pith. machine review for the scientific record. sign in

arxiv: 2605.09125 · v1 · submitted 2026-05-09 · 📡 eess.SY · cs.LG· cs.SY· math.OC

Recognition: 2 theorem links

· Lean Theorem

Transfer Learning of Multiobjective Indirect Low-Thrust Trajectories Using Diffusion Models and Markov Chain Monte Carlo

Jannik Graebner , Ryne Beeson
Authors on Pith no claims yet

Pith reviewed 2026-05-12 02:19 UTC · model grok-4.3

classification 📡 eess.SY cs.LGcs.SYmath.OC
keywords transfer learningdiffusion modelsMarkov chain Monte Carlolow-thrust trajectoriesindirect optimal controlmultiobjective optimizationhomotopyspacecraft mission design
0
0 comments X

The pith

Transfer learning with MCMC-sampled costates and a mass-conditioned diffusion model generates 40 percent more feasible low-thrust trajectories and a better Pareto front than standard indirect methods.

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

The paper shows how to reuse past trajectory data across different spacecraft masses by combining homotopy in the mass parameter with Markov chain Monte Carlo sampling in costate space. This produces training sets that let a diffusion model learn the distribution of high-quality initial costates, which are then used to generate new solutions rapidly. The approach converts the multiobjective optimal-control problem into sampling from an unnormalized target distribution, avoiding repeated expensive gradient-based solves for each new mass value. For a test planar multi-revolution transfer, the resulting framework yields substantially more feasible trajectories and improves the quality of the Pareto front relative to adjoint-control-transformation baselines.

Core claim

By performing MCMC sampling on an unnormalized target distribution defined in costate space and using homotopy in the system mass, the method efficiently assembles training data that fine-tunes a diffusion model conditioned on mass; the conditioned model then samples initial costates that, when integrated, produce feasible low-thrust trajectories at higher rates and with superior Pareto quality than a state-of-the-art indirect optimizer.

What carries the argument

MCMC sampling from an unnormalized target distribution in costate space, combined with mass-parameter homotopy, to generate training data for a mass-conditioned diffusion model.

If this is right

  • Gradient-based MCMC variants give the best trade-off between sample quality and computational cost for generating the training data.
  • The trained diffusion model produces a global representation of the solution distribution across the mass range.
  • New solutions for different mass values can be generated without repeating the full indirect optimization process.
  • The framework directly addresses the need for rapid generation of solutions when mission parameters are still uncertain.

Where Pith is reading between the lines

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

  • The same sampling-plus-transfer strategy could be applied to other varying parameters such as thrust level or transfer time without retraining from scratch.
  • The generated costate distributions might be used to initialize direct methods or reinforcement-learning policies for the same class of transfers.
  • If the MCMC bias remains small across wider parameter ranges, the approach could support on-board replanning when spacecraft mass changes due to propellant use.

Load-bearing premise

The MCMC samples drawn from the unnormalized target distribution in costate space are sufficiently unbiased and representative to serve as high-quality training data for the diffusion model.

What would settle it

Running the fine-tuned diffusion model on a new mass value outside the training range and finding that the fraction of feasible trajectories drops below the baseline indirect optimizer or that the Pareto front is dominated by the baseline.

read the original abstract

Preliminary low-thrust spacecraft mission design is a global search problem characterized by a complex solution landscape, multiple objectives, and numerous local minima. During this phase, mission parameters are often not yet fully defined, requiring new solutions to be generated at a high cadence across varying parameter values. When combined with the indirect approach to optimal control, diffusion models can accelerate this search by learning distributions that represent high-quality initial costates. However, generating training data remains expensive, and opportunities exist to better exploit past data. We propose a transfer-learning framework that combines homotopy in a mission parameter with Markov chain Monte Carlo (MCMC) to generate training data more efficiently. The approach reformulates a multiobjective optimization problem as sampling from an unnormalized target distribution in costate space. We compare three MCMC algorithms on a planar multi-revolution transfer in the circular restricted three-body problem, with homotopy in the system mass parameter. The results show that gradient-based MCMC variants achieve the best trade-off between sample quality and computational cost. For the test transfer, the proposed framework generates 40 % more feasible solutions and achieves a higher-quality Pareto front than a state-of-the-art indirect approach based on adjoint control transformations and gradient-based optimization. Finally, the MCMC-generated samples are used to fine-tune a diffusion model conditioned on the mass parameter, enabling it to learn a global representation of the underlying solution distribution and efficiently generate new solutions. These findings establish the transfer-learning framework as a practical method for efficiently solving indirect trajectory optimization problems with varying parameters.

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.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a transfer-learning framework for multiobjective indirect low-thrust trajectory optimization that combines homotopy continuation in the spacecraft mass parameter with MCMC sampling to generate training data for a mass-conditioned diffusion model. The multiobjective problem is reformulated as sampling from an unnormalized target distribution over initial costates. On a planar multi-revolution transfer in the CR3BP, gradient-based MCMC variants are reported to offer the best quality-cost trade-off; the resulting samples yield 40% more feasible solutions and a superior Pareto front relative to a state-of-the-art adjoint-control-transformation baseline. These MCMC samples are then used to fine-tune the diffusion model, enabling efficient generation of new trajectories across mass values.

Significance. If the central claims hold after verification, the work offers a practical route to accelerate preliminary low-thrust mission design by reusing past solutions via transfer learning and generative models, thereby reducing the repeated global-search cost when mission parameters vary. The integration of MCMC-generated costate distributions with diffusion models for indirect optimal control is a novel methodological contribution that could extend to other parameter-dependent trajectory problems.

major comments (2)
  1. [Methods (MCMC subsection)] Methods section on MCMC sampling: the central performance claim rests on the quality of samples drawn from the unnormalized target distribution in costate space, yet no convergence diagnostics (Gelman-Rubin statistic, effective sample size, autocorrelation times, or trace plots) or mixing assessments are reported. Without these, it remains unclear whether the empirical distribution is sufficiently unbiased to serve as reliable training data for the diffusion model, especially given the high-dimensional costate manifold and discrete homotopy steps in mass.
  2. [Results (comparison to baseline)] Results section (comparison paragraph and associated table/figure): the headline result of '40% more feasible solutions' and 'higher-quality Pareto front' is presented without the number of independent trials, the precise definition of feasibility, the quantitative metric used for Pareto-front quality (e.g., hypervolume indicator), or any statistical significance test. These omissions make it impossible to evaluate whether the reported advantage over the adjoint-control baseline is robust.
minor comments (2)
  1. [Problem formulation] The exact functional form of the unnormalized target density (including any objective-weighting or scaling constants) should be written explicitly, as these choices directly affect the sampled distribution.
  2. [Figures] Figures depicting Pareto fronts or sampled costate distributions would benefit from overlaying multiple independent runs or confidence bands to illustrate variability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which highlight important aspects of reproducibility and rigor in our MCMC and comparative results sections. We address each major comment below and will incorporate the suggested improvements in the revised manuscript.

read point-by-point responses

  1. Referee: [Methods (MCMC subsection)] Methods section on MCMC sampling: the central performance claim rests on the quality of samples drawn from the unnormalized target distribution in costate space, yet no convergence diagnostics (Gelman-Rubin statistic, effective sample size, autocorrelation times, or trace plots) or mixing assessments are reported. Without these, it remains unclear whether the empirical distribution is sufficiently unbiased to serve as reliable training data for the diffusion model, especially given the high-dimensional costate manifold and discrete homotopy steps in mass.

    Authors: We agree that convergence diagnostics are important for validating the MCMC samples used as training data. The original manuscript did not include them, relying instead on downstream performance metrics from the diffusion model and trajectory optimization. In the revision we will add effective sample sizes, autocorrelation times, and representative trace plots for the gradient-based MCMC variants to demonstrate mixing and convergence on the costate manifold. revision: yes

  2. Referee: [Results (comparison to baseline)] Results section (comparison paragraph and associated table/figure): the headline result of '40% more feasible solutions' and 'higher-quality Pareto front' is presented without the number of independent trials, the precise definition of feasibility, the quantitative metric used for Pareto-front quality (e.g., hypervolume indicator), or any statistical significance test. These omissions make it impossible to evaluate whether the reported advantage over the adjoint-control baseline is robust.

    Authors: We acknowledge that the comparison results would benefit from greater detail on experimental protocol and quantitative evaluation. The manuscript did not report the number of independent trials, an explicit feasibility definition, or a formal Pareto metric such as hypervolume. In the revision we will add the number of independent trials performed, a precise definition of feasibility (convergence of the two-point boundary-value problem within a specified tolerance), the hypervolume indicator for Pareto-front quality, and appropriate statistical significance tests across trials. revision: yes

Circularity Check

0 steps flagged

No significant circularity; results anchored by external baseline comparison

full rationale

The paper's derivation chain consists of reformulating the multiobjective indirect optimal control problem as MCMC sampling from an unnormalized target density in costate space (a standard technique), generating samples via homotopy in the mass parameter, training/fine-tuning a mass-conditioned diffusion model on those samples, and empirically comparing the resulting feasible solutions and Pareto fronts against a state-of-the-art adjoint-control baseline. No quoted step reduces by construction to its own inputs, no fitted parameters are relabeled as predictions, and no load-bearing claim rests solely on self-citation. The 40% feasibility gain and superior Pareto front are presented as measured outcomes relative to an independent external method, providing an external anchor rather than a tautology.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities can be identified. The target distribution is described as unnormalized but its exact construction is not detailed.

pith-pipeline@v0.9.0 · 5586 in / 1225 out tokens · 35313 ms · 2026-05-12T02:19:06.765239+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

47 extracted references · 47 canonical work pages · 2 internal anchors

  1. [1]

    Survey of Numerical Methods for Trajectory Optimization,

    J. T. Betts, “Survey of Numerical Methods for Trajectory Optimization,”Journal of Guidance, Control, and Dynamics, V ol. 21, No. 2, 1998, pp. 193–207, 10.2514/2.4231

    work page doi:10.2514/2.4231 1998

  2. [2]

    Low- Thrust Trajectory Design as a Constrained Global Optimization Problem

    C. H. Yam, D. D. Lorenzo, and D. Izzo, “Low-thrust trajectory design as a constrained global optimization problem,”Proceedings of the Institution of Mechanical Engineers, Part G: Journal of Aerospace Engineering, V ol. 225, No. 11, 2011, pp. 1243–1251, 10.1177/0954410011401686

    work page doi:10.1177/0954410011401686 2011

  3. [3]

    Global Search of Optimal Spacecraft Trajectories using Amortiza- tion and Deep Generative Models,

    R. Beeson, A. Li, and A. Sinha, “Global Search of Optimal Spacecraft Trajectories using Amortiza- tion and Deep Generative Models,” arXiv preprinthttps://arxiv.org/abs/2412.20023, 2024. 10.48550/arXiv.2412.20023

    work page doi:10.48550/arxiv.2412.20023 2024

  4. [4]

    Global Search for Optimal Low Thrust Spacecraft Trajectories Using Diffusion Models and the Indirect Method,

    J. Graebner and R. Beeson, “Global Search for Optimal Low Thrust Spacecraft Trajectories Using Diffusion Models and the Indirect Method,”The Journal of the Astronautical Sciences, V ol. 72, No. 6, 2025, p. 62, 10.1007/s40295-025-00535-1

    work page doi:10.1007/s40295-025-00535-1 2025

  5. [5]

    Deep Unsupervised Learning using Nonequi- librium Thermodynamics,

    J. Sohl-Dickstein, E. Weiss, N. Maheswaranathan, and S. Ganguli, “Deep Unsupervised Learning using Nonequi- librium Thermodynamics,”Proceedings of the 32nd International Conference on Machine Learning(F. Bach and D. Blei, eds.), V ol. 37 ofProceedings of Machine Learning Research, Lille, France, PMLR, 07–09 Jul 2015, pp. 2256–2265. 37

    work page 2015

  6. [6]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,”Proceedings of the 34th International Conference on Neural Information Processing Systems, NIPS ’20, Red Hook, NY , USA, Curran Associates Inc., 2020

    work page 2020

  7. [7]

    R. Douc, E. Moulines, P. Priouret, and P. Soulier,Markov Chains. Springer Cham, 1 ed., 2018, https://doi.org/10.1007/978-3-319-97704-1

    work page doi:10.1007/978-3-319-97704-1 2018

  8. [8]

    Primer Vector Theory Applied to Global Low-Thrust Trade Studies,

    R. P. Russell, “Primer Vector Theory Applied to Global Low-Thrust Trade Studies,”Journal of Guidance, Con- trol, and Dynamics, V ol. 30, No. 2, 2007, pp. 460–472, 10.2514/1.22984

    work page doi:10.2514/1.22984 2007

  9. [9]

    Muiltiobjective optimization using nondominated sorting in genetic algorithms,

    N. Srinivas and K. Deb, “Muiltiobjective optimization using nondominated sorting in genetic algorithms,”Evol. Comput., V ol. 2, Sept. 1994, p. 221–248, 10.1162/evco.1994.2.3.221

    work page doi:10.1162/evco.1994.2.3.221 1994

  10. [10]

    Meyarivan

    K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, “A fast and elitist multiobjective genetic algorithm: NSGA-II,” IEEE Transactions on Evolutionary Computation, V ol. 6, No. 2, 2002, pp. 182–197, 10.1109/4235.996017

    work page doi:10.1109/4235.996017 2002

  11. [11]

    Multiobjective optimization of low-thrust trajectories using a genetic algorithm hybrid,

    M. Vavrina and K. Howell, “Multiobjective optimization of low-thrust trajectories using a genetic algorithm hybrid,”AAS/AIAA Space Flight Mechanics Meeting, 2009, pp. 09–141

    work page 2009

  12. [12]

    Optimal multi-objective low-thrust spacecraft trajec- tories,

    V . Coverstone-Carroll, J. Hartmann, and W. Mason, “Optimal multi-objective low-thrust spacecraft trajec- tories,”Computer Methods in Applied Mechanics and Engineering, V ol. 186, No. 2, 2000, pp. 387–402, https://doi.org/10.1016/S0045-7825(99)00393-X

    work page doi:10.1016/s0045-7825(99)00393-x 2000

  13. [13]

    Optimal Interplanetary Spacecraft Trajectories via a Pareto Genetic Algorithm,

    J. W. Hartmann, V . L. Coverstone-Carroll, and S. N. Williams, “Optimal Interplanetary Spacecraft Trajectories via a Pareto Genetic Algorithm,”The Journal of the Astronautical Sciences, V ol. 46, No. 3, 1998, pp. 267–282, 10.1007/BF03546237

    work page doi:10.1007/bf03546237 1998

  14. [14]

    A multi-objective approach to the design of low thrust space trajectories using optimal control,

    M. Dellnitz, S. Ober-Bl ¨obaum, M. Post, O. Sch¨utze, and B. Thiere, “A multi-objective approach to the design of low thrust space trajectories using optimal control,”Celestial Mechanics and Dynamical Astronomy, V ol. 105, No. 1, 2009, pp. 33–59, 10.1007/s10569-009-9229-y

    work page doi:10.1007/s10569-009-9229-y 2009

  15. [15]

    Designing optimal low-thrust gravity-assist trajec- tories using space pruning and a multi-objective approach,

    O. Sch ¨utze, M. Vasile, O. Junge, M. Dellnitz, and D. Izzo, “Designing optimal low-thrust gravity-assist trajec- tories using space pruning and a multi-objective approach,”Engineering Optimization, V ol. 41, No. 2, 2009, pp. 155–181, 10.1080/03052150802391734

    work page doi:10.1080/03052150802391734 2009

  16. [16]

    Designing Low-Thrust Transfers near Earth–Moon L2 via Multi-Objective Reinforcement Learning,

    C. J. Sullivan, N. Bosanac, and R. L. Anderson, “Designing Low-Thrust Transfers near Earth–Moon L2 via Multi-Objective Reinforcement Learning,”Journal of Spacecraft and Rockets, V ol. 60, No. 2, 2023, pp. 634– 647, 10.2514/1.A35463

    work page doi:10.2514/1.a35463 2023

  17. [17]

    Global Optimization by Basin-Hopping and the Lowest Energy Structures of Lennard-Jones Clusters Containing up to 110 Atoms,

    D. J. Wales and J. P. K. Doye, “Global Optimization by Basin-Hopping and the Lowest Energy Structures of Lennard-Jones Clusters Containing up to 110 Atoms,”The Journal of Physical Chemistry A, V ol. 101, No. 28, 1997, pp. 5111–5116, 10.1021/jp970984n

    work page doi:10.1021/jp970984n 1997

  18. [18]

    A. C. Englander,Speeding-Up a Random Search for the Global Minimum of a Non-Convex, Non-Smooth Objec- tive Function. Ph.d. dissertation, University of New Hampshire, 2021. Doctoral Dissertations

    work page 2021

  19. [19]

    Amortized Global Search for Efficient Preliminary Trajectory Design with Deep Generative Models,

    A. Li, A. Sinha, and R. Beeson, “Amortized Global Search for Efficient Preliminary Trajectory Design with Deep Generative Models,”AAS/AIAA Astrodynamics Specialist Conference, No. 23-352, Big Sky, Montana, 8 2023. arXiv preprint 10.48550/arXiv.2308.03960

    work page doi:10.48550/arxiv.2308.03960 2023

  20. [20]

    DiffuSolve: Diffusion-based Solver for Non-convex Trajectory Optimization,

    A. Li, Z. Ding, A. B. Dieng, and R. Beeson, “DiffuSolve: Diffusion-based Solver for Non-convex Trajectory Optimization,”Proceedings of the 7th Annual Learning for Dynamics & Control Conference(N. Ozay, 38 L. Balzano, D. Panagou, and A. Abate, eds.), V ol. 283 ofProceedings of Machine Learning Research, PMLR, 04–06 Jun 2025, pp. 45–58

    work page 2025

  21. [21]

    Learning Optimal Control and Dynamical Structure of Global Trajectory Search Problems with Diffusion Models,

    J. Graebner, A. Li, A. Sinha, and R. Beeson, “Learning Optimal Control and Dynamical Structure of Global Trajectory Search Problems with Diffusion Models,”AAS/AIAA Astrodynamics Specialist Conference, No. 24- 213, Broomfield, Colorado, 8 2024. arXiv preprint 10.48550/arXiv.2410.02976

    work page doi:10.48550/arxiv.2410.02976 2024

  22. [22]

    Real-Time Guidance for Low-Thrust Transfers Using Deep Neural Networks,

    D. Izzo and E. ¨Ozt¨urk, “Real-Time Guidance for Low-Thrust Transfers Using Deep Neural Networks,”Journal of Guidance, Control, and Dynamics, V ol. 44, No. 2, 2021, pp. 315–327, 10.2514/1.G005254

    work page doi:10.2514/1.g005254 2021

  23. [23]

    Low Thrust Minimum-Fuel Orbital Transfer: A Homotopic Ap- proach,

    T. Haberkorn, P. Martinon, and J. Gergaud, “Low Thrust Minimum-Fuel Orbital Transfer: A Homotopic Ap- proach,”Journal of Guidance, Control, and Dynamics, V ol. 27, No. 6, 2004, pp. 1046–1060, 10.2514/1.4022

    work page doi:10.2514/1.4022 2004

  24. [24]

    Solution space exploration of low-thrust minimum-time tra- jectory optimization by combining two homotopies,

    J. Zhang, Q. Xiao, and L. Li, “Solution space exploration of low-thrust minimum-time tra- jectory optimization by combining two homotopies,”Automatica, V ol. 148, 2023, p. 110798, https://doi.org/10.1016/j.automatica.2022.110798

    work page doi:10.1016/j.automatica.2022.110798 2023

  25. [25]

    New smoothing techniques for solving bang–bang optimal control problems— numerical results and statistical interpretation,

    R. Bertrand and R. Epenoy, “New smoothing techniques for solving bang–bang optimal control problems— numerical results and statistical interpretation,”Optimal Control Applications and Methods, V ol. 23, No. 4, 2002, pp. 171–197, 10.1002/oca.709

    work page doi:10.1002/oca.709 2002

  26. [26]

    Enhanced Smoothing Technique for Indirect Optimization of Minimum-Fuel Low-Thrust Trajectories,

    E. Taheri, I. Kolmanovsky, and E. Atkins, “Enhanced Smoothing Technique for Indirect Optimization of Minimum-Fuel Low-Thrust Trajectories,”Journal of Guidance, Control, and Dynamics, V ol. 39, No. 11, 2016, pp. 2500–2511, 10.2514/1.G000379

    work page doi:10.2514/1.g000379 2016

  27. [27]

    Double-Homotopy Method for Solving Optimal Control Problems,

    B. Pan, P. Lu, X. Pan, and Y . Ma, “Double-Homotopy Method for Solving Optimal Control Problems,”Journal of Guidance, Control, and Dynamics, V ol. 39, No. 8, 2016, pp. 1706–1720, 10.2514/1.G001553

    work page doi:10.2514/1.g001553 2016

  28. [28]

    Finding Best Solution in Low-Thrust Trajectory Optimization by Two-Phase Homo- topy,

    B. Pan, X. Pan, and P. Lu, “Finding Best Solution in Low-Thrust Trajectory Optimization by Two-Phase Homo- topy,”Journal of Spacecraft and Rockets, V ol. 56, No. 1, 2019, pp. 283–291, 10.2514/1.A34144

    work page doi:10.2514/1.a34144 2019

  29. [29]

    Smoothing Homotopy Methods for Solving Nonlinear Optimal Control Prob- lems,

    B. Pan, Y . Ni, Y . Ma, and P. Lu, “Smoothing Homotopy Methods for Solving Nonlinear Optimal Control Prob- lems,”Journal of Guidance, Control, and Dynamics, V ol. 46, No. 8, 2023, pp. 1470–1484, 10.2514/1.G006977

    work page doi:10.2514/1.g006977 2023

  30. [30]

    W., Rosenbluth M

    N. Metropolis, A. W. Rosenbluth, M. N. Rosenbluth, A. H. Teller, and E. Teller, “Equation of State Calcula- tions by Fast Computing Machines,”The Journal of Chemical Physics, V ol. 21, No. 6, 1953, pp. 1087–1092, 10.1063/1.1699114

    work page doi:10.1063/1.1699114 1953

  31. [31]

    Optimal Scaling of Discrete Approximations to Langevin Diffusions,

    G. O. Roberts and J. S. Rosenthal, “Optimal Scaling of Discrete Approximations to Langevin Diffusions,” Journal of the Royal Statistical Society Series B: Statistical Methodology, V ol. 60, No. 1, 1998, pp. 255–268, 10.1111/1467-9868.00123

    work page doi:10.1111/1467-9868.00123 1998

  32. [32]

    MCMC using Hamiltonian dynamics,

    R. Neal, “MCMC using Hamiltonian dynamics,”Handbook of Markov Chain Monte Carlo, 06 2012, 10.1201/b10905-6

    work page doi:10.1201/b10905-6 2012

  33. [33]

    D. E. Kirk,Optimal control theory: An introduction. Dover Books on Electrical Engineering Ser, Mineola, N.Y: Dover Publications, first published in 2004, unabridged republication of the thirteenth printing ed., 2004

    work page 2004

  34. [34]

    D. F. Lawden,Optimal trajectories for space navigation. Butterworths, 1964

    work page 1964

  35. [35]

    Liberzon,Calculus of variations and optimal control theory: A concise introduction

    D. Liberzon,Calculus of variations and optimal control theory: A concise introduction. Princeton, NJ and Woodstock: Princeton University Pres, 2012, 10.1515/9781400842643

    work page doi:10.1515/9781400842643 2012

  36. [36]

    L. S. Pontryagin,Mathematical Theory of Optimal Processes. Routledge, 2018, 10.1201/9780203749319. 39

    work page doi:10.1201/9780203749319 2018

  37. [37]

    B. A. Conway,Spacecraft Trajectory Optimization, V ol. v.29 ofCambridge Aerospace Series. New York: Cam- bridge University Press, 2010

    work page 2010

  38. [38]

    Classifier-Free Diffusion Guidance,

    J. Ho and T. Salimans, “Classifier-Free Diffusion Guidance,”NeurIPS 2021 Workshop on Deep Generative Models and Downstream Applications, 2021. Available at:https://openreview.net/forum?id= qw8AKxfYbI

    work page 2021

  39. [39]

    Aligning Text-to-Image Models using Human Feedback

    K. Lee, H. Liu, M. Ryu, O. Watkins, Y . Du, C. Boutilier, P. Abbeel, M. Ghavamzadeh, and S. S. Gu, “Aligning Text-to-Image Models using Human Feedback,” arXiv preprinthttps://arxiv.org/abs/2302.12192,

    work page internal anchor Pith review arXiv

  40. [40]

    10.48550/arXiv.2302.12192

    work page internal anchor Pith review doi:10.48550/arxiv.2302.12192

  41. [41]

    Monte Carlo sampling methods using Markov chains and their applications,

    W. K. Hastings, “Monte Carlo sampling methods using Markov chains and their applications,”Biometrika, V ol. 57, No. 1, 1970, pp. 97–109, 10.1093/biomet/57.1.97

    work page doi:10.1093/biomet/57.1.97 1970

  42. [42]

    A Conceptual Introduction to Hamiltonian Monte Carlo

    M. Betancourt, “A Conceptual Introduction to Hamiltonian Monte Carlo,” arXiv preprinthttps://arxiv. org/abs/1701.02434, 2018. 10.48550/arXiv.1701.02434

    work page Pith review doi:10.48550/arxiv.1701.02434 2018

  43. [43]

    Communications of the ACM , month = sep, pages =

    J. L. Bentley, “Multidimensional binary search trees used for associative searching,”Communications of the ACM, V ol. 18, No. 9, 1975, pp. 509–517, 10.1145/361002.361007

    work page doi:10.1145/361002.361007 1975

  44. [44]

    Dynamically Leveraged Automated (N) Multibody Trajectory Optimization (DyLAN),

    R. Beeson, A. Sinha, B. Jagannatha, D. Bunce, and D. L. Carroll, “Dynamically Leveraged Automated (N) Multibody Trajectory Optimization (DyLAN),”AAS/AIAA Space Flight Mechanics Conference. American As- tronautical Society. Charlotte, NC, Aug. 2022

    work page 2022

  45. [45]

    SIAM Rev

    P. E. Gill, W. Murray, and M. A. Saunders, “SNOPT: An SQP Algorithm for Large-Scale Constrained Optimiza- tion,”SIAM Review, V ol. 47, No. 1, 2005, pp. 99–131, 10.1137/S0036144504446096

    work page doi:10.1137/s0036144504446096 2005

  46. [46]

    Multiobjective Evolutionary Algorithms: A Comparative Case Study and the Strength Pareto Approach,

    E. Zitzler and L. Thiele, “Multiobjective evolutionary algorithms: a comparative case study and the strength Pareto approach,”IEEE Transactions on Evolutionary Computation, V ol. 3, No. 4, 1999, pp. 257–271, 10.1109/4235.797969

    work page doi:10.1109/4235.797969 1999

  47. [47]

    Adjoint-control transformations for solving practical op- timal control problems,

    L. C. W. Dixon and M. C. Bartholomew-Biggs, “Adjoint-control transformations for solving practical op- timal control problems,”Optimal Control Applications and Methods, V ol. 2, No. 4, 1981, pp. 365–381, 10.1002/oca.4660020405. 40

    work page doi:10.1002/oca.4660020405 1981