Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Deep Space One”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 19 records
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

Bridging the gap between experiments and simulations using machine learning

The physics of inertial confinement fusion is rich and complex. Simulation codes that are used to design experiments are computationally expensive and lack the predictive capability required for extensive parameter exploration in search of a high-performing design for laser direct drive. In this work we use deep learning to build a fast emulator of experiments. To facilitate the development of the deep-learning model, an autoencoder is used to reduce the dimensionality of the input space. Two deep learning models are developed. One model is trained on a vast array of simulation data and is subsequently calibrated to expensive and limited experimental data using a technique known as “transfer learning.” The other model is trained on a statistical model and is subsequently calibrated using experimental data. A comparative study of the two predictive models is carried out. The models potentially reproduce key experimental observables with high accuracy and unprecedented inference times relative to those achieved with simulation codes. These models facilitate rapid exploration of a high dimensional input parameter space.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2024

CubeSat Interplanetary Exploration: A Deep Dive into Nuclear Propulsion and Astrodynamics for Small Satellites

CubeSats are small satellites most commonly used by universities, however, nowadays places like NASA are looking to CubeSats for interplanetary missions. Because of their universal form fit it makes them cost effective, economical, and therefor desirable for deep space. To achieve deep space exploration, we must explore the applications of propulsion and astrodynamics on small satellites. This paper will touch on one scaling Ion and chemical propulsion as well as nuclear kilo reactors. Two choreographing a flight path for these small satellites and lastly designing an efficient cost-effective platform for CubeSats to get to deep space territory.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗
DOE OSTI2024

Uncertainty Quantification and Sensitivity Analysis of Low-Dimensional Manifold via Co-Kurtosis PCA in Combustion Modeling

For multi-scale multi-physics applications e.g., the turbulent combustion code Pele, robust and accurate dimensionality reduction is crucial to solving problems at exascale and beyond. A recently developed technique, Co-Kurtosis based Principal Component Analysis (CoK-PCA) which leverages principal vectors of co-kurtosis, is a promising alternative to traditional PCA for complex chemical systems. To improve the effectiveness of this approach, we employ Artificial Neural Networks for reconstructing thermo-chemical scalars, species production rates, and overall heat release rates corresponding to the full state space. Our focus is on bolstering confidence in this deep learning based non-linear reconstruction through Uncertainty Quantification (UQ) and Sensitivity Analysis (SA). UQ involves quantifying uncertainties in inputs and outputs, while SA identifies influential inputs. One of the noteworthy challenges is the computational expense inherent in both endeavors. To address this, we employ the Monte Carlo methods to effectively quantify and propagate uncertainties in our reduced spaces while managing computational demands. Our research carries profound implications not only for the realm of combustion modeling but also for a broader audience in UQ. By showcasing the reliability and robustness of CoK-PCA in dimensionality reduction and deep learning predictions, we empower researchers and decision-makers to navigate complex combustion systems with greater confidence.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2025

Alpha Decay Chains as Thermal Power Sources: Analysis and Applications for RTGs

Radioactive sources can provide power in remote and environmentally harsh locations such as the arctic or space. The generators powered by such sources are rugged and can withstand extreme temperatures, lack of sunlight, and require no human intervention for multiple years. Radioisotopes are used in thermoelectric generators to provide power at remote sites and deep in space. Isotopes like Pu-238, Cm-244, and Am-241 are used in these generators by NASA for power in space probes and spacecrafts. These power sources deliver a steady supply of energy over extended periods of time. Alpha particles created during decay do not travel far in a material. Their kinetic energy is transferred to heat that we can then convert into energy. Unlike beta and gamma decay, the slower-moving alpha particles stop in the material, making their energy available for use. Energy from these natural decay processes provides a reliable source of power. Spontaneous fission is rare and unreliable, and unlike induced fission processes, alpha decay occurs naturally and does not require external management or ignition. The ideal properties of an isotope for use as a power source depend upon the intended use. For use in an Arctic research base over a period of several years, but less than a decade, an isotope that provides high power output over a shorter lifespan may be the most suitable option. Whereas, for deep space missions where a consistent power source for decades or perhaps more than 100 years is needed that would require a very different isotope. One with a much longer half-life that would provide consistent power throughout that time and survive in that state in for these extended periods of time. These examples represent two extreme sides in terms of time frames. By analyzing the power produced by different radioactive decay processes over time, we can evaluate the suitability of various isotope decay chains for specific uses. Some unstable isotopes undergo a series of radioactive decays, transforming into different isotopes at each step and resulting in a stable isotope. The lists of isotopes in these decay processes are known as decay chains. Some of these chains, illustrated in the figures below, are currently being investigated for use in radioisotope thermoelectric generators (RTGs) designed for a range of operational durations.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2024

Neural network representations of multiphase Equations of State

Abstract Equations of State model relations between thermodynamic variables and are ubiquitous in scientific modelling, appearing in modern day applications ranging from Astrophysics to Climate Science. The three desired properties of a general Equation of State model are adherence to the Laws of Thermodynamics, incorporation of phase transitions, and multiscale accuracy. Analytic models that adhere to all three are hard to develop and cumbersome to work with, often resulting in sacrificing one of these elements for the sake of efficiency. In this work, two deep-learning methods are proposed that provably satisfy the first and second conditions on a large-enough region of thermodynamic variable space. The first is based on learning the generating function (thermodynamic potential) while the second is based on structure-preserving, symplectic neural networks, respectively allowing modifications near or on phase transition regions. They can be used either “from scratch” to learn a full Equation of State, or in conjunction with a pre-existing consistent model, functioning as a modification that better adheres to experimental data. We formulate the theory and provide several computational examples to justify both approaches, highlighting their advantages and shortcomings.

Science & Technology - Other Topics↗
DOE OSTI2024

On the Training and Generalization of Deep Operator Networks

Here, we present a novel training method for deep operator networks (DeepONets), one of the most popular neural network models for operators. DeepONets are constructed by two subnetworks, namely the branch and trunk networks. Typically, the two subnetworks are trained simultaneously, which amounts to solving a complex optimization problem in a high dimensional space. In addition, the nonconvex and nonlinear nature makes training very challenging. To tackle such a challenge, we propose a two-step training method that trains the trunk network first and then sequentially trains the branch network. The core mechanism is motivated by the divide-and-conquer paradigm and is the decomposition of the entire complex training task into two subtasks with reduced complexity. Therein the Gram–Schmidt orthonormalization process is introduced which significantly improves stability and generalization ability. On the theoretical side, we establish a generalization error estimate in terms of the number of training data, the width of DeepONets, and the number of input and output sensors. Numerical examples are presented to demonstrate the effectiveness of the two-step training method, including Darcy flow in heterogeneous porous media.

deep operator networks↗
DOE OSTI2025

Hierarchical Gaussian process-based Bayesian optimization for materials discovery in high entropy alloy spaces

Bayesian optimization (BO) is a powerful and data-efficient method for iterative materials discovery and design, particularly valuable when prior knowledge is limited, underlying functional relationships are complex or unknown, and the cost of querying the materials space is significant. Traditional BO methodologies typically utilize conventional Gaussian Processes (cGPs) to model the relationships between material inputs and properties, as well as correlations within the input space. However, cGP-BO approaches often fall short in multi-objective optimization scenarios, where they are unable to fully exploit correlations between distinct material properties. Leveraging these correlations can significantly enhance the discovery process, as information about one property can inform and improve predictions about others. Here, this study addresses this limitation by employing advanced kernel structures to capture and model multi-dimensional property correlations through multi-task (MTGPs) or deep Gaussian Processes (DGPs), thus accelerating the discovery process. We demonstrate the effectiveness of MTGP-BO and DGP-BO in rapidly and robustly solving complex materials design challenges that occur within the context of complex multi-objective optimization over FCC FeCrNiCoCu high entropy alloy (HEA) spaces, where traditional cGP-BO approaches fail. Furthermore, we highlight how the differential costs associated with querying various material properties can be strategically leveraged to make the materials discovery process more cost-efficient.

36 MATERIALS SCIENCE↗
DOE OSTI2026

Recurrent convolutional neural networks for modeling nonadiabatic dynamics of quantum-classical systems

Recurrent neural networks (RNNs) have recently been extensively applied to model the time evolution in fluid dynamics, weather predictions, and even chaotic systems due to their ability to capture temporal dependencies and sequential patterns in data. Here we present an RNN model based on convolutional neural networks for modeling the nonlinear nonadiabatic dynamics of hybrid quantum-classical systems. The dynamical evolution of the hybrid systems is governed by equations of motion for classical degrees of freedom and von Neumann equation for electrons. The Physics-Aware Recurrent Convolution (PARC) neural network structure incorporates a differentiator-integrator architecture that inductively models the spatiotemporal dynamics of generic physical systems. Here, we apply our RNN approach to learn the space-time evolution of a one-dimensional semiclassical Holstein model after an interaction quench. For shallow quenches (small changes in electron-lattice coupling), the deterministic dynamics can be accurately captured using a single-CNN-based recurrent network. In contrast, deep quenches induce chaotic evolution, making long-term trajectory prediction significantly more challenging. Nonetheless, we demonstrate that the PARC-CNN architecture can effectively learn the statistical climate of the Holstein model under deep-quench conditions.

Holstein model↗
DOE OSTI2025

Energy-efficient, Large-scale Molecular Dynamics Simulations via Hardware- and Algorithm-level Optimization

This work aims to develop a framework for energy-efficient computing that will enable molecular dynamics (MD) simulations of large-scale phenomena with atomic precision and simultaneously remove computational bottlenecks limiting the speed of MD simulations. We seek to implement such an approach through the development of surrogate models for the interatomic force calculation combined with the use of mixed numerical precision formats. For a model system of neutral atoms (only pairwise interactions), significant force calculation efficiency improvements were achieved, without detrimental effects on atomic structures or average energies, using single precision, by developing a surrogate model (deep neural network), and by quantizing this surrogate model. For a model system of charged atoms, the reciprocal-space calculation of electrostatic interactions was identified as the main bottleneck, and the development of a surrogate model should be pursued to achieve an estimated one-order-of-magnitude additional speedup.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
DOE OSTI2026

Neural-Network Inverse Design of SRF Cavities and Transmons for Bosonic Quantum Computation

Three-dimensional superconducting radio-frequency (SRF) cavities provide exceptionally long-lived electromagnetic modes and, when coupled to nonlinear elements such as transmon qubits, become promising architectures for bosonic quantum information processing. The inverse design of such systems, i.e., recovering device geometries that produce specified electromagnetic and coupling targets, is generally a one-to-many problem. The qubit-cavity coupling strength depends sensitively on both the transmon geometry and its position within the cavity's electromagnetic field. As these systems scale up and their design parameter spaces grow, the cost of conventional iterative simulation becomes prohibitive. We present two deep neural network (DNN) approaches that address this inverse-design problem at complementary levels of the design stack. The first proposes SRF cavity geometries that produce target cavity observables. The second proposes transmon qubit designs that produce target qubit-cavity parameters - the coupling rate, qubit frequency, and anharmonicity $(g, ν_q, α)$. The recovered candidate designs match the targets to within ~5% (cavity) and ~2% (transmon), confirmed by end-to-end re-simulation. Both approaches map desired device behavior directly to candidate designs, a fast alternative to the iterative simulation studies usually required.

Yaker, Joseph [Fermilab; Northwestern U.]↗
DOE OSTI2025

Diffusion power spectra as a window into dynamic materials architecture

Chemical recycling of commodity and specialty polymers presents a multifaceted challenge for industrial societies. On one hand, macromolecular architectures must be engineered to yield durable products that, on the other hand, rapidly deconstruct to recyclable monomers under pre-determined conditions. Polymer deconstruction is a chemical process that requires deep understanding of molecular reactivity in heterogeneous media, where porous material architectures evolve in both space and time. To build this understanding, we develop herein experimental and analytical methods describing sets of diffusive eigenmodes that exist within time-varying, non-Euclidean boundary conditions, a situation commonly encountered in the reactive deconstruction of polymers where chain fragments splay, alter their local dynamics, and evolve in their confinement of reacting media. Diffusion power spectra, discerned experimentally by NMR, yield polymer and solvent frequency-domain velocity autocorrelation functions that are analyzed in the context of physical models for chemical reactions parameterized with fractal mathematics. The results connect local motion in polymers to chemical reactivity during acidolysis of circular elastomers.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2026

Performance and transport in the ARC tokamak

The ARC TM tokamak, a high-field (𝐵 𝑇 = 11.4 T) fusion power plant, under development by Commonwealth Fusion Systems, is studied using a suite of integrated modelling tools to predict its fusion power generation (𝑃𝑓⁡𝑢⁢𝑠), transport and confinement properties. Analysis is based off an ARC operational point scoped first with zero-dimensional (0-D) plasma operational contour (POPCON) modelling to produce 1.13 GW of fusion power. A suite of integrated modelling tools (TRANSP, ASTRA and TORAX) were applied to predict the performance and kinetic profiles of the ARC design point, yielding a range of predicted performance spanning from ∼900 to 1300 MW in rough quantitative agreement with POPCON predictions. The sensitivity of these results to uncertain modelling inputs was probed using scans of pedestal boundary conditions around EPED-predicted values (total pressure and temperature ratios), tungsten concentration and seperatrix density around their nominal assumptions. Pedestal pressure and pedestal top (𝑇 𝑖 /𝑇 𝑒 ) play a large role in 1.5-dimensional performance predictions, able to modify the predicted 𝑃 𝑓⁡𝑢⁢𝑠 by a factor of 2 within reasonable assumptions. High-fidelity core nonlinear gyrokinetic profile predictions, performed using CGYRO (Candy et al. 2016 J. Comput. Phys., vol. 324, pp. 73–93) coupled with the PORTALS (Rodriguez-Fernandez et al. 2024 Nucl. Fusion, vol. 64, 076034; Phys. Plasmas, vol. 31, 2024, 062501) framework, yield substantially lower performance (𝑃 𝑓⁡𝑢⁢𝑠 =677 MW) compared with 0-D and medium-fidelity modelling for nominal assumptions, showing that there is non-negligible uncertainty between models and that future work on SPARC may help resolve discrepancies. Lower overall performance results from significantly reduced volume-averaged densities and temperatures, along with reduced levels of density and temperature peaking. Turbulence and transport are largely dominated by ion temperature gradient across the profile, confirmed by both linear stability and the response of the nonlinear fluxes to changes in gradients, with some impact of kinetic ballooning modes in the deep core. This work represents one of the most complete scoping of potential fusion power plant conditions performed to date. The extensive integrated modelling provides confidence in ARC performance approaching 1 GW, while nonlinear gyrokinetic modelling results in open questions into the physics of density and temperature peaking in fusion-power-plant-relevant operational space. A discussion of results and the role that the SPARC tokamak (Creely et al. 2020 J. Plasma Phys., vol. 86, 865860502) will play in informing ARC design, performance and operation is presented.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2026

A New Window into the Baryon Cycle at Cosmic Noon with Line Intensity Mapping: Forecasts for auto- and cross-correlations in [CII]-158$μ$m, HI 21 cm, CO$_{J+1\rightarrow J}$, and H$α$ galaxies

Across the peak of cosmic star formation at $z\sim1-2$, inflow, processing, and feedback drive rapid changes in the spatial distribution and chemical composition of baryons in galaxies and surrounding reservoirs; this baryon cycle can be tomographically mapped by line intensity mapping (LIM) of atomic hydrogen, ionized carbon, and carbon monoxide. We present a simulation-based forecasting framework for detecting auto- and cross-power spectra between spectroscopic surveys of four such tracers at $z\sim0.5-1.7$ mapping the same deep field - TIM, EoRSpec/FYST, MeerKAT, & Euclid. We forward-model 3-D distributions for these tracers from magnetohydrodynamic simulations, directly capturing the two-halo, one-halo, and shot statistics without relying on analytical decompositions. We further detail a signal-to-noise formalism, tailored to LIM surveys with highly anisotropic geometries and Fourier-space coverage. We demonstrate that galaxy cross-correlations will be the dominant discovery channel for current-generation surveys. These instruments will detect the auto-spectra for CO and HI 21 cm and the CO $\times$ 21 cm cross-spectrum at modest S/N $\sim 1-10$, while placing upper limits on the [CII]-158$μ$m signals. [CII], CO, and HI LIM will be $\sim3-30\times$ ($0.5-1.5$ dex) more sensitive to cross-correlation with the Euclid survey, however, than their respective auto-correlations, constraining all three models of line emission at high significance (S/N $\sim 10-40$) within this decade. Finally, we formulate a staged instrumental trajectory with planned or reasonable improvements, including the as-proposed SKA-Mid. We forecast advancing the per-$k$-mode sensitivities of each auto-, galaxy-line, and line-line spectrum by several orders of magnitude, enabling new percent- and sub-percent level constraints on cosmology and the redshift evolution of star formation and the baryon cycle.

Agrawal, Shubh [Pennsylvania U., Dept. Math.]↗
DOE OSTI2024

Distributed-Memory Sparse Deep Neural Network Inference Using Global Arrays

Partitioned Global Address Space (PGAS) models exhibit tremendous promise in developing efficient and productive distributed-memory parallel applications. They have been used extensively in scientific computations due to conveniently offering a ``shared-memory''-like model and convenient interfaces that separate communication with synchronization. Traditionally, PGAS communication models have been applied to dense/contiguously distributed data, but most modern applications depict varied levels of sparsity. Existing PGAS models require certain adaptations to support distributed sparse computations, since associated computations often require matrix arithmetic, in addition to data movement. The Global Arrays toolkit from Pacific Northwest National Laboratory (PNNL) is one of the earliest PGAS models to combine one-sided data communication and distributed matrix operations and is still used in the popular NWChem quantum chemistry suite. Recently, we have expanded the Global Arrays toolkit to support common sparse operations, like sparse matrix-dense matrix multiplies (SpMM), sparse matrix-sparse matrix multiplication (SpGEMM) and Sampled Dense-Dense Matrix Multiplication (SDDMM). As it turns out, these operations are the bedrock of sparse Deep Learning (DL); sparse deep neural networks and Graph Neural Networks (GNNs) have gained increasing attention recently in achieving speedups on training and inference with reduced memory footprints. Unlike scientific applications in High Performance Computing (HPC), modern (distributed-memory capable) DL toolkits often rely on non-standardized and closed-source vendor software optimizations, creating challenges in software-hardware co-design at scale. Our goal is to support a variety of distributed-memory sparse matrix operations and helper functions in the newly created Sparse Global Arrays (SGA), such that it is possible to build portable and productive Machine Learning scenarios for algorithm/software and hardware codesign purposes. Contemporary data-parallel schemes for training/inference are undergoing a major overhaul since model replication limits scalability and causes resource inefficiencies. As such, we have adopted tensor parallelism in decomposing the model and inputs, to mitigate memory issues. Current implementation is built on top of MPI and uses CPUs to maximize the portability across the platforms.

Distributed computing, machine learning↗
DOE OSTI2025

Osmotic control of the spacing of parallel shear cracks in shale growing subcritically in geologic past

The geological genesis of natural cracks in sedimentary rocks such as shale is a problem that needs to be understood to improve the technology of hydraulic fracturing as well as deep sequestration of harmful fluids. Why are the vertical natural cracks roughly parallel and equidistant, and why is the spacing roughly 10 cm rather than 1 cm or 100 cm? Fracture mechanics of critical cracks cannot answer this question. Neither can the material heterogeneity. The growth of critical parallel cracks is impossible because the relative crack face displacements would immediately localize into one crack, leading to an earthquake. The cracks must have formed, on the tectonic time scale, by a slow growth of subcritical shear cracks governed by the Charles-Evans law. The idea advanced here is that what controls the crack spacing is the balance between the reduction, due to shear dilatancy, of the concentration of ions such as Na + and Cl - in each fracture process zone (PFZ), which decelerates the cracks, and the restoration of ion concentration by diffusion of ions from the space between the cracks into the FPZ. This diffusion of water is driven mainly by the osmotic pressure gradient, which offsets the deceleration and depends strongly on the crack spacing. A simple analytical solution of the steady state is rendered possible by approximating the ion concentration profiles between adjacent cracks by parabolic arcs. Applying this theory to Woodford shale yields the approximate crack spacing of 10 cm, which is realistic. Furthermore, the stability of unlimited parallel mode II frictional crack growth is proven by examining the second variation of the free energy. Water concentration drop in the FPZ due to shear dilatancy and its restoration by water diffusion from the inter-crack space have similar effect, although probably much weaker.

42 ENGINEERING↗
DOE OSTI2025

Diffusion Model-Guided Inverse Design of Bimetallic Catalysts for Ammonia Decomposition

In the past decade, artificial intelligence and deep learning have played increasingly prominent roles in materials design and discovery. Among these, generative AI models, known for their ability to create unique and complex structures, have emerged as state-of-the-art tools for materials screening due to their high efficiency and low computational cost. In catalysis, one of the major challenges is identifying promising material candidates within an immense chemical space. This challenge can be addressed using generative approaches, such as diffusion-based inverse design models. In this study, we present a machine learning-guided workflow that employed a diffusion model for the inverse design of bimetallic alloy catalysts for low-carbon ammonia decomposition, a key reaction for ammonia emission control and sustainable hydrogen production. Catalyst candidates were evaluated using nitrogen adsorption energy as the key descriptor, inspired by multiscale modeling. The proposed workflow identified low-cost, environmentally friendly catalysts with excellent catalytic performance, which have been validated theoretically and experimentally. Our framework decoupled the generative and property-prediction components, enhancing both flexibility and accuracy in the catalytic material design process.

Adsorption↗
DOE OSTI2026

Dark Energy Survey Year 3 results: Simulation-based 𝑤CDM inference from weak lensing and galaxy clustering maps with deep learning: Analysis design

Data-driven approaches using deep learning are emerging as powerful techniques to extract non-Gaussian information from cosmological large-scale structure. Here, this work presents the first simulation-based inference (SBI) pipeline that combines weak lensing and galaxy clustering maps in a realistic Dark Energy Survey Year 3 (DES Y3) configuration and serves as preparation for a forthcoming analysis of the survey data. We develop a scalable forward model based on the CosmoGridV1 suite of N-body simulations to generate over one million self-consistent mock realizations of DES Y3 at the map level. Leveraging this large dataset, we train deep graph convolutional neural networks on the full survey footprint in spherical geometry to learn low-dimensional features that approximately maximize mutual information with target parameters. These learned compressions enable neural density estimation of the implicit likelihood via normalizing flows in a ten-dimensional parameter space spanning cosmological 𝑤CDM, intrinsic alignment, and linear galaxy bias parameters, while marginalizing over baryonic, photometric redshift, and shear bias nuisances. To ensure robustness, we extensively validate our inference pipeline using synthetic observations derived from both systematic contaminations in our forward model and independent Buzzard galaxy catalogs. Our forecasts yield significant improvements in cosmological parameter constraints, achieving 2−3× higher figures of merit in the 𝛺 𝑚 − 𝑆 8 plane relative to our implementation of baseline two-point statistics and effectively breaking parameter degeneracies through probe combination. These results demonstrate the potential of SBI analyses powered by deep learning for upcoming Stage-IV wide-field imaging surveys.

Thomsen, A. [Zurich, ETH] (ORCID:0000000203099021)↗
DOE OSTI2025

Toward Complete Merger Identification at Cosmic Noon with Deep Learning

As we enter the era of large imaging surveys such as Roman, Rubin, and Euclid, a deeper understanding of potential biases and selection effects in optical astronomical catalogs created with the use of ML-based methods is paramount. This work focuses on a deeper understanding of the performance and limitations of deep learning-based classifiers as tools for galaxy merger identification. We train a ConvNeXT-Pico model on mock HST CANDELS images from the IllustrisTNG50 simulation. Our focus is on a more challenging classification of galaxy mergers and non-mergers at higher redshifts 1 < z < 1.5, including minor mergers and lower mass galaxies down to the stellar mass of 108M⊙. We demonstrate, for the first time, that a deep learning model, such as the one developed in this work, can successfully identify even minor and low mass mergers even at these redshifts. Our model achieves overall accuracy, purity, and completeness of over 73%. We show that some galaxy mergers can only be identified from certain observation angles, leading to a potential upper limit in overall accuracy. Using Grad-CAMs and UMAPs, we more deeply examine the performance and observe a visible gradient in the latent space with stellar mass and specific star formation rate, but no visible gradient with merger mass ratio or merger stage.

Schechter, Aimee L. [U. Colorado, Boulder]↗