{"id":"404b72bd-388b-44bc-9f95-ee8ee61b1823","arxiv_id":"1908.04082","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A UAV and a reconfigurable intelligent surface can be jointly optimized, with a closed-form phase alignment and a successive-convex-approximation trajectory design, to raise the average downlink rate above trajectory-only or surface-only benchmarks.","lead":"This paper designs a joint UAV flight path and reflecting-surface phase schedule to maximize the average data rate when the direct link to a ground user is blocked. It shows that jointly optimizing the path and the passive surface beats hovering above the user or using the surface without path optimization.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Algorithm can decrease the true rate because SCA slack variables are not reset to the actual distances, and the monotonicity claim in Section III-C is therefore unsupported.","rationale":"The reader's weakest_assumption identified the incomplete proof of Lemma 1 as the main correctness risk. I investigated Lemma 1 and found that the lemma is actually true: for functions of the form log(1+sum K_i x^{-a_i} y^{-b_i}), the Hessian in the original coordinates is PSD because the log-sum-exp Hessian in log-coordinates is a covariance matrix plus a positive diagonal term. Thus the convexity gap in the appendix is a proof presentation issue, not a fundamental flaw. The more serious issue is the non-tightness of the SCA surrogate with respect to the true rate. The slack variables u_k, v_k produced by solving (18) are generally larger than the actual distances d_UG(q_k), d_UR(q_k). Consequently, the linearized objective is not a global underestimator of F(q) that touches at the current trajectory, and the standard monotone convergence argument from [3] does not apply. The provided toy counterexample shows a strict decrease in the true rate in a single iteration, which directly contradicts the claim in Section III-C. This concern is load-bearing because the central claim of the paper—that the joint design 'considerably improves' the average achievable rate—relies on the proposed algorithm reliably increasing the objective; if the algorithm can move from a better trajectory to a worse one, the numerical improvement may depend on initialization or on implicitly resetting slack variables, and the comparison with benchmarks becomes unreliable. The issue is addressable: the algorithm should reset u and v to the actual distances before linearization, or the convergence proof should be modified to account for the slack variables. Because the central numerical result may still hold after this fix, the verdict remains CONDITIONAL rather than REJECT, but the required revision is concrete and specific.","tokens_in":11708,"tokens_out":51479,"duration_ms":496755,"concrete_test":"Run a one-slot, no-mobility version of Algorithm 1 with two links at q=2 and q=8, distances d1(q)=sqrt(1+(q-2)^2), d2(q)=sqrt(1+(q-8)^2), A=B=1, gamma0=1, and initialize q0=5.5, u0=v0=100. Compute the true rate Rbar_0=F(5.5)≈0.279, solve problem (18) to obtain q1,u1,v1, and compute Rbar_1=F(q1). If Rbar_1=d_UG(q[n]) and v[n]>=d_UR(q[n]). However, the standard SCA monotonicity argument also requires the surrogate to be tight at the previous iterate, i.e., the linearization point should satisfy u_{k-1}=d_UG(q_{k-1}) and v_{k-1}=d_UR(q_{k-1}). In Algorithm 1 the slack variables are not reset to the actual distances after the trajectory update, so u_k from (18) equals (d_UG(q_k)^2+u_{k-1}^2)/(2u_{k-1}), which is strictly larger than d_UG(q_k) whenever u_{k-1} differs from d_UG(q_k), and similarly for v. Consequently the surrogate value at the previous point is strictly below the true rate, and the chain F(q_k) >= R_slack(u_k,v_k) >= L_k(u_k,v_k) >= L_k(u_{k-1},v_{k-1}) = R_slack(u_{k-1},v_{k-1}) <= F(q_{k-1}) does not imply F(q_k) >= F(q_{k-1}). A concrete single-slot instance with d_1(q)=sqrt(1+(q-2)^2), d_2(q)=sqrt(1+(q-8)^2), A=B=1, gamma0=1, q0=5.5 and u0=v0=100 gives a true rate that strictly decreases after one iteration (from log2(1.4175) about 0.503 to log2(1.4) about 0.485). The algorithm should be modified to reset u_k and v_k to the exact distances d_UG(q_k), d_UR(q_k) after each trajectory update, and the monotonicity proof should be restated under that rule.","section":"Section III-C, Algorithm 1, problem (18)"},{"comment":"The convexity proof is incomplete. In the main text the positive definiteness of the Hessian is asserted without demonstration; the supplementary material expands eta^2 f_xx f_yy into a long list of positive monomials and then concludes that the negative terms in (28) are dominated by the positive terms in (29) on the basis of a blue-part comparison, but no explicit algebraic inequality among the monomial coefficients is actually provided. Since Lemma 1 is the foundation for the global underestimator property (15) and hence for the validity of the SCA step (18), this is a load-bearing gap. The authors should either supply a complete factorized proof of the Hessian determinant inequality or verify the lemma rigorously for the parameter ranges used in the simulations (e.g., kappa=3.5). If Lemma 1 were false in some region, the first-order Taylor expression could overestimate R_slack, and the minorization argument would collapse.","section":"Appendix, Lemma 1"},{"comment":"The numerical results do not specify the number of channel realizations over which the average rate is computed, nor do they provide error bars or confidence intervals. Because h_tilde in (2) and h_NLoS in (4) are random, the achievable rate is a random variable; without Monte Carlo averaging, the reported 'considerable improvement' of JT&PB over the benchmarks could be within statistical fluctuation. The authors should state the number of random realizations used and, ideally, show confidence intervals or multiple realization curves to support the magnitude of the claimed gain.","section":"Section IV, Figs. 2-3"}],"minor_comments":[{"comment":"The initialization of u0 and v0 is not specified in Algorithm 1. This matters because the SCA tightness issue in the first major comment is aggravated when the initial slack values are loose; the authors should explicitly set u0[n]=d_UG(q0[n]) and v0[n]=d_UR(q0[n]).","section":"Algorithm 1, initialization"},{"comment":"There are several typos and formatting issues: 'matirx' in Section II-B, 'repate' in Algorithm 1, 'Sine' in the supplementary proof (should be 'Since'), and the figure label 'versus T' should be 'versus T' consistently formatted.","section":"Throughout"},{"comment":"The complexity statement O(Kite N^3.5) is given without derivation or reference; please clarify whether it is per-iteration complexity or total complexity, and state the complexity of solving the convex problem (18) with a standard solver.","section":"Section III-C, complexity"}],"recommendation":"major_revision","confidential_remarks":"The paper is a letter-style contribution whose main evidence is the numerical comparison in Figs. 2-3. The unsupported monotonicity claim and the incomplete proof of Lemma 1 are fixable with a resetting step and a more rigorous appendix, so I do not recommend rejection. I would encourage the authors to provide the simulation script or a detailed description of the random channel generation, as this would materially increase confidence in the reported gains."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is a straightforward blend of two established toolkits: RIS phase alignment (the closed-form co-phasing from [6]) and SCA-type UAV trajectory optimization (as in [3]). What's new is the specific joint design and a convexity lemma for the two-distance rate expression. If the algorithm works as advertised, it's a useful engineering result for RIS-assisted UAV downlinks, but it's not a conceptual leap.\n\nThe good: the phase-shift derivation is correct, the alternating structure is sensible, and the simulations show the joint design outperforms the single-optimization benchmarks. The convexity lemma, if true, is a nice piece of analysis. The paper cites the relevant prior work and doesn't try to hide that the components are known.\n\nWhere it gets soft: the proof of Lemma 1 in the appendix is not convincing. It expands the Hessian determinants and then asserts that \"the blue part in (28) is smaller than the blue part in (29)\" without showing the comparison. That's a gap a referee would need closed.\n\nMore importantly, the stress-test note is correct. The monotonicity claim in Section III-C, citing [3], does not follow from the SCA argument as written. The linearized surrogate in (18) is a global underestimator of R_slack(u,v), but not of the true rate F(q) = R_slack(d_UG(q), d_UR(q)), because the slack variables are not reset to the actual distances at the current trajectory. The counterexample in the note (single slot, A=B=1, q0=5.5, u0=v0=100) gives a strict decrease in the true rate after the first iteration. This means the reported \"non-decreasing over iterations\" guarantee is not established. It might be fixable by resetting u and v to the distances after each trajectory update, or by proving monotonicity for the slack-objective and then translating back, but as written the claim is unsupported.\n\nAlso minor: the simulations appear to be single realizations—no Monte Carlo averaging or error bars—so the magnitude of the reported gain isn't statistically grounded. No code is provided.\n\nOverall, this is a reasonable engineering paper with a real but addressable theoretical flaw. The core idea is sound and the combination is timely for the RIS-UAV literature. A serious referee could send it back for revision with specific requests: complete the Lemma 1 proof, fix the monotonicity claim or the algorithm, and add Monte Carlo statistics. I'd engage with it as a reviewer, but I wouldn't cite it until the convergence claim is sorted out.","headline":"A competent but incremental RIS-UAV combination; the SCA monotonicity claim doesn't hold as written because the slack variables are never reset.","tokens_in":12646,"tokens_out":8808,"would_cite":false,"duration_ms":73435,"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":"Jointly optimizing a UAV's flight path and a reconfigurable surface's reflecting phases markedly raises the average achievable rate in blocked urban links.","keywords":["reconfigurable intelligent surface","UAV communication","trajectory design","passive beamforming","successive convex approximation","phase alignment","average achievable rate","Rician fading"],"falsifier":"Compute the Hessian determinant of $f(x,y)=\\log_2(1+K_1 x^{-\\kappa}+K_2 y^{-2}+K_3 x^{-\\kappa/2}y^{-1})$ at a grid of positive $x,y$ using the paper's parameter values; a single negative determinant disproves Lemma 1 and would invalidate the SCA lower bound. A second test is to run the joint algorithm with imperfect channel estimates (e.g., 10-20% phase error at the RIS) and compare against trajectory-only optimization: if the rate gap shrinks sharply, the reported benefit depends on perfect phase alignment rather than on joint optimization alone.","tokens_in":11530,"feed_emoji":"📡","tokens_out":8398,"duration_ms":82315,"temperature":0.7,"pith_summary":"This paper studies a wireless downlink in which a drone serving a ground user has its direct line of sight blocked, and a wall-mounted reconfigurable intelligent surface (RIS) reflects the drone's signal to the user. Its claim is that maximizing the user's average data rate requires the drone's flight path and the RIS's reflecting phases to be designed together. For any fixed path, the paper gives a closed-form phase setting that makes all reflected paths add coherently at the user, and it then solves the remaining path-planning problem by successive convex approximation. Numerical results show this joint design achieves a considerably higher average rate than optimizing either the path or the phases alone.","feed_headline":"UAV path plus smart reflection lifts urban link rate","feed_subtitle":"Jointly tuning drone trajectory and reflective-surface phases beats optimizing either one alone.","key_machinery":"The load-bearing object is the phase-alignment identity (11): choosing each RIS phase as $\\theta_i[n]=\\arg(\\tilde h)+\\omega_i+\\frac{2\\pi}{\\lambda}d(i-1)\\varphi_{UR}[n]$ makes every reflected path arrive at the user with the same phase, so the SNR becomes a squared sum of magnitudes instead of a random phasor sum. This converts the non-convex joint problem into an optimization over two distances, $d_{UG}[n]$ and $d_{UR}[n]$. The second piece is Lemma 1, which asserts that $f(x,y)=\\log_2(1+K_1 x^{-\\kappa}+K_2 y^{-2}+K_3 x^{-\\kappa/2}y^{-1})$ is convex for $x,y>0$; that property justifies using first-order Taylor expansions as global underestimators inside the successive-convex-approximation loop. With slack variables $u[n]$, $v[n]$ bounding the distances, the trajectory subproblem is convex and can be solved by standard tools.","core_discovery":"The central claim is that a UAV serving a ground user through a building-mounted RIS should not simply hover above the user; it should fly an arc that balances the direct UAV-user link against the UAV-RIS-user reflected link, with the RIS phases set to align the two signal components at the receiver. For a fixed trajectory, phase alignment across all reflecting elements---equation (11), $\\theta_i[n]=\\arg(\\tilde{h})+\\omega_i+\\frac{2\\pi}{\\lambda}d(i-1)\\varphi_{UR}[n]$---maximizes the received signal power and collapses the joint design into a trajectory-only problem. The trajectory problem is then solved with slack variables and successive convex approximation, using the lemma that the rate expression is convex in the two distances $d_{UG}$ and $d_{UR}$. In the paper's simulations the joint algorithm beats benchmarks that optimize only one side of the problem, and the optimized trajectory is an arc whose hovering point sits away from directly above the user.","pith_inferences":["Beyond the paper: the phase-alignment identity is structural and should carry over to rectangular RIS arrays or multi-antenna UAVs, since it only needs the product form of the reflected channel.","Beyond the paper: a direct sensitivity test would add channel-estimation noise to the phases in (11); the paper assumes perfect CSI, so the real-world gap between joint and one-sided design may narrow when alignment errors grow.","Beyond the paper: the convexity lemma, if valid for all positive parameters, is a reusable tool for other RIS-UAV rate maximizations and could save repeated convergence proofs.","Beyond the paper: the simulations suggest a design heuristic---move the hover point toward the RIS as the direct link weakens and toward the user as the reflected link weakens---that could be tested as a low-complexity substitute for full optimization."],"forward_implications":["A UAV serving a ground user in a blocked environment should fly an arc that balances the direct and reflected links; hovering directly above the user is suboptimal when a RIS is available.","System designers should treat UAV path planning and RIS configuration as one coupled problem; optimizing either alone leaves a measurable rate gain on the table.","Because the phases are closed-form once the path is known, the RIS update is computationally cheap and can be applied in real time as the UAV moves.","The trajectory subproblem is a disciplined convex program, so standard solvers can find each update, with monotone improvement and $O(K_{\\mathrm{ite}}N^{3.5})$ complexity.","The result supports deploying RISs as a low-power way to restore connectivity in UAV-to-ground links when the line of sight is blocked, without adding transmit power."],"supporting_citations":[{"why":"Supplies the alternating-optimization and successive-convex-approximation template whose monotone convergence the proposed algorithm inherits.","marker":"[3]"},{"why":"Supplies the phase-alignment passive beamforming approach and the path-loss/channel model used for the RIS-assisted links.","marker":"[6]"},{"why":"Provides the product-channel and ULA reflection model that makes the closed-form phase identity (11) valid.","marker":"[10]"},{"why":"Cited as the method by which the channel state information of the RIS links is obtained.","marker":"[12]"},{"why":"Also supports the CSI availability assumption for passive-surface links.","marker":"[13]"}],"fun_headline_variants":["Joint drone path and smart-surface phases lift urban rates","Drone arc plus aligned RIS reflections beats hovering spot","Joint design of UAV trajectory and RIS phases raises rate","Optimizing both drone path and reflective phases beats one-sided tuning"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The algorithm's monotone improvement rests on Lemma 1's claim that the per-slot rate expression is bowl-shaped in the two distances, so each successive-approximation step is a valid lower bound; if that convexity fails, the trajectory update may not improve the rate and the reported joint gain is not guaranteed.","fun_headline_variants_meta":{"raw":{"variants":["Joint drone path and smart-surface phases lift urban rates","Drone arc plus aligned RIS reflections beats hovering spot","Joint design of UAV trajectory and RIS phases raises rate","Optimizing both drone path and reflective phases beats one-sided tuning"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000402,"raw_usage":{"total_tokens":2089,"prompt_tokens":931,"completion_tokens":1158,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":547,"completion_tokens_details":{"reasoning_tokens":1093}},"tokens_in":547,"tokens_out":1158,"duration_ms":11715,"temperature":1.0,"reasoning_tokens":1093,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:51:46.467740+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the Hessian determinant of $f(x,y)=\\log_2(1+K_1 x^{-\\kappa}+K_2 y^{-2}+K_3 x^{-\\kappa/2}y^{-1})$ at a grid of positive $x,y$ using the paper's parameter values; a single negative determinant disproves Lemma 1 and would invalidate the SCA lower bound. A second test is to run the joint algorithm with imperfect channel estimates (e.g., 10-20% phase error at the RIS) and compare against trajectory-only optimization: if the rate gap shrinks sharply, the reported benefit depends on perfect phase alignment rather than on joint optimization alone.","supporting_citations":[{"cited_title":"Securing UA V commun i- cations via joint trajectory and power control,","cited_arxiv_id":null,"evidence_quote":"Supplies the alternating-optimization and successive-convex-approximation template whose monotone convergence the proposed algorithm inherits."},{"cited_title":"Wireless communications through reconfigurable intelligent surfaces,","cited_arxiv_id":null,"evidence_quote":"Provides the product-channel and ULA reflection model that makes the closed-form phase identity (11) valid."},{"cited_title":"Channel estimation and low -complexity beamforming design for passive intelligent surface assist ed MISO wire- less energy transfer","cited_arxiv_id":null,"evidence_quote":"Also supports the CSI availability assumption for passive-surface links."}],"review_version":1}