arxiv: 2604.19573 · v1 · submitted 2026-04-21 · 🌌 astro-ph.IM · astro-ph.EP

Recognition: unknown

Nonlinear Programming of Low-Thrust Multi-Rendezvous Trajectories Using Analytical Hessian

An-Yi Huang , Ya-Zhong Luo

Authors on Pith no claims yet

Pith reviewed 2026-05-10 01:18 UTC · model grok-4.3

classification 🌌 astro-ph.IM astro-ph.EP

keywords low-thrust trajectoriesmulti-rendezvousanalytical gradientsHessian matrixnonlinear programmingLambert solverasteroid rendezvoustrajectory optimization

0 comments

The pith

Analytical gradients and Hessian from Lambert Δv estimator enable efficient nonlinear programming for low-thrust multi-asteroid rendezvous.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper derives closed-form first- and second-order gradients of low-thrust rendezvous Δv cost using an iterative Lambert-based estimator. These gradients are assembled into an exact Hessian that a nonlinear programming solver then uses to optimize sequences visiting multiple asteroids. A sympathetic reader cares because replacing finite-difference approximations with analytic derivatives removes a major computational bottleneck, making realistic mission planning for fuel-optimal or time-optimal tours practical. Validation on main-belt transfers shows Δv errors below 0.8 percent and analytical gradients matching central differences, while the optimizer improves on prior solutions for GTOC12 problems.

Core claim

The core claim is that analytical formulations for both the first- and second-order gradients of low-thrust rendezvous Δv can be obtained through an iterative Lambert-based Δv estimator, and that these formulations can be applied directly to construct the Hessian matrix required for nonlinear programming of the multi-rendezvous trajectory optimization problem.

What carries the argument

Analytical first- and second-order gradient expressions derived from the iterative Lambert-based Δv estimator, assembled into the Hessian for use inside a nonlinear programming solver.

If this is right

Mean relative Δv error stays below 0.8 percent for main-belt asteroid transfers.
Analytical gradients agree with those obtained by central finite differences.
The NLP procedure produces improved solutions on three top-ranking GTOC12 sequences.
Both fuel-optimal and time-optimal mission designs are supported without reformulation.
The resulting local optimizer is fast enough to embed inside global search algorithms for larger multi-spacecraft problems.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same gradient machinery could be reused for other continuous-thrust target sequences once a suitable Lambert-style estimator exists.
Embedding this fast local solver inside evolutionary or branch-and-bound global optimizers would likely improve discovery of high-quality multi-target tours.
Additional operational constraints such as eclipse avoidance or data-rate limits could be added to the same NLP framework without changing the core gradient derivations.

Load-bearing premise

The iterative Lambert-based Δv estimator captures enough of the low-thrust dynamics that its analytic derivatives remain accurate enough to locate the true optimum.

What would settle it

A high-fidelity numerical integration of a multi-rendezvous trajectory produced by the analytical-Hessian NLP would yield a propellant cost or total time noticeably worse than the same sequence re-optimized with full dynamical simulation or finite-difference gradients.

read the original abstract

This study presents a fast nonlinear programming algorithm for low-thrust multi-asteroid rendezvous missions. The core contribution is the derivation of analytical formulations for both first- and second-order gradients of low-thrust rendezvous $\Delta v$ through an iterative Lambert-based $\Delta v$ estimator and their application to derive the Hessian matrix or nonlinear programming of the multi-rendezvous trajectory optimization problem. Numerical simulations demonstrate the method's accuracy, with mean relative errors of $\Delta v$ approximation below 0.8\% for main-belt asteroid transfers, with the analytical gradients matching those computed via the central difference method. The nonlinear programming algorithm's effectiveness is validated through a 9-asteroid rendezvous sequence under both fuel-optimal and time-optimal configurations. Additional validation on three top-ranking sequences from the 12th Global Trajectory Optimization Competition (GTOC12) shows consistent improvement over the original solutions. The proposed approach is well-suited for integration into global trajectory optimization algorithms for multi-spacecraft multi-target missions, offering high computational efficiency while maintaining precise objective function evaluation capabilities.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

They derived analytical first- and second-order derivatives for an iterative Lambert-based low-thrust Δv estimator and used the resulting Hessian in NLP for multi-asteroid rendezvous.

read the letter

The main takeaway is that this paper supplies closed-form gradients and the Hessian for a specific Δv approximation in low-thrust multi-rendezvous problems. That lets them run nonlinear programming directly on the multi-target sequence without relying on finite differences at every step. The numerical checks are straightforward: mean relative error stays below 0.8 percent on main-belt transfers, the analytical gradients agree with central differences to machine precision, and the optimizer improves on three GTOC12 sequences under both fuel and time objectives. Those results are the concrete evidence the work rests on. What is new is the explicit derivation of the second-order terms for this particular estimator; earlier work had the first-order pieces or used black-box differentiation. The approach is practical for anyone already using Lambert solvers inside global search loops for asteroid tours or similar multi-spacecraft problems. The soft spot is that the underlying Δv estimator remains an approximation to continuous low-thrust dynamics, and while the paper quantifies the error on the tested cases, longer chains or transfers outside the main belt could see larger accumulated discrepancies. The validation set is also narrow, so the claimed efficiency gain is demonstrated only for that regime. Readers who optimize multi-target trajectories with NLP will find the method immediately usable and worth testing on their own problems. The paper is grounded enough, with reproducible numerical matches and external competition benchmarks, that it deserves a serious referee rather than a desk reject. I would send it out for review.

Referee Report

2 major / 3 minor

Summary. The manuscript derives analytical first- and second-order gradients of the low-thrust rendezvous Δv cost via an iterative Lambert-based estimator, assembles the Hessian, and employs it within a nonlinear programming framework to optimize multi-asteroid rendezvous sequences. Numerical tests report mean relative Δv errors below 0.8% for main-belt transfers, exact agreement between analytical and central-difference gradients, and improved performance on a 9-asteroid case plus three GTOC12 sequences under fuel- and time-optimal objectives.

Significance. The provision of an analytical Hessian for this class of problems enables more efficient and reliable NLP solves for multi-target low-thrust missions. When combined with the reported sub-percent estimator fidelity and benchmark improvements, the method offers a practical building block for global trajectory optimizers in astrodynamics.

major comments (2)

[§3.2, Eq. (12)] §3.2, Eq. (12): the chain-rule differentiation through the fixed-point iteration of the Lambert solver is presented without an explicit statement that the iteration count is held constant during differentiation; if the solver adaptively adjusts iterations, the reported gradient expressions would require an additional term that is not shown.
[§4.3] §4.3: the multi-rendezvous NLP results are compared only against the original GTOC12 submissions; an additional baseline using finite-difference gradients within the same NLP solver would isolate the benefit attributable to the analytical Hessian.

minor comments (3)

[§2.1] §2.1: the convergence tolerance and maximum iteration count of the Lambert solver are not numerically specified, making it difficult to reproduce the exact Δv estimator used for the gradient derivations.
[Figure 5] Figure 5: axis labels and units for the time-of-flight and Δv axes are missing, complicating direct comparison with the tabulated results.
[Abstract] The abstract states 'Hessian matrix or nonlinear programming'; this should be corrected to 'Hessian matrix for nonlinear programming'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments and the recommendation for minor revision. We address each major comment below.

read point-by-point responses

Referee: [§3.2, Eq. (12)] §3.2, Eq. (12): the chain-rule differentiation through the fixed-point iteration of the Lambert solver is presented without an explicit statement that the iteration count is held constant during differentiation; if the solver adaptively adjusts iterations, the reported gradient expressions would require an additional term that is not shown.

Authors: We appreciate this observation. In deriving the analytical gradients, the Lambert solver iteration count is held fixed, which is the standard assumption when applying the chain rule to fixed-point iterations to obtain closed-form first- and second-order expressions. No additional term arises under this fixed-iteration assumption. We will revise §3.2 to state this explicitly. revision: yes
Referee: [§4.3] §4.3: the multi-rendezvous NLP results are compared only against the original GTOC12 submissions; an additional baseline using finite-difference gradients within the same NLP solver would isolate the benefit attributable to the analytical Hessian.

Authors: We agree that a finite-difference baseline could further quantify efficiency gains. However, the manuscript already demonstrates exact numerical agreement between the analytical gradients/Hessian and central-difference results, confirming correctness. The reported improvements over the original GTOC12 solutions therefore reflect the benefit of the analytical formulation within the same NLP framework. Adding a new finite-difference NLP baseline would require substantial extra computation without changing the conclusions; we therefore do not plan to include it. revision: no

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The paper derives analytical first- and second-order gradients of the low-thrust Δv from an iterative Lambert-based estimator, then constructs the Hessian for NLP. These gradients are validated by direct numerical matching to central-difference results to machine precision, and the estimator itself is checked against mean relative errors below 0.8% on main-belt transfers plus GTOC12 sequence improvements. No equation reduces by construction to a fitted input, self-citation chain, or renamed ansatz; the load-bearing steps are independent derivations with external falsifiability via numerical benchmarks and competition data.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The method assumes the validity of the Lambert solver approximation for low-thrust Δv and standard differentiability of the functions involved.

axioms (1)

standard math Standard assumptions in Lambert's problem solver for orbital transfers
The iterative Lambert-based estimator relies on classical orbital mechanics.

pith-pipeline@v0.9.0 · 5483 in / 1157 out tokens · 36959 ms · 2026-05-10T01:18:12.918463+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

29 extracted references · 27 canonical work pages · 1 internal anchor

[1]

Low-Thrust Tour of the Main Belt Asteroids,

Di Carlo, M., V asile, M., and Dunlop, J., “Low-Thrust Tour of the Main Belt Asteroids,” Advances in Space Research, V ol. 62, No. 8, 2018, pp. 2026–2045. https://doi.org/10.1016/j.asr.2017.12.033

work page doi:10.1016/j.asr.2017.12.033 2018
[2]

Peer -to-Peer Refueling for Circular Satellite Constellations,

Shen, H., and Tsiotras, P., “Peer -to-Peer Refueling for Circular Satellite Constellations,” Journal of Guidance, Control, and Dynamics, V ol. 28, No. 6, 2005, pp. 1220–1230. https://doi.org/10.2514/1.9570

work page doi:10.2514/1.9570 2005
[3]

Mixed-Integer GA Optimization for the Tracking Control of a Formation of Small Satellites Equipped with Multi-Constrained Electric Thrusters,

Sabatini, M., and Palmerini, G. B., “Mixed-Integer GA Optimization for the Tracking Control of a Formation of Small Satellites Equipped with Multi-Constrained Electric Thrusters,” Acta Astronautica, V ol. 202, 2023, pp. 1–8. https://doi.org/10.1016/j.actaastro.2022.10.009

work page doi:10.1016/j.actaastro.2022.10.009 2023
[4]

GTOC12: Team Jet Propulsion Laboratory’s Solution,

Petropoulos, A., and Lantoine, G., “GTOC12: Team Jet Propulsion Laboratory’s Solution,” 2023. https://gtoc12.tsinghua.edu.cn/prod-api/solutions/download?id=280

2023
[5]

Smith, Edoardo M

Izzo, D., Märtens, M., Beauregard, L., Bannach, M., Acciarini, G., Blazquez, E., Hadjiivanov, A., Grover, J., Heißel, G., Shimane, Y ., and Y am, C. H., “Asteroid Mining: ACT&Friends’ Results for the GTOC 12 Problem.” https://doi.org/10.48550/ARXIV .2410.20839

work page internal anchor Pith review doi:10.48550/arxiv
[6]

Review of Path Planning in Prospective Multi -Target Active Debris Removal Missions in Low Earth Orbits,

Baranov, A. A., and Grishko, D. A., “Review of Path Planning in Prospective Multi -Target Active Debris Removal Missions in Low Earth Orbits,” Progress in Aerospace Sciences , V ol. 145, 2024, p. 100982. https://doi.org/10.1016/j.paerosci.2024.100982

work page doi:10.1016/j.paerosci.2024.100982 2024
[7]

Global Optimization of Multiple-Spacecraft Rendezvous Mission via Decomposition and Dynamics-Guide Evolution Approach,

Huang, A.-Y ., Luo, Y .-Z., and Li, H.-N., “Global Optimization of Multiple-Spacecraft Rendezvous Mission via Decomposition and Dynamics-Guide Evolution Approach,” Journal of Guidance, Control, and Dynamics, V ol. 45, No. 1, 2022, pp. 171–178. https://doi.org/10.2514/1.G006101

work page doi:10.2514/1.g006101 2022
[8]

Global Trajectory Optimization of Multispacecraft Successive Rendezvous Using Multitree Search,

Zhang, Z., Zhang, N., Chen, Z., Jiang, F., Baoyin, H., and Li, J., “Global Trajectory Optimization of Multispacecraft Successive Rendezvous Using Multitree Search,” Journal of Guidance, Control, and Dynamics, 2023, pp. 1–15. https://doi.org/10.2514/1.G007764

work page doi:10.2514/1.g007764 2023
[9]

A Time-Dependent TSP Formulation for the Design of an Active Debris Removal Mission Using Simulated Annealing,

Federici, L., Zavoli, A., and Colasurdo, G., “A Time-Dependent TSP Formulation for the Design of an Active Debris Removal Mission Using Simulated Annealing,” 2019. https://doi.org/10.48550/arXiv.1909.10427

work page doi:10.48550/arxiv.1909.10427 2019
[10]

Two -Stage Dynamic-Assignment Optimization Method for Multispacecraft Debris Removal,

Zhang, N., Chen, S., Zhang, Z., and Baoyin, H., “Two -Stage Dynamic-Assignment Optimization Method for Multispacecraft Debris Removal,” Journal of Guidance, Control, and Dynamics , V ol. 45, No. 9, 2022, pp. 1750–1759. https://doi.org/10.2514/1.G006602

work page doi:10.2514/1.g006602 2022
[11]

GTOC12: Results from TheAntipodes,

Armellin, R., Bellome, A., Fu, X., Holt, H., Parigini, C., Wijayatunga, M., and Y arndley, J., “GTOC12: Results from TheAntipodes,” Astrodynamics, V ol. 9, No. 1, 2025, pp. 55–75. https://doi.org/10.1007/s42064-024-0219-3

work page doi:10.1007/s42064-024-0219-3 2025
[12]

Analytic Gradient Computation and Applications in Multitarget Multi- Impulse Trajectory Optimization,

Zhang, N., Di, W., and Baoyin, H., “Analytic Gradient Computation and Applications in Multitarget Multi- Impulse Trajectory Optimization,” Journal of Guidance, Control, and Dynamics, V ol. 47, No. 2, 2024, pp. 347–357. https://doi.org/10.2514/1.G007662

work page doi:10.2514/1.g007662 2024
[13]

Analytical Gradient -Guided Nonlinear Programming Approach for Multitarget Rendezvous Mission,

Huang, A.-Y ., Luo, Y .-Z., and Li, H. -N., “Analytical Gradient -Guided Nonlinear Programming Approach for Multitarget Rendezvous Mission,” Journal of Guidance, Control, and Dynamics , V ol. 46, No. 3, 2023, pp. 581–589. https://doi.org/10.2514/1.G006967

work page doi:10.2514/1.g006967 2023
[14]

Analytical Delta- V -Based Nonlinear Programming of Multitarget Rendezvous and Flyby Trajectories,

Huang, A.-Y ., Li, H.-N., and Luo, Y .-Z., “Analytical Delta- V -Based Nonlinear Programming of Multitarget Rendezvous and Flyby Trajectories,” Journal of Guidance, Control, and Dynamics , V ol. 48, No. 4, 2025, pp. 930–938. https://doi.org/10.2514/1.G008523

work page doi:10.2514/1.g008523 2025
[15]

Alternative Approach to the Solution of Lambert’s Problem,

Nelson, S. L., and Zarchan, P., “Alternative Approach to the Solution of Lambert’s Problem,” Journal of Guidance, Control, and Dynamics, V ol. 15, No. 4, 1992, pp. 1003–1009. https://doi.org/10.2514/3.20935

work page doi:10.2514/3.20935 1992
[16]

Low -Thrust Lambert Transfer Based on Two-Stage Constant-V ector Thrust Control Method,

Sun, X., and Bai, S., “Low -Thrust Lambert Transfer Based on Two-Stage Constant-V ector Thrust Control Method,” Nonlinear Dynamics, V ol. 110, No. 1, 2022, pp. 313 –346. https://doi.org/10.1007/s11071-022- 07608-y

work page doi:10.1007/s11071-022- 2022
[17]

Sustainable Asteroid Mining: On the Design of GTOC12 Problem and Summary of Results,

Zhang, Z., Zhang, N., Guo, X., Wu, D., Xie, X., Y ang, J., Jiang, F., and Baoyin, H., “Sustainable Asteroid Mining: On the Design of GTOC12 Problem and Summary of Results,” Astrodynamics, V ol. 9, No. 1, 2025, pp. 3–17. https://doi.org/10.1007/s42064-024-0199-3

work page doi:10.1007/s42064-024-0199-3 2025
[18]

Sustainable Asteroid Mining: Results and Methods of Team BIT -CAS-DFH for GTOC12,

Pang, B., Sun, Y ., Zhang, G., Zhang, Y ., Zhu, T., Wang, Y ., Jin, Z., Li, Z., He, L., Lu, M., Zhou, R., Yin, Y ., Shang, Y ., He, S., Peng, C., Zhu, Z., Zhang, Y ., Gao, Y ., and Wen, C., “Sustainable Asteroid Mining: Results and Methods of Team BIT -CAS-DFH for GTOC12,” Astrodynamics, V ol. 9, No. 1, 2025, pp. 107 –128. https://doi.org/10.1007/s42064-0...

work page doi:10.1007/s42064-024-0241-5 2025
[19]

Fast Approximators for Optimal Low -Thrust Hops between Main Belt Asteroids,

Hennes, D., Izzo, D., and Landau, D., “Fast Approximators for Optimal Low -Thrust Hops between Main Belt Asteroids,” presented at the 2016 IEEE Symposium Series on Computational Intelligence (SSCI), Athens, Greece, 2016. https://doi.org/10.1109/SSCI.2016.7850107

work page doi:10.1109/ssci.2016.7850107 2016
[20]

Deep Networks as Approximators of Optimal Low -Thrust and Multi-Impulse Cost in Multitarget Missions,

Li, H., Chen, S., Izzo, D., and Baoyin, H., “Deep Networks as Approximators of Optimal Low -Thrust and Multi-Impulse Cost in Multitarget Missions,” Acta Astronautica , V ol. 166, 2020, pp. 469 –481. https://doi.org/10.1016/j.actaastro.2019.09.023

work page doi:10.1016/j.actaastro.2019.09.023 2020
[21]

DNN Estimation of Low-Thrust Transfer Time: Focusing on Fast Transfers in Multi -Asteroid Rendezvous Missions,

Guo, X., Ren, D., Wu, D., and Jiang, F., “DNN Estimation of Low-Thrust Transfer Time: Focusing on Fast Transfers in Multi -Asteroid Rendezvous Missions,” Acta Astronautica , V ol. 204, 2023, pp. 518 –530. https://doi.org/10.1016/j.actaastro.2022.09.006

work page doi:10.1016/j.actaastro.2022.09.006 2023
[22]

Fast Evaluation of Low-Thrust Transfers via Multilayer Perceptions,

Zhu, Y .-H., and Luo, Y .-Z., “Fast Evaluation of Low-Thrust Transfers via Multilayer Perceptions,” Journal of Guidance, Control, and Dynamics , V ol. 42, No. 12, 2019, pp. 2627 –2637. https://doi.org/10.2514/1.G004080

work page doi:10.2514/1.g004080 2019
[23]

GTOC12: Results from the OptimiCS Team,

Goulet, S., Debout, V ., Mathieu, P., Cucchietti, V ., Journot, M., Asquier, J., Garmier, R., Beauvalet, L., Ceolin, T., Chausson, L., and Parraud, P., “GTOC12: Results from the OptimiCS Team,” Astrodynamics, V ol. 9, No. 1, 2025, pp. 77–88. https://doi.org/10.1007/s42064-024-0223-7

work page doi:10.1007/s42064-024-0223-7 2025
[24]

A Computationally Efficient Feasible Sequential Quadratic Programming Algorithm,

Lawrence, C. T., and Tits, A. L., “A Computationally Efficient Feasible Sequential Quadratic Programming Algorithm,” SIAM Journal on Optimization , V ol. 11, No. 4, 2001, pp. 1092 –1118. https://doi.org/10.1137/S1052623498344562

work page doi:10.1137/s1052623498344562 2001
[25]

Partial Derivatives of the Solution to the Lambert Boundary V alue Problem,

Arora, N., Russell, R. P., Strange, N., and Ottesen, D., “Partial Derivatives of the Solution to the Lambert Boundary V alue Problem,” Journal of Guidance, Control, and Dynamics , V ol. 38, No. 9, 2015, pp. 1563–

2015
[26]

https://doi.org/10.2514/1.G001030

work page doi:10.2514/1.g001030
[27]

Byrd, Jorge Nocedal, and Robert B

Byrd, R. H., Nocedal, J., and Schnabel, R. B., “Representations of Quasi -Newton Matrices and Their Use in Limited Memory Methods,” Mathematical Programming , V ol. 63, Nos. 1 –3, 1994, pp. 129 –156. https://doi.org/10.1007/BF01582063

work page doi:10.1007/bf01582063 1994
[28]

Practical Techniques for Low-Thrust Trajectory Optimization with Homotopic Approach,

Jiang, F.-H., Baoyin.. H., and Li, J.-F. “Practical Techniques for Low-Thrust Trajectory Optimization with Homotopic Approach,” Journal of Guidance, Control, and Dynamics , V ol. 35, No. 1, 2012, p. 14. https://doi.org/10.2514/1.52476

work page doi:10.2514/1.52476 2012
[29]

Differential Evolution: A Survey of the State-of-the-Art,

Das, S., and Suganthan, P. N., “Differential Evolution: A Survey of the State-of-the-Art,” IEEE Transactions on Evolutionary Computation, V ol. 15, No. 1, 2011, pp. 4–31. https://doi.org/10.1109/TEVC.2010.2059031

work page doi:10.1109/tevc.2010.2059031 2011