03165","evidence_quote":"Adds discrete phase-shift constraints and the idea of mapping continuous solutions to discrete values, which the algorithms adopt in their final step."},{"cited_title":"Quali ty of service and max-min fair transmit beamforming to multiple cochanne l multicast groups,","cited_arxiv_id":null,"evidence_quote":"Provides the SDR method and max-min fairness scaling for multicast transmit beamforming that the paper applies to the W-side subproblem."},{"cited_title":"On approximating compl ex quadratic optimization problems via semidefinite programming relaxa tions,","cited_arxiv_id":null,"evidence_quote":"Supplies Gaussian randomization as the post-processing that converts rank-relaxed SDP solutions into feasible rank-one beamformers and phase vectors."},{"cited_title":"Transmit beamforming for physical-layer multicasting,","cited_arxiv_id":null,"evidence_quote":"Provides the multicast beamforming SDR-with-randomization approach whose randA, randB, and randC variants the paper cites for post-processing."},{"cited_title":"Semidefin ite relaxation of quadratic optimization problems,","cited_arxiv_id":null,"evidence_quote":"Justifies semidefinite relaxation of quadratic optimization problems, the theoretical basis for solving the relaxed subproblems."}],"review_version":1}