{"id":"34519945-d70f-4fd7-b9d0-91bd411999dc","arxiv_id":"1908.04887","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"A Lyapunov-based two-scale algorithm for joint user scheduling, base station sleeping, and beamforming is proposed for grid-energy cost minimization, but its optimality proof is deferred.","lead":"This paper proposes a two-scale algorithm that schedules users, sleeps small-cell base stations, and computes beamforming to reduce grid energy costs in networks powered by solar energy and the smart grid. The key theoretical guarantee, asymptotic optimality, is stated but not proven in this preprint.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Scheduling rule (23) only minimizes the rate term (22), not the full drift-plus-penalty expression (19), so the asymptotic optimality bound (33) is unsupported.","rationale":"The reader's verdict of REJECT rests on the two explicitly omitted proofs: the activeness of relaxation (27) and the proof of Proposition 2. Those omissions are real and load-bearing, and the manuscript itself flags both. I agree that the paper cannot support its central theoretical claim as submitted. However, I would sharpen the criticism further: the scheduling rule (23) appears to be derived by minimizing only the rate-dependent term (22), not the full drift-plus-penalty upper bound (19). The scheduled indicator a determines which UEs appear in the beamforming problem, and through (8) and (13) it directly affects the grid-energy expenditure inside V E{G[k]}. It also determines whether the convex problem (32) is feasible at all. The paper provides no argument that the threshold rule (23) is the minimizer of the coupled objective, and the omission of Proposition 2's proof means this step is never checked. The numerical results show a plausible energy-delay tradeoff, but they do not compare against the claimed optimal value G* and therefore cannot substitute for the missing proof. This is not a disagreement with the broader research direction; it is an internal gap in the argument. Because the main claim is both unproven and connected to a specific non-sequitur in the derivation, the appropriate verdict for the submitted preprint is REJECT, or at most CONDITIONAL on supplying a complete proof with feasibility guarantees. I therefore recommend REJECT, matching the reader's verdict while identifying a more concrete weakness than the two deferred proofs alone.","tokens_in":10273,"tokens_out":5385,"duration_ms":64965,"concrete_test":"For a minimal network (e.g., M=2 ScBSs, one or two UEs each), instantiate the frame-level scheduling set from (23) and test feasibility of the convex problem (32) with constraints (12), (27), (28) over many random channel realizations using the paper's parameters. If any realization makes the (23)-selected set infeasible while a smaller subset is feasible, Algorithm 1 has no defined output and the asymptotic claim cannot hold. Separately, for feasible realizations, exhaustively evaluate the full per-frame objective (26) over all 2^N scheduling subsets; if any subset excluded by (23) yields a strictly lower objective, then (23) is not the minimizer of (26) and the Lyapunov optimality argument is invalid.","verdict_should_be":"REJECT","load_bearing_attack":"The central claim, Proposition 2's bound (33), is explicitly deferred, but the gap is not merely missing detail. The derivation of the optimal scheduled UE indicator a* in (23) minimizes only the rate-dependent term (22), ignoring that a also enters V E{G[k]} through P_SC (8) and the power balance (13), and that rates are coupled across UEs through beamforming and the proportional-rate constraints. Minimizing one summand of the RHS of (19) does not minimize the entire upper bound. Moreover, (23) schedules every UE with qU<qA regardless of channel quality or transmit power limits; the resulting per-slot problem (32) may be infeasible while a strict subset of UEs is feasible, and the paper gives no feasibility condition. Since Algorithm 1 executes (32) exactly for the set fixed by (23), an infeasible instance breaks the algorithm before any optimality gap can be stated. Both the activeness of (27) and the proof of Proposition 2 are explicitly postponed to an extended version, so the submitted manuscript contains no proof of its principal claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies joint beamforming, user scheduling, and small-cell base station sleeping in a two-scale (frame/slot) small-cell network powered by smart grid and renewable energy. It formulates a long-term grid-energy expenditure minimization problem with proportional-rate fairness constraints, then applies Lyapunov drift-plus-penalty optimization to decouple the frame-level scheduling decisions from slot-level beamforming. The main theoretical claim is Proposition 2, which states that the proposed algorithm asymptotically achieves the optimal grid-energy expenditure within an additive gap Ψ/V while keeping queues stable. Numerical results based on solar irradiation data are reported to show a delay-energy tradeoff.","tokens_in":10510,"tokens_out":3140,"duration_ms":31040,"significance":"If fully established, the proposed two-scale cross-layer framework would be a useful contribution to energy-aware small-cell network design, as it targets a practical timescale separation between scheduling and beamforming. The problem formulation and the use of Lyapunov optimization are reasonable, and the numerical study with realistic renewable-energy data is a positive feature. However, the paper's principal theoretical guarantees are not proven: Proposition 2 is explicitly deferred, and the tightness of the convex relaxation that underlies the per-slot optimization is also deferred. In addition, the scheduling rule appears to minimize only a component of the drift-plus-penalty expression, so the optimality claim is not supported. Because the core claims are unverified, the contribution as submitted is limited.","major_comments":[{"comment":"The main theoretical result, the asymptotic optimality bound in (33), is stated without proof; the text says 'the detailed proof is omitted and it will be provided in the extended version.' This is not a local appendix omission but the central claim of the paper. The bound (33) is the basis for saying the proposed algorithm approaches the optimal grid-energy expenditure, so the manuscript does not currently provide the stated guarantee.","section":"Section III, Proposition 2 (Eq. 33)"},{"comment":"The proof that the relaxed problem (31) has the same optimal value as the original problem (26) is also deferred: 'The detailed proof of the activeness of (27) will be provided in the extended version.' The relaxation replaces the non-convex proportional-rate constraints (11) with (27)-(28). If the relaxation is not tight, the algorithm solves a different problem, and the bound (33) does not apply to the original problem (15). This is a load-bearing step that cannot be left as a promise.","section":"Section III-B, after Eq. (31)"},{"comment":"The scheduling rule a* in (23) minimizes only the rate term (22), which is one summand in the upper bound (19) of the drift-plus-penalty function. It does not account for the V E{G[k]} term, which depends on the scheduled user set through P_SC in (8) and the power balance (13), nor for the coupling between user rates and beamforming through the proportional-rate constraints. Minimizing one component of the RHS of (19) does not minimize the full upper bound, so the optimality of (23) is not established. Moreover, (23) schedules all users with q^U < q^A without checking channel quality or transmit power limits, so the per-slot problem (32) may become infeasible; the paper provides no feasibility condition or recovery mechanism.","section":"Section III-A, Eq. (23)"},{"comment":"The numerical results cannot compensate for the missing theoretical guarantees because the algorithm's decisions, particularly the scheduling rule (23), are not shown to minimize the drift-plus-penalty expression. The paper reports performance trends but does not test the claimed bound (33) against an optimal or benchmark solution. Without the missing proofs, the numerical section demonstrates only that the proposed heuristic behaves sensibly in the simulated scenario.","section":"Section IV"}],"minor_comments":[{"comment":"Equation (37) appears to have a typo: the left-hand side is written as ∆X + E{G[k]}, but the drift-plus-penalty function defined in (18) has the penalty weighted by V, so the LHS should likely be ∆X + V E{G[k]}. Also, the second summation on the right-hand side appears to have a duplicated 'N_m ∑ n=1' index.","section":"Appendix, Eq. (37)"},{"comment":"The rate expression r_{m,n}(t_k) = log(1 + SINR_{m,n}(t_k)) is used without stating the base of the logarithm; the numerical section uses nats/slot/Hz, so the base should be stated earlier to avoid ambiguity.","section":"Section II-B"},{"comment":"The paper repeatedly refers to an 'extended version' for missing proofs and for the non-ideal energy trading case (Remark 1), but no reference or indication is given that such a version exists. The manuscript should either include the proofs or clearly state that the results are conjectural.","section":"General"}],"recommendation":"reject","confidential_remarks":"The manuscript is a conference-length paper with two of its most important proofs (Proposition 2 and the tightness of the relaxation) explicitly deferred to an 'extended version.' Even setting aside the deferred proofs, the scheduling rule in (23) appears to minimize only a partial expression rather than the full drift-plus-penalty bound, which raises a correctness concern that would not be resolved by merely adding the missing proofs. In my view, the paper cannot be accepted in its current form, and the issues are substantial enough to recommend rejection rather than a request for minor revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Zev, quick take on 1908.04887. The central theorem—Proposition 2's asymptotic optimality bound (33)—is explicitly deferred to an extended version, and the proof of the relaxation tightness (activeness of (27)) is likewise postponed. That alone makes the paper incomplete as a submission. The deeper problem is that the scheduling rule (23) does not actually minimize the full drift-plus-penalty upper bound (19). It minimizes only the rate-dependent term (22), while the grid-energy expenditure term V E{G[k]} also depends on the scheduling decisions through the power consumption in (8) and the power balance (13). So the claimed optimality gap is unsupported on two levels.\n\nWhat is genuinely new: the joint design of user scheduling, ScBS sleeping, and beamforming under proportional-rate constraints in a two-scale, smart-grid-powered small-cell network. Each component is known (Lyapunov drift-plus-penalty, convex relaxation of proportional rates), but the specific combination and the frame/slot decomposition are not in the prior work I know. The system model is carefully laid out, and the Lyapunov drift bound in Proposition 1 is standard and correctly derived. The numerical results show the expected tradeoff between delay and energy expenditure, but there are only two figures and no comparison to any baseline algorithm, so they are suggestive rather than convincing.\n\nThere is also a practical feasibility issue that the stress-test note flags: rule (23) schedules every UE with qU<qA regardless of channel quality or transmit power limits. The per-slot problem (32) may then be infeasible, and Algorithm 1 has no feasibility check or fallback. This is not a minor gap; it means the algorithm can fail on instances where a strict subset of UEs would be feasible.\n\nTo be fair: the paper is not a sham. It is clearly written, the authors are explicit about the missing proofs, and the citation pattern is fine—the self-citations are for standard results. But the load-bearing arguments are absent, so the scientific contribution as submitted is incomplete.\n\nFor peer review: I would not accept this as is. But it deserves a serious technical referee rather than a desk reject, because the problem is real and the flaws are subtle enough that an expert can pinpoint them constructively. The authors need to provide the missing proofs and fix the scheduling rule before the claims can be taken seriously.","headline":"The paper's asymptotic optimality claim is unproven and the scheduling rule looks like it ignores part of the objective, but the two-scale problem is well posed and the writing is honest.","tokens_in":10977,"tokens_out":2743,"would_cite":false,"duration_ms":27950,"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":"A two-scale scheduling and beamforming algorithm asymptotically achieves the minimum grid-energy expenditure in renewable-powered small-cell networks.","keywords":["small-cell networks","beamforming","user scheduling","base station sleeping","smart grid","Lyapunov optimization","energy harvesting","two-scale resource allocation"],"falsifier":"Take a small two-user, one-base-station instance with fixed channels and solve both the original problem (26) with proportional-rate constraints (11) and the relaxed problem (31) with constraints (27)-(28). If the optimal values differ, or if at the optimum of (31) some constraint in (27) is slack, the claimed tightness fails. A second check is to simulate the full algorithm over many frames and see whether the sample-mean grid-energy expenditure actually lies within $\\Psi/V$ of the optimum for large $K$.","tokens_in":10089,"feed_emoji":"⚡","tokens_out":5723,"duration_ms":55252,"temperature":0.7,"pith_summary":"This paper claims that a small-cell network running on renewable energy plus the smart grid can minimize its long-term electricity bill by acting on two time scales: once per frame it decides which users to serve and which base stations to sleep, and once per slot it computes the beamforming vectors. The authors prove, via Lyapunov optimization, that the resulting grid-energy expenditure is bounded between the true optimum and the optimum plus $\\Psi/V$, a gap that shrinks as the control parameter $V$ grows, while all user queues remain stable. If the claim holds, operators gain a principled knob for trading end-to-end delay against energy cost without reconfiguring schedules at every slot. The tradeoff is that two load-bearing proofs---the exactness of the convex relaxation and the optimality bound---are deferred to an extended version.","feed_headline":"Two-scale algorithm approaches optimal grid-energy cost","feed_subtitle":"Frame-level user scheduling and sleeping plus slot-level beamforming keep electricity bills near the minimum.","key_machinery":"The machinery is the Lyapunov drift-plus-penalty framework applied to a two-scale system, supplemented by a convex reformulation of the non-convex proportional-rate constraints. The drift bound in Proposition 1 separates the scheduling decisions, which enter through the queue-backlog difference $q^U_{m,n}[k]-q^A_{m,n}[k]$, from the beamforming and energy-trading decisions, so the integer indicators can be optimized in closed form at frame granularity. The remaining slot-level problem is made convex by introducing the auxiliary variable $\\varphi(t_k)$ and the constraints (27)-(28), and the paper argues the relaxation is tight because the optimal beamformers satisfy (27) with equality; a one-dimensional search over $\\varphi(t_k)$ completes the per-slot solution.","core_discovery":"The central claim is that the long-term grid-energy expenditure minimization problem, which couples integer user-scheduling indicators with continuous beamforming vectors, can be solved near-optimally by a two-scale greedy algorithm. Scheduling follows a simple backlog rule: a user is scheduled only when its access-queue backlog is positive and smaller than its processing-queue backlog, and a base station sleeps when none of its users are scheduled. Beamforming and grid trading are then obtained per slot from a convex optimization problem, parameterized by a one-dimensional variable $\\varphi(t_k)$, that replaces the proportional-rate constraints with the convex constraints (27)-(28). The paper asserts that the optimal beamformers make these constraints active, so the relaxed problem has the same value as the original, and that the Lyapunov drift-plus-penalty method yields the asymptotic bound $G^* \\le \\frac{1}{K}\\sum_{k=0}^{K-1}\\mathbb{E}\\{G[k]\\} \\le \\frac{\\Psi}{V}+G^*$ together with queue stability. Both proofs are stated but deferred.","pith_inferences":["A natural extension would be to prove the activeness of (27) directly: if the rate function $r_{m,n}=\\psi_{m,n}\\varphi(t_k)$ is increasing in $\\varphi$, the optimal solution should push $\\varphi$ to its upper bound, which is exactly the regime where (27) binds; supplying this argument would close the main gap.","The same two-scale decomposition could be applied to other cross-layer problems, such as joint caching and beamforming, where integer content-placement decisions are made at frame level and precoding at slot level.","One testable prediction is that the scheduling threshold (23) remains near-optimal even when renewable-energy arrivals are correlated across frames, since the drift-plus-penalty analysis only needs the expectation over random sources, not temporal independence.","If the relaxation is not always tight, a practical fix would be to add a small penalty that encourages the constraints (27) toward equality, or to use sequential convex programming; numerical results would reveal whether such a fix changes the cost."],"forward_implications":["An operator can tune $V$ to move continuously between low delay and low grid-energy expenditure, with the suboptimality gap bounded by $\\Psi/V$.","Scheduling and base-station sleeping need only be updated once per frame, so the overhead and reliability concerns of frequent switching are avoided.","The backlog-based scheduling rule (23) automatically provides proportional fairness, since users with larger access-queue backlogs get proportionally higher service rates.","Renewable-energy volatility is absorbed at frame granularity while beamforming tracks fast channel variations at slot granularity."],"supporting_citations":[{"why":"Supplies the power-consumption model and the single-cell long-term grid-energy expenditure formulation that this work extends.","marker":"[10]"},{"why":"Demonstrates online beamforming with asymptotic optimality in a smart-grid setting, establishing the baseline the multicell result must beat.","marker":"[11]"},{"why":"Introduces the two-scale stochastic control idea that this paper adapts to scheduling and sleeping.","marker":"[14]"},{"why":"Provides the Lyapunov drift-plus-penalty technique used to decouple scheduling from beamforming.","marker":"[18]"},{"why":"Is the source of the argument that optimal beamformers make the relaxed constraints active and of the one-dimensional search method.","marker":"[19]"},{"why":"Is the convex solver used to implement the per-slot beamforming optimization.","marker":"[20]"}],"fun_headline_variants":["Two-scale scheduling and beamforming approach grid-energy optimum","Smart-grid small cells near optimal energy via two-scale design","Lyapunov-based two-scale method nears optimal grid energy","Cross-layer design approaches optimal grid-energy cost in small cells"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument rests on assuming the relaxed per-slot problem (31), which replaces proportional-rate constraints by the convex set (27)-(28), has exactly the same optimal value as the original problem (26); if the constraints (27) are ever inactive at the optimum, the asymptotic grid-energy bound applies only to the relaxed problem, not to the true one.","fun_headline_variants_meta":{"raw":{"variants":["Two-scale scheduling and beamforming approach grid-energy optimum","Smart-grid small cells near optimal energy via two-scale design","Lyapunov-based two-scale method nears optimal grid energy","Cross-layer design approaches optimal grid-energy cost in small cells"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000587,"raw_usage":{"total_tokens":2737,"prompt_tokens":907,"completion_tokens":1830,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":523,"completion_tokens_details":{"reasoning_tokens":1763}},"tokens_in":523,"tokens_out":1830,"duration_ms":18338,"temperature":1.0,"reasoning_tokens":1763,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:28:49.719336+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a small two-user, one-base-station instance with fixed channels and solve both the original problem (26) with proportional-rate constraints (11) and the relaxed problem (31) with constraints (27)-(28). If the optimal values differ, or if at the optimum of (31) some constraint in (27) is slack, the claimed tightness fails. A second check is to simulate the full algorithm over many frames and see whether the sample-mean grid-energy expenditure actually lies within $\\Psi/V$ of the optimum for large $K$.","supporting_citations":[{"cited_title":"Dynamic cross- layer beamforming in hybrid powered communication systems with harvest-use-trade strategy,","cited_arxiv_id":null,"evidence_quote":"Supplies the power-consumption model and the single-cell long-term grid-energy expenditure formulation that this work extends."},{"cited_title":"Dynamic energy management for smart-grid-powered coordinated multipoint systems,","cited_arxiv_id":null,"evidence_quote":"Demonstrates online beamforming with asymptotic optimality in a smart-grid setting, establishing the baseline the multicell result must beat."},{"cited_title":"Two-scale stochastic control for integrated multipoint communication systems with renewables,","cited_arxiv_id":null,"evidence_quote":"Introduces the two-scale stochastic control idea that this paper adapts to scheduling and sleeping."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the Lyapunov drift-plus-penalty technique used to decouple scheduling from beamforming."},{"cited_title":"Robust energy efficient beamforming in MISOME-SWIPT systems with proportional secrecy rate,","cited_arxiv_id":null,"evidence_quote":"Is the source of the argument that optimal beamformers make the relaxed constraints active and of the one-dimensional search method."},{"cited_title":"CVX: Matlab software for disciplined convex programming, version 2.1,","cited_arxiv_id":null,"evidence_quote":"Is the convex solver used to implement the per-slot beamforming optimization."}],"review_version":1}