{"id":"18149965-3047-4aa3-ab28-a6219f129915","arxiv_id":"1908.04689","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A joint completion-time and energy minimization framework for multi-group NOMA-based mobile-edge computing, with iterative, bisection, and convex-equivalent algorithms.","lead":"This paper designs algorithms for allocating time, power, and computing resources in a mobile-edge computing network where groups of users upload tasks via non-orthogonal multiple access. It jointly minimizes completion time and energy use, and reports that the proposed NOMA approach beats conventional TDMA and FDMA schemes in simulations.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 1's equivalence is unsupported when power caps bind: the proof requires increasing a stronger user's power, and a two-user example shows Problem (11) can then be feasible while Problem (10) is not.","rationale":"The reader's weakest assumption identifies exactly the load-bearing step. Re-deriving Appendix A shows the issue is not merely a missing technical detail: when the stronger user's power cap binds, the proposed perturbation cannot be performed, and the inequality relaxation can be feasible where the original equality problem is not, as the two-user example shows. This makes Lemma 1 false as stated, so Algorithm 1 and the infinite-cloud convex equivalence in Theorem 2 are not yet justified under finite power limits. The omega = 1 bisection path (Theorem 1 and Lemma 6) is constructed directly from the original equations and does not share this flaw, and the numerical comparisons do not provide evidence for the correctness of the general algorithm. Because the flaw is localized and could in principle be repaired by adding a power-headroom condition or by supplying a different tightening argument, the reader's CONDITIONAL verdict remains appropriate; this stress-test therefore leaves the verdict unchanged.","tokens_in":20324,"tokens_out":24141,"duration_ms":257486,"concrete_test":"Solve, with a global optimizer or dense grid, a one-group, two-user instance with B = 1, sigma^2 B = 1, h1 = h2 = 1, P1 = 0.5, P2 = 2, R1 = log2(1.5), R2 = 1, C_j = 1 cycle/bit, F_j = 10^-3 cycles/s, edge capacity F = infinity, and omega = 0.9. The tiny local capacities force d_j = R_j at any acceptable completion time. Compute the global optima of Problem (10) and Problem (11). If Problem (11) is feasible at tau = 0.9, T = 0.9 with p1 = 0.5 and p2 about 1.89, while Problem (10) is infeasible there because the equality (10d) for the weaker user forces p2 near 1.160 and then the stronger user needs p1 near 1.23 > P1, the equivalence in Lemma 1 fails and Algorithm 1 is solving a relaxation rather than the stated MEC problem.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Lemma 1 (Section III-A, proof in Appendix A) is the hinge of the paper's general and infinite-cloud results. It claims that Problem (10) and Problem (11) are equivalent because any slack in (11d) can be removed by decreasing p_ij and, for j>1, increasing p_{i(j-1)} by (h_ij/h_{i(j-1)})epsilon so the summed SIC rate constraints remain tight. This increase is possible only if p_{i(j-1)} < P_{i(j-1)}; footnote 2 merely assumes the maximal power is large enough, but Lemma 1 itself carries no such condition. When that cap binds, Problem (11) is a strict relaxation, not an equivalent problem. Concretely, take one group with two users, B = sigma^2 B = 1, h1 = h2 = 1, tau = 1, d1 = log2(1.5), d2 = log2(2), P1 = 0.5, and P2 >= 1.5. The original equalities (10d) force p2 = 1 and then p1 = 1 > P1, so Problem (10) is infeasible at these data; the relaxed inequalities (11d) are feasible with p1 = 0.5 and p2 = 1.5. The tightening step is therefore impossible exactly in the power-limited regime the paper claims to address. Algorithm 1 solves Problem (11)/(17) and reconstructs t_i = tau_i/x_i; if (17c) is not tight, the output can violate the original equality constraints (10d). The omega = 1 bisection result is built from the original equations and does not inherit this flaw, but the general iterative algorithm and the infinite-cloud convex equivalence both depend on the unsupported equivalence.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper considers an uplink NOMA-based mobile-edge computing network in which users are partitioned into groups that share time via TDMA, and within each group users transmit simultaneously using NOMA with SIC at the base station. The objective is to minimize a weighted sum of the completion time and the total energy consumption, including offloading transmission energy and local computation energy, subject to local latency, offloading rate, time-sharing, power, and edge-cloud capacity constraints. The main contributions are: (i) a transformation of the nonconvex problem into a supposedly equivalent relaxed form and an iterative algorithm with closed-form per-step solutions; (ii) a bisection algorithm that is claimed to find the optimal completion time for the special case with only completion-time minimization (omega=1); and (iii) a proof that for infinite edge-cloud capacity the problem is equivalent to a convex problem whose global optimum can be found. Numerical results compare the proposed scheme with TDMA, FDMA, and exhaustive-search variants under two-user-per-group NOMA.","tokens_in":20723,"tokens_out":7321,"duration_ms":81256,"significance":"If the central equivalence were valid, the paper would provide a useful and fairly general resource-allocation framework for multi-group NOMA-based MEC, with the attractive feature of closed-form updates in the iterative algorithm and global optimality guarantees in two special cases. The derivations in Lemmas 4 and 6, the perspective-function convexity argument in Appendix G, and the bisection feasibility conditions are largely sound and are presented carefully. However, the load-bearing equivalence in Lemma 1 is not correct under binding maximal-power constraints, and both the general iterative algorithm and the infinite-cloud convex equivalence depend on it. The special-case results for omega=1 are built on the original formulation and appear to survive this issue, but the general claims and the infinite-cloud convex equivalence need substantial repair before the paper's headline conclusions are supported.","major_comments":[{"comment":"The claimed equivalence of Problem (10) and Problem (11) is not established when maximal-power constraints bind. The proof lowers p_ij and, for j>1, raises p_{i(j-1)} by (h_ij/h_{i(j-1)})epsilon, but this operation requires p_{i(j-1)} < P_{i(j-1)}; footnote 2 merely assumes the maximal power is large enough, while Lemma 1 is stated unconditionally. The failure is concrete: take one group with B = sigma^2 B = 1, h1 = h2 = 1, x1 = t1 = 1, d1 = log2(1.5), d2 = log2(2), P1 = 0.5, and P2 = 1.5. The original equalities (10d) force p2 = 1 and then p1 = 1, violating P1, so Problem (10) is infeasible. The relaxed inequalities (11d) are feasible with p1 = 0.5 and p2 = 1.5. Thus Problem (11) is a strict relaxation rather than an equivalent problem in the power-limited regime. Since Algorithm 1 solves Problem (11)/(17) and reconstructs t_i = tau_i/x_i only after optimization, the output can violate the original equality constraints (10d).","section":"Section III-A, Lemma 1 and Appendix A (Eqs. (10d), (11d))"},{"comment":"The convex equivalence for infinite edge-cloud capacity inherits the unsupported equivalence of Lemma 1. Theorem 2 is proved by starting from Problem (11), replacing p_ij with q_ij = tau_i p_ij, and dropping the individual constraints tau_i <= T x_i in favor of sum_i tau_i <= T. If Problem (11) is a strict relaxation of Problem (10) under power caps, then an optimal point of the convex problem (25) need not correspond to a feasible point of the original Problem (10). Therefore the assertion that solving Problem (25) yields the global optimum of Problem (10) with F = infinity is not supported unless Lemma 1 is repaired or an independent proof of equivalence is supplied.","section":"Section III-C, Theorem 2 and Appendix G"},{"comment":"Algorithm 1 requires an initial feasible solution of Problem (10), but no procedure for obtaining one is given. In the power-limited regime where the strict-relaxation failure of Lemma 1 occurs, finding such a feasible starting point is nontrivial, and the monotone-convergence argument only applies if every iterate is feasible for Problem (10). The manuscript should either provide an initialization method, prove that a feasible point always exists under stated assumptions, or explicitly acknowledge that the algorithm is only defined when a feasible point is available.","section":"Algorithm 1 and Section III-A"}],"minor_comments":[{"comment":"The infeasible branch of the bisection update should read 'set Tmin = T', not 'set T = Tmin'; as written, the algorithm would not shrink the interval correctly.","section":"Algorithm 2, step 3"},{"comment":"The constraint in the convex feasibility set is written as sum_{i=1}^N sum_{j=1}^2 f_ij <= F, but group i contains M_i users in the general model; the upper limit should be M_i unless the authors intend to restrict the setup to two users per group.","section":"Eq. (20d)"},{"comment":"The appendix is titled 'Proof of Lemma 5', but it proves Lemma 6; the numbering should be corrected.","section":"Appendix F"},{"comment":"There are several typographical errors, e.g., 'To concur the nonconvexity' should be 'To circumvent the nonconvexity', 'ther are some examples' should be 'there are some examples', and 'spacial case' should be 'special case'. These do not affect the technical content.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The central issue is the unproven equivalence in Lemma 1 when power limits bind. This is not a cosmetic flaw: it undermines the general iterative algorithm and the infinite-cloud convex equivalence, which are major advertised contributions. The completion-time-only results (Algorithm 2 and Lemma 6) appear to be built directly from the original problem and may be salvageable as the paper's core contribution. I would encourage the authors to either add explicit power-feasibility conditions to Lemma 1 and Algorithm 1, or reframe the paper around the special cases that remain valid."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things you should know. The paper really does extend the authors' conference work to multiple NOMA groups with TDMA among groups, and the special-case completion-time minimization (Algorithm 2) is derived from the original equality constraints and looks solid. But the main equivalence in Lemma 1 is not proven as stated: the proof requires increasing a stronger user's power to compensate for decreasing a weaker user's power, and footnote 2 simply assumes max power is large enough. A two-user counterexample shows the claim can fail exactly in the power-limited regime the paper claims to address: with h1=h2=1, B=1, sigma^2 B=1, tau=1, d1=log2(1.5), d2=log2(2), P1=0.5, P2>=1.5, the original equalities force p2=1 and then p1=1>P1, so Problem (10) is infeasible, while the relaxed inequalities (11d) are feasible with p1=0.5, p2=1.5. So Problem (11) is a strict relaxation, not an equivalent problem, and Algorithm 1, which solves (11)/(17), can output a point violating the original equality constraints. The omega=1 bisection result is built directly from (10) and does not inherit this flaw; the infinite-cloud convex equivalence and the general iterative algorithm both do. A secondary, smaller gap: Algorithm 1 needs a feasible initialization for (10), which is nontrivial when (10) is infeasible, and none is specified. What is genuinely good: the multi-group model, the cumulative-rate transformation itself (even if the equality claim is unsupported), the closed-form feasibility conditions in Lemma 6, the bisection algorithm, and the convexity argument for the infinite-cloud special case. Numerical comparisons against TDMA, FDMA, and exhaustive search are standard and appropriate. The literature coverage is honest, and the self-citation to the conference paper is prior work, not circular. This is a paper for researchers in NOMA-MEC optimization who want the multi-group formulation and the completion-time result. I would send it to a serious referee because the model is useful and the flaw is concrete and potentially fixable, but the referee should ask for a corrected Lemma 1 (e.g., explicit conditions under which the tightening step is feasible) and a feasible-initialization procedure. If that is repaired, the paper would be solid; as is, the general algorithm is solving a relaxation.","headline":"The multi-group NOMA-MEC model is a real extension and the completion-time result is clean, but the central Lemma 1 equivalence fails when power caps bind, and the general and infinite-cloud algorithms inherit that break.","tokens_in":21201,"tokens_out":2587,"would_cite":false,"duration_ms":26439,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that a nonconvex NOMA mobile-edge-computing resource-allocation problem can be reformulated exactly and solved by low-complexity alternating optimization, with optimal solutions for two important special cases.","keywords":["mobile-edge computing","NOMA","resource allocation","completion time minimization","energy minimization","successive interference cancellation","convex optimization","time division multiple access"],"falsifier":"Take a small network with two users in one group, set their maximum powers just barely high enough to meet the data-rate constraints, and compare the optimum of the original nonconvex Problem (10), found by exhaustive grid search, with the solution produced by the proposed iterative method on the reformulated Problem (11). If the reformulated solution has a strictly lower completion time or energy than anything feasible in Problem (10), the claimed equivalence has failed.","tokens_in":20153,"feed_emoji":"📡","tokens_out":7221,"duration_ms":71068,"temperature":0.7,"pith_summary":"The paper studies an uplink network in which users, grouped into clusters, offload computation to an edge cloud using NOMA for simultaneous transmission within a group and time sharing among groups. It minimizes a weighted sum of the task completion time and total user energy, covering both transmission energy and local computation energy, subject to latency, rate, time-sharing, power, and edge-capacity constraints. The central claim is that this nonconvex problem, Problem (10), is equivalent to a reformulated problem, Problem (11), whose two natural block subproblems are convex, so an iterative algorithm with closed-form solutions at each step converges. For the special case of completion-time-only minimization, a bisection method is proven optimal; for infinite edge-cloud capacity, the problem is shown equivalent to a convex problem giving the global optimum. A sympathetic reader should care because this offers a low-complexity way to balance delay and energy in a realistic multi-group NOMA-MEC deployment.","feed_headline":"A low-cost algorithm balances time and energy in NOMA-MEC networks","feed_subtitle":"Iterative closed-form updates handle the general case; pure completion time gets an optimal bisection search.","key_machinery":"The load-bearing mechanism is the substitution $\\tau_i = x_i t_i$ together with the cumulative sum-rate reformulation $B\\tau_i \\log_2\\left(1 + \\frac{\\sum_{l=j}^{M_i} p_{il} h_{il}}{\\sigma^2 B}\\right) \\ge \\sum_{l=j}^{M_i} d_{il}$. This substitution separates the time-sharing variable $x_i$ from the transmission duration $t_i$ and makes the rate constraints convex in the power variables and in the new rate variables. The second central mechanism is a recursive power formula, equation (22), that for fixed data and time computes the minimum per-group transmit time needed to respect each user's maximum power; it turns the completion-time feasibility problem into the convex set (20) with closed-form feasibility checks. The third is the perspective function from convex analysis, which converts the infinite-capacity problem into a convex one. These mechanisms together carry the proof that the hard-looking nonconvex problems become solvable.","core_discovery":"On the paper's own terms, the discovery is that the joint optimization of how much data each user offloads, how long each group transmits, how the edge's CPU capacity is split, and what power each user uses can be reorganized into a tractable form. The key step (Lemma 1) replaces the variable pair of transmit time and time-sharing fraction by a product variable, $\\tau_i = x_i t_i$, and replaces individual data-rate equality constraints by cumulative-rate inequalities; the paper proves that at an optimum these inequalities are active, so the reformulation loses nothing. The resulting Problem (11) is then solved by alternating between two convex blocks: allocating $\\boldsymbol{\\tau}, \\mathbf{f}, T$ with data and powers fixed, and allocating data, time-sharing, and powers with $\\boldsymbol{\\tau}, \\mathbf{f}, T$ fixed. For completion-time-only minimization the paper derives an equivalent convex feasibility set and two closed-form feasibility conditions, making the bisection search optimal. For infinite edge-cloud capacity, a perspective-function transformation turns the whole problem into a convex problem whose global optimum can be obtained by standard dual methods.","pith_inferences":["Inference: The equivalence proof leans on a power-headroom assumption; a testable extension is exhaustive search on small instances with tight power caps to check whether the reformulation ever understates the true optimum.","Inference: The closed-form feasibility conditions (23) and (24) could be reused as a fast admission or scheduling test for dynamic edge networks, independently of the full optimization.","Inference: The principle that transmitting with maximal time saves energy suggests that the energy-delay tradeoff frontier is governed by the product of time and power, so other schedulers for NOMA-MEC should exhibit the same decreasing energy-versus-time shape.","Inference: The numerical finding that strong-strong pairing works best among the compared methods, while one big group is best when decoding complexity is ignored, points toward joint user-grouping-and-resource-allocation design as a natural next step."],"forward_implications":["For the completion-time-only case, the optimal minimum time can be found by bisection over $T$, checking only two closed-form feasibility conditions rather than solving the full nonconvex problem.","With an infinite-capacity edge cloud, the global optimum of the weighted time-energy objective is computable by convex optimization, and at that optimum all available transmission time is used because longer transmission lowers required power and energy.","In the general finite-capacity case, the proposed alternating algorithm converges because each block update is optimal and the objective is bounded below, with per-step closed-form or dual solutions whose complexity grows linearly with the number of users.","NOMA outperforms TDMA and FDMA in the reported numerical regime, with the largest gains at high edge-cloud capacity and low maximal transmit power.","The BS can run the algorithms centrally with overhead growing linearly in the number of users, making the scheme implementable as users join or leave."],"supporting_citations":[{"why":"Supplies the TDMA/FDMA completion-time-minimization model that the paper extends and uses as a numerical baseline.","marker":"[5]"},{"why":"Supplies the energy-minimization MEC framework and baseline that motivate the weighted objective.","marker":"[6]"},{"why":"Provides the local-computing energy and time model adopted in the problem formulation.","marker":"[11]"},{"why":"Gives the earlier single-group NOMA-MEC offloading formulation that this work generalizes to multiple groups.","marker":"[30]"},{"why":"Supplies analytical results on NOMA-MEC latency and energy that motivate NOMA offloading.","marker":"[31]"},{"why":"Provides the prior completion-time minimization for NOMA-MEC with deadlines that the bisection result extends.","marker":"[32]"},{"why":"Provides the prior joint power-and-time allocation for NOMA-MEC that the general algorithm builds on.","marker":"[33]"},{"why":"Supplies the rationale for small NOMA groups due to decoding complexity and error propagation, justifying the multi-group time-sharing model.","marker":"[34]"},{"why":"Supplies the user-pairing methods used to generate numerical comparisons.","marker":"[37]"},{"why":"Provides the perspective-function and KKT machinery used to prove convexity and solve subproblems.","marker":"[42]"}],"fun_headline_variants":["Low-cost NOMA-MEC: closed-form updates for time-energy","Optimal bisection for completion time in NOMA-MEC","Convex reformulation for infinite cloud in NOMA-MEC","NOMA-MEC: joint offload and power via convex blocks","Balancing time and energy in NOMA-MEC with iterations"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The equivalence proof assumes each user has enough spare transmit power that, when one message's power is reduced to keep a rate constraint tight, the earlier (stronger) user in the decoding order can compensate by raising its power; if that headroom is absent, the reformulated problem may not be equivalent to the original.","fun_headline_variants_meta":{"raw":{"variants":["Low-cost NOMA-MEC: closed-form updates for time-energy","Optimal bisection for completion time in NOMA-MEC","Convex reformulation for infinite cloud in NOMA-MEC","NOMA-MEC: joint offload and power via convex blocks","Balancing time and energy in NOMA-MEC with iterations"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00033,"raw_usage":{"total_tokens":1842,"prompt_tokens":952,"completion_tokens":890,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":568,"completion_tokens_details":{"reasoning_tokens":802}},"tokens_in":568,"tokens_out":890,"duration_ms":9397,"temperature":1.0,"reasoning_tokens":802,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:47:20.457914+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a small network with two users in one group, set their maximum powers just barely high enough to meet the data-rate constraints, and compare the optimum of the original nonconvex Problem (10), found by exhaustive grid search, with the solution produced by the proposed iterative method on the reformulated Problem (11). If the reformulated solution has a strictly lower completion time or energy than anything feasible in Problem (10), the claimed equivalence has failed.","supporting_citations":[{"cited_title":"Efficient resource a llocation in mobile-edge computation offloading: Completi on time minimization,","cited_arxiv_id":null,"evidence_quote":"Supplies the TDMA/FDMA completion-time-minimization model that the paper extends and uses as a numerical baseline."},{"cited_title":"Energy-efficient resource allocation for mobile-edge computation offloadin g,","cited_arxiv_id":null,"evidence_quote":"Supplies the energy-minimization MEC framework and baseline that motivate the weighted objective."},{"cited_title":"Multiuser resource allocation for mobile-edge computation offloading,","cited_arxiv_id":null,"evidence_quote":"Provides the local-computing energy and time model adopted in the problem formulation."},{"cited_title":"Optimized multiuser comput ation offloading with multi-antenna NOMA,","cited_arxiv_id":null,"evidence_quote":"Gives the earlier single-group NOMA-MEC offloading formulation that this work generalizes to multiple groups."},{"cited_title":"Impact of non-orthogona l multiple access on the offloading of mobile edge computing,","cited_arxiv_id":null,"evidence_quote":"Supplies analytical results on NOMA-MEC latency and energy that motivate NOMA offloading."},{"cited_title":"Delay min imization for NOMA-MEC offloading,","cited_arxiv_id":null,"evidence_quote":"Provides the prior completion-time minimization for NOMA-MEC with deadlines that the bisection result extends."},{"cited_title":"Joint power and t ime allocation for NOMA-MEC offloading,","cited_arxiv_id":null,"evidence_quote":"Provides the prior joint power-and-time allocation for NOMA-MEC that the general algorithm builds on."},{"cited_title":"O n multiple users scheduling using superposition coding ove r rayleigh fading channels,","cited_arxiv_id":null,"evidence_quote":"Supplies the rationale for small NOMA groups due to decoding complexity and error propagation, justifying the multi-group time-sharing model."}],"review_version":1}