{"id":"80dca572-b8bd-4564-9716-faf737508d2a","arxiv_id":"2507.04657","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A joint user-association and resource-allocation algorithm, DAUR, maximizes processed bits per unit delay-energy in a blockchain-enabled Metaverse wireless system, and simulations report higher DPE than simple baselines.","lead":"This paper proposes DAUR, an algorithm that jointly decides user-server connections and allocates wireless and computing resources to improve a new metric, data processing efficiency, for blockchain-enabled Metaverse networks. It also reports simulations in which DAUR outperforms simple baselines, though the theoretical transformation is not fully rigorous.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 2's transformation to P3 is not an equivalence: P3 is unbounded in α and ϑ, and at the KKT values claimed in (17)-(20) the P3 objective is identically zero, so the convexification theorem is unsupported.","rationale":"The reader's formal weakest_assumption was Lemma 4's γ = 1/2, and that is a real limitation. My check, however, targets a more fundamental step: Lemma 2 is the hinge between the ratio problem and the summation problem that the rest of the algorithm solves. Since P3 as written is unbounded and its objective at the claimed KKT point is zero, the advertised convexification is not merely missing a mild assumption; it is an invalid equivalence. This moves the center of gravity of the review from 'γ is pre-fixed' to 'the chain of reformulations is not established.' I do not escalate to REJECT because the numerical contribution is separable: Algorithm 1 can be read as a fixed-point/heuristic method that fixes α and ϑ between subproblems, and the reported 87.87 M bits/(s·J) against weak baselines is neither internally contradicted nor independently verified. A conditional acceptance requiring a corrected Lemma 2/Theorem 1, plus code/data and stronger baselines, remains the right gate. Hence UNCHANGED relative to the reader's CONDITIONAL, with partial agreement: the reader mentioned the P3 defect in the rationale but did not make it the weakest_assumption.","tokens_in":29631,"tokens_out":7334,"duration_ms":84191,"concrete_test":"Analytic check: instantiate N = M = 1 with positive constants c_1, c_{1,1}, d_1, and feasible x = 1, ϕ = 0.5, φ, ρ, ζ, ψ. Write P3 exactly as Eq. (16) without adding constraints (15a)-(15b). First, observe the objective is affine and increasing in α^s_{1,1} when ϑ^s_{1,1} = 0, so it is unbounded above; hence P3 has no finite optimum. Second, plug the KKT values from Eqs. (17)-(20) into P3 and verify every term is zero while P2's objective is positive. If either check holds, Lemma 2 does not establish equivalence, and Theorem 1 needs restatement.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that DAUR transforms P1 into a solvable convex problem rests on Lemma 2, and that step is not sound as written. P3 (Eq. 16) maximizes over α and ϑ with no bound on these variables and no constraint linking them to costs. For any feasible x, ϕ, γ, φ, ρ, ζ, ψ, choose a term with x_{n,m} ϕ_n d_n > 0, set ϑ^s_{n,m} = 0, and let α^s_{n,m} → +∞; the objective goes to +∞. Thus P3 has no finite maximum and cannot be equivalent to P2 or P1. Moreover, Appendix B derives the stationarity and complementary-slackness conditions of P2, not P3. Substituting the claimed KKT values (17)-(20) into the P3 objective makes every term α(num − ϑ·cost) equal to zero, whereas P2's objective at the same point is the sum of DPE ratios. The gap can be repaired by presenting P3 as a fixed-parameter subproblem in which α and ϑ are held fixed and updated externally, as Algorithm 1 does, and then proving convergence of that fixed-point iteration to a stationary point of P1; but that is not what Lemma 2 states. The γ = 1/2 issue in Lemma 4 is secondary: it only affects optimality of one variable, while the invalid P3 equivalence undercuts the main theoretical claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a data processing efficiency (DPE) metric for blockchain-enabled Metaverse wireless systems and proposes the DAUR algorithm to jointly optimize user association, work offloading, task-specific computing resource distribution, bandwidth, transmit power, and computing allocations. The optimization problem P1 is a nonconvex sum-of-ratios problem. The authors claim to transform P1 into a solvable convex problem through a sequence of auxiliary variables and relaxations (P2, P3, ..., P12), solved by an alternating fractional-programming and QCQP/SDR-with-DC-penalty approach. Simulations over several user/server configurations show that DAUR achieves a higher DPE (87.87 M bits/(s·J)) than four baseline schemes (80.38–84.82).","tokens_in":29915,"tokens_out":2961,"duration_ms":35024,"significance":"If the transformations and convergence claims are sound, the paper offers a practical joint user-association and resource-allocation algorithm for a timely application domain, and it proposes a meaningful efficiency metric (processed bits per delay-plus-energy) that generalizes earlier utility-cost metrics. The manuscript is unusually transparent: it clearly lists differences from the authors' conference version [1], provides an extensive appendix with proofs, reports running times and iteration counts, and compares several rounding techniques. The numerical study is self-contained and shows consistent improvements across bandwidth, server capacity, user capacity, transmit power, and weight-ratio sweeps. The central weakness is that the main theoretical transformation (Lemma 2) is not a valid equivalence as written; this undermines the paper's stated central claim, though the algorithmic framework and simulations may remain useful if the theoretical gap can be repaired.","major_comments":[{"comment":"The claimed equivalence of P2 and P3 is not correct as stated. In Problem P3, the variables α and ϑ appear only in the objective and have no constraints or bounds. For any feasible (x, ϕ, γ, φ, ρ, ζ, ψ, T), pick a term with x_{n,m} ϕ_n d_n > 0, set ϑ^s_{n,m} = 0, and let α^s_{n,m} → +∞; the objective tends to +∞. Hence P3 is unbounded above and cannot be equivalent to P2 or P1. Moreover, substituting the KKT values (17)–(20) into the P3 objective makes each bracketed term α(·)(num − ϑ·cost) equal to zero, so the P3 objective value at those points is zero, not the sum of DPE ratios of P2. The proof in Appendix B derives stationarity and complementary slackness of P2, not a transformation of P2 into P3. The intended statement appears to be that P3 is a fixed-parameter subproblem when α and ϑ are held constant and updated externally, as done in Algorithm 1; that formulation can be a valid algorithmic device, but Lemma 2 and the surrounding text must be rewritten to state and prove convergence of that fixed-point iteration rather than an equivalence.","section":"Section V, Lemma 2, Eq. (16)–(20), Appendix B"},{"comment":"Lemma 4 sets ω_b = 1 and treats the validation delay T^sv as negligible to conclude γ*? = 1/2. For general ω_b, the expression γ^2 + ω_b(1−γ)^2 is minimized at γ = ω_b/(1+ω_b), and the sum T^sp + T^sg is minimized at γ = 1/(1+ω_b), neither of which equals 1/2 unless ω_b = 1. The validation delay T^sv is not negligible in general, and it depends on the γ values of other servers, so the optimal γ would couple across servers. Algorithm 1 initializes γ_{n,m} = 0.5 and never updates it, so the claimed joint optimization of the task-specific computing resource split (variable γ) is not actually performed. This issue affects the validity of Theorem 3 and the interpretation of the simulation results, which set ω_b = 1 by default.","section":"Section V-B, Lemma 4, Appendix D, Algorithm 1"},{"comment":"The paper claims that the alternating block-coordinate update converges to a stationary point of Problem P3, but no convergence proof is provided for the outer loop of Algorithm 1 or for the overall DAUR procedure. Given that P3 as currently stated is unbounded, the stationarity claim is not meaningful without a corrected formulation. If the authors adopt the fixed-point interpretation of P3, they should prove that the sequence produced by the α/ϑ updates and the inner FP and QCQP loops converges to a stationary point of the original P1 (or of a properly defined surrogate), with explicit assumptions on the penalty parameter and rounding step.","section":"Section V, paragraphs after Lemma 2 and Algorithm 1"}],"minor_comments":[{"comment":"The phrase 'transforms the nonconvex problem of maximizing the sum of DPE ratios into a solvable convex optimization problem' is repeated in the abstract, introduction, and Section V; given the issues in Lemma 2, this claim should be softened or made conditional on the fixed-point interpretation until the theory is repaired.","section":"Abstract and Section V"},{"comment":"The notation σ^2 φ_{n,m} b_m in the Shannon formula is introduced without defining σ^2 as a noise power spectral density; consider clarifying whether the bandwidth scaling is intentional.","section":"Table I and Eq. (1)"},{"comment":"The proof of Lemma 3 shows equality of partial derivatives between F and G at the specific value υ^s = 1/(2 x ρ p φ d r), which is a local first-order argument; the text should explain more precisely that this is a standard fractional-programming surrogate rather than an exact problem transformation, since the equality of functions is only shown at the stationary update value.","section":"Appendix C, Lemma 3"},{"comment":"In Fig. 5(c), the vertical axis labels 'delay(s)', 'energy(J)', and 'DPE(Mbits/(s·J))' are combined in one plot, but the delay and energy values appear to be on different scales; clarifying the scaling or using separate subplots would improve readability.","section":"Section VII and Fig. 5(c)"},{"comment":"There are several typographical and formatting issues, including 'Karush-–Kuhn—Tucker' in the statement of Lemma 2, and inconsistent use of 'different-of-convex' versus 'difference-of-convex' in Section I; a careful pass for language and notation is needed.","section":"Overall"}],"recommendation":"major_revision","confidential_remarks":"The paper is a straightforward extension of the authors' conference version [1] and closely related to their prior work [22], but the manuscript is honest about these connections. The main reason for major revision is the invalid equivalence in Lemma 2; this is a load-bearing theoretical step, not a cosmetic issue. The simulations themselves appear internally consistent and the comparisons are fair, so a corrected version with a proper convergence proof for a fixed-point formulation would be publishable in principle. I would not recommend rejection, as the algorithmic framework is plausible and the numerical gains are consistently reported, but the current theoretical narrative overclaims."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nQuick take: this paper has a useful incremental result hiding behind an overclaimed equivalence proof. The new material compared with the MobiHoc version—DC-based rank-one retention, five rounding methods, updated simulations—is real and mostly described in enough detail to reimplement. The rounding comparison is the most valuable part: it gives concrete evidence for choosing rank-1 approximation over Hungarian/randomized/greedy, and the convergence plots match what the algorithm text says. The authors are also transparent about what changed from [1] and about the centralized limitation. That is honest work.\n\nThe problem is Lemma 2. As written, P3 is not equivalent to P2, and it is not even bounded: α and ϑ appear with no box or linking constraints, so for any feasible point with a positive term, you can drive that term to infinity. The \"KKT values\" in (17)-(20) are derived from the Lagrangian of P2, not P3, and plugging them into P3 makes every α(num−ϑ·cost) term zero, whereas P2's objective at the same point is the sum of DPE ratios. The proof in Appendix B therefore supports neither the transformation nor the claimed optimality. This is the load-bearing wall behind the abstract's \"transforms into a solvable convex problem,\" and it currently does not stand. The fix is straightforward in spirit—present P3 as a fixed-parameter subproblem where α and ϑ are updated externally, as Algorithm 1 actually does, and prove convergence of that fixed-point iteration—but that is not what Lemma 2 states.\n\nThe γ=1/2 issue is real but secondary. Lemma 4 sets ωb=1 and ignores T^sv to get γ=1/2; Algorithm 1 initializes γ=0.5 and never updates it. So the \"joint optimization\" of task-specific computing split is not actually performed for ωb≠1 or nontrivial validation delay. Minor-to-moderate: the numerical conclusions use ωb=1, so at least the simulations are internally consistent with the assumption.\n\nI would not call this a fundamental rejection of the engineering. The simulations are self-contained, the gains over the four baselines are modest but consistent across resource sweeps, and the method is plausibly useful for blockchain-MEC resource allocation. But the baselines are weak—random/greedy/average combinations—and there are no error bars or code, so the 80-to-88 Mbit/(s·J) story is a lower bound on what a skeptical reader needs.\n\nBottom line: worth a serious referee, but only with a clear instruction that Lemma 2 must be rewritten, γ justified, and stronger baselines added. I would not cite the theorem as it stands. I'd bring it to reading group only if someone wants to workshop the repair.","headline":"A useful rounding-comparison study and a plausible simulation story are currently attached to a central equivalence proof that does not hold as written; worth a referee, but only with a clear demand to fix Lemma 2.","tokens_in":30488,"tokens_out":2468,"would_cite":false,"duration_ms":28679,"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 the DAUR algorithm turns the nonconvex problem of maximizing data processing efficiency in blockchain-enabled Metaverse wireless systems into a solvable convex optimization, and that in simulation it reaches 87.87…","keywords":["data processing efficiency","blockchain-enabled Metaverse","user association","resource allocation","fractional programming","semidefinite relaxation","difference-of-convex programming","wireless communications"],"falsifier":"Take the same DPE maximization with $\\omega_b = 2$ and with non-negligible validation delay, allow $\\gamma$ to vary over $(0,1)$ while recomputing the optimal association and resources, and compare the resulting DPE with the value obtained at the fixed split $\\gamma = 1/2$; if the variable-$\\gamma$ solution is materially higher, Lemma 4's premise fails and DAUR's reported optimum holds only for the special case.","tokens_in":29326,"feed_emoji":"📡","tokens_out":6675,"duration_ms":67929,"temperature":0.7,"pith_summary":"This paper introduces data processing efficiency (DPE), the ratio of processed data bits to the sum of delay and energy consumption, as a system-level objective for blockchain-enabled Metaverse wireless networks where users offload NFT-related tasks to servers. The central claim is that the DAUR algorithm jointly optimizes user association, work offloading ratios, task-specific server resource splits, bandwidth, transmit power, and computing allocations by transforming the nonconvex sum-of-ratios problem into a sequence of solvable convex subproblems. The transformation alternates a fractional-programming block for communication and computing resources with a QCQP/SDP block, keeping the rank-one constraint and recovering discrete association by rank-1 approximation. Simulation results reported in the paper show DAUR achieving DPE 87.87 Mbit/(s·J), above RUCAA (80.78), GUCAA (80.38), AAUCO (83.25), and GUCRO (84.82). A sympathetic reader would take the paper's claim as: coordinated user-server association and resource allocation can be made tractable and measurably raises DPE, provided the server resource split is fixed at its assumed optimum.","feed_headline":"87.87 Mbit/(s·J): DAUR beats four Metaverse baselines","feed_subtitle":"Joint user-association and resource allocation lifts DPE above RUCAA, GUCAA, AAUCO, and GUCRO in simulation.","key_machinery":"The central objects are the DPE ratio itself, DPE := processed bits/(delay + energy), and the DAUR algorithm that alternates two optimization blocks. The first block fixes user association $x$, offloading ratio $\\phi$, and server split $\\gamma$, and uses the auxiliary variable $\\upsilon^{(s)}_{n,m} = 1/(2 x_{n,m}\\rho_n p_n \\phi_n d_n r_{n,m})$ to turn the nonconvex transmission-cost ratio into a convex surrogate. The second block fixes bandwidth, power, and computing allocations and relaxes the remaining mixed-integer QCQP into an SDR problem, whose rank-one constraint is handled by the DC penalty $\\mathrm{Tr}(S)-\\|S\\|_2$ and linearized as Problem P12; rank-1 approximation then recovers the discrete association. Lemma 4 does key work: it sets the data size changing ratio $\\omega_b = 1$ and declares the validation delay $T^{sv}$ negligible, so the optimal server resource split is $\\gamma^* = 1/2$.","core_discovery":"The central claim is that the nonconvex problem P1, maximizing the sum of DPE ratios over discrete user association and continuous offloading and resource variables, can be relaxed into convex subproblems without losing solution quality. The argument proceeds by a fractional-programming transformation that rewrites each ratio with auxiliary variables $\\alpha$ and $\\vartheta$, then splits optimization into two blocks: with user association, offloading ratio, and server split fixed, the communication and computing block becomes concave through the auxiliary variable $\\upsilon^{(s)}_{n,m}$; with the communication block fixed, the residual QCQP over $x$ and $\\phi$ is relaxed by semidefinite relaxation, and the rank-one constraint is retained through a difference-of-convex penalty $\\mathrm{Tr}(S)-\\|S\\|_2$ that is linearized into a convex problem. The discrete association is recovered by rank-1 approximation, which the paper's numerical comparison selects over Hungarian, randomized, secondary-discrete, and greedy rounding. The headline numeric result is DPE 87.87 Mbit/(s·J) at default settings, versus 80.78, 80.38, 83.25, and 84.82 for the four baselines.","pith_inferences":["If $\\omega_b \\neq 1$ or the validation delay is not negligible, the fixed split $\\gamma = 1/2$ is no longer provably optimal, so the claimed joint optimization would need $\\gamma$ as a decision variable; rerunning the simulations with $\\omega_b = 2$ and optimizing over $\\gamma$ is a direct test.","The reported DPE values scale with the preference weights $c_n$ and $c_{n,m}$; at low preference settings the paper reports DPE dropping to 17.29, so the absolute numbers are preference-dependent even if the ordering over baselines persists.","The validation delay $T^{sv}$ is modeled but then declared negligible; real systems with many validator servers may see block-generation and validation costs dominate, a regime the default simulation does not stress.","Because Algorithm 1 initializes $\\gamma_{n,m} = 0.5$ and never updates it, the 'joint' optimization in the code is effectively over the other variables only; a version that treats $\\gamma$ as an optimized variable would separate the paper's algorithmic contribution from its strongest simplifying assumption."],"forward_implications":["Under the paper's simulation settings, DAUR achieves DPE 87.87 Mbit/(s·J), exceeding RUCAA, GUCAA, AAUCO, and GUCRO by about 7.09, 7.49, 4.62, and 3.05 Mbit/(s·J), respectively.","Retaining the rank-one constraint in the SDR step yields a slightly higher objective than dropping it, improving the conference-version values from 81.87 and 86.48 to 83.25 and 87.87.","The DAUR algorithm converges within one outer iteration for the tested (10,2), (20,3), and (30,4) setups, with total runtimes from 0.22 s to 228.86 s.","The paper states DAUR extends beyond DPE to energy-efficiency and utility-cost problems, with successive convex approximation available for nonconcave utility functions.","The method requires a centralized optimizer, which the paper identifies as a privacy risk and a target for future decentralized implementation."],"supporting_citations":[{"why":"The conference version supplies the previous drop-rank-plus-Hungarian method and the baseline DPE values 81.87 and 86.48 that the journal version improves.","marker":"[1]"},{"why":"Provides the blockchain delay model, including block propagation and validation delay expressions, that the paper adapts for $T^{bp}_{n,m}$ and $T^{sv}_{n,m}$.","marker":"[7]"},{"why":"Supplies the joint offloading and user-association baseline and the worst-case complexity expression used in the complexity analysis.","marker":"[8]"},{"why":"Underpins the difference-of-convex algorithm and the claim that the linearized iteration reaches a critical point solution.","marker":"[26]"},{"why":"Provides the Hungarian algorithm used as one of the rounding techniques compared against rank-1 approximation.","marker":"[27]"},{"why":"Supplies the semidefinite relaxation theory and the rank-1 approximation justification used to recover the discrete association.","marker":"[28]"}],"fun_headline_variants":["DAUR algorithm transforms nonconvex DPE problem into convex solution","Metaverse data efficiency hits 87.87 Mbit/(s·J) with DAUR","DAUR scheme beats four baselines in blockchain Metaverse data processing","User association and resource allocation boost Metaverse DPE to 87.87"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The paper's argument assumes that the data size changing ratio $\\omega_b = 1$ and that the blockchain validation delay $T^{sv}$ is negligible; if either assumption fails, fixing the server resource split at $\\gamma = 1/2$ is not optimal and the claimed joint optimization is not actually being carried out.","fun_headline_variants_meta":{"raw":{"variants":["DAUR algorithm transforms nonconvex DPE problem into convex solution","Metaverse data efficiency hits 87.87 Mbit/(s·J) with DAUR","DAUR scheme beats four baselines in blockchain Metaverse data processing","User association and resource allocation boost Metaverse DPE to 87.87"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000211,"raw_usage":{"total_tokens":1425,"prompt_tokens":970,"completion_tokens":455,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":586,"completion_tokens_details":{"reasoning_tokens":372}},"tokens_in":586,"tokens_out":455,"duration_ms":4878,"temperature":1.0,"reasoning_tokens":372,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T19:45:19.345144+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the same DPE maximization with $\\omega_b = 2$ and with non-negligible validation delay, allow $\\gamma$ to vary over $(0,1)$ while recomputing the optimal association and resources, and compare the resulting DPE with the value obtained at the fixed split $\\gamma = 1/2$; if the variable-$\\gamma$ solution is materially higher, Lemma 4's premise fails and DAUR's reported optimum holds only for the special case.","supporting_citations":[{"cited_title":"Data processing efficiency aware use r association and resource allocation in blockchain enabled Metaverse ov er wireless communications,","cited_arxiv_id":null,"evidence_quote":"The conference version supplies the previous drop-rank-plus-Hungarian method and the baseline DPE values 81.87 and 86.48 that the journal version improves."},{"cited_title":"Joint optimiza tion of radio and computational resources allocation in blockchain-ena bled mobile edge computing systems,","cited_arxiv_id":null,"evidence_quote":"Provides the blockchain delay model, including block propagation and validation delay expressions, that the paper adapts for $T^{bp}_{n,m}$ and $T^{sv}_{n,m}$."},{"cited_title":"Joint computati on offloading and user association in multi-task mobile edge computing,","cited_arxiv_id":null,"evidence_quote":"Supplies the joint offloading and user-association baseline and the worst-case complexity expression used in the complexity analysis."},{"cited_title":"Over-the-air computation via inte lligent reflecting surfaces,","cited_arxiv_id":null,"evidence_quote":"Underpins the difference-of-convex algorithm and the claim that the linearized iteration reaches a critical point solution."},{"cited_title":"The Hungarian method for the assignment pro blem,","cited_arxiv_id":null,"evidence_quote":"Provides the Hungarian algorithm used as one of the rounding techniques compared against rank-1 approximation."},{"cited_title":"Sem idefinite relaxation of quadratic optimization problems,","cited_arxiv_id":null,"evidence_quote":"Supplies the semidefinite relaxation theory and the rank-1 approximation justification used to recover the discrete association."}],"review_version":1}