pith. sign in

arxiv: 2602.10712 · v2 · submitted 2026-02-11 · 💻 cs.GR · astro-ph.EP· astro-ph.IM

Photons x Force: Differentiable Radiation Pressure Modeling

Charles Constant , Elizabeth Bates , Santosh Bhattarai , Marek Ziebart , Tobias Ritschel This is my paper

Pith reviewed 2026-05-16 03:38 UTC · model grok-4.3

classification 💻 cs.GR astro-ph.EPastro-ph.IM
keywords radiation pressureneural network proxydifferentiable modelingspacecraft designMonte Carlo simulationforce optimizationparametric design
0
0 comments X

The pith

Neural networks serve as fast differentiable proxies for radiation pressure forces after training on Monte Carlo simulations.

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

The paper develops a system for optimizing spacecraft designs under radiation pressure, the dominant non-conservative force above roughly 800 km altitude. It starts with an efficient Monte Carlo simulator that handles entire families of designs in parallel by borrowing computer graphics techniques such as importance sampling and next-event estimation. From those simulation results the authors train a neural network that maps design parameters directly to forces. Because the network is differentiable, it supports gradient-based optimization routines that can search for geometry, materials, or operating rules meeting goals like shorter travel time or lower fuel use, all at speeds orders of magnitude above repeated full simulations.

Core claim

We introduce neural networks as a representation of forces from design parameters. This neural proxy model, learned from simulations, is inherently differentiable and can query forces orders of magnitude faster than a full MC simulation. We demonstrate optimizing inverse radiation pressure designs, such as finding geometry, material or operation parameters that minimizes travel time, maximizes proximity given a desired end-point, minimize thruster fuel, trains mission control policies or allocated compute budget in extraterrestrial compute.

What carries the argument

A neural network proxy that takes design parameters as input and outputs radiation pressure forces, trained on data from a parallel Monte Carlo simulator using importance sampling and next-event estimation.

If this is right

  • Spacecraft geometry and materials can be optimized directly to minimize travel time under radiation pressure.
  • Design parameters can be adjusted to maximize proximity to a target point while respecting radiation forces.
  • Thruster fuel budgets can be reduced by gradient searches over the differentiable force model.
  • Mission control policies can be trained end-to-end using the fast, differentiable force queries.
  • Compute resources for extraterrestrial operations can be allocated on the basis of rapid force evaluations.

Where Pith is reading between the lines

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

  • The same training-plus-proxy pattern could accelerate other expensive physics-based design loops in aerospace.
  • Combining the proxy with additional differentiable simulators would enable complete mission-level optimization pipelines.
  • Real-time force estimates from the network might improve tracking models used in space situational awareness.

Load-bearing premise

The neural network must approximate the true Monte Carlo radiation pressure forces closely enough across the design space that optimization does not settle on wrong parameters.

What would settle it

Take the final designs produced by optimization with the neural proxy, run independent full Monte Carlo simulations on them, and check whether the force values differ by more than the tolerance needed to change the optimum.

Figures

Figures reproduced from arXiv: 2602.10712 by Charles Constant, Elizabeth Bates, Marek Ziebart, Santosh Bhattarai, Tobias Ritschel.

Figure 1
Figure 1. Figure 1: We devise a system for the automatic design of spacecraft (satellite, left) such that they arrive at a desired target (flag, right) along the blue path in the [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Main flow of our system: The input is a design space of spacecraft, shown here with different panel configurations as well as a design goal symbolized [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Radiation pressure converting light to force: Radiation arrives from [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Evaluation dependency. Edges are works-if relationships. Nodes are evaluation steps. Grey components are parts of the system, but we trust they were verified elsewhere. An overview of all methods is shown in Tab. 1. We will progres￾sively detail all aspects of a method as they become relevant in each step of the evaluation [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 8
Figure 8. Figure 8: BVH time and space (see text). Experimental evidence for this is seen in [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 7
Figure 7. Figure 7: Convergence (see text). A detailed breakdown for different satellites is seen in Tab. 4. We use two sets of meshes: high resolution meshes as provided by the group at TU Delft for GRACE￾FO, Swarm and CHAMP [March et al. 2019], and a second model for GPS2F, which we de￾veloped ourselves using openly accessible data about the spacecraft [Fisher and Ghassemi 1999; International GNSS Service 2025; Montenbruck … view at source ↗
Figure 5
Figure 5. Figure 5: Force maps for different spacecraft geometry and materials. In a force map, each pixel corresponds to one light direction in the spacecraft-fixed [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visualization of sampling a one-dimensional design space of chang [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 9
Figure 9. Figure 9: Point Approach task (Sec. 5.1): spacecraft of different reflectance [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 12
Figure 12. Figure 12: Radiation pressure maps for two hypothetical flexible satellite [PITH_FULL_IMAGE:figures/full_fig_p012_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Collision avoidance task from Sec. 5.4. A satellite in a near-circular [PITH_FULL_IMAGE:figures/full_fig_p012_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Atmospheric density varies with altitude, latitude, and longitude [PITH_FULL_IMAGE:figures/full_fig_p013_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Formation flight result from Sec. 5.6. We optimize the time-varying [PITH_FULL_IMAGE:figures/full_fig_p014_15.png] view at source ↗
Figure 17
Figure 17. Figure 17: Main result of Sec. 5.9, where a 500 kg box-wing spacecraft is equipped with one GPU that emits heat (a) on each of the six sides of the bus. The spacecraft surface area is 80.85 m2 . This radiation results in deviation (vertical axis) from an orbit without compute; the dashed lines show this deviation for four arbitrary compute patterns (b) varying over time (horizontal axis). We optimize the allocation … view at source ↗
read the original abstract

We propose a system to optimize parametric designs subject to radiation pressure, \ie the effect of light on the motion of objects. This is most relevant in the design of spacecraft, where radiation pressure presents the dominant non-conservative forcing mechanism, which is the case beyond approximately 800 km altitude. Despite its importance, the high computational cost of high-fidelity radiation pressure modeling has limited its use in large-scale spacecraft design, optimization, and space situational awareness applications. We enable this by offering three innovations in the simulation, in representation and in optimization: First, a practical computer graphics-inspired Monte-Carlo (MC) simulation of radiation pressure. The simulation is highly parallel, uses importance sampling and next-event estimation to reduce variance and allows simulating an entire family of designs instead of a single spacecraft as in previous work. Second, we introduce neural networks as a representation of forces from design parameters. This neural proxy model, learned from simulations, is inherently differentiable and can query forces orders of magnitude faster than a full MC simulation. Third, and finally, we demonstrate optimizing inverse radiation pressure designs, such as finding geometry, material or operation parameters that minimizes travel time, maximizes proximity given a desired end-point, minimize thruster fuel, trains mission control policies or allocated compute budget in extraterrestrial compute.

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 paper proposes a Monte Carlo simulation for radiation pressure on parametric spacecraft designs that supports families of designs and uses importance sampling, a neural network surrogate trained on MC outputs to provide fast differentiable force queries, and demonstrations of its use for inverse optimization tasks such as minimizing travel time, maximizing proximity, or reducing fuel use.

Significance. If the neural proxy proves sufficiently accurate, the work could enable scalable gradient-based optimization of spacecraft under radiation pressure, a dominant force beyond low Earth orbit, by replacing expensive MC evaluations with fast differentiable queries. This combination of physics-based MC simulation and learned surrogates addresses a practical bottleneck in design and space situational awareness applications.

major comments (2)
  1. [Abstract] Abstract: the claim that the neural proxy 'can query forces orders of magnitude faster' and enables optimization of geometry, material, and operation parameters is unsupported by any error metrics, held-out validation accuracy, or re-evaluation of proxy-optimized designs against full MC ground truth.
  2. [Optimization demonstrations] Optimization demonstrations: the pipeline uses the neural proxy for gradient-based search on objectives such as travel-time minimization, yet no quantitative assessment is given of how approximation error propagates into the discovered optima or whether re-simulating those optima with the original MC simulator recovers comparable performance.
minor comments (2)
  1. Provide the neural-network architecture details, training-set size, loss function, and regularization used to fit the proxy to MC data.
  2. Clarify the exact sampling strategy and next-event estimation implementation in the MC simulator, including any variance-reduction parameters.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We agree that the current manuscript lacks quantitative support for the neural proxy claims and optimization reliability. We will revise to include held-out validation metrics, speedup measurements, error propagation analysis, and MC re-evaluation of optimized designs.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that the neural proxy 'can query forces orders of magnitude faster' and enables optimization of geometry, material, and operation parameters is unsupported by any error metrics, held-out validation accuracy, or re-evaluation of proxy-optimized designs against full MC ground truth.

    Authors: We accept this criticism. The revised manuscript will report specific held-out validation accuracy (MAE and relative error on force vectors), query-time benchmarks showing the speedup factor versus MC, and re-simulation of proxy-optimized designs with the original MC simulator to confirm recovered performance. revision: yes

  2. Referee: [Optimization demonstrations] Optimization demonstrations: the pipeline uses the neural proxy for gradient-based search on objectives such as travel-time minimization, yet no quantitative assessment is given of how approximation error propagates into the discovered optima or whether re-simulating those optima with the original MC simulator recovers comparable performance.

    Authors: We agree this assessment is missing. In revision we will add: (1) sensitivity analysis showing how proxy error affects the discovered optima, and (2) direct MC re-evaluation of the final designs to compare objective values and verify that the proxy-guided optima remain competitive under ground-truth simulation. revision: yes

Circularity Check

0 steps flagged

No significant circularity in surrogate-based optimization pipeline

full rationale

The paper's chain runs from an independent Monte Carlo radiation pressure simulator (physics-based, parallelized with importance sampling) to training a neural network surrogate on its outputs, then using the differentiable surrogate for inverse design optimization. This is standard surrogate modeling; the MC outputs constitute external data generation rather than a self-referential loop. The optimization result is not equivalent to the training inputs by construction, nor does any step rename a fit as a first-principles prediction. No self-citations, uniqueness theorems, or ansatzes imported from prior author work appear in the abstract or description to bear load on the central claims. The pipeline remains self-contained against the MC benchmark.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

Approach rests on standard Monte Carlo assumptions for radiation pressure and the universal approximation capability of neural networks; no new physical entities are introduced.

free parameters (1)
  • Neural network weights and biases
    Parameters of the proxy model are fitted to Monte Carlo simulation outputs.
axioms (2)
  • domain assumption Monte Carlo ray tracing with importance sampling and next-event estimation produces accurate radiation pressure forces
    Invoked as the basis for generating training data for the neural proxy.
  • domain assumption Neural networks can represent the mapping from design parameters to forces with sufficient fidelity for optimization
    Required for the proxy to be usable in gradient-based design search.

pith-pipeline@v0.9.0 · 5540 in / 1385 out tokens · 109157 ms · 2026-05-16T03:38:43.610561+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

55 extracted references · 55 canonical work pages

  1. [1]

    Moritz Bächer, Emily Whiting, Bernd Bickel, and Olga Sorkine-Hornung

    Physical design using differentiable learned simulators.arXiv preprint arXiv:2202.00728(2022). Moritz Bächer, Emily Whiting, Bernd Bickel, and Olga Sorkine-Hornung

    work page arXiv 2022

  2. [2]

    Spin- it: Optimizing moment of inertia for spinnable objects.ACM Trans. Graph. (Proc. SIGGRAPH)33, 4 (2014), 1–10. Jeremy Banik and Paul Hausgen

    work page 2014

  3. [3]

    Santosh Bhattarai, Marek Ziebart, Tim Springer, Francisco Gonzalez, and Guillermo Tobias

    Demonstrating developments in high-fidelity analytical radiation force modelling methods for spacecraft with a new model for GPS IIR/IIR-M.J Geodesy93 (2019), 1515–1528. Santosh Bhattarai, Marek Ziebart, Tim Springer, Francisco Gonzalez, and Guillermo Tobias

    work page 2019

  4. [4]

    James Bradbury, Roy Frostig, Peter Hawkins, Matthew James Johnson, Chris Leary, Dougal Maclaurin, George Necula, Adam Paszke, Jake VanderPlas, Skye Wanderman- Milne, and Qiao Zhang

    High-precision physics-based radiation force models for the Galileo spacecraft.Advances in Space Research69, 12 (2022), 4141–4154. James Bradbury, Roy Frostig, Peter Hawkins, Matthew James Johnson, Chris Leary, Dougal Maclaurin, George Necula, Adam Paszke, Jake VanderPlas, Skye Wanderman- Milne, and Qiao Zhang. 2018.JAX: composable transformations of Pyth...

    work page 2022

  5. [5]

    Ricky TQ Chen, Yulia Rubanova, Jesse Bettencourt, and David K Duvenaud

    Machine learning in orbit estimation: A survey.Acta Astronautica(2024). Ricky TQ Chen, Yulia Rubanova, Jesse Bettencourt, and David K Duvenaud

    work page 2024

  6. [6]

    Charles Constant, Shaylah M

    Neural ordinary differential equations.NeurIPS31 (2018). Charles Constant, Shaylah M. Mutschler, Santosh Bhattarai, and Marcin Pilinski

    work page 2018

  7. [7]

    Per K Enge

    End-to-end differentiable physics for learning and control.NeurIPS31 (2018). Per K Enge

    work page 2018

  8. [8]

    Daniel P

    The global positioning system: Signals, measurements, and perfor- mance.Int J Wireless Information Networks1 (1994), 83–105. Daniel P. Engelhart, R. Cooper, Heather Cowardin, J. Maxwell, Elena Plis, D. Ferguson, D. Barton, S. Schiefer, and R. Hoffmann

    work page 1994

  9. [9]

    Geometric Calibration of the Orion Optical Navigation Camera using Star Field Images,

    Space Weathering Experiments on Spacecraft Materials.J Astronautical Sciences66 (2019), 210–223. doi:10.1007/s40295- 019-00175-2 Michael Fischer and Tobias Ritschel

    work page doi:10.1007/s40295- 2019

  10. [10]

    SIGGRAPH)43, 4 (2024), 1–15

    ZeroGrads: Learning Local Surrogates for Non-Differentiable Graphics.ACM Trans Graph (Proc. SIGGRAPH)43, 4 (2024), 1–15. S. C. Fisher and K. Ghassemi

    work page 2024

  11. [11]

    87, 1 (1999), 24–47

    GPS IIF-the Next Generation. 87, 1 (1999), 24–47. Moritz Geilinger, David Hahn, Jonas Zehnder, Moritz Bächer, Bernhard Thomaszewski, and Stelian Coros

    work page 1999

  12. [12]

    Add: Analytically differentiable dynamics for multi-body systems with frictional contact.ACM Trans. Graph. (Proc. SIGGRAPH Asia)39, 6 (2020), 1–15. Peter E Glaser

    work page 2020

  13. [13]

    Jeffrey Goldsmith and John Salmon

    Power from the sun: Its future.Science162, 3856 (1968), 857–861. Jeffrey Goldsmith and John Salmon

    work page 1968

  14. [14]

    Steven J

    Automatic creation of object hierarchies for ray tracing.IEEE Comp Graph & App7, 5 (1987), 14–20. Steven J. Gortler, Radek Grzeszczuk, Richard Szeliski, and Michael F. Cohen

    work page 1987

  15. [15]

    Neuroanima- tor: Fast neural network emulation and control of physics-based models. InProc. SIGGRAPH. 9–20. A Rupert Hall. 2012.From Galileo to Newton. Courier Corporation. Isadore Harris and Wolfgang Priester

    work page 2012

  16. [16]

    Gerald S Hawkins

    Time-dependent structure of the upper atmosphere.Journal of Atmospheric Sciences19, 4 (1962), 286–301. Gerald S Hawkins

    work page 1962

  17. [17]

    NA Hładczuk, Jose van den IJssel, Timothy Kodikara, Christian Siemes, and Pieter Visser

    Stonehenge decoded.Nature200, 4904 (1963), 306–308. NA Hładczuk, Jose van den IJssel, Timothy Kodikara, Christian Siemes, and Pieter Visser

    work page 1963

  18. [18]

    Philipp Holl, Vladlen Koltun, and Nils Thuerey

    GRACE-FO radiation pressure modelling for accurate density and crosswind retrieval.Advances in Space Research73, 5 (2024), 2355–2373. Philipp Holl, Vladlen Koltun, and Nils Thuerey

    work page 2024

  19. [19]

    Yuanming Hu, Luke Anderson, Tzu-Mao Li, Qi Sun, Nathan Carr, Jonathan Ragan- Kelley, and Frédo Durand

    Learning to control PDEs with differentiable physics.ICLR(2020). Yuanming Hu, Luke Anderson, Tzu-Mao Li, Qi Sun, Nathan Carr, Jonathan Ragan- Kelley, and Frédo Durand

    work page 2020

  20. [20]

    International GNSS Service

    DiffTaichi: Differentiable programming for physical simulation.ICLR(2020). International GNSS Service

    work page 2020

  21. [21]

    http: //acc.igs.org/orbits/ Accessed: 2025-01-23

    IGS Orbits - Analysis Center Coordinator. http: //acc.igs.org/orbits/ Accessed: 2025-01-23. Victor Isakov. 2006.Inverse problems for partial differential equations. Vol

    work page 2025

  22. [22]

    Les Johnson, Roy Young, Edward Montgomery, and Dean Alhorn

    Optimality principles in spacecraft neural guidance and control.Science Robotics9, 91 (2024). Les Johnson, Roy Young, Edward Montgomery, and Dean Alhorn

    work page 2024

  23. [23]

    Gary Johnston, Anna Riddell, and Grant Hausler

    Status of solar sail technology within NASA.Adv Space Res48, 11 (2011), 1687–1694. Gary Johnston, Anna Riddell, and Grant Hausler

    work page 2011

  24. [24]

    Springer Handbook of Global Navigation Satellite Systems(2017), 967–982

    The International GNSS Service. Springer Handbook of Global Navigation Satellite Systems(2017), 967–982. Tero Karras and Timo Aila

    work page 2017

  25. [25]

    Fast parallel construction of high-quality bounding volume hierarchies. InProc. High-Performance Graphics. 89–99. Patrick William Kenneally. 2019.Faster than Real-Time GPGPU Radiation Pressure Modeling Methods. Ph. D. Dissertation. University of Colorado at Boulder. Patrick W Kenneally and Hanspeter Schaub

    work page 2019

  26. [26]

    Pyotr N Lebedev

    Fast spacecraft solar radiation pressure modeling by ray tracing on graphics processing unit.Advances in Space Research65, 8 (2020), 1951–1964. Pyotr N Lebedev

    work page 2020

  27. [27]

    Phys6, 433 (1901), 10–1002

    Experimental examination of light pressure.Ann. Phys6, 433 (1901), 10–1002. Jaakko Lehtinen, Jacob Munkberg, Jon Hasselgren, Samuli Laine, Tero Karras, Miika Aittala, and Timo Aila

    work page 1901

  28. [28]

    Tzu-Mao Li, Miika Aittala, Frédo Durand, and Jaakko Lehtinen

    Noise2Noise: Learning image restoration without clean data (2018).ICML(2018). Tzu-Mao Li, Miika Aittala, Frédo Durand, and Jaakko Lehtinen. 2018a. Differentiable Monte Carlo ray tracing through edge sampling.ACM Trans. Graph. (Proc. SIG- GRAPH Asia)37, 6 (2018), 1–11. Zhen Li, Marek Ziebart, Santosh Bhattarai, David Harrison, and Stuart Grey. 2018b. Fast ...

    work page 2018

  29. [29]

    doi:10.2514/1.62986 Matthew M Loper and Michael J Black

    Space Object Shape Characterization and Tracking Using Light Curve and Angles Data.J of Guidance, Control, and Dynamics37, 1 (2014), 13–25. doi:10.2514/1.62986 Matthew M Loper and Michael J Black

    work page doi:10.2514/1.62986 2014

  30. [30]

    Reparameterizing discontinuous integrands for differentiable rendering.ACM Trans. Graph. (Proc. SIGGRAPH Asia)38, 6 (2019), 1–14. J David MacDonald and Kellogg S Booth

    work page 2019

  31. [31]

    Luc Maisonobe and Veronique Pommier-Maurussane

    Heuristics for ray tracing using space subdivision.The Visual Computer6 (1990), 153–166. Luc Maisonobe and Veronique Pommier-Maurussane

    work page 1990

  32. [32]

    Darren McKnight, Rachel Witner, Francesca Letizia, Stijn Lemmens, Luciano Anselmo, Carmen Pardini, Alessandro Rossi, Chris Kunstadter, Satomi Kawamoto, Vladimir Aslanov, et al

    High-fidelity geometry models for improving the consistency of CHAMP, GRACE, GOCE and Swarm thermospheric density data sets.Adv Space Res63 (2019), 213–238. Darren McKnight, Rachel Witner, Francesca Letizia, Stijn Lemmens, Luciano Anselmo, Carmen Pardini, Alessandro Rossi, Chris Kunstadter, Satomi Kawamoto, Vladimir Aslanov, et al

    work page 2019

  33. [33]

    Antoine McNamara, Adrien Treuille, Zoran Popović, and Jos Stam

    Identifying the 50 statistically-most-concerning derelict objects in LEO.Acta Astronautica181 (2021), 282–291. Antoine McNamara, Adrien Treuille, Zoran Popović, and Jos Stam

    work page 2021

  34. [34]

    Graph.23, 3 (2004), 449–456

    Fluid control using the adjoint method.ACM Trans. Graph.23, 3 (2004), 449–456. Sharon K. R. Miller and Bruce A. Banks

    work page 2004

  35. [35]

    doi:10.1557/mrs2010.612 O

    Degradation of Spacecraft Materials in the Space Environment.MRS Bulletin35, 1 (2010), 20–24. doi:10.1557/mrs2010.612 O. Montenbruck and E. Gill. 2000.Satellite Orbits: Models, Methods, and Applications. Springer Berlin Heidelberg. Oliver Montenbruck, Ralf Schmid, Flavien Mercier, Peter Steigenberger, Carey Noll, Roman Fatkulin, Satoshi Kogure, and Aiylam...

    work page doi:10.1557/mrs2010.612 2010

  36. [36]

    ACM Trans

    GNSS satellite geometry and attitude models.Adv Space Res56, 6 (2015), 1015–1029. ACM Trans. Graph., Vol. 45, No. 4, Article

    work page 2015

  37. [37]

    Space Weather21, 1 (2023)

    Physics- Based Approach to Thermospheric Density Estimation Using CubeSat GPS Data. Space Weather21, 1 (2023). Nvidia. 2023.NVIDIA H100 Tensor Core GPU Architecture. Daniel J O’Shaughnessy, James V McAdams, Peter D Bedini, Andrew B Calloway, Kenneth E Williams, and Brian R Page

    work page 2023

  38. [38]

    Ryan S Park, William M Folkner, James G Williams, and Dale H Boggs

    MESSENGER’s use of solar sailing for cost and risk reduction.Acta Astronautica93 (2014), 483–489. Ryan S Park, William M Folkner, James G Williams, and Dale H Boggs

    work page 2014

  39. [39]

    The JPL planetary and lunar ephemerides DE440 and DE441.The Astronomical J161, 3 (2021),

    work page 2021

  40. [40]

    David Pérez and Riccardo Bevilacqua

    The development and evaluation of the Earth Gravitational Model 2008 (EGM2008).J of geophysical research: solid earth117, B4 (2012). David Pérez and Riccardo Bevilacqua

    work page 2008

  41. [41]

    Matt Pharr, Wenzel Jakob, and Greg Humphreys

    Neural Network based calibration of atmospheric density models.Acta Astronautica110 (2015), 58–76. Matt Pharr, Wenzel Jakob, and Greg Humphreys. 2023.Physically based rendering: From theory to implementation. MIT Press. L.S. Pontryagin, V. G. Boltyanskii, R. V. Gamkrelidze, and E. F. Mishchenko. 1962.The mathematical theory of optimal processes.Wiley, NY....

    work page 2015

  42. [42]

    Make it stand: balancing shapes for 3D fabrication.ACM Trans Graph. (Proc. SIG- GRAPH)32, 4 (2013), 1–10. Jacqueline A Reyes and Darren Cone

    work page 2013

  43. [43]

    CJ Rodriguez-Solano, U Hugentobler, and P Steigenberger. 2012a. Adjustable box-wing model for solar radiation pressure impacting GPS satellites.Adv Space Res49, 7 (2012), 1113–1128. CJ Rodriguez-Solano, U Hugentobler, and P Steigenberger. 2012b. Impact of albedo radiation on GPS satellites. InProc. Geodesy for Planet Earth. Springer, 113–119. Carlos Javie...

    work page 2012

  44. [44]

    Hanspeter Schaub and John L

    XLA: Compiling machine learning for peak performance.Google Res(2020). Hanspeter Schaub and John L. Junkins. 2018.Analytical Mechanics of Space Systems (4th ed.). American Institute of Aeronautics and Astronautics, Reston, VA. Thomas Schildknecht

    work page 2020

  45. [45]

    The Red Book

    Optical surveys for space debris.The Astronomy and Astrophysics Review14 (2007), 41–111. Sheldahl. 2015.Sheldahl Optical Properties Handbook. Sheldahl, Northfield, Minnesota, USA. Revision E. Commonly known as "The Red Book". TA Springer, G Beutler, and M Rothacher

    work page 2007

  46. [46]

    Marc A Toussaint, Kelsey Rebecca Allen, Kevin A Smith, and Joshua B Tenenbaum

    A new solar radiation pressure model for GPS satellites.GPS solutions2 (1999), 50–62. Marc A Toussaint, Kelsey Rebecca Allen, Kevin A Smith, and Joshua B Tenenbaum

    work page 1999

  47. [47]

    Robotics: Science and systems foundation(2018)

    Differentiable physics and stable modes for tool-use and manipulation planning. Robotics: Science and systems foundation(2018). Slava G Turyshev, Viktor T Toth, Gary Kinsella, Siu-Chun Lee, Shing M Lok, and Jordan Ellis

    work page 2018

  48. [48]

    Eric Veach

    Support for the thermal origin of the Pioneer anomaly.Physical Review Letters108, 24 (2012). Eric Veach. 1998.Robust Monte Carlo methods for light transport simulation. Stanford University. Delio Vicini, Sébastien Speierer, and Wenzel Jakob

    work page 2012

  49. [49]

    SIGGRAPH)40, 4 (2021), 108:1–108:14

    Path Replay Backpropagation: Differentiating Light Paths using Constant Memory and Linear Time.ACM Trans Graph (Proc. SIGGRAPH)40, 4 (2021), 108:1–108:14. Pascal Vincent, Hugo Larochelle, Yoshua Bengio, and Pierre-Antoine Manzagol

    work page 2021

  50. [50]

    Graph.26, 1 (2007)

    Ray tracing deformable scenes using dynamic bounding volume hierarchies.ACM Trans. Graph.26, 1 (2007). Gerhard Wanner and Ernst Hairer. 1996.Solving ordinary differential equations II. Vol

    work page 2007

  51. [51]

    W Wollenhaupt

    Clouds and the Earth’s Radiant Energy System (CERES): An earth observing system experiment.Bulletin of the American Meteorological Society77, 5 (1996), 853–868. W Wollenhaupt

    work page 1996

  52. [52]

    Radosław Zajdel, Salim Masoumi, Krzysztof Sośnica, Filip Gałdyn, Dariusz Strugarek, and Grzegorz Bury

    Temperature aware workload management in geo-distributed datacenters.ACM SIGMETRICS performance evaluation review41, 1 (2013), 373–374. Radosław Zajdel, Salim Masoumi, Krzysztof Sośnica, Filip Gałdyn, Dariusz Strugarek, and Grzegorz Bury

    work page 2013

  53. [53]

    Combination and SLR validation of IGS Repro3 orbits for ITRF2020.Journal of Geodesy97, 10 (2023),

    work page 2023

  54. [54]

    Adv Space Res36, 3 (2005), 424–430

    Combined radiation pressure and thermal modelling of complex satellites: Algorithms and on-orbit tests. Adv Space Res36, 3 (2005), 424–430. ACM Trans. Graph., Vol. 45, No. 4, Article

    work page 2005

  55. [55]

    Publication date: July 2026

    work page 2026