Pith. sign in

REVIEW 3 major objections 5 minor 40 references

Efficient Training of Physics-enhanced Neural ODEs via Direct Collocation and Nonlinear Programming

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper shows that training physics-enhanced neural ODEs can be recast as one large nonlinear program solved by direct collocation, cutting training time from hours to minutes while matching reference dynamics.

desk verdict A solid, honest extension of direct-collocation NODE training to physics-enhanced models; the central stability claim is plausible but only tested on smooth, hand-gridded benchmarks. read the letter →

arxiv 2505.03552 v2 pith:62OOEDM2 submitted 2025-05-06 cs.LG math.DSmath.OC

Linus Langenkamp , Philip Hannebohm , Bernhard Bachmann This is my paper · ORCID
classification cs.LGmath.DSmath.OC MSC 65L6065L0649M3768T07
keywords physics-enhancedneuralODEsdirectcollocationnonlinearprogrammingRadauIIAsimultaneousoptimizationordinarydifferentialequationsdynamictrajectory
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that training physics-enhanced neural ODEs can be cast as a single dynamic optimization problem and solved by direct collocation: discretize both the known physical dynamics and the embedded neural network on a high-order implicit Runge-Kutta grid, then let a nonlinear optimizer adjust state trajectories and network parameters simultaneously. This avoids repeated forward simulation and backpropagation through an ODE solver, which the paper argues is the main bottleneck of current training. If the claim is right, small hybrid surrogates that combine first-principles models with learned force terms can be trained on a laptop in minutes while matching reference dynamics, as demonstrated on a quarter-vehicle suspension model and a Van-der-Pol oscillator. It also makes physical knowledge, such as a learned force vanishing at zero displacement, a hard constraint rather than a penalty.

What carries the argument

The load-bearing object is the Radau IIA collocation scheme at flipped Legendre-Gauss-Radau points, embedded as equality constraints in a nonlinear program. On each interval the states are Lagrange polynomials that must satisfy the ODE at the collocation nodes, while the loss integral is evaluated with the matching Radau quadrature. The central mechanism is simultaneous optimization: the NLP variables include every discretized state value plus the network parameters, so a single interior-point solver updates trajectories and weights together using analytic first and second derivatives of the sparse constraint system, rather than propagating gradients through an ODE solver.

What would settle it

Train the same architecture on a stiff or multi-timescale system, e.g. the Van-der-Pol oscillator at large $\mu$ or a suspension with a fast tire mode, using the paper's fixed 500-interval, five-point Radau grid; if the learned surrogate fails to reproduce a validation trajectory through the fast region, or works only after hand-refining the grid, the claimed step-size advantage over solver-based training would not hold outside smooth benchmarks.

Watch

Extended reading notes

Core claim

On its own terms, the central discovery is that the PeN-ODE training problem—minimize an integral loss subject to the augmented differential equation—is an instance of a dynamic optimization problem, and that transcribing it with flipped Legendre-Gauss-Radau collocation (equivalently Radau IIA, an $A$-, $B$-, and $L$-stable method of order $2m-1$) yields a large, sparse nonlinear program in which the state values at all collocation nodes and the network weights are unknowns. Solving that program with an interior-point optimizer gives simultaneous access to first and second derivatives and lets the optimizer leave intermediate iterates infeasible, which the paper credits with faster, stabler convergence than solver-based training. The paper reports training times under seven minutes for the quarter-vehicle model, versus hours for the comparable baseline, and a few seconds for the Van-der-Pol oscillator, with learned vector fields that match the reference even under strong noise.

Load-bearing premise

The load-bearing premise is that a hand-chosen time grid and collocation order adequately resolve the dynamics; Section 2.5.1 concedes the grid must be fixed a priori, and the benchmarks use constant 500- or 2500-interval grids with five Radau points per interval and no mesh adaptation.

Editorial extensions

If this is right

  • Training sessions for small neural surrogates drop from hours to minutes on a laptop; the quarter-vehicle benchmark completes in under seven minutes.
  • Because the loss is approximated with high-order Radau quadrature instead of a first-order sample mean, the discretization accuracy of the integrator is preserved in the objective.
  • Physical priors such as zero crossings of learned force elements become hard constraints in the NLP, not penalty terms that can distort the optimum.
  • The same machinery handles non-neural surrogates such as Chebyshev rational functions, matching the learned behavior with only 32 parameters in under 35 seconds.
  • The method is designed to extend to neural differential-algebraic equations inside an equation-based modeling environment, removing export and external-training steps.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: if the speed advantage persists on stiffer problems, collocation-based training could become the default for small hybrid models, with ODE-solver backpropagation reserved for very large networks where exact Hessians are prohibitive.
  • Editorial inference: the fixed-grid requirement suggests the natural next test is adaptive mesh refinement; the paper itself flags this, and success there would extend the method to systems with localized fast events.
  • Editorial inference: because intermediate NLP iterates may violate the dynamics, the approach opens the door to adding physics-consistency penalties or regularization on the trajectory itself, which are awkward in forward-simulation training.
  • Editorial inference: the same transcription should apply directly to neural differential-algebraic equations once an equation-based environment provides index-reduced DAEs, a step the paper says is under development but does not yet demonstrate.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes training Physics-enhanced Neural ODEs (PeN-ODEs) by expressing the training problem as a dynamic optimization problem and transcribing it into a large-scale NLP via direct collocation at flipped Legendre-Gauss-Radau points. The state trajectories and neural-network parameters are optimized simultaneously with Ipopt, using a custom, parallelized open-source implementation (an extension of GDOPT). The method is demonstrated on a quarter-vehicle model with two neural force surrogates and on a Van-der-Pol oscillator learned as a pure NODE, reporting training times of minutes or seconds on a laptop, good surrogate accuracy, generalization to an unseen road profile, and a 100-run sensitivity analysis to measurement noise. The paper also outlines an intended integration into OpenModelica for Neural DAE training. The central claim is that this collocation-NLP formulation overcomes key limitations of ODE-solver-based training in terms of order, stability, accuracy, and allowable step size.

Significance. If the claims hold, this is a practical contribution to the training of small PeN-ODEs: it gives an alternative to ODE-solver backpropagation that is fast, stable on smooth problems, supports hard physical constraints such as zero crossings, and is accompanied by a publicly available, parallelized implementation. The mathematical transcription from DOP to NLP is standard and appears correctly presented, and the open-source code and reproducible experiments are concrete strengths. However, the demonstrated advantages are conditional on fixed, hand-selected collocation grids that resolve the dynamics, and the speed comparison against ODE-solver-based training is not a controlled one. The paper is transparent about the grid-selection limitation in Section 2.5.1, but the abstract and conclusion state the advantage in more general terms than the evidence supports.

major comments (3)
  1. [Section 2.5.1 and Conclusion] The conclusion states that the approach "overcomes key limitations of ODE solver-based training in terms of order, stability, accuracy, and allowable step size," but the experiments use only fixed, hand-chosen equidistant grids: 2500 intervals for the quarter-vehicle model (Section 4.1.2) and 500 intervals for the Van-der-Pol oscillator (Section 4.2). Section 2.5.1 explicitly says that the collocation scheme and grid are embedded into the NLP and must be given a-priori, with adaptive mesh refinement left to future work. Consequently, the step-size and stability advantage is conditional on the grid resolving the true dynamics; for stiff dynamics or localized fast transients the fixed discretization can miss behavior, and a uniformly fine grid may erase the runtime advantage. Please qualify the central claim to the resolved-grid, smooth regime, or add a stiff/localized-dynamics experiment or an adaptive mesh refinement study.
  2. [Section 4.1.3 and Table 1] The speed comparison against ODE-solver-based training is not controlled. The reported 4.5 hours from [2] were obtained on different hardware, with different network sizes, and in a different paper; Table 1 gives only absolute training times on a laptop. The claim of "superior accuracy, speed, generalization" compared with other training techniques would be substantially strengthened by a head-to-head baseline using ODE-solver-based backpropagation on the same machine, network architecture, data, and initialization. Without such a baseline, the runtime advantage is indicative but not directly demonstrated.
  3. [Section 4.2.1 and Appendix (Figure 10)] The robustness claim "even under severe noise" rests on a single high-noise run in Section 4.2.1, while the Appendix's 100-run sensitivity analysis shows that at sigma = 0.5 several runs converge to poor local optima or fail to converge, producing period mismatches or trajectory collapse. The main text should report the failure rate and the median/quantile bands from Figure 10, and should temper the statement that the method is robust under high noise, since the single displayed run is not representative of the distribution of outcomes.
minor comments (5)
  1. [Section 2.3, Eq. (8)-(9)] The relation between the interval-local Lagrange polynomials l_j(t) in Eq. (7) and the reference polynomials l_tilde_k(tau) in Eq. (9) is implicit; explicitly stating l_j(t) = l_tilde_j((t - t_i)/Delta t_i) would remove ambiguity.
  2. [Section 4.2, Table 3] The table heading "Total Ipopt Callbacks #Epochs" is ambiguous because it is unclear which columns are times and which are counts; adding units such as "Total [s], Ipopt [s], Callbacks [s], #Epochs" would improve readability.
  3. [Section 4.2.1] The phrase "the optimization terminates prematurely, since the optimality tolerance is fulfilled" is contradictory; "prematurely" should be replaced by wording such as "terminates early" or "terminates at a local optimum" to reflect that termination is due to optimality, not an error.
  4. [Section 5.3] The statement that OpenModelica's index reduction and BLT transformations "restructure DAEs into semi-explicit ODE form with index 1" is imprecise; the standard result is a transformation into a semi-explicit index-1 DAE form, not generally into an ODE form.
  5. [Section 4.1.1] The road input derivative dot z_r = u appears in the model equations, but the link between the ISO 8608 road profile generation and the input u is not formally specified in the data-generation paragraph; a brief statement would clarify the setup.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the training derivation is a standard collocation transcription, and the benchmark claims are validated fits against reference data rather than predictions forced by construction.

full rationale

The paper's derivation chain is self-contained. Training is formulated as a dynamic optimization problem (Eq. 6) and transcribed into an NLP via Radau IIA direct collocation (Eqs. 8, 12); this is a standard numerical transcription, not a result whose conclusion is embedded in its assumptions. The objective is a data-fit loss, and the benchmarks validate the fitted surrogates against reference models and unseen road profiles (Type C validation after Type D training), which is legitimate empirical validation rather than a renamed prediction. The stability, order, and step-size claims are supported by standard properties of Radau IIA and by external references, not by a self-citation chain. The authors do cite their own prior work ([20], OpenModelica-related papers), but those citations concern implementation details, quadrature construction explanation, or planned future integration, and none carries the central training claim. The paper also honestly discloses its main limitations: the collocation grid must be fixed a-priori (Section 2.5.1), the QVM benchmark uses a hand-selected 2500-interval grid chosen because of very fast dynamics (Section 4.1.2), and the Appendix shows that high noise can lead to poor local optima or failed convergence. These are scope and robustness caveats, not circular reductions. There is no equation, fitted parameter, or uniqueness invocation that is equivalent by construction to the claimed result, so no circular step is identified.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

No new physical entities are introduced. The central claim rests on standard collocation theory, on hand-selected hyperparameters (grids, regularizer, network sizes), and on the assumption that the NLP solver converges from the chosen guesses. The method contributes the reformulation and implementation, while the benchmark accuracy depends on settings chosen by the authors.

free parameters (5)
  • QVM neural surrogate weights for F_fr^NN and F_pr^NN = 92 weights, values not reported in text
    Trained by the NLP in strategies I and II; all QVM benchmark plots depend on these weights.
  • QVM rational function coefficients = 32 coefficients (N=D=7)
    Trained in strategy III; used to show that non-NN surrogates also work.
  • Van-der-Pol NODE weights = 102 weights
    Trained on noisy trajectory data; central to the VdP benchmark.
  • Regularization weight lambda = 1e-4 for sigma=0 and 0.1; 1e-3 for sigma=0.5
    Chosen by hand per noise level; affects the objective and convergence behavior.
  • Collocation grid configuration = QVM: 2500 intervals, 5 fLGR nodes; VdP: 500 intervals, 5 fLGR nodes; global run: 1 interval, 70 nodes
    Grid and node count are a-priori choices identified as critical in Section 2.5.1 and directly affect accuracy and runtime.
assumptions (4)
  • domain assumption All model functions, including neural activations, are twice continuously differentiable.
    Required for Ipopt's second-order information; Section 2.1 states this, and the authors choose squareplus and sigmoid activations. This excludes ReLU-style nonsmooth components from the proposed training pipeline.
  • standard math fLGR/Radau IIA collocation yields a high-order, stable discretization for smooth ODEs.
    Textbook numerical analysis result, invoked in Section 2.3. It underpins the claim that the discretized NLP faithfully represents the continuous ODE.
  • ad hoc to paper The fixed a-priori grids (2500 or 500 intervals) resolve the true dynamics.
    Section 2.5.1 says the grid must be given a-priori; QVM uses 2500 intervals and VdP uses 500 intervals. The reported accuracy is conditional on these hand-picked grids.
  • ad hoc to paper From the chosen initialization, Ipopt converges to a suitable local optimum.
    The paper's own Section 2.5.2 and the high-noise VdP sensitivity analysis show that poor local optima occur; the runtime and stability claims assume good convergence for the tested cases.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Efficient Training of Physics-enhanced Neural ODEs via Direct Collocation and Nonlinear Programming." pith.science (2026). https://pith.science/paper/62OOEDM2

@misc{pith2026250503552,
  author       = {Pith},
  title        = {Pith review of: Efficient Training of Physics-enhanced Neural ODEs via Direct Collocation and Nonlinear Programming},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/62OOEDM2}},
  note         = {Machine review of arXiv:2505.03552}
}
read the original abstract

We propose a novel approach for training Physics-enhanced Neural ODEs (PeN-ODEs) by expressing the training process as a dynamic optimization problem. The full model, including neural components, is discretized using a high-order implicit Runge-Kutta method with flipped Legendre-Gauss-Radau points, resulting in a large-scale nonlinear program (NLP) efficiently solved by state-of-the-art NLP solvers such as Ipopt. This formulation enables simultaneous optimization of network parameters and state trajectories, addressing key limitations of ODE solver-based training in terms of stability, runtime, and accuracy. Extending on a recent direct collocation-based method for Neural ODEs, we generalize to PeN-ODEs, incorporate physical constraints, and present a custom, parallelized, open-source implementation. Benchmarks on a Quarter Vehicle Model and a Van-der-Pol oscillator demonstrate superior accuracy, speed, generalization with smaller networks compared to other training techniques. We also outline a planned integration into OpenModelica to enable accessible training of Neural DAEs.

Figures

Figures reproduced from arXiv: 2505.03552 by the authors.

Figure 1
Figure 1. Modelica Models of the Linear (without boxes) and Neural (with boxes) QVM. Modified from T. Kamp. [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Objective History with Respect to Training Time [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Neural and Reference Damper Characteristics [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (5 more)
Figure 5
Figure 5. Figure 5: Body Accelerations ab for Simulations of PeN-ODE and Standard Models on a Type C Road [34] By having principal knowledge of the underlying characteristics shown in [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Simulation Results of the Neural and Reference Models as well as the Data for [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 8
Figure 8. Figure 8: Scalar Fields Representing the 2-Norm Error Between the Neural (σ = 0.5, σ = 0.1) and Reference Vector Fields (Values > 6 are white) In addition to the local collocation approach with 500 intervals, we also reproduce the results in [12] using a global, spectral colloca…
Figure 9
Figure 9. Figure 9: OpenModelica Workflow for PeN-ODE Training (under development) [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]
Figure 10
Figure 10. Figure 10: Sensitivity analysis of the learned Van-der-Pol (VdP) Neural ODEs to different levels of Gaussian noise. For [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 22 canonical work pages

  1. [12]

    Training neural odes using fully discretized simultaneous optimization,

    Mariia Shapovalova and Calvin Tsay. Training neural odes using fully discretized simultaneous optimization,

    work page

  2. [2]

    Closing the sim-to-real gap with physics-enhanced neural odes

    Tobias Kamp, Johannes Ultsch, and Jonathan Brembeck. Closing the sim-to-real gap with physics-enhanced neural odes. In Guiseppina Gini, Henk Nijmeijer, and Dimitar Filev, editors, 20th International Conference on Informatics in Control, Automation and Robotics, ICINCO 2023, volume 2 ofProceedings of the 20th International Conference on Informatics in Cont...

    work page 2023

  3. [1]

    Neural ordinary differential equations

    Tian Qi Chen, Yulia Rubanova, Jesse Bettencourt, and David Duvenaud. Neural ordinary differential equations. CoRR, abs/1806.07366, 2018. URL http://arxiv.org/abs/1806.07366

    arXiv 2018

  4. [3]

    Universal differential equations for scientific machine learning

    Christopher Rackauckas, Yingbo Ma, Julius Martensen, Collin Warner, Kirill Zubov, Rohit Supekar, Dominic Skinner, and Ali Jasim Ramadhan. Universal differential equations for scientific machine learning. CoRR, abs/2001.04385, 2020. URL https://arxiv.org/abs/2001.04385

    arXiv 2001

  5. [4]

    Physics-enhanced neural ordinary differential equations: Application to industrial chemical reaction systems

    Farshud Sorourifar, You Peng, Ivan Castillo, Juan Venegas, and Joel Paulson. Physics-enhanced neural ordinary differential equations: Application to industrial chemical reaction systems. Industrial & Engineering Chemistry Research, 62, 09 2023. doi:10.1021/acs.iecr.3c01471

    doi:10.1021/acs.iecr.3c01471 2023

  6. [5]

    Modia and julia for grey box modeling

    Frederic Bruder and Lars Mikelsons. Modia and julia for grey box modeling. In Proceedings of the 14th International Modelica Conference, pages 87–95, 09 2021. doi:10.3384/ecp2118187

    work page doi:10.3384/ecp2118187 2021

  7. [6]

    Neuralfmu: Presenting a workflow for in- tegrating hybrid neuralodes into real-world applications

    Tobias Thummerer, Johannes Stoljar, and Lars Mikelsons. Neuralfmu: Presenting a workflow for in- tegrating hybrid neuralodes into real-world applications. Electronics, 11(19), 2022. ISSN 2079-9292. doi:10.3390/electronics11193202. URL https://www.mdpi.com/2079-9292/11/19/3202

    doi:10.3390/electronics11193202 2022

  8. [7]

    Maximizing information from chemical engineering data sets: Applications to machine learning

    Alexander Thebelt, Johannes Wiebe, Jan Kronqvist, Calvin Tsay, and Ruth Misener. Maximizing information from chemical engineering data sets: Applications to machine learning. Chemical Engineering Science, 252: 117469, 02 2022. doi:10.1016/j.ces.2022.117469

    arXiv 2022

Show all 40 references
  1. [8]

    Capturing missing physics in climate model parameterizations using neural differential equations, 2023

    Ali Ramadhan, John Marshall, Andre Souza, Xin Kai Lee, Ulyana Piterbarg, Adeline Hillier, Gregory LeClaire Wagner, Christopher Rackauckas, Chris Hill, Jean-Michel Campin, and Raffaele Ferrari. Capturing missing physics in climate model parameterizations using neural differenti...

    2023 arXiv

  2. [9]

    Formulating data-driven surrogate models for process optimization

    Ruth Misener and Lorenz Biegler. Formulating data-driven surrogate models for process optimization. Computers & Chemical Engineering , 179:108411, 2023. ISSN 0098-1354. doi:https://doi.org/10.1016/j.compchemeng.2023.108411. URL https://www.sciencedirect.com/ science/article/pi...

    2023

  3. [10]

    Accelerating neural odes using model order reduction

    Mikko Lehtimäki, Lassi Paunonen, and Marja-Leena Linne. Accelerating neural odes using model order reduction. IEEE Transactions on Neural Networks and Learning Systems, 35(1):519–531, January 2024. ISSN 2162-2388. doi:10.1109/tnnls.2022.3175757. URL http://dx.doi.org/10.1109/T...

    2024

  4. [11]

    Collocation based training of neural ordinary differential equations

    Elisabeth Roesch, Chris Rackauckas, and Michael Stumpf. Collocation based training of neural ordinary differential equations. Statistical Applications in Genetics and Molecular Biology, 20, 07 2021. doi:10.1515/sagmb- 2020-0025. 14 Efficient Training of PeN-ODEs via Direct Col...

    2021 doi

  5. [13]

    Andreas Wächter and Lorenz T. Biegler. On the implementation of an interior-point filter line-search algorithm for large-scale nonlinear programming. Mathematical Programming, 106(1):25–57, March 2006. ISSN 1436-4646. doi:10.1007/s10107-004-0559-y. URL https://doi.org/10.1007/...

    2006 doi

  6. [14]

    The openmodelica integrated environment for modeling, simulation, and model-based development

    Peter Fritzson, Adrian Pop, Karim Abdelhak, et al. The openmodelica integrated environment for modeling, simulation, and model-based development. Modeling, Identification and Control: A Norwegian Research Bulletin, 41:241–295, 10 2020. doi:10.4173/mic.2020.4.1

    2020 doi

  7. [15]

    V . M. Becerra. Solving complex optimal control problems at no cost with psopt. In 2010 IEEE International Sym- posium on Computer-Aided Control System Design, pages 1391–1396, 2010. doi:10.1109/CACSD.2010.5612676

    2010

  8. [16]

    CasADi – A software framework for nonlinear optimization and optimal control

    Joel A E Andersson, Joris Gillis, Greg Horn, James B Rawlings, and Moritz Diehl. CasADi – A software framework for nonlinear optimization and optimal control. Mathematical Programming Computation, 11(1):1–36,

  9. [17]

    Patterson and Anil V

    Michael A. Patterson and Anil V . Rao. Gpops-ii: A matlab software for solving multiple-phase optimal control problems using hp-adaptive gaussian quadrature collocation methods and sparse nonlinear programming. ACM Trans. Math. Softw., 41(1), October 2014. ISSN 0098-3500. doi:...

    2014 doi

  10. [18]

    Efficient imple- mentation of collocation methods for optimization using openmodelica and adol-c

    Vitalij Ruge, Willi Braun, Bernhard Bachmann, Andrea Walther, and Kshitij Kulshreshtha. Efficient imple- mentation of collocation methods for optimization using openmodelica and adol-c. In Proceedings of the 10th International Modelica Conference, March 10-12, 2014, Lund, Swed...

    2014 doi

  11. [19]

    Dynamic optimization in jmodelica.org

    Fredrik Magnusson and Johan Åkesson. Dynamic optimization in jmodelica.org. Processes, 3(2):471–496, 2015. ISSN 2227-9717. doi:10.3390/pr3020471. URL https://www.mdpi.com/2227-9717/3/2/471

    2015 doi

  12. [20]

    Adaptively refined mesh for collocation-based dynamic optimization, 12 2024

    Linus Langenkamp. Adaptively refined mesh for collocation-based dynamic optimization, 12 2024. URL https://dx.doi.org/10.13140/RG.2.2.18499.72484. Master’s thesis

    2024

  13. [21]

    Schneider and W

    C. Schneider and W. Werner. Some new aspects of rational interpolation. Math. Comp., 47(175):285–299, 1986. doi:10.1090/S0025-5718-1986-0842136-8. URL https://www.ams.org/journals/mcom/1986-47-175/ S0025-5718-1986-0842136-8/

    1986 doi

  14. [22]

    P. E. Gill, W. Murray, M. A. Saunders, and Elizabeth Wong. Snopt 7.7 user’s manual. Technical Report CCoM Technical Report 18-1, Center for Computational Mathematics, University of California, San Diego, San Diego, CA, 2007

    2007

  15. [23]

    Gill, Walter Murray, and Michael A

    Philip E. Gill, Walter Murray, and Michael A. Saunders. Snopt: An sqp algorithm for large-scale constrained optimization. SIAM Rev., 47(1):99–131, January 2005. ISSN 0036-1445. doi:10.1137/S0036144504446096. URL https://doi.org/10.1137/S0036144504446096

    2005 doi

  16. [24]

    Byrd, Jorge Nocedal, and Richard A

    Richard H. Byrd, Jorge Nocedal, and Richard A. Waltz.Knitro: An Integrated Package for Nonlinear Optimization, pages 35–59. Springer US, Boston, MA, 2006. ISBN 978-0-387-30065-8. doi:10.1007/0-387-30065-1_4. URL https://doi.org/10.1007/0-387-30065-1_4

    2006 doi

  17. [25]

    Amestoy, Iain S

    Patrick R. Amestoy, Iain S. Duff, Jean-Yves L’Excellent, and Jacko Koster. A fully asynchronous multifrontal solver using distributed dynamic scheduling. SIAM Journal on Matrix Analysis and Applications, 23(1):15–41, 2001

    2001

  18. [26]

    HSL: A collection of fortran codes for large-scale scientific computation

    HSL. HSL: A collection of fortran codes for large-scale scientific computation. Available at http://www.hsl. rl.ac.uk, 2013. Accessed: 15-04-2025

    2013

  19. [27]

    L. T. Biegler. Nonlinear Programming. Concepts, Algorithms, and Applications to Chemical Processes, volume 10 of MOS-SIAM Series on Optimization. Society for Industrial and Applied Mathematics (SIAM), Philadelphia, PA,

  20. [28]

    Dynamic optimization using local collocation methods and improved multiresolution technique

    Jisong Zhao and Teng Shang. Dynamic optimization using local collocation methods and improved multiresolution technique. Applied Sciences, 8:1680, 09 2018. doi:10.3390/app8091680

    2018 doi

  21. [29]

    Adaptive mesh refinement method for optimal control us- ing nonsmoothness detection and mesh size reduction

    Fengjin Liu, William Hager, and Anil Rao. Adaptive mesh refinement method for optimal control us- ing nonsmoothness detection and mesh size reduction. Journal of the Franklin Institute , 47, 05 2015. doi:10.1016/j.jfranklin.2015.05.028

    2015 doi

  22. [30]

    Distributed optimization and statistical learning via the alternating direction method of multipliers

    Stephen Boyd, Neal Parikh, Eric Chu, Borja Peleato, and Jonathan Eckstein. Distributed optimization and statistical learning via the alternating direction method of multipliers. Foundations and Trends® in Machine Learning, 3(1): 1–122, 2011. ISSN 1935-8237. doi:10.1561/2200000...

    2011 doi

  23. [31]

    Lueg, Victor Alves, Daniel Schicksnus, John R

    Laurens R. Lueg, Victor Alves, Daniel Schicksnus, John R. Kitchin, Carl D. Laird, and Lorenz T. Biegler. A simultaneous approach for training neural differential-algebraic systems of equations, 2025. URL https: //arxiv.org/abs/2504.04665

    2025

  24. [32]

    Generation of sparse jacobians for the function mock-up interface 2.0

    Johan Åkesson, Willi Braun, Petter Lindholm, and Bernhard Bachmann. Generation of sparse jacobians for the function mock-up interface 2.0. In Book of abstracts / 9th International Modelica Conferenc , volume 76 of Linköping Electronic Conference Proceedings, pages 185–196. Lin...

    2012 doi

  25. [33]

    Parallel program- ming in OpenMP

    Rohit Chandra, Leo Dagum, David Kohr, Ramesh Menon, Dror Maydan, and Jeff McDonald. Parallel program- ming in OpenMP. Morgan Kaufmann, 2001

    2001

  26. [34]

    Simulated road profiles according to iso 8608 in vibration analysis

    Peter Mú ˇcka. Simulated road profiles according to iso 8608 in vibration analysis. Journal of Testing and Evaluation, 46:20160265, 01 2018. doi:10.1520/JTE20160265

    2018 doi

  27. [35]

    Neural Network Library in Modelica

    Fabio Codecà and Francesco Casella. Neural Network Library in Modelica. InProceedings of the 5th International Modelica Conference, volume 2, pages 549–557, Vienna, Austria, September 2006. Modelica Association

    2006

  28. [36]

    Pseudo array causalization

    Karim Abdelhak, Francesco Casella, and Bernhard Bachmann. Pseudo array causalization. In Proceedings of the 15th International Modelica Conference, pages 177–188, Aachen, Germany, 12 2023. doi:10.3384/ecp204177

    2023 doi

  29. [37]

    Compiler status and development of the new backend, 2025

    Karim Abdelhak and Bernhard Bachmann. Compiler status and development of the new backend, 2025. URL https://openmodelica.org/events/openmodelica-workshop/2025. Presentation at the 17th OpenMod- elica Annual Workshop - February 3, 2025, accessed on 2025-04-25. 16 Efficient Trai...

    2025

  30. [2010]

    URL https://doi.org/10.1137/1.9780898719383

    doi:10.1137/1.9780898719383. URL https://doi.org/10.1137/1.9780898719383

    doi

  31. [2019]

    doi:10.1007/s12532-018-0139-4

    doi

  32. [2025]

    URL https://arxiv.org/abs/2502.15642

    arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.