{"id":"7b773f9c-8e5a-4c75-98a8-1552628d995c","arxiv_id":"2608.06910","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":3,"one_line_summary":"An adjoint general synthetic iterative scheme accelerates shape optimization of oscillatory rarefied gas flows, cutting iteration counts from tens of thousands to dozens.","lead":"Scientists built a faster solver that reshapes tiny vibrating sensors to reduce the drag force from surrounding gas. It couples the kinetic gas equations with an adjoint method, reaching design improvements in tens of optimization steps rather than tens of thousands.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The adjoint GSIS central claim rests on the Chapman-Enskog closure of Eq. (34); its small-temporal-Knudsen assumption is not exercised by any test with S comparable to delta_rp.","rationale":"In good faith, the derivation is careful, the adjoint formulation is first-principles, and the sensitivity checks plus the experiment-matched primal benchmark are genuine supporting evidence. The concern is not an internal inconsistency; the paper explicitly states the epsilon, epsilon*S << 1 assumption. It is load-bearing because fast convergence and asymptotic preservation are the two pillars of the claimed practical advance, and the temporal-Knudsen regime where the closure is unjustified is precisely the high-frequency MEMS regime the method is meant to serve. The reader identified the same assumption as weakest; I agree with that identification but weight it slightly more strongly, because the paper's framing ('across various Knudsen and Strouhal numbers') invites extrapolation beyond the tested S/delta_rp range. The proposed test is inexpensive and would either confirm the generality of the method or force a regime-qualified statement. A CONDITIONAL verdict is therefore appropriate pending that test or an explicit limitation statement, rather than an unqualified ACCEPT.","tokens_in":19665,"tokens_out":12310,"duration_ms":132501,"concrete_test":"Run the Fourier stability analysis of Section 3.3 and the oscillating-cylinder adjoint GSIS for (delta_rp, S) = (100, 100) and (10, 10), where Kn_t = 1. Record rho(G) and iteration counts to the 1e-6 residual tolerance, and compare one FFD sensitivity with the central-difference FDM of Eq. (59). If rho(G) remains below 0.5, GSIS converges within dozens of iterations, and the sensitivity matches FDM, the concern is resolved; otherwise the fast-convergence claim should be restricted to Kn_t << 1.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Section 3.2 closes the adjoint moment system with the Chapman-Enskog relations (34), justified only when epsilon = 1/delta_rp and epsilon*S = S/delta_rp = Kn_t are both small. This is the key assumption behind the spectral-radius estimate rho(G) = O(delta_rp^-2) in Eq. (40). Because the method is advertised for oscillatory MEMS, where S can be comparable to delta_rp, the synthetic equations then lose their diffusive NS closure and the promised halving of error per iteration is not guaranteed. The reported convergence and sensitivity tests cover S = 1 and S = 0.1 with delta_rp = 1, 10, 100, so Kn_t = S/delta_rp is at most 1 and only reaches order one in the rarefied delta_rp = 1 cases; no near-continuum case with S/delta_rp of order one is demonstrated. The converged solution could still be accurate because the high-order kinetic corrections in Eq. (32) are exact, but the fast-convergence pillar of the central claim is an extrapolation for high-frequency near-continuum operation.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a fast-converging, asymptotic-preserving adjoint shape optimization method for small-amplitude oscillatory rarefied gas flows. It derives a mesoscopic adjoint of the frequency-domain linearized Shakhov kinetic equation from a Lagrangian, constructs macroscopic synthetic equations via velocity moments with a Chapman-Enskog closure and high-order kinetic corrections, and uses Fourier stability analysis to argue that the resulting adjoint GSIS has spectral radius below 0.5. The shape sensitivity is extracted from discretized boundary integrals and passed through an FFD parameterization to a gradient-based optimizer. The method is validated on an oscillating cylinder and a biaxial accelerometer, with sensitivities checked against finite differences and CIS, the primal solver matched to experimental damping data, and reported force reductions of 33.1–84.0% and 35.5–52.6%, respectively.","tokens_in":19889,"tokens_out":9896,"duration_ms":100107,"significance":"If the claims hold, this is a substantial step: it makes adjoint-based shape optimization at the kinetic level practical for vibrating MEMS by replacing slow kinetic iterations with a few dozen GSIS iterations and by permitting meshes coarser than the mean free path in near-continuum regimes. The strengths of the paper are explicit: the adjoint equation is obtained from a Lagrangian rather than postulated; the Chapman-Enskog closure is derived rather than fitted; the stability analysis is carried out analytically; and the sensitivities are cross-checked against both finite-difference and CIS reference results, with the primal solver validated against experiment. These checks give confidence that the derived sensitivities and optimization results are not artifacts of the discretization. The main caveat is that the fast-convergence guarantee is proved and demonstrated only in regimes where the temporal Knudsen number is small, which is narrower than the broad 'vibrating MEMS' framing in the title and abstract.","major_comments":[{"comment":"The spectral-radius estimate ρ(G)=O(δrp^-2) and the associated 'convergence within dozens of iterations' claim are derived under the assumption ε=1/δrp and εS=S/δrp=Kn_t both small. In the numerical demonstrations, the near-continuum cases keep Kn_t≤0.01 (δrp=100 with S=1 in Section 5, δrp=100 with S=0.1 in Section 6), and the only case with Kn_t=O(1) is at δrp=1, where the flow is rarefied rather than near-continuum. The regime S∼δrp with large δrp, which is physically relevant for high-frequency MEMS operation, is therefore neither analyzed nor tested. Because the fixed point of the GSIS remains exact via the high-order corrections in Eq. (32), the accuracy of converged solutions is not in question; the concern is specifically the advertised fast convergence in the near-continuum high-frequency regime. I ask the authors either to add a numerical test with δrp large and S/δrp of order one, or to explicitly restrict the fast-convergence claim and the abstract/title wording to low temporal Knudsen numbers.","section":"Sections 3.2–3.3, Eqs. (34) and (40)"},{"comment":"The objective is the squared modulus |J|^2 of the complex force amplitude. For a harmonic oscillation, the real part of J is proportional to the time-averaged dissipated power (the damping component), while the imaginary part is the stiffness/inertia component. Minimizing |J|^2 does not in general minimize drag or damping. The paper repeatedly describes the results as 'drag reduction' and claims relevance to quality factor and damping, but the optimizer is not constrained to reduce Re{J}; in the reported cases Re{J} does decrease, but this is an outcome, not an enforced objective. The authors should either justify |J|^2 as the relevant MEMS objective for the specific applications, or reformulate the objective (for instance, minimize Re{J} or a weight between the real and imaginary parts) and re-verify the sensitivities and optimizations.","section":"Section 2.3, Eqs. (14)–(15), and abstract/title"}],"minor_comments":[{"comment":"The reduced two-dimensional adjoint system is introduced as a direct construction from the reduced primal system; a short statement explaining why this is equivalent to reducing the three-dimensional adjoint equations would help readers verify the reduction.","section":"Appendix A.3"},{"comment":"The caption lists the Strouhal values used and the range of δrp over which the spectral radius is below 0.5; without this information the 'below 0.5' claim in the abstract cannot be checked against the plot.","section":"Figure 2"},{"comment":"The finite-difference perturbation is fixed at ε=10^-3; reporting the sensitivity of the comparison to this step size would strengthen the gradient verification.","section":"Section 5.2, Eq. (59)"},{"comment":"The iteration in Eq. (22) is labeled CIS, but the left-hand side already contains the implicit δrp ϕ term; clarifying that the 'conventional' scheme uses the same implicit treatment as the GSIS kinetic step would avoid ambiguity in the convergence comparison.","section":"Section 3.1, Eq. (22)"},{"comment":"The phrase 'It's worth to note' is informal; it should read 'It is worth noting'.","section":"Section 5.2"}],"recommendation":"major_revision","confidential_remarks":"The paper is clearly within the scope of the journal and the technical work is careful. My main concern is that the advertised fast-convergence property is not demonstrated in the temporally rarefied near-continuum regime that motivates the MEMS application; this should be fixed by adding a test or by narrowing the claims. The objective-function issue also needs clarification before publication. If the authors address these points, the paper would be a solid contribution."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper delivers a real new tool: an adjoint GSIS for the frequency-domain linearized kinetic equation, including adjoint synthetic equations, a Fourier stability analysis, and an incremental boundary treatment. The derivation is careful, and the sensitivities are cross-checked against finite differences and against the conventional iterative scheme. The primal solver matches experiments on a MEMS accelerometer, which grounds the whole pipeline. This is not a repackaging of prior work; the adjoint machinery is genuinely new relative to the authors' own frequency-domain GSIS and steady-state adjoint papers.\n\nThe soft spots are real but limited. The Chapman-Enskog closure behind the synthetic equations assumes small temporal Knudsen number, S/delta_rp. The paper demonstrates convergence for S=1 with delta_rp up to 100 and S=0.1 with delta_rp up to 100, so the regime where S/delta_rp is order one in a near-continuum setting is not exercised. That leaves the advertised fast convergence for high-frequency, near-continuum oscillations as an extrapolation. The stability analysis is also only for infinite domains, so wall-bounded behavior is shown numerically rather than analytically. Neither point undermines the central claim for the tested cases, and the paper is honest about the low-frequency focus.\n\nThe absence of code is a practical limitation for reproducibility, but the detailed numerical schemes in the appendix and the standard benchmarks make the results credible. The self-citation pattern is appropriate: the frequency-domain primal GSIS is the direct foundation, and the comparison against it is meaningful.\n\nI agree with the reader's accept recommendation. The math and data look solid within the stated regime, the novelty is clear, and the paper opens a class of shape optimization problems that were previously impractical. It deserves a serious referee. I would be happy to see this in a computational physics or microfluidics journal, and I would cite it for the adjoint-GSIS construction.","headline":"A solid adjoint-GSIS extension for oscillatory rarefied flows, with real derivations and honest validation; the high-frequency near-continuum regime is an extrapolation but not a fatal gap.","tokens_in":20423,"tokens_out":715,"would_cite":true,"duration_ms":8739,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"An adjoint kinetic solver makes shape optimization of oscillatory rarefied gas flows fast and accurate, with force reductions up to 84 percent.","keywords":["adjoint optimization","rarefied gas dynamics","general synthetic iterative scheme","MEMS gas damping","frequency-domain kinetic equations","shape optimization","drag reduction","asymptotic-preserving scheme"],"falsifier":"Repeat the adjoint GSIS convergence test and shape-optimization run at $\\delta_{rp}=100$ with Strouhal numbers $S=10$ and $S=100$, so the temporal Knudsen number $S/\\delta_{rp}$ reaches $0.1$ and $1$. If the measured spectral radius climbs toward unity or the adjoint sensitivities drift away from the finite-difference gradients, the Chapman-Enskog closure used in the adjoint synthetic equations is the failing component.","tokens_in":19452,"feed_emoji":"📉","tokens_out":7752,"duration_ms":64339,"temperature":0.7,"pith_summary":"The paper aims to establish that adjoint shape optimization can be made practical for vibrating micro-electro-mechanical systems by solving the linearized Boltzmann kinetic equation in the frequency domain with a general synthetic iterative scheme for both the forward and adjoint problems. The method targets gas damping on oscillating MEMS, where rarefaction effects invalidate continuum Navier-Stokes optimization. The key claim is that the adjoint GSIS converges with spectral radius below 0.5, so primal and adjoint solutions settle in dozens of iterations, while remaining asymptotic-preserving, that is, accurate even when spatial cells are much larger than the molecular mean free path. On two test cases the optimized geometries reduce the gas-force amplitude by 33.1% to 84.0% for an oscillating cylinder and 35.5% to 52.6% for a biaxial accelerometer, with adjoint sensitivities verified against finite differences.","feed_headline":"Adjoint scheme cuts MEMS gas drag by up to 84 percent","feed_subtitle":"A kinetic adjoint solver converges in dozens of iterations and slashes gas damping of vibrating micro-devices.","key_machinery":"The central object is the adjoint general synthetic iterative scheme (adjoint GSIS): a fixed-point iteration that alternates one update of the mesoscopic adjoint kinetic equation with a solve of macroscopic synthetic equations. Those equations are velocity moments of the adjoint equation, closed by Chapman-Enskog constitutive relations and corrected by high-order terms extracted from the distribution function, so that diffusive information propagates across the whole domain instead of only within a few mean free paths. Free-form deformation with B-spline control points maps boundary-node sensitivities to a low-dimensional design vector, so the gradient is computed from converged primal and adjoint fields at a cost nearly independent of the number of design variables.","core_discovery":"The paper presents an adjoint form of the frequency-domain linearized Shakhov kinetic equation with diffuse-reflection boundary conditions, then closes the moment system by splitting adjoint stress and heat flux into Navier-Stokes constitutive parts plus high-order kinetic corrections. With the Chapman-Enskog closure, the adjoint synthetic macroscopic equations accelerate the kinetic iteration and remain valid in the near-continuum, low-frequency regime. Fourier stability analysis in an infinite domain gives a spectral radius below 0.5, while conventional iteration has a spectral radius approaching unity, and wall-bounded tests confirm convergence within dozens of steps and asymptotic-preserving behavior on cells coarser than the mean free path. The authors present this as a unified adjoint optimization tool for oscillatory rarefied gas flows, and they validate the gradient accuracy against finite differences before reporting the force reductions.","pith_inferences":["Editorial extension: the asymptotic-preserving closure is demonstrated at Strouhal numbers $S=1$ and $S=0.1$; at higher frequencies where the temporal Knudsen number $\\mathrm{Kn}_t=S/\\delta_{rp}$ is no longer small, the Chapman-Enskog closure of the adjoint synthetic equations would need re-examination and the speedup may degrade.","Editorial extension: the optimized cylinder profiles at high $\\delta_{rp}$ converge toward a common elongated shape, suggesting a possible regime-independent low-damping template that could be tested without re-optimization.","Editorial extension: since the conventional primal and adjoint iterations share the same spectral radius, the adjoint GSIS acceleration should transfer to other kinetic models with the same moment structure, for example, a simplified relaxation-time collision model or the full Boltzmann collision operator, where the sensitivity accuracy can be checked against finite differences."],"forward_implications":["Solving both primal and adjoint kinetic equations together takes dozens of iterations instead of tens of thousands; at $(\\delta_{rp},S)=(100,0.1)$ the speedup over conventional iteration is 42.5 times.","Because the scheme is asymptotic-preserving, shape optimization can use spatial meshes whose cells are far larger than the molecular mean free path, which is what makes near-continuum MEMS cases affordable.","For the oscillating cylinder, the optimized elongated shape reduces horizontal gas-force amplitude by 33.1%, 62.8%, and 84.0% at $\\delta_{rp}=1,10,100$ ($S=1$).","For the biaxial accelerometer, the constrained optimization reduces damping-force amplitude by 35.5%, 44.3%, and 52.6% at $\\delta_{rp}=1,10,100$ ($S=0.1$), and the baseline damping force reproduces the experimental value.","The adjoint sensitivities agree with finite-difference sensitivities in all tested cases, so the reported reductions come from an accurate gradient."],"supporting_citations":[{"why":"Supplies the frequency-domain GSIS framework and its constitutive treatment that the adjoint scheme extends to the adjoint equation.","marker":"[6]"},{"why":"Establishes the mesoscopic adjoint formulation and discrete boundary sensitivity evaluation used here.","marker":"[10]"},{"why":"Provides the fast-converging asymptotic-preserving adjoint shape optimization framework this work builds on.","marker":"[11]"},{"why":"Shows that GSIS finds steady-state kinetic solutions within dozens of iterations, the property extended to the adjoint solver.","marker":"[19]"},{"why":"Gives the convergence-rate and asymptotic-preserving analysis that justifies the spectral-radius comparison.","marker":"[21]"},{"why":"Defines the Shakhov kinetic model used for the frequency-domain linearized gas description.","marker":"[23]"},{"why":"Supplies the free-form deformation parameterization that maps shape changes to FFD control points.","marker":"[25, 26]"},{"why":"Provides the biaxial accelerometer geometry and experimental damping-force data used for validation.","marker":"[35]"}],"fun_headline_variants":["Adjoint method slashes MEMS drag in oscillatory gas flows","Fast adjoint solver cuts gas damping in vibrating devices","84% drag cut for MEMS via adjoint shape optimization","Asymptotic-preserving adjoint method for MEMS drag reduction","Kinetic adjoint halves error, converges fast, cuts drag"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The acceleration rests on a Chapman-Enskog closure that assumes the mean free path is small compared with the device size and the oscillation period is long compared with the molecular collision time; if the oscillation is fast enough that the temporal Knudsen number is not small, this closure and the fast convergence can break down.","fun_headline_variants_meta":{"raw":{"variants":["Adjoint method slashes MEMS drag in oscillatory gas flows","Fast adjoint solver cuts gas damping in vibrating devices","84% drag cut for MEMS via adjoint shape optimization","Asymptotic-preserving adjoint method for MEMS drag reduction","Kinetic adjoint halves error, converges fast, cuts drag"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000294,"raw_usage":{"total_tokens":1698,"prompt_tokens":917,"completion_tokens":781,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":533,"completion_tokens_details":{"reasoning_tokens":695}},"tokens_in":533,"tokens_out":781,"duration_ms":7000,"temperature":1.0,"reasoning_tokens":695,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T18:31:58.031651+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the adjoint GSIS convergence test and shape-optimization run at $\\delta_{rp}=100$ with Strouhal numbers $S=10$ and $S=100$, so the temporal Knudsen number $S/\\delta_{rp}$ reaches $0.1$ and $1$. If the measured spectral radius climbs toward unity or the adjoint sensitivities drift away from the finite-difference gradients, the Chapman-Enskog closure used in the adjoint synthetic equations is the failing component.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the frequency-domain GSIS framework and its constitutive treatment that the adjoint scheme extends to the adjoint equation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the mesoscopic adjoint formulation and discrete boundary sensitivity evaluation used here."},{"cited_title":"Zhang, R","cited_arxiv_id":null,"evidence_quote":"Provides the fast-converging asymptotic-preserving adjoint shape optimization framework this work builds on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the convergence-rate and asymptotic-preserving analysis that justifies the spectral-radius comparison."},{"cited_title":"Shakhov, Approximate kinetic equations in rarefied gas theory, Fluid Dynamics 3 (1) (1968) 112–115","cited_arxiv_id":null,"evidence_quote":"Defines the Shakhov kinetic model used for the frequency-domain linearized gas description."},{"cited_title":"Frangi, A","cited_arxiv_id":null,"evidence_quote":"Provides the biaxial accelerometer geometry and experimental damping-force data used for validation."}],"review_version":1}