{"id":"24f503e5-a54a-4737-a6a8-9f71a67d2411","arxiv_id":"1908.04712","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A continuous-adjoint shape derivative for the Eulerian particle erosion model is derived and used to optimize a 90-degree pipe bend, reducing predicted erosion for a range of Stokes numbers.","lead":"The authors derive a shape-derivative formula for minimizing particle erosion in a pipe bend and use gradient descent to optimize a 3D geometry. The optimized bend reduces predicted erosion by 76% for the target particle size and by over 20% for a range of other sizes.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The quantitative erosion-reduction claims are only validated with the same Eulerian model that overpredicts at the target Stokes number; no independent check of the optimized geometry is provided.","rationale":"I read the paper as making two connected claims: (i) a formal shape derivative (Theorem 2) for the Eulerian erosion model, and (ii) a numerical demonstration that gradient descent on this derivative yields a bend with substantially reduced erosion. Claim (i) appears internally consistent; the proof follows the standard Lagrangian/adjoint route and I found no obvious algebraic inconsistency. The load-bearing part of the paper is claim (ii), and its support is thin exactly where the model is weakest. The validation in Figure 3 shows agreement with published Lagrangian and experimental data only for Stk > 0.43, but the optimization target is Stk=0.33, in the acknowledged overprediction regime. The optimized geometry is then judged by the same biased model, so the 76% max-erosion reduction and the ≥20% reductions for Stk>0.2 are predictions of that model, not demonstrated physical properties. This is a correctness risk, not an internal inconsistency or a disagreement with consensus; the authors honestly state the model's limitations. A Lagrangian particle-tracking cross-check on the optimized geometry would settle whether the optimized bend is genuinely better or merely better according to the Eulerian model. Since this is the same concern the reader identified and the appropriate response is to require that check before accepting the quantitative claims, I leave the CONDITIONAL verdict unchanged.","tokens_in":15382,"tokens_out":3902,"duration_ms":42226,"concrete_test":"Run an independent Lagrangian particle-tracking (DPM) simulation on the initial and optimized geometries (or at least on the geometry at iteration 18) with the same flow field, same drag law and same Oka erosion parameters, for the Stokes numbers in Table 1. If the Lagrangian model does not reproduce a 76% reduction of maximal erosion at Stk=0.33 and reductions of at least 20% for Stk>0.2, then the claimed physical improvement is not supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central numerical claim is that Algorithm 1 lowers the predicted maximal erosion at Stk=0.33 by 76% (Section 5.2, Figure 5) and integrated erosion by at least 20% for every tested Stk > 0.2 (Figure 8a). Both numbers are computed with the same single-velocity Eulerian model that is used in the optimization. Section 5.1 and Figure 3 show that this model agrees with reference data only for Stk > 0.43, while the optimization target Stk=0.33 lies in the range where the authors themselves state the model overpredicts impact rates. The optimized geometry is then evaluated only with the same model, on a deformed geometry for which no validation exists. Because a shape optimizer can exploit systematic model bias (e.g., in predicted impact angle/velocity distributions or in artificial diffusion) without producing a physically better bend, the headline reductions are not established. The paper's own Section 2.1 acknowledges that the univariate Eulerian model cannot represent crossing particle trajectories, a limitation that is known to be relevant in bends; this further raises the risk that the optimized shape is tuned to the model's errors.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript derives a continuous adjoint shape calculus for minimizing particle erosion in a 3D pipe bend using a one-way coupled, single-velocity Eulerian particle model. The main theoretical result is Theorem 2, Eq. (26), which gives the Eulerian semi-derivative of the reduced cost functional, together with the adjoint equations (22)-(25). The authors implement the derivative via a linear-elasticity Riesz projection and a gradient descent method (Algorithm 1), validate the forward model's impact rates against experimental and numerical references for Stokes numbers above 0.43, and optimize a 90-degree bend for Stk = 0.33. They report a 76% reduction in maximal erosion at the target Stokes number and at least a 20% reduction in integrated erosion for every tested Stokes number above 0.2.","tokens_in":15590,"tokens_out":4116,"duration_ms":43262,"significance":"If the central claims hold, this is a useful first demonstration of gradient-based shape optimization for an Eulerian erosion model on a 3D pipe bend, and Eq. (26) is a nontrivial theoretical contribution that can be adapted to other erosion models. The authors deserve credit for deriving the adjoint system explicitly, for validating the forward impact rates against independent experimental and numerical references, and for using externally calibrated erosion constants from the literature. The strength of the quantitative conclusions is, however, limited by the fact that the optimized shape is evaluated only with the same model that overpredicts in the target regime; this is a correctness-risk concern rather than an internal inconsistency.","major_comments":[{"comment":"The optimization target Stk = 0.33 lies in the range Stk <= 0.43 where the authors state that the Eulerian model overpredicts impact rates, and the optimized geometry is evaluated only with the same Eulerian model. Because a gradient-based method can exploit systematic model bias (for example in impact-rate magnitude, impact-angle distribution, or artificial diffusion) to reduce the predicted objective without improving the physical erosion profile, the reported 76% and greater-than-20% reductions are not established as physical improvements. The authors should validate the optimized geometry with an independent method, such as Lagrangian particle tracking or experiments, or at minimum quantify the sensitivity of the optimized shape to the model discrepancy at Stk = 0.33.","section":"Sections 5.1 and 5.2, Figures 3, 5, and 8"},{"comment":"The reduced cost functional J(Omega) is defined under an unproved assumption that Eq. (7) has a unique solution, and Eq. (28) additionally requires the forward state to be shape differentiable. The nonlinear Schiller-Naumann drag term and the sign-dependent boundary conditions for the volume fraction alpha make these assumptions nontrivial. Since Theorem 2 and Algorithm 1 rest on this formal calculus, the paper should either prove the needed well-posedness and differentiability in a suitable setting or explicitly mark the derivation as formal and state how this limitation affects the validity of the optimality conditions.","section":"Section 2.3 and Theorem 2"},{"comment":"The authors acknowledge in Section 2.1 that a single particle velocity cannot represent crossing particle trajectories, a limitation known to be relevant in pipe bends. The optimized geometry is a strongly deformed bend, and no evidence is given that this limitation does not become worse on the deformed geometry. This is not a circularity, but it is a concrete correctness risk for the reported reductions; the paper should address it, for example by comparing the optimized geometry against a Lagrangian particle model or by demonstrating that trajectory crossing is negligible in the deformed bend.","section":"Sections 2.1 and 5.2"}],"minor_comments":[{"comment":"The symbol g is used both for the gravity vector in the momentum equations and for the erosion integrand in the cost functional (10); this notational conflict should be resolved for clarity.","section":"Equations (8)-(10) and Section 2.1"},{"comment":"The derivation of the adjoint equations omits a detailed discussion of the boundary terms on the sign-dependent set Gamma^-(qv); because the function space Z(Omega) depends on the forward solution, the treatment of these terms should be clarified.","section":"Adjoint equations (22)-(24)"},{"comment":"The stopping criterion 'until converged' is vague; the authors should specify the tolerance or the precise test used on the relative decrease of J and the gradient norm shown in Figure 4a.","section":"Algorithm 1"},{"comment":"There are several language issues, including 'bended' for 'bent' and 'the later of which' for 'the latter of which'; a careful proofreading pass is recommended.","section":"Throughout"},{"comment":"The statement that the deviation at low Stokes numbers is 'within the range of deviations among the reference studies' is qualitative; reporting the numerical values of the deviations would make the validation more informative.","section":"Figure 3 and Section 5.1"},{"comment":"The paper models only half of the bend by symmetry; it should state explicitly whether the deformable boundary Gamma_f and the optimized geometry are constrained to respect that symmetry.","section":"Section 5.1 and Figure 2"}],"recommendation":"major_revision","confidential_remarks":"This is a borderline case. The formal shape calculus is a solid contribution, but the numerical validation of the optimized geometry is too weak for the quantitative claims as stated. I would encourage major revision rather than rejection: an independent check of the optimized bend with a Lagrangian model or experiments is feasible and would substantially raise confidence. I also note that the manuscript is an arXiv preprint with no reproducibility artifacts; providing the mesh, solver parameters, and post-processing scripts would strengthen the paper."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe short version: this paper derives a shape derivative for erosion reduction in a bended pipe using an Eulerian particle model, and shows a predicted 76% drop in peak erosion at the target Stokes number. The derivative formula in Theorem 2 appears to be new, and the numerical pipeline is sensible. But the validation has a real gap: the model overpredicts impact rates for Stk <= 0.43, which is exactly the regime of the optimization target (Stk=0.33). The optimized geometry is then judged with the same model that produced it. That does not invalidate the paper, but it means the headline numbers are model artifacts until confirmed independently.\n\nWhat is genuinely new: the continuous adjoint derivation for the Oka erosion model coupled with a one-way Eulerian particle transport. I checked the references and this is not in the earlier work. The authors also take care with the Willmore regularization and the gradient projection, and they compare impact rates against several published datasets. The multi-Stokes evaluation of the optimized geometry is a nice touch, even if it is model-on-model.\n\nThe soft spots are as I said. The derivation is formal and assumes uniqueness of the forward solution without proof; that is common in this literature and not disqualifying. The bigger issue is the Stokes mismatch. Their own Figure 3 shows the Eulerian model departing from the reference data below 0.43, and they choose 0.33 for the optimization. They argue the deviation is within the scatter of references, which is fair, but it is still a bias that a shape optimizer can exploit. And because no code or data is provided, I cannot check whether the 76% is robust to mesh size or stabilization parameters. The paper would be stronger with an open-source implementation or at least a second validation on the deformed geometry with a Lagrangian model.\n\nWho this is for: people working on PDE-constrained shape optimization, especially with Eulerian particle or droplet models. They will find the derivative formula and the numerical setup useful. The erosion community might be more skeptical about the low-Stokes predictions. I would send it to peer review, with a request for the code and for a paragraph on the model's known low-Stokes bias.","headline":"A solid formal contribution to shape optimization for erosion, but the 76% reduction claim rests on the same model that overpredicts at the target Stokes number.","tokens_in":16116,"tokens_out":3198,"would_cite":true,"duration_ms":30981,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["49Q10","49K20","76D55"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper derives a closed-form shape derivative for an Eulerian erosion model and shows that gradient-based reshaping of a pipe bend cuts predicted peak erosion by 76% at the design particle size and at least 20% across a range of sizes.","keywords":["shape optimization","particle erosion","Eulerian particle model","continuous adjoint","shape derivative","pipe bend","Stokes number","Dean vortices"],"falsifier":"Measure or simulate wall erosion on the initial and optimized bends using Lagrangian particle tracking or an experimental aerosol flow with the design particle size (Stk ≈ 0.33). If the optimized geometry does not show substantially lower erosion than the initial bend—or if the gap is much smaller than the model's 76%—then the reported reduction is an artifact of the Eulerian approximation rather than a physical property of the shape.","tokens_in":15159,"feed_emoji":"🌀","tokens_out":6474,"duration_ms":60772,"temperature":0.7,"pith_summary":"Erosion in bent pipes is usually studied by tracking individual particles; this paper instead treats the particle cloud as a continuous field and derives, through continuous adjoint calculus, the shape derivative of a wall-erosion cost functional. That derivative tells how predicted erosion responds to small deformations of the bend surface, which turns erosion reduction into a PDE-constrained shape optimization problem solved by gradient descent. Applied to a 90° pipe bend, the method produces a deformed geometry whose predicted peak erosion is 76% lower for the design particle size (Stk = 0.33) and whose total erosion is at least 20% lower for every tested Stokes number above 0.2. The point of the exercise is that the optimized shape is not hand-tuned; it emerges from a systematic, gradient-based procedure, and the same adjoint formulas extend to other erosion models by replacing one function.","feed_headline":"Gradient-designed pipe bend cuts erosion by 76%","feed_subtitle":"Optimized with an Eulerian particle model, the new bend also lowers predicted erosion by 20% or more for other particle sizes.","key_machinery":"The load-bearing object is the continuous adjoint shape derivative $dJ(\\Omega)[V]$ in Eq. (26): a formula for how the reduced cost changes when the bend is deformed by a vector field $V$. It is assembled from the forward solution $(u,p,v,\\alpha)$ and adjoint variables $(\\lambda_u,\\lambda_p,\\lambda_v,\\lambda_\\alpha)$, with volume terms for each PDE constraint and surface terms for erosion and curvature regularization; the derivative is then projected into a smooth mesh deformation via the linear-elasticity saddle-point system (Eqs. (30)–(33)) so each gradient step gives a legitimate geometry change.","core_discovery":"The paper's central claim is that the Eulerian semi-derivative of the reduced erosion cost functional exists in closed form (Theorem 2, Eq. (26)) and can be evaluated from the forward fluid/particle state and a decoupled set of adjoint equations (Eqs. (22)–(25)). The derivative consists of boundary integrals over the deformable wall—capturing the erosion model, the normal convection of impact rate, and the Willmore curvature regularization—plus volume integrals encoding how the Navier-Stokes, particle-velocity, and volume-fraction equations respond to domain changes. Embedded in Algorithm 1, with the derivative projected onto mesh deformations through linear elasticity and a restricted-gradient correction, it drives a gradient descent that stops at a locally optimal shape. For the benchmark 90° bend, the optimized geometry lowers the integrated erosion by at least 20% for all tested Stokes numbers above 0.2 and by 76% for the design species, and the improvement is attributed not to fewer impacts but to more favorable impact angles and lower impact speeds.","pith_inferences":["A natural testable extension is to build or simulate the optimized bend with Lagrangian particle tracking or experiments, since the 76% figure is a prediction of an Eulerian model that itself overpredicts small-particle impacts at the validation stage.","The adjoint-derivative structure suggests the same shape optimization could be run for other target Stokes numbers or for a multi-species cost; the reported persistence of improvement across Stokes numbers hints that a single robust shape may serve a whole particle-size distribution.","Because the cost functional can be modified without re-deriving the full system—only the erosion model's derivatives change—this framework could be coupled with surrogate or space-mapping strategies for turbulent erosion, which the paper names as future work.","The same shape calculus could be applied to multi-velocity or moment-based particle models once their adjoints are derived, potentially addressing particle trajectory crossing that the single-velocity model cannot represent."],"forward_implications":["If the central claim is right, pipe bends can be erosion-optimized with a gradient method rather than by geometrical intuition or trial-and-error parameter studies.","The optimized bend's benefit is not limited to the design particle size: predicted erosion falls by at least 20% for every Stokes number above 0.2 in the tested range.","Because only the partial derivatives of the erosion function enter the adjoint equations, the same derivative machinery applies to other erosion models (e.g., Finnie or E/CRC) with minor substitutions.","The improvement mechanism is identifiable: high impact rates, steep angles, and high speeds are spatially separated on the optimized wall, so erosion hot spots are dispersed.","The optimized geometry is locally optimal within the chosen deformation class, as indicated by the decrease of the cost and projected gradient norms over iterations."],"supporting_citations":[{"why":"supplies the Eulerian particle transport model used for the particulate phase.","marker":"[4]"},{"why":"supplies the finite element treatment and boundary conditions of the Eulerian impingement model.","marker":"[5]"},{"why":"supplies the Oka erosion model used in the cost functional.","marker":"[28]"},{"why":"supplies the material-dependent erosion parameters for stainless steel.","marker":"[27]"},{"why":"supplies the experimental 90° bend test case used for validation.","marker":"[31]"},{"why":"supplies reference impact-rate data and mesh-resolution guidance for the validation test case.","marker":"[46]"},{"why":"supplies Lagrangian/LES reference results for comparing predicted impact rates.","marker":"[6]"},{"why":"supplies the shape calculus framework used to derive the shape derivative.","marker":"[10]"},{"why":"supplies the restricted mesh deformation correction used in the gradient projection.","marker":"[17]"},{"why":"supplies the derivative of the Willmore functional used in the curvature regularization.","marker":"[3]"}],"fun_headline_variants":["Adjoint-based shape design cuts pipe bend erosion by 76%","Gradient-optimized pipe bend reduces erosion 20%+ for all tested particle sizes","Eulerian adjoint shape optimization cuts pipe erosion 76%","Pipe bend optimized cuts erosion 76% and 20%+ for other sizes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The entire 76% reduction is computed with a simplified particle model that, by the authors' own validation, overpredicts impact rates for the very small particles around the design condition; if that error grows after the geometry changes, the optimized bend may not be physically better.","fun_headline_variants_meta":{"raw":{"variants":["Adjoint-based shape design cuts pipe bend erosion by 76%","Gradient-optimized pipe bend reduces erosion 20%+ for all tested particle sizes","Eulerian adjoint shape optimization cuts pipe erosion 76%","Pipe bend optimized cuts erosion 76% and 20%+ for other sizes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001509,"raw_usage":{"total_tokens":6011,"prompt_tokens":869,"completion_tokens":5142,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":485,"completion_tokens_details":{"reasoning_tokens":5059}},"tokens_in":485,"tokens_out":5142,"duration_ms":34907,"temperature":1.0,"reasoning_tokens":5059,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:33:46.289925+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure or simulate wall erosion on the initial and optimized bends using Lagrangian particle tracking or an experimental aerosol flow with the design particle size (Stk ≈ 0.33). If the optimized geometry does not show substantially lower erosion than the initial bend—or if the gap is much smaller than the model's 76%—then the reported reduction is an artifact of the Eulerian approximation rather than a physical property of the shape.","supporting_citations":[{"cited_title":"Three-dimensional Eulerian approach to droplet impingement simu- lation using fensap-ice, part 1: model, algorithm, and validation","cited_arxiv_id":null,"evidence_quote":"supplies the Eulerian particle transport model used for the particulate phase."},{"cited_title":"A finite element method study of Eulerian droplets impingement models","cited_arxiv_id":null,"evidence_quote":"supplies the finite element treatment and boundary conditions of the Eulerian impingement model."},{"cited_title":"Practical estimation of erosion damage caused by solid particle impact: Part 1: Effects of impact parameters on a predictive equation","cited_arxiv_id":null,"evidence_quote":"supplies the Oka erosion model used in the cost functional."},{"cited_title":"The impact angle dependence of erosion damage caused by solid particle impact","cited_arxiv_id":null,"evidence_quote":"supplies the material-dependent erosion parameters for stainless steel."},{"cited_title":"Experimental study of particle deposition in bends of circular cross section","cited_arxiv_id":null,"evidence_quote":"supplies the experimental 90° bend test case used for validation."},{"cited_title":"Analysis of particle transport and deposition of micron-sized particles in a 90 ◦ bend using a two-fluid Eulerian–Eulerian approach","cited_arxiv_id":null,"evidence_quote":"supplies reference impact-rate data and mesh-resolution guidance for the validation test case."},{"cited_title":"Prediction of aerosol deposition in 90 ◦ bends using LES and an efficient Lagrangian tracking method","cited_arxiv_id":null,"evidence_quote":"supplies Lagrangian/LES reference results for comparing predicted impact rates."},{"cited_title":"Shapes and geometries: metrics, analysis, differential calculus, and optimization","cited_arxiv_id":null,"evidence_quote":"supplies the shape calculus framework used to derive the shape derivative."},{"cited_title":"First and Second Order Shape Optimization based on Restricted Mesh Deformations","cited_arxiv_id":"1810.10313","evidence_quote":"supplies the restricted mesh deformation correction used in the gradient projection."},{"cited_title":"Parametric FEM for geometric biomembranes","cited_arxiv_id":null,"evidence_quote":"supplies the derivative of the Willmore functional used in the curvature regularization."}],"review_version":1}