pith. sign in

arxiv: 2605.16330 · v1 · pith:ZTUXZKIVnew · submitted 2026-05-05 · ⚛️ physics.chem-ph · cond-mat.mtrl-sci· physics.comp-ph

A Data-Driven Parametric Reduced-Order Chemical Kinetics Model Derived from Atomistic Simulations

Michael N. Sakano , Alejandro Strachan This is my paper

Pith reviewed 2026-05-21 00:10 UTC · model grok-4.3

classification ⚛️ physics.chem-ph cond-mat.mtrl-sciphysics.comp-ph
keywords reduced-order modelingchemical kineticsautoencoderatomistic simulationsenergetic materialsmachine learningparametric modelinterpretability
0
0 comments X

The pith

A parametric autoencoder yields an interpretable reduced-order chemical kinetics model from atomistic simulations across temperatures.

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

The paper introduces a machine learning method to derive simplified models of chemical reactions directly from detailed molecular simulations. It builds a single neural network that accounts for how reactions change with temperature by using an autoencoder with special constraints to keep the internal representations tied to real chemical species. The network also learns how fast reactions happen and how much heat they release at the same time. This matters for a sympathetic reader because it could allow simulations of large-scale events like explosions or fires without needing to track every atom, making them practical on computers.

Core claim

The authors present a parametric, temperature-dependent autoencoder framework that learns a unified reduced-order description of chemical decomposition across a wide range of temperatures within a single model. Physical interpretability is enforced through non-negativity constraints and a softmax activation, enabling the latent variables to be directly associated with additive chemical components and their relative contributions. Reaction kinetics and heat-release parameters are optimized simultaneously within the neural-network architecture, providing a self-consistent coupling between chemical evolution and energetics. This yields significantly improved reconstruction accuracy compared to,

What carries the argument

Parametric temperature-dependent autoencoder with non-negativity constraints and softmax activation for learning additive chemical components.

If this is right

  • Chemical evolution and energy release remain consistent because they are optimized together in the network.
  • The model can describe decomposition at many temperatures using one set of parameters rather than separate models.
  • Latent representations stay interpretable, allowing direct links to physical chemical species.
  • Reconstruction of chemical data is more accurate than with conventional dimensionality reduction techniques.

Where Pith is reading between the lines

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

  • Applying this framework to other materials or reaction types could extend its use beyond energetic materials to fields like catalysis or biochemistry.
  • Validation against experimental measurements of reaction rates at various temperatures would strengthen confidence in the model's predictions.
  • Embedding additional physical constraints, such as conservation laws, might enhance the model's reliability for long-term simulations.

Load-bearing premise

The non-negativity constraints and softmax activation will produce latent variables that correspond to real additive chemical components with optimizable kinetics and energetics.

What would settle it

A direct comparison showing whether the model's predicted species evolution and heat release match atomistic simulation results at a temperature outside the training set, or if the latent components fail to align with expected chemical species identities.

Figures

Figures reproduced from arXiv: 2605.16330 by Alejandro Strachan, Michael N. Sakano.

Figure 1
Figure 1. Figure 1: Representative bonding environments (BEs) and schematic overview of the feedforward autoencoder architecture. (a) Examples of local bonding environments shown using both pictorial representations and corresponding alphabetic notations (left). The associated time series of all 280 BE descriptors obtained from a molecular dynamics simulation are shown on the right. (b) Architecture of the autoencoder, which … view at source ↗
Figure 2
Figure 2. Figure 2: Temperature-specific autoencoder models and construction of temperature-dependent kernel weights. (a) Encoded latent variables obtained from autoencoder models trained indepen￾dently at each isothermal temperature, showing the evolution of three interpretable components associated with reactants, intermediates, and products. (b) Schematic illustrating the extraction of temperature-dependent kernel weights:… view at source ↗
Figure 3
Figure 3. Figure 3: Performance of the parametric, interpretable autoencoder. (a) Encoded latent variables obtained from the temperature-dependent autoencoder, in which the isothermal temperature is provided as an explicit input to construct the encoder and decoder kernel weights for each compo￾nent. The resulting latent-space trajectories exhibit consistent, physically interpretable behavior across temperatures. (b) Parity p… view at source ↗
Figure 4
Figure 4. Figure 4: (a) compares the fitted ROCK model predictions (dashed lines) with the autoencoder￾encoded components (solid lines). The optimized kinetic parameters are Z1 = 35.481 ps−1 , E1 = 24.301 kcal mol−1 Z2 = 7.693 ps−1 , E2 = 23.000 kcal mol−1 Relative to Ref. 26, the prefactor Z1 is modestly larger, while Z2 is approximately four times smaller. Because the activation energies remain comparable, these differences… view at source ↗
Figure 5
Figure 5. Figure 5: Adiabatic temperature evolution comparing molecular dynamics simulations with pre￾dictions from the reduced-order chemical kinetics model coupled with heat release. Across all initial temperatures, the model accurately captures both the transient temperature evolution and the overall exothermicity, demonstrating the suitability of the chosen functional form for the heat￾release parameters. To assess model … view at source ↗
Figure 6
Figure 6. Figure 6: Time-lagged, stacked parametric autoencoder for isothermal dynamics. The model takes the component concentrations at time t, the isothermal temperature T, and the timestep ∆t as inputs and iteratively propagates the system forward to predict concentrations at t + ∆t, where N denotes the number of stacked prediction steps. Multi-step temporal evolution is achieved through repeated application of a single-st… view at source ↗
Figure 7
Figure 7. Figure 7: Time-lagged, stacked parametric autoencoder for adiabatic dynamics. The model takes the component concentrations at time t, the temperature T, and the timestep ∆t as inputs and iteratively propagates both concentrations and temperature to t + n∆t, where N is the number of stacked prediction steps. Adiabatic temperature evolution is incorporated through the coupled kinetics–heat-release formulation. Finally… view at source ↗
Figure 8
Figure 8. Figure 8: Self-consistent time-lagged, stacked parametric autoencoder. The architecture follows the construction shown in [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
read the original abstract

Coarse-grained modeling in molecular simulations serves not only to extend accessible time and length scales beyond atomistic limits, but also to reduce high-dimensional chemical data to low-dimensional representations that expose the underlying latent structure. In the context of energetic materials, reduced-order chemical kinetics models are essential for describing thermally driven decomposition, deflagration, and detonation. Recent data-driven approaches based on machine learning and dimensionality reduction have shown promise for constructing such models directly from atomistic simulations; however, when reaction pathways vary strongly with thermodynamic conditions, these methods can produce latent representations that are difficult to interpret physically or extrapolate reliably. Here, we introduce a parametric, temperature-dependent autoencoder framework that learns a unified reduced-order description of chemical decomposition across a wide range of temperatures within a single model. Physical interpretability is enforced through non-negativity constraints and a softmax activation, enabling the latent variables to be directly associated with additive chemical components and their relative contributions. Reaction kinetics and heat-release parameters are optimized simultaneously within the neural-network architecture, providing a self-consistent coupling between chemical evolution and energetics. The proposed approach yields significantly improved reconstruction accuracy compared to a state-of-the-art dimensionality-reduction method, as quantified by reductions in mean-squared error, while preserving a physically meaningful latent representation. These results demonstrate that parametric, interpretable machine-learning models can provide robust reduced-order chemical kinetics suitable for multiscale modeling of complex reactive systems.

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 introduces a parametric, temperature-dependent autoencoder that learns a unified reduced-order model of chemical decomposition from atomistic simulations of energetic materials. Non-negativity constraints and softmax activation are used to enforce that latent variables represent additive chemical components; reaction kinetics and heat-release parameters are optimized jointly inside the network. The central claim is that this yields lower mean-squared reconstruction error than a state-of-the-art dimensionality-reduction baseline while preserving a physically meaningful latent representation suitable for multiscale modeling.

Significance. If the interpretability and accuracy claims are substantiated, the method would supply a self-consistent, temperature-extrapolatable reduced-order kinetics model that couples chemistry and energetics, addressing a practical need in simulations of decomposition and detonation.

major comments (2)
  1. [§4.3 and Table 2] §4.3 and Table 2: the reported MSE reductions versus the baseline method are presented without error bars, without the exact baseline implementation details, and without a clear statement of the train/test split across temperatures; these omissions make it impossible to judge whether the improvement is statistically robust or merely an artifact of the particular data partitioning.
  2. [§3.2, Eq. (8)] §3.2, Eq. (8) and the accompanying text: the claim that non-negativity plus softmax produces latent variables that 'directly correspond to additive chemical components' is asserted from the architecture alone; no quantitative comparison to known species concentrations or reaction pathways from the underlying atomistic trajectories is shown, leaving open the possibility that the constraints are satisfied without recovering chemically meaningful decomposition channels.
minor comments (2)
  1. [Figure 3] Figure 3 caption: the temperature values used for the extrapolation test should be stated explicitly rather than referred to only as 'outside the training range.'
  2. [Methods] Notation: the symbol T_p for the parametric temperature input is introduced without a clear definition of its range or normalization; this should be added to the methods section.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading of the manuscript and for the constructive comments. We address each of the major points below and indicate the revisions we will make to strengthen the presentation.

read point-by-point responses

  1. Referee: [§4.3 and Table 2] §4.3 and Table 2: the reported MSE reductions versus the baseline method are presented without error bars, without the exact baseline implementation details, and without a clear statement of the train/test split across temperatures; these omissions make it impossible to judge whether the improvement is statistically robust or merely an artifact of the particular data partitioning.

    Authors: We agree that the current presentation of the MSE results would benefit from additional statistical context. In the revised manuscript we will add error bars computed from multiple independent training runs with different random seeds, provide a detailed description of the baseline implementation (including any specific hyperparameters and software references), and explicitly state the train/test partitioning procedure used across the temperature range. These additions will allow readers to evaluate the robustness of the reported improvements. revision: yes

  2. Referee: [§3.2, Eq. (8)] §3.2, Eq. (8) and the accompanying text: the claim that non-negativity plus softmax produces latent variables that 'directly correspond to additive chemical components' is asserted from the architecture alone; no quantitative comparison to known species concentrations or reaction pathways from the underlying atomistic trajectories is shown, leaving open the possibility that the constraints are satisfied without recovering chemically meaningful decomposition channels.

    Authors: The non-negativity constraint together with the softmax activation mathematically guarantees that each latent vector consists of non-negative entries that sum to one, thereby representing additive fractional contributions by construction. This architectural choice is what enables the direct association with chemical components. While the manuscript does not contain a side-by-side quantitative match between the learned latent variables and specific species concentrations extracted from the atomistic trajectories, the physical utility of the representation is evidenced by the joint optimization with kinetics and heat-release terms and by the improved reconstruction fidelity. We will add a brief discussion of this point and, if feasible, a supplementary comparison in the revised version. revision: partial

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The paper trains a parametric autoencoder on atomistic simulation data to produce a temperature-dependent reduced-order kinetics model. Reconstruction accuracy is measured via MSE against a baseline dimensionality-reduction method, and interpretability is imposed via non-negativity and softmax constraints within the network architecture. No equation or claim reduces the reported improvement or physical meaning to a fitted quantity defined by the same inputs, a self-citation chain, or a renaming of known results. The central results rest on empirical performance of the trained model rather than tautological re-expression of the training data.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Review performed on abstract only; therefore the ledger records only the explicit modeling choices stated in the abstract and notes that all numerical details, training data, and validation procedures remain unknown.

axioms (1)
  • domain assumption Non-negativity constraints and softmax activation enforce that latent variables represent additive chemical components.
    Stated directly in the abstract as the mechanism for physical interpretability.

pith-pipeline@v0.9.0 · 5787 in / 1173 out tokens · 52810 ms · 2026-05-21T00:10:41.931963+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

68 extracted references · 68 canonical work pages · 5 internal anchors

  1. [1]

    Journal of Physics: Conference Series , volume=

    Shock-induced hotspot formation and chemical reaction initiation in PETN containing a spherical void , author=. Journal of Physics: Conference Series , volume=. 2014 , organization=

    work page 2014

  2. [2]

    The Journal of Physical Chemistry C , volume=

    Ultrafast chemistry under nonequilibrium conditions and the shock to deflagration transition at the nanoscale , author=. The Journal of Physical Chemistry C , volume=. 2015 , publisher=

    work page 2015

  3. [3]

    AIP conference proceedings , volume=

    Ignition chemistry in HMX from thermal explosion to detonation , author=. AIP conference proceedings , volume=. 2002 , organization=

    work page 2002

  4. [4]

    Combustion Theory and Modelling , volume=

    Detonation waves in PBX 9501 , author=. Combustion Theory and Modelling , volume=. 2006 , publisher=

    work page 2006

  5. [5]

    The Journal of Physical Chemistry , volume=

    Critical conditions for impact-and shock-induced hot spots in solid explosives , author=. The Journal of Physical Chemistry , volume=. 1996 , publisher=

    work page 1996

  6. [6]

    Combustion and Flame , volume=

    Thermal decomposition models for HMX-based plastic bonded explosives , author=. Combustion and Flame , volume=. 2004 , publisher=

    work page 2004

  7. [7]

    AIP Conference Proceedings , volume=

    Modeling thermal ignition and the initial conditions for internal burning in PBX 9501 , author=. AIP Conference Proceedings , volume=. 2009 , organization=

    work page 2009

  8. [8]

    Combustion and flame , volume=

    Evaluation of reaction kinetics models for meso-scale simulations of hotspot initiation and growth in HMX , author=. Combustion and flame , volume=. 2020 , publisher=

    work page 2020

  9. [9]

    Comparison of ReaxFF, DFTB, and DFT for phenolic pyrolysis. 2. Elementary reaction paths , author=. The Journal of Physical Chemistry A , volume=. 2013 , publisher=

    work page 2013

  10. [10]

    Proceedings of the National Academy of Sciences , volume=

    First-principles--based reaction kinetics from reactive molecular dynamics simulations: Application to hydrogen peroxide decomposition , author=. Proceedings of the National Academy of Sciences , volume=. 2019 , publisher=

    work page 2019

  11. [11]

    Journal of chemical theory and computation , volume=

    Mechanism of graphene formation via detonation synthesis: a DFTB nanoreactor approach , author=. Journal of chemical theory and computation , volume=. 2019 , publisher=

    work page 2019

  12. [12]

    Combustion and Flame , volume=

    Reactive molecular dynamics simulation and chemical kinetic modeling of pyrolysis and combustion of n-dodecane , author=. Combustion and Flame , volume=. 2011 , publisher=

    work page 2011

  13. [13]

    Journal of chemical theory and computation , volume=

    Automated discovery of reaction pathways, rate constants, and transition states using reactive molecular dynamics simulations , author=. Journal of chemical theory and computation , volume=. 2015 , publisher=

    work page 2015

  14. [14]

    Journal of chemical information and modeling , volume=

    Automated chemical kinetic modeling via hybrid reactive molecular dynamics and quantum chemistry simulations , author=. Journal of chemical information and modeling , volume=. 2018 , publisher=

    work page 2018

  15. [15]

    Nature chemistry , volume=

    Discovering chemistry with an ab initio nanoreactor , author=. Nature chemistry , volume=. 2014 , publisher=

    work page 2014

  16. [16]

    Journal of chemical theory and computation , volume=

    Automated discovery and refinement of reactive molecular dynamics pathways , author=. Journal of chemical theory and computation , volume=. 2016 , publisher=

    work page 2016

  17. [17]

    Chemical Science , volume=

    The non-adiabatic nanoreactor: towards the automated discovery of photochemistry , author=. Chemical Science , volume=. 2021 , publisher=

    work page 2021

  18. [18]

    The Journal of Physical Chemistry A , volume=

    Nitromethane decomposition via automated reaction discovery and an ab initio corrected kinetic model , author=. The Journal of Physical Chemistry A , volume=. 2021 , publisher=

    work page 2021

  19. [19]

    SMILES, a chemical language and information system. 1. Introduction to methodology and encoding rules , author=. Journal of chemical information and computer sciences , volume=. 1988 , publisher=

    work page 1988

  20. [20]

    SLAS Discovery , volume=

    Development of CYP3A4 inhibition models: comparisons of machine-learning techniques and molecular descriptors , author=. SLAS Discovery , volume=. 2005 , publisher=

    work page 2005

  21. [21]

    Chemometrics and Intelligent Laboratory Systems , volume=

    Large-scale QSAR study of aromatase inhibitors using SMILES-based descriptors , author=. Chemometrics and Intelligent Laboratory Systems , volume=. 2014 , publisher=

    work page 2014

  22. [22]

    Chemical science , volume=

    Learning continuous and data-driven molecular descriptors by translating equivalent chemical representations , author=. Chemical science , volume=. 2019 , publisher=

    work page 2019

  23. [23]

    Molecular Systems Design & Engineering , volume=

    Deep learning for molecular design—a review of the state of the art , author=. Molecular Systems Design & Engineering , volume=. 2019 , publisher=

    work page 2019

  24. [24]

    The Journal of chemical physics , volume=

    Mirrored continuum and molecular scale simulations of the ignition of high-pressure phases of RDX , author=. The Journal of chemical physics , volume=. 2016 , publisher=

    work page 2016

  25. [25]

    Combustion and Flame , volume=

    Mirrored continuum and molecular scale simulations of deflagration in a nano-slab of HMX , author=. Combustion and Flame , volume=. 2020 , publisher=

    work page 2020

  26. [26]

    Bulletin of the American Physical Society , year=

    Unsupervised learning-based multiscale model of thermochemistry in 1, 3, 5-trinitro-1, 3, 5-triazine (RDX) , author=. Bulletin of the American Physical Society , year=

    work page

  27. [27]

    Propellants, Explosives, Pyrotechnics , volume=

    Developing reaction chemistry models from reactive molecular dynamics: TATB , author=. Propellants, Explosives, Pyrotechnics , volume=. 2022 , publisher=

    work page 2022

  28. [28]

    Chemical Reviews , volume=

    Estimation of heats of formation of organic compounds by additivity methods , author=. Chemical Reviews , volume=. 1993 , publisher=

    work page 1993

  29. [29]

    AIChE Journal , volume=

    New group contribution method for estimating properties of pure compounds , author=. AIChE Journal , volume=. 1994 , publisher=

    work page 1994

  30. [30]

    The Journal of Physical Chemistry A , volume=

    Atomic-level features for kinetic monte carlo models of complex chemistry from molecular dynamics simulations , author=. The Journal of Physical Chemistry A , volume=. 2021 , publisher=

    work page 2021

  31. [31]

    The Journal of Chemical Physics , volume=

    Heuristics for chemical species identification in dense systems , author=. The Journal of Chemical Physics , volume=. 2020 , publisher=

    work page 2020

  32. [32]

    2007 , institution=

    Exploring network structure, dynamics, and function using NetworkX , author=. 2007 , institution=

    work page 2007

  33. [33]

    Journal of Chemical Theory and Computation , volume=

    A quantum-based approach to predict primary radiation damage in polymeric networks , author=. Journal of Chemical Theory and Computation , volume=. 2020 , publisher=

    work page 2020

  34. [34]

    SIAM review , volume=

    Finding structure with randomness: Probabilistic algorithms for constructing approximate matrix decompositions , author=. SIAM review , volume=. 2011 , publisher=

    work page 2011

  35. [35]

    Applied and Computational Harmonic Analysis , volume=

    A randomized algorithm for the decomposition of matrices , author=. Applied and Computational Harmonic Analysis , volume=. 2011 , publisher=

    work page 2011

  36. [36]

    nature , volume=

    Learning the parts of objects by non-negative matrix factorization , author=. nature , volume=. 1999 , publisher=

    work page 1999

  37. [37]

    Proceedings of the 17th International Conference on Pattern Recognition, 2004

    Application of non-negative and local non negative matrix factorization to facial expression recognition , author=. Proceedings of the 17th International Conference on Pattern Recognition, 2004. ICPR 2004. , volume=. 2004 , organization=

    work page 2004

  38. [38]

    IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics) , volume=

    Graph-preserving sparse nonnegative matrix factorization with application to facial expression recognition , author=. IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics) , volume=. 2010 , publisher=

    work page 2010

  39. [39]

    Statistical Analysis and Data Mining: The ASA Data Science Journal , volume=

    Nonnegative tensor decomposition with custom clustering for microphase separation of block copolymers , author=. Statistical Analysis and Data Mining: The ASA Data Science Journal , volume=. 2019 , publisher=

    work page 2019

  40. [40]

    Machine Learning: Science and Technology , volume=

    A neural network for determination of latent dimensionality in non-negative matrix factorization , author=. Machine Learning: Science and Technology , volume=. 2021 , publisher=

    work page 2021

  41. [41]

    From Principal Subspaces to Principal Components with Linear Autoencoders

    From principal subspaces to principal components with linear autoencoders , author=. arXiv preprint arXiv:1804.10253 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  42. [42]

    PloS one , volume=

    Deep neural networks show an equivalent and often superior performance to dermatologists in onychomycosis diagnosis: Automatic construction of onychomycosis datasets by region-based convolutional deep neural network , author=. PloS one , volume=. 2018 , publisher=

    work page 2018

  43. [43]

    European Journal of Cancer , volume=

    Deep neural networks are superior to dermatologists in melanoma image classification , author=. European Journal of Cancer , volume=. 2019 , publisher=

    work page 2019

  44. [44]

    arXiv preprint arXiv:2209.01862 , year=

    Kinetics parameter optimization via neural ordinary differential equations , author=. arXiv preprint arXiv:2209.01862 , year=

    work page arXiv

  45. [45]

    The Journal of chemical physics , volume=

    Algorithmic dimensionality reduction for molecular structure analysis , author=. The Journal of chemical physics , volume=. 2008 , publisher=

    work page 2008

  46. [46]

    Dimensionality reduction methods for molecular simulations

    Dimensionality reduction methods for molecular simulations , author=. arXiv preprint arXiv:1710.10629 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  47. [47]

    The Journal of chemical physics , volume=

    Collective variable discovery and enhanced sampling using autoencoders: Innovations in network architecture and error function design , author=. The Journal of chemical physics , volume=. 2018 , publisher=

    work page 2018

  48. [48]

    Nature communications , volume=

    Deep learning for universal linear embeddings of nonlinear dynamics , author=. Nature communications , volume=. 2018 , publisher=

    work page 2018

  49. [49]

    The Journal of chemical physics , volume=

    Time-lagged autoencoders: Deep learning of slow collective variables for molecular kinetics , author=. The Journal of chemical physics , volume=. 2018 , publisher=

    work page 2018

  50. [50]

    Physical Review E , volume=

    Variational encoding of complex dynamics , author=. Physical Review E , volume=. 2018 , publisher=

    work page 2018

  51. [51]

    Journal of chemical theory and computation , volume=

    Transferable neural networks for enhanced sampling of protein dynamics , author=. Journal of chemical theory and computation , volume=. 2018 , publisher=

    work page 2018

  52. [52]

    Neural networks , volume=

    Stacked generalization , author=. Neural networks , volume=. 1992 , publisher=

    work page 1992

  53. [53]

    AIChE Journal , volume=

    Process modeling using stacked neural networks , author=. AIChE Journal , volume=. 1996 , publisher=

    work page 1996

  54. [54]

    The Journal of Physical Chemistry A , volume=

    Coupled thermal and electromagnetic induced decomposition in the molecular explosive HMX; a reactive molecular dynamics study , author=. The Journal of Physical Chemistry A , volume=. 2014 , publisher=

    work page 2014

  55. [55]

    The Journal of Physical Chemistry C , volume=

    Reactive molecular dynamics simulations to investigate the shock response of liquid nitromethane , author=. The Journal of Physical Chemistry C , volume=. 2019 , publisher=

    work page 2019

  56. [56]

    The Journal of Physical Chemistry A , volume=

    Insight into the chemistry of PETN under shock compression through ultrafast broadband mid-infrared absorption spectroscopy , author=. The Journal of Physical Chemistry A , volume=. 2020 , publisher=

    work page 2020

  57. [57]

    parallel computing , volume=

    Parallel reactive molecular dynamics: Numerical methods and algorithmic techniques , author=. parallel computing , volume=. 2012 , publisher=

    work page 2012

  58. [58]

    Journal of computational physics , volume=

    Fast parallel algorithms for short-range molecular dynamics , author=. Journal of computational physics , volume=. 1995 , publisher=

    work page 1995

  59. [59]

    The Journal of Physical Chemistry C , volume=

    Role of Molecular Disorder on the Reactivity of RDX , author=. The Journal of Physical Chemistry C , volume=. 2018 , publisher=

    work page 2018

  60. [60]

    Structure Reports , volume=

    Redetermination of cyclo-trimethylenetrinitramine , author=. Structure Reports , volume=. 2008 , publisher=

    work page 2008

  61. [61]

    IEICE transactions on fundamentals of electronics, communications and computer sciences , volume=

    Fast local algorithms for large scale nonnegative matrix and tensor factorizations , author=. IEICE transactions on fundamentals of electronics, communications and computer sciences , volume=. 2009 , publisher=

    work page 2009

  62. [62]

    Neural computation , volume=

    Algorithms for nonnegative matrix factorization with the -divergence , author=. Neural computation , volume=. 2011 , publisher=

    work page 2011

  63. [63]

    Nature methods , volume=

    SciPy 1.0: fundamental algorithms for scientific computing in Python , author=. Nature methods , volume=. 2020 , publisher=

    work page 2020

  64. [64]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Xception: Deep learning with depthwise separable convolutions , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    work page

  65. [65]

    TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems

    Tensorflow: Large-scale machine learning on heterogeneous distributed systems , author=. arXiv preprint arXiv:1603.04467 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  66. [66]

    A Critical Review of Recurrent Neural Networks for Sequence Learning

    A critical review of recurrent neural networks for sequence learning , author=. arXiv preprint arXiv:1506.00019 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  67. [67]

    Learning Adversary-Resistant Deep Neural Networks

    Learning adversary-resistant deep neural networks , author=. arXiv preprint arXiv:1612.01401 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  68. [68]

    Strang, in Linear Algebra and Its Applications, 2nd ed

    G. Strang, in Linear Algebra and Its Applications, 2nd ed. (Academic Press, Inc., Orlando, FL, 1980), pp. 139--142

    work page 1980