{"id":"5e55e6f4-05de-4690-91bf-cdfe469ce66b","arxiv_id":"2412.07960","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A new smoothing method, GRASHS, extends the RASHS trajectory optimization approach to handle arbitrary boolean logic (AND, OR, NOT) by converting it to disjunctive normal form and embedding it in smooth equations of motion.","lead":"This paper introduces a way to optimize multi-phase trajectories, like a Mars landing, when the switch between flight phases is controlled by any combination of logical conditions. It smooths those logical conditions into continuous math, turning a hard multi-point boundary value problem into a simpler two-point problem.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Unproven convergence to the OR-logic optimum plus circular validation leaves the central GRASHS optimality claim unsupported.","rationale":"The reader identified the load-bearing premise as the unproven convergence of the smoothed TPBVP to the original OR-logic problem, and I agree. The paper's strongest claim, that GRASHS 'embeds' arbitrary discrete logic and implicitly satisfies interior-point conditions, is exactly what would need to be proven or independently validated. The paper provides neither a convergence theorem nor an independent benchmark; the comparison with RASHS and MPBVP is circular because those comparison solutions are derived from GRASHS's own trigger selection. This is not an external-consensus disagreement; it is a missing correctness argument internal to the paper's logic. The proposed test, solving the alternate trigger branch and comparing costs, would directly reveal whether GRASHS selects the truly optimal mode sequence. Given these gaps, the conditional verdict remains appropriate: the method is plausible and the numerics are consistent, but the central optimality claim is not yet fully supported. No change to the reader's verdict is needed.","tokens_in":22015,"tokens_out":6288,"duration_ms":65137,"concrete_test":"For each mission profile, solve the alternative trigger assumption that GRASHS did not return: for Profile 1, an altitude-triggered RASHS/MPBVP with hP enforced and vP ignored; for Profile 2, a velocity-triggered RASHS/MPBVP with vP enforced and hP ignored. Compare the true optimal cost and trajectory of each alternative with the GRASHS solution. If GRASHS matches the lower-cost branch (or the alternative is infeasible), the OR-logic optimality claim gains support; if the alternative has strictly lower cost, GRASHS converged to a suboptimal mode sequence and the central claim fails. As a stronger independent check, solve the same EDL problem with a direct transcription method that encodes the OR logic explicitly (for example, by enumerating both mode sequences or using binary variables) and compare against GRASHS.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that Eq. (13) reduces the MPBVP to a TPBVP whose solution, as s and zeta grow, approximates the true optimum of the original OR-logic problem. This claim rests on two connected gaps. First, Section 2 asserts without proof that s and zeta can be made 'arbitrarily large' to make the TPBVP 'an arbitrarily close approximation' of the MPBVP; no theorem, error bound, or analysis of the co-state jump conditions in the limit is provided. The smoothed problem is a different optimal control problem, and it is not shown that its solutions satisfy the interior-point conditions of Eq. (19) when the active OR minterm is unknown. Second, the numerical validation in Sections 5.2 and 5.3 is circular: the RASHS and MPBVP comparison solutions are built using the trigger knowledge obtained from the GRASHS solution itself, as the paper states explicitly. Agreement therefore only demonstrates consistency with the mode sequence GRASHS happened to select; it cannot detect a wrong mode sequence. A concrete failure mode is that the tanh-saturated OR weight can activate a segment when no minterm is fully true but the sum of sigmoid products is positive, and homotopy may drive the solution to a suboptimal branch that the circular benchmark would still confirm.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a Generalized Relaxed Autonomously Switched Hybrid System (GRASHS) approach for indirect multi-phase trajectory optimization when the active-segment logic is an arbitrary Boolean expression over predicates on states and time. The method converts each segment's Boolean condition to disjunctive normal form, smooths each predicate with a sigmoid, represents the OR of minterms as the hyperbolic tangent of their sum, and embeds the smoothed switching weights into the equations of motion and Lagrangian. This reduces the necessary conditions of optimality from a multi-point boundary value problem (MPBVP) to a two-point boundary value problem (TPBVP). The approach is demonstrated on a Mars entry, descent, and landing problem with a parachute-deployment condition that combines AND and OR logic, for two mission profiles in which the trigger is either velocity-first or altitude-first. The GRASHS solutions are compared with RASHS and MPBVP solutions that are built using the trigger information obtained from the GRASHS solution.","tokens_in":22274,"tokens_out":5305,"duration_ms":55676,"significance":"If the central claim is correct, GRASHS would fill a genuine gap in indirect optimization of autonomously switched systems, extending RASHS from AND-only to arbitrary Boolean logic without requiring a priori knowledge of the active trigger. The construction in Section 2 is mathematically clean: DNF conversion, sigmoid smoothing of predicates, and tanh saturation of the OR are all well-defined and the resulting equations are smooth. The paper also provides a demanding aerospace test case with two distinct trigger modes and reproduces the expected co-state jump pattern in the numerical results. However, the work's significance is strongly tempered by two gaps: there is no proof or quantitative error bound for convergence of the smoothed optimal solution to the original OR-logic optimum, and the numerical validation is circular because the comparison baselines are derived from the GRASHS solution itself. As a result, the paper establishes that the smoothed problem is solvable and self-consistent, but it does not yet establish that the GRASHS solution is optimal for the original arbitrary-logic problem.","major_comments":[{"comment":"The central assertion that increasing s and zeta \"arbitrarily large values\" makes the TPBVP \"an arbitrarily close approximation of the MPBVP\" is stated without proof. No theorem or error bound is given for the convergence of the smoothed optimal controls, trajectories, or co-states to the solution of the original OR-logic problem. In particular, the paper does not show that the interior-point jump conditions of Eq. (19) are recovered in the limit, nor that the sequence of minimizers of the smoothed problems converges to a minimizer of the original problem. This is load-bearing because the claimed reduction from MPBVP to TPBVP for arbitrary discrete logic rests entirely on this convergence. The authors should provide a rigorous convergence analysis, an epi-convergence argument, or a carefully constructed numerical convergence study that explicitly demonstrates the approach of co-state jumps and mode sequences to the original problem as s and zeta grow.","section":"Section 2, after Eq. (13)"},{"comment":"The validation is circular. The RASHS and MPBVP comparison solutions are constructed by removing the OR-logic trigger that was identified from the GRASHS solution; the paper states this explicitly when it says the interior-point boundary conditions \"will be obtained from the solution of the GRASHS approach\" (Section 3.1) and when it removes hP or vP based on that knowledge (Sections 5.2 and 5.3). Consequently, agreement between GRASHS and these baselines only demonstrates consistency with a mode sequence that GRASHS itself selected; it cannot detect whether GRASHS picked the wrong trigger or whether the solution satisfies the true OR-logic optimality conditions. The original MPBVP with the full OR-logic interior-point conditions is never formulated or solved. The examples therefore do not independently confirm the central claim. A non-circular validation would compare against a direct method (e.g., SCP with state-triggered constraints), a combinatorial search over all feasible trigger orders, or an MPBVP that includes the genuinely disjunctive switching conditions.","section":"Sections 5.2 and 5.3"},{"comment":"The smoothed dynamics in Eq. (13) do not form a partition of unity over the m flight segments. For finite s and zeta, the sum of the smoothed segment weights can be greater than or less than 1 in transition regions, because the weights are independent sigmoid products and tanh-saturated sums rather than normalized convex weights. This means the smoothed equations of motion are not a convex combination of the segment dynamics, and the implied co-state dynamics may not correspond to any true switching system even in the limit of large slopes. The paper does not address this non-unit partition or quantify its effect on the claimed implicit interior-point conditions. The authors should either prove that the error vanishes as s and zeta tend to infinity in the appropriate sense or reformulate the smoothing so that the weights sum to one identically.","section":"Section 2, Eq. (13)"}],"minor_comments":[{"comment":"The product index in the smoothed equations uses \"gi,k\" instead of \"gi,j,k\"; the subscript j is missing.","section":"Eq. (13)"},{"comment":"The paper contains several typographical inconsistencies, including \"GRASH\" in Section 5.2 where \"GRASHS\" is meant, and reference entries such as \"665, 683\" with a comma instead of a range dash.","section":"General"},{"comment":"The notation tP DI in Eq. (27) is used without a clear definition of the subscript spacing; it should be typeset consistently as tPDI or t_PDI throughout.","section":"Section 5.1"},{"comment":"Several reference entries have formatting issues, including missing spaces after commas and inconsistent page ranges, which would need to be corrected for journal production.","section":"References"},{"comment":"The homotopy description would benefit from a figure or pseudocode that clarifies the five steps and the exact scheduling of s, zeta, and the boundary-condition continuation, since this is the practical means by which the claimed convergence is achieved.","section":"Section 3.2"}],"recommendation":"major_revision","confidential_remarks":"The paper is a plausible extension of RASHS, but the two gaps identified - the missing convergence proof and the circular validation - are substantial. The numerical results themselves would be convincing if the baselines were not built from the GRASHS solution. I would encourage the editor to seek a revised version that either supplies a convergence theorem or an independent validation, and that formally addresses the non-unit partition issue."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"GRASHS is a useful, well-motivated extension of RASHS. The DNF-plus-sigmoid/tanh construction is clean: AND becomes a product of sigmoids, OR becomes tanh of the sum, and the need to know which OR-trigger fires first disappears. The reduction from MPBVP to TPBVP is structurally sound, and the Mars EDL example does what it claims — the same smoothed equations produce two different trigger modes when the parachute deployment altitude changes, with no reformulation. The paper is also honest about the limits of its comparison: it states that the RASHS and MPBVP baselines are constructed using trigger knowledge obtained from the GRASHS solution.\n\nThe real soft spot is the missing convergence proof. The paper asserts (Section 2, after Eq. 13) that s and ζ can be made arbitrarily large to approximate the MPBVP arbitrarily well, but no theorem or error bound is given. The smoothed problem is a different optimal control problem; without a limit argument it is not established that its optimum, or even its mode sequence, is the original OR-logic optimum. This is a load-bearing gap, not a technicality.\n\nThe validation is weaker than it first appears because of the circularity: agreement with a baseline built from the GRASHS trigger only confirms consistency with the mode sequence GRASHS happened to select. It cannot detect a wrong mode sequence. The two-profile test is still a useful practical demonstration of the no-apriori-knowledge workflow, but it does not fill the proof gap. There is also a minor (but real) concern with the tanh OR smoothing: for finite s, each sigmoid product is never exactly zero, so the sum of minterms can be small positive even when no minterm is fully true; with large ζ, tanh can saturate and spuriously activate a segment. This deserves discussion.\n\nThe derivation itself is correct as far as it goes, and I have no reason to think the basic idea fails. The paper is a legitimate contribution to the indirect trajectory optimization literature. I would send it to peer review — the topic is timely, the writing is clear, and a good referee could request a convergence analysis or an independent benchmark (e.g., brute-force enumeration of all mode sequences) without asking for a rewrite from scratch. I'd cite it as related work in the meantime, but I would not rely on its optimality claim without further analysis.","headline":"GRASHS is a clean, useful extension of RASHS for OR logic, but the paper leaves the convergence question open and its validation is partly circular, so the optimality claim is not yet established.","tokens_in":22787,"tokens_out":4075,"would_cite":true,"duration_ms":40368,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["49K15","65L10"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that arbitrary boolean switching logic in multi-phase trajectory optimization can be smoothed into continuous dynamics, reducing the necessary conditions from a multi-point boundary value problem to a two-point boundary…","keywords":["trajectory optimization","indirect methods","entry, descent, and landing","hybrid systems","mixed integer programming","GRASHS","sigmoid smoothing","disjunctive normal form"],"falsifier":"Solve a minimal two-mode, two-trigger example, such as switching when $x<0$ or $y<0$, exactly as a multi-point boundary value problem and compare its states and co-state jumps with the GRASHS limit as $s$ and $\\zeta$ are increased; if the limits do not match for some initial and boundary data, the central convergence claim fails.","tokens_in":21725,"feed_emoji":"🚀","tokens_out":5746,"duration_ms":54069,"temperature":0.7,"pith_summary":"Multi-phase trajectories, such as a Mars entry, descent, and landing profile, switch between flight segments when conditions on velocity, altitude, or time are met. When those conditions involve OR logic, the standard indirect optimization approach forces the planner to guess which trigger fires first. This paper claims that arbitrary boolean logic can be converted to disjunctive normal form and smoothed, using products of sigmoids for AND and a hyperbolic tangent of the sum for OR, so that the piecewise equations of motion and Lagrangian become continuous and differentiable. The payoff is that the necessary conditions of optimality reduce from a multi-point boundary value problem, whose interior-point conditions are hard to guess and enforce, to a simpler two-point boundary value problem. In two Mars EDL example profiles, the same smoothed formulation recovers both a velocity-triggered and an altitude-triggered parachute deployment, matching the separately solved multi-point solution.","feed_headline":"Smoothing turns OR-logic flight switches into one continuous problem","feed_subtitle":"A new indirect method, GRASHS, handles arbitrary boolean switching conditions with no prior knowledge of which trigger fires first.","key_machinery":"The central object is Eq. (13), the smoothed switching-function embedding. It converts any boolean segment-activation expression to disjunctive normal form, represents each AND minterm as a product of sigmoids $1/(1+e^{s\\,g_{i,j,k}(X,t)})$, and represents the OR of minterms as $\\tanh\\!\\big(\\zeta\\sum_i \\prod_j [\\cdot]\\big)$. The sigmoid slope $s$ and the hyperbolic-tangent slope $\\zeta$ act as homotopy parameters pushed to large values, so the smooth two-point boundary value problem is asserted to approximate the original multi-point problem arbitrarily closely while the interior-point jump conditions are automatically encoded in the dynamics and cost.","core_discovery":"The paper's central discovery is that the discrete switching function for a flight segment can be represented exactly as the signum of a sum of products of horizontally flipped unit step functions, one product per minterm of the segment's boolean condition in disjunctive normal form, and then approximated smoothly. Replacing each unit-step factor by a sigmoid $1/(1+e^{s g})$ and the signum by $\\tanh(\\zeta \\cdot)$ yields Eq. (13), a smooth version of $\\dot{X}=\\sum_k \\xi_k f_k$ and $L=\\sum_k \\xi_k L_k$. Because the smoothed dynamics embed the switching conditions, the interior-point boundary conditions at segment transitions are implicitly satisfied, and the necessary conditions for optimality become a two-point boundary value problem. The paper demonstrates that, in a Mars EDL problem with parachute descent active when $(v<v_P \\text{ or } h<h_P)$ and $h\\ge h_{PDI}$, the same GRASHS equations reproduce the velocity-triggered solution when $h_P=3.5$ km and the altitude-triggered solution when $h_P=6.5$ km, without any prior knowledge of which trigger fires first.","pith_inferences":["If the convergence of the smoothed problem to the discontinuous problem as $s,\\zeta\\to\\infty$ were proven with an error bound, the same embedding could likely be extended to non-autonomous switched systems where the discrete mode is itself a control decision.","A natural testable extension is to study simple two-mode examples with overlapping OR triggers and compare the smoothed limit with the exact multi-point solution for all initial and boundary data, checking whether the limit always selects a physically valid mode sequence.","The paper's reliance on manual homotopy suggests that adaptive or stabilized continuation schemes could replace the hand-tuned homotopy steps and broaden the method's reliability.","Because the smoothing slopes act as continuation parameters, the method could also be used as a mechanism for homotopy between different trigger logics within one optimization framework."],"forward_implications":["The same GRASHS equations apply to any boolean combination of predicates across flight segments, including NOT operations, which are handled by replacing the predicate with its sign-flipped form.","Interior-point co-state jumps, which in the multi-point formulation must be explicitly computed through Lagrange multipliers, are produced automatically by the smoothed equations of motion and Lagrangian.","For EDL design, a planner no longer needs trial-and-error analysis of trigger thresholds to guarantee that every flight segment activates.","The GRASHS solution can serve as a high-quality initial guess for the exact multi-point boundary value problem, making the hard solve tractable.","The approach avoids the exponential growth in the number of trajectory optimizations that results from enumerating which OR-trigger fires first."],"supporting_citations":[{"why":"Introduces the RASHS sigmoid-smoothing approach for AND-only switching that GRASHS generalizes to arbitrary boolean logic.","marker":"Saranathan and Grant [2018a]"},{"why":"Supplies the disjunctive normal form representation used to convert arbitrary boolean logic into sums of products.","marker":"Pahl and Damrath [1984]"},{"why":"Provides the multi-point boundary value problem necessary conditions used as the exact comparison baseline.","marker":"Bryson and Ho [1975a]"},{"why":"Provides the two-point boundary value problem necessary conditions that the GRASHS-smoothed problem satisfies.","marker":"Bryson and Ho [1975b]"},{"why":"Supplies the hyperbolic tangent smoothing technique used to approximate the signum function in the OR logic.","marker":"Taheri and Junkins [2018]"},{"why":"Supplies the homotopy continuation framework used to evolve the solution while raising the smoothing slopes.","marker":"Grant and Braun [2015]"},{"why":"Supplies the boundary value problem solver used to compute the numerical solutions in the demonstration.","marker":"Kierzenka and Shampine [2001]"}],"fun_headline_variants":["GRASHS method smooths arbitrary boolean flight switches","OR-logic triggers now handled in indirect trajectory optimization","Mars EDL: boolean logic no longer blocks optimal paths","New smoothing unifies AND and OR switching into one BVP"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that letting the smoothing slopes $s$ and $\\zeta$ grow without bound makes the smoothed two-point problem's solution converge to the solution of the original piecewise multi-point problem; the paper asserts this via homotopy but supplies no theorem or error bound.","fun_headline_variants_meta":{"raw":{"variants":["GRASHS method smooths arbitrary boolean flight switches","OR-logic triggers now handled in indirect trajectory optimization","Mars EDL: boolean logic no longer blocks optimal paths","New smoothing unifies AND and OR switching into one BVP"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000349,"raw_usage":{"total_tokens":1966,"prompt_tokens":1063,"completion_tokens":903,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":679,"completion_tokens_details":{"reasoning_tokens":837}},"tokens_in":679,"tokens_out":903,"duration_ms":9545,"temperature":1.0,"reasoning_tokens":837,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T18:21:41.738345+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Solve a minimal two-mode, two-trigger example, such as switching when $x<0$ or $y<0$, exactly as a multi-point boundary value problem and compare its states and co-state jumps with the GRASHS limit as $s$ and $\\zeta$ are increased; if the limits do not match for some initial and boundary data, the central convergence claim fails.","supporting_citations":[{"cited_title":"Mathematical Foundations of Computational Engineering, volume II, chapter 1","cited_arxiv_id":null,"evidence_quote":"Supplies the disjunctive normal form representation used to convert arbitrary boolean logic into sums of products."},{"cited_title":"Rapid indirect trajectory optimization for conceptual design of hypersonic missions","cited_arxiv_id":null,"evidence_quote":"Supplies the homotopy continuation framework used to evolve the solution while raising the smoothing slopes."}],"review_version":1}