Search NASA⌕ Search

SEARCH · Search NASA

Results for “adjoint problem”

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 217 records · Page 12

Development of a Library for Conducting Monte Carlo Tallies on Heterogeneous Systems

A library is introduced that enables users to conduct Monte Carlo tallies on heterogeneous computing systems. The library can be built with or linked to a Monte Carlo transport code to provide tally capabilities for the project. The library provides a set of tools and an interface for constructing tallies that can be computed on CPUs or GPUs regardless of where the sample data was generated. The library can be incorporated into existing Monte Carlo transport codes at both the research and production level with minimal intrusion into the transport code. The tally library is capable of computing a variety of tallies including track-length histogram (mesh) tallies, collisionbased kernel density estimators, functional expansion tallies, eigenvalue sensitivity coefficients via iterated fission probability, and sensitivities of reaction rates or adjoint-weighted quantities via differential operator sampling. The parallelization methods are discussed and speedups are shown for 2-D and 3-D problems with comparisons to both the tally library CPU implementations as well as the tally capabilities in the host transport code. Performance results and examples are shown for MCNP, MCATK, and OpenMC.

97 MATHEMATICS AND COMPUTING↗

Nonlinear Optimal Control of Electron Dynamics Within Hartree-Fock Theory

Consider the problem of determining the optimal applied electric field to drive a molecule from an initial state to a desired target state. For even moderately sized molecules, solving this problem directly using the exact equations of motion—the time-dependent Schrödinger equation (TDSE)—is numerically intractable. Here, we present a solution of this problem within time-dependent Hartree-Fock (TDHF) theory, a mean field approximation of the TDSE. Optimality is defined in terms of minimizing the total control effort while maximizing the overlap between desired and achieved target states. We frame this problem as an optimization problem constrained by the nonlinear TDHF equations; we solve it using trust region optimization with gradients computed via a custom-built adjoint state method. For three molecular systems, we show that with very small neural network parametrizations of the control, our method yields solutions that achieve desired targets within acceptable constraints and tolerances.

97 MATHEMATICS AND COMPUTING↗

Dynamic Learning of Correlation Potentials for a Time-Dependent Kohn-Sham System

We develop methods to learn the correlation potential for a time-dependent Kohn-Sham (TDKS) system in one spatial dimension. We start from a low-dimensional two-electron system for which we can numerically solve the time-dependent Schr¨odinger equation; this yields electron densities suitable for training models of the correlation potential. We frame the learning problem as one of optimizing a least-squares objective subject to the constraint that the dynamics obey the TDKS equation. Applying adjoints, we develop efficient methods to compute gradients and thereby learn models of the correlation potential. Our results show that it is possible to learn values of the correlation potential such that the resulting electron densities match ground truth densities. We also show how to learn correlation potential functionals with memory, demonstrating one such model that yields reasonable results for trajectories outside the training set.

97 MATHEMATICS AND COMPUTING↗

Quantum vacuum energy in graphs and billiards

The vacuum (Casimir) energy in quantum field theory is a problem relevant both to new nanotechnology devices and to dark energy in cosmology. The crucial question is the dependence of the energy on the system geometry. Despite much progress since the first prediction of the Casimir effect in 1948 and its subsequent experimental verification in simple geometries, even the sign of the force in nontrivial situations is still a matter of controversy. Mathematically, vacuum energy fits squarely into the spectral theory of second-order self-adjoint elliptic linear differential operators. Specifically one promising approach is based on the small-t asymptotics of the cylinder kernel e{sup -t{radical}(H)}, where H is the self-adjoint operator under study. In contrast with the well-studied heat kernel e{sup -tH}, the cylinder kernel depends in a non-local way on the geometry of the problem. We discuss some results by the Louisiana-Oklahoma-Texas collaboration on vacuum energy in model systems, including quantum graphs and two-dimensional cavities. The results may shed light on general questions, including the relationship between vacuum energy and periodic or closed classical orbits, and the contribution to vacuum energy of boundaries, edges, and corners.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Methods for Detector Placement and Analysis of Criticality Accident Alarm Systems

Determining the optimum placement to minimize the number of detectors for a criticality accident alarm system (CAAS) in a large manufacturing facility is a complex problem. There is typically a target for the number of detectors that can be used over a given zone of the facility. A study to optimize detector placement typically begins with some initial guess at the placement of the detectors and is followed by either predictive calculations of accidents at specific locations or adjoint calculations based on preferred detector locations. Within an area of a facility, there may be a large number of potential criticality accident sites. For any given placement of the detectors, the list of accident sites can be reduced to a smaller number of locations at which accidents may be difficult for detectors to detect. Developing the initial detector placement and determining the list of difficult accident locations are both based on the practitioner's experience. Simulations following fission particles released from an accident location are called 'forward calculations.' These calculations can be used to answer the question 'where would an alarm be triggered?' by an accident at a specified location. Conversely, 'adjoint calculations' start at a detector site using the detector response function as a source and essentially run in reverse. These calculations can be used to answer the question 'where would an accident be detected?' by a specified detector location. If the number of accidents, P, is much less than the number of detectors, Q, then forward simulations may be more convenient and less time-consuming. If Q is large or the detectors are not placed yet, then a mesh tally of dose observed by a detector at any location must be computed over the entire zone. If Q is much less than P, then adjoint calculations may be more efficient. Adjoint calculations employing a mesh tally can be even more advantageous because they do not rely on a list of specific difficult-to-detect accident sites, which may not have included every possible accident location. Analog calculations (no biasing) simply follow particles naturally. For sparse buildings and line-of-sight calculations, analog Monte Carlo (MC) may be adequate. For buildings with internal walls or large amounts of heavy equipment (dense geometry), variance reduction may be required. Calculations employing the CADIS method use a deterministic calculation to create an importance map and a matching biased source distribution that optimize the final MC to quickly calculate one specific tally. Calculations employing the FW-CADIS method use two deterministic calculations (one forward and one adjoint) to create an importance map and a matching biased source distribution that are designed to make the MC calculate a mesh tally with more uniform uncertainties in both high-dose and low-dose areas. Depending on the geometry of the problem, the number of detectors, and the number of accident sites, different approaches to CAAS placement studies can be taken. These are summarized in Table I. SCALE 6.1 contains the MAVRIC sequence, which can be used to perform any of the forward-based approaches outlined in Table I. For analog calculations, MAVRIC simply calls the Monaco MC code. For CADIS and FW-CADIS, MAVRIC uses the Denovo discrete ordinates (SN) deterministic code to generate the importance map and biased source used by Monaco. An adjoint capability is currently being added to Monaco and should be available in the next release of SCALE. An adjoint-based approach could be performed with Denovo alone - although fine meshes, large amounts of memory, and long computation times may be required to obtain accurate solutions. Coarse-mesh SN simulations could be employed for adjoint-based scoping studies until the adjoint capability in Monaco is complete. CAAS placement studies, especially those dealing with mesh tallies, require some extra utilities to aid in the analysis. Detectors must receive a minimum dose rate in order to alarm; therefore, a simple yes/no plot could be more useful to the analyst than a standard dose rate contour plot. Alarm systems that require several detectors to alarm simultaneously for a given accident would need to combine several yes/no plots in order to show accident sites where multiple detectors would be triggered. This could require a plot of 'areas of coverage' that is mapped over the building geometry. Several new utilities (which will be part of SCALE 6.2) were created to help with CAAS analysis.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Training material models using gradient descent algorithms

High temperature design requires accurate constitutive models to describe material inelastic deformation and failure behavior. Oftentimes, calibrating accurate models devolves into the problem of fitting the model parameters against experimental test data. Here, we present the pyopmat package, an open source framework for calibrating constitutive models against experiment data subjected to various loading conditions using machine learning techniques. The package calculates the exact gradient of the model response with respect to the parameters using a combination of automatic differentiation and the adjoint method. Given this exact gradient, we compare the performance of several gradient-based optimization techniques in fitting realistic constitutive models against data. Here, we demonstrate the efficiency and accuracy of our package through example problems using both synthetic data, generated using known parameter sets, under monotonic and cyclic loading conditions and also with an example applying the techniques developed here to actual high temperature creep-fatigue test data.

36 MATERIALS SCIENCE↗

AMOC sensitivity to surface buoyancy fluxes: Stronger ocean meridional heat transport with a weaker volume transport?

Oceanic northward heat transport is commonly assumed to be positively correlated with the Atlantic meridional overturning circulation (AMOC). For example, in numerical "water-hosing" experiments, imposing anomalous freshwater fluxes in the northern Atlantic leads to a slow-down of the AMOC and the corresponding reduction of oceanic northward heat transport. Here, we study the sensitivity of the ocean heat and volume transports to surface heat and freshwater fluxes using a generalized stability analysis. For the sensitivity to surface freshwater fluxes, we find that, while the direct relationship between the AMOC volume and heat transports holds on shorter time scales, it can reverse on timescales longer than 500 years or so. That is, depending on the model surface boundary conditions, reduction in the AMOC volume transport can potentially lead to a stronger heat transport on long timescales, resulting from the gradual increase in ocean thermal stratification. Finally, we discuss the implications of these results for the problem of steady state (statistical equilibrium) in ocean and climate GCM as well as paleoclimate problems including millennial climate variability.

54 ENVIRONMENTAL SCIENCES↗

Automated Weight-Window Generation for Threat Detection Applications Using ADVANTG

Deterministic transport codes have been used for some time to generate weight-window parameters that can improve the efficiency of Monte Carlo simulations. As the use of this hybrid computational technique is becoming more widespread, the scope of applications in which it is being applied is expanding. An active source of new applications is the field of homeland security--particularly the detection of nuclear material threats. For these problems, automated hybrid methods offer an efficient alternative to trial-and-error variance reduction techniques (e.g., geometry splitting or the stochastic weight window generator). The ADVANTG code has been developed to automate the generation of weight-window parameters for MCNP using the Consistent Adjoint Driven Importance Sampling method and employs the TORT or Denovo 3-D discrete ordinates codes to generate importance maps. In this paper, we describe the application of ADVANTG to a set of threat-detection simulations. We present numerical results for an 'active-interrogation' problem in which a standard cargo container is irradiated by a deuterium-tritium fusion neutron generator. We also present results for two passive detection problems in which a cargo container holding a shielded neutron or gamma source is placed near a portal monitor. For the passive detection problems, ADVANTG obtains an O(10{sup 4}) speedup and, for a detailed gamma spectrum tally, an average O(10{sup 2}) speedup relative to implicit-capture-only simulations, including the deterministic calculation time. For the active-interrogation problem, an O(10{sup 4}) speedup is obtained when compared to a simulation with angular source biasing and crude geometry splitting.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Adaptive immersed isogeometric level-set topology optimization

Here, this paper presents for the first time an adaptive immersed approach for level-set topology optimization using higher-order truncated hierarchical B-spline discretizations for design and state variable fields. Boundaries and interfaces are represented implicitly by the iso-contour of one or multiple level-set functions. An immersed finite element method, the eXtended IsoGeometric Analysis, is used to predict the physical response. The proposed optimization framework affords different adaptively refined higher-order B-spline discretizations for individual design and state variable fields. The increased continuity of higher-order B-spline discretizations together with local refinement enables direct control over the accuracy of the representation of each field while simultaneously reducing computational cost compared to uniformly refined discretizations. A flexible mesh adaptation strategy enables local refinement based on geometric measures or physics-based error indicators. These adaptive discretization and analysis approaches are integrated into gradient-based optimization schemes, evaluating the design sensitivities using the adjoint method. Numerical studies illustrate the features of the proposed framework with static, linear elastic, multi-material, two- and three-dimensional problems. The examples provide insight into the effect of refining the design variable field on the optimization result and the convergence rate of the optimization process. Using coarse higher-order B-spline discretizations for level-set fields promotes the development of smooth designs and suppresses the emergence of small features. Moreover, adaptive mesh refinement for state variable fields results in a reduction of overall computational cost. Higher-order B-spline discretizations are especially interesting when evaluating gradients of state variable fields due to their higher inter-element continuity.

36 MATERIALS SCIENCE↗

Optimization of Periodic Composite Structures for Sub-Wavelength Focusing

Recently, there has been plenty of work in designing and fabricating materials with an effective negative refractive index. Veselago realized that a slab of material with a refractive index of -1 would act as a lens. Pendry suggested that the Veselago lens would act as a superlens, providing a perfect image of an object in contrast to conventional lenses which are only able to focus a point source to an image having a diameter of the order of the wavelength of the incident field.Recent work has shown that similar focusing effects can be obtained with certain slabs of 'conventional' periodic composite materials: photonic crystals. The present work seeks to answer the question of what periodic dielectric composite medium (described by dielectric coefficient with positive real part) gives an optimal image of a point source. An optimization problem is formulated and it is shown that a solution exists provided the medium has small absorption. Solutions are characterized by an adjoint-state gradient condition, and several numerical examples illustrate both the plausibility of this design approach, and the possibility of obtaining smaller image spot sizes than with typical photonic crystals.

97 MATHEMATICS AND COMPUTING↗

Application of Ensemble Sensitivity Analysis to Observation Targeting for Short-term Wind Speed Forecasting

The operators of electrical grids, sometimes referred to as Balancing Authorities (BA), typically make critical decisions on how to most reliably and economically balance electrical load and generation in time frames ranging from a few minutes to six hours ahead. At higher levels of wind power generation, there is an increasing need to improve the accuracy of 0- to 6-hour ahead wind power forecasts. Forecasts on this time scale have typically been strongly dependent on short-term trends indicated by the time series of power production and meteorological data from a wind farm. Additional input information is often available from the output of Numerical Weather Prediction (NWP) models and occasionally from off-site meteorological towers in the region surrounding the wind generation facility. A widely proposed approach to improve short-term forecasts is the deployment of off-site meteorological towers at locations upstream from the wind generation facility in order to sense approaching wind perturbations. While conceptually appealing, it turns out that, in practice, it is often very difficult to derive significant benefit in forecast performance from this approach. The difficulty is rooted in the fact that the type, scale, and amplitude of the processes controlling wind variability at a site change from day to day if not from hour to hour. Thus, a location that provides some useful forecast information for one time may not be a useful predictor a few hours later. Indeed, some processes that cause significant changes in wind power production operate predominantly in the vertical direction and thus cannot be monitored by employing a network of sensors at off-site locations. Hence, it is very challenging to determine the type of sensors and deployment locations to get the most benefit for a specific short-term forecast application. Two tools recently developed in the meteorological research community have the potential to help determine the locations and parameters to measure in order to get the maximum positive impact on forecast performance for a particular site and short-term look-ahead period. Both tools rely on the use of NWP models to assess the sensitivity of a forecast for a particular location to measurements made at a prior time (i.e. the look-ahead period) at points surrounding the target location. The fundamental hypothesis is that points and variables with high sensitivity are good candidates for measurements since information at those points are likely to have the most impact on the forecast for the desired parameter, location and look-ahead period. One approach is called the adjoint method (Errico and Vukicevic, 1992; Errico, 1997) and the other newer approach is known as Ensemble Sensitivity Analysis (ESA; Ancell and Hakim 2007; Torn and Hakim 2008). Both approaches have been tested on large-scale atmospheric prediction problems (e.g. forecasting pressure or precipitation over a relatively large region 24 hours ahead) but neither has been applied to mesoscale space-time scales of winds or any other variables near the surface of the earth. A number of factors suggest that ESA is better suited for short-term wind forecasting applications. One of the most significant advantages of this approach is that it is not necessary to linearize the mathematical representation of the processes in the underlying atmospheric model as required by the adjoint approach. Such a linearization may be especially problematic for the application of short-term forecasting of boundary layer winds in complex terrain since non-linear shifts in the structure of boundary layer due to atmospheric stability changes are a critical part of the wind power production forecast problem. The specific objective of work described in this paper is to test the ESA as a tool to identify measurement locations and variables that have the greatest positive impact on the accuracy of wind forecasts in the 0- to 6-hour look-ahead periods for the wind generation area of California's Tehachapi Pass during the warm (high generation) season. The paper is organized as follows. Section 2 highlights the methodology, Section 3 presents results, and Section 4 concludes with a summary and brief discussion of future work.

17 WIND ENERGY↗

Topology optimization of additively manufactured fluidic components free of internal support structures

This report integrates projection-based approaches for implementing overhang constraints with fluid topology optimization to design additively manufactured fluidic components that do not require internal support structures. Internal support structures are challenging, potentially impossible to remove in complex fluid networks, yet would degrade fluidic performance if left within the part. This presents a major challenge in coupling topology optimization with additive manufacturing in the design of fluid components. The proposed approach leverages past work in overhang constraints for solid mechanics, including projection formulations and adjoint sensitivity analysis, and considers incompressible Navier–Stokes equations for laminar fluid flow. The approach is demonstrated on 2D and 3D fluid topology optimization problems to minimize pressure drop and constrain mass flow rates in pipes and manifolds. Resulting designs are crisp, logical, satisfy performance constraints, and satisfy overhang constraints, eliminating the need for internal support structures.

42 ENGINEERING↗

SENSMG: First-Order Sensitivities of Neutron Reaction Rates, Reaction-Rate Ratios, Leakage, k eff , and α Using PARTISN

SENSMG is a tool for computing first-order sensitivities of neutron reaction rates, reaction-rate ratios, leakage, k eff , and α using the PARTISN multigroup discrete-ordinates code. SENSMG computes sensitivities to all of the transport cross sections and data (total, fission, nu, chi, and all scattering moments), two edit cross sections (absorption and capture), and the density for every isotope and energy group. It also computes sensitivities to the mass density for every material and derivatives with respect to all interface locations. The tool can be used for one-dimensional spherical (r) and two-dimensional cylindrical (r-z) geometries. The tool can be used for fixed-source and eigenvalue problems. The tool implements Generalized Perturbation Theory (GPT) as discussed by Williams and Stacey. Section II of this report describes the theory behind adjoint-based sensitivities, gives the equations that SENSMG solves, and defines the sensitivities that are output. Section III describes the user interface, including the input file and command line options. Section IV describes the output. Section V gives some notes about the coding that may be of interest. Section VI discusses verification, which is ongoing. Section VII lists needs and ideas for future work. Appendix A lists all of the input files whose results are presented in Sec. VI.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Gas-Cooled High-Temperature Pebble-Bed Reactor Reference Plant Model

This report details a fully coupled neutronics thermal hydraulics reference plant model for a gas-cooled high-temperature pebble-bed reactor. The multiphysics model is developed on the Nuclear Regulatory Commission's Comprehensive Reactor Analysis Bundle (BlueCRAB) available on the Idaho National Laboratory high performance computer, which natively and seamlessly couples Griffin, Pronghorn, and the BISON MOOSE-based applications. Griffin provides the reactor physics capabilities including depletion to the equilibrium core, k-eigenvalue, adjoint, and transient solutions. Pronghorn solves the porous medium equations for the fluid regions and conduction in the solid regions. BISON solves thermal conduction problems for the average pebble and TRISO in the pebble-bed core thus providing the fuel and moderator spatial fields for each pebble burnup group. The neutronics feedback relies primarily on fuel and moderator temperatures. We present results for the coupled steady state equilibrium core and a protected loss of flow event. Although this model is to be considered prototypical with regard to the capabilities in BlueCRAB, its results are consistent with published work by the Institute of Nuclear and New Energy Technology (INET) in China and other research entities.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Adjoint-Based Sensitivity and Uncertainty Analysis for Density and Composition: A User’s Guide

The ability to perform sensitivity analyses using adjoint-based first-order sensitivity theory has existed for decades. This paper provides guidance on how adjoint sensitivity methods can be used to predict the effect of material density and composition uncertainties in critical experiments, including when these uncertain parameters are correlated or constrained. Two widely used Monte Carlo codes, MCNP6 (Ref. 2) and SCALE 6.2 (Ref. 3), are both capable of computing isotopic density sensitivities in continuous energy and angle. Additionally, Perkó et al. have shown how individual isotope density sensitivities, easily computed using adjoint methods, can be combined to compute constrained first-order sensitivities that may be used in the uncertainty analysis. This paper provides details on how the codes are used to compute first-order sensitivities and how the sensitivities are used in an uncertainty analysis. Constrained first-order sensitivities are computed in a simple example problem.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

Nonlinear Diffusion Acceleration of the Least-Squares Transport Equation in Geometries with Voids

In this paper we show the extension of the Nonlinear-Diffusion Acceleration (NDA) to geometries containing small voids using a weighted least-squares (WLS) high order equation. Even though the WLS equation is well defined in voids, the low-order drift diffusion equation was not defined in materials with a zero cross section. This paper derives the necessary modifications to the NDA algorithm. We show that a small change to the NDA closure term and a non-local definition of the diffusion coefficient solve the problems for voids regions. These changes do not affect the algorithm for optical thick material regions, while making the algorithm well defined in optically thin ones. We use a Fourier analysis to perform an iterative analysis to confirm that the modifications result in a stable and efficient algorithm. Numerical results of our method will be presented in the second part of the paper. We test this formulation with a small, one-dimensional test problem. Additionally we present results for a modified version of the C5G7 benchmark containing voids as a more complex, reactor like problem. We compared our results to PDT, Texas A&M’s transport code, utilizing a first order discontinuous formulation as reference and the self-adjoint angular flux equation with void treatment (SAAFτ ), a different second order form. The results indicate, that the NDA WLS performed comparably or slightly worse then the asymmetric SAAFτ , while maintaining a symmetric discretization matrix.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Progress on Optimizing Wind Farms and Rotor Designs Using Adjoints

Modern wind plants are increasingly tasked with multiple performance objectives. In addition to designing plants that maximize power output and minimize the levelized cost of energy (LCOE), the design and operation of wind plants is increasingly influenced by challenges regarding grid integration of variable generation renewables. This places a growing emphasis on making wind plants more controllable and predictable. WindSE is a Reynolds-averaged Navier-Stokes (RANS) model designed around analytical gradient and adjoint methods, with the ability to capture terrain-induced effects, as shown in Figure 1. The recent addition of an unsteady solver with an actuator line method (ALM) and ongoing work to enable massively parallel optimizations gives it a unique niche to explore coupled plant-level controls and design problems. This code is an open source python package built on the FEniCS framework that utilizes fast, parallel PETSc solvers to model fluid flow throughout wind-farm scale domains. Two recent studies performed using WindSE demonstrate the capability to optimize under a wide variety of flow conditions and objective functions. In the first, we present an optimization focused on modifying the layout of a wind farm with a fixed number of turbines for maximum total power output [1]. This study highlights the ability to quickly perform simulations using the steady Navier-Stokes solver combined with rotors represented as actuator disks while also stressing the importance of capturing terrain-induced effects. Gradient-based optimization using the RANS equations is viable due to the inclusion of efficiently computed adjoint derivatives. We interpret the physical results of the optimal layout and also discuss the computational cost of scaling to larger problems. In the second study, we present the capabilities of the unsteady Navier-Stokes solver, where rotor-blade profiles represented by actuator lines are optimized to enhance wake steering effects and overall power production [2]. We quantify the wind plant performance gains obtained from this type of simultaneous control co-design optimization as compared to optimizing the blade design and yaw independently. Figure 2 shows the differences between a baseline two-turbine system and an optimized system where we fine-tune the blade chord profile. Results and challenges from each study are quickly summarized and used to motivate the current development efforts within WindSE. Current and future work is focused on enabling higher-resolution studies with more degrees of freedom through parallelization of both the simulation and optimization algorithms. We present benchmarking results to show that WindSE performs well in both weak- and strong-scaling tests and further demonstrate that the optimizer obtains the same convergence rates in both shared- and distributed-memory environments. Using larger wind farms, we can study deep-array effects within an optimization context, allowing the use of objective functions that have been previously unstudied. As an example, we present ongoing work on a blockage metric which characterizes the loss of available kinetic energy due to wake effects from multiple upstream turbines.

adjoint optimization↗

Neural-adjoint method for the inverse design of all-dielectric metasurfaces

All-dielectric metasurfaces exhibit exotic electromagnetic responses, similar to those obtained with metal-based metamaterials. Research in all-dielectric metasurfaces currently uses relatively simple unit-cell designs, but increased geometrical complexity may yield even greater scattering states. Although machine learning has recently been applied to the design of metasurfaces with impressive results, the much more challenging task of finding a geometry that yields a desired spectra remains largely unsolved. We propose and demonstrate a method capable of finding accurate solutions to ill-posed inverse problems, where the conditions of existence and uniqueness are violated. A specific example of finding the metasurface geometry which yields a radiant exitance matching the external quantum efficiency of gallium antimonide is demonstrated. We also show how the neural-adjoint method can intelligently grow the design search space to include designs that increasingly and accurately approximate the desired scattering response. The neural-adjoint method is not restricted to the case demonstrated and may be applied to plasmonics, photonic crystal, and other artificial electromagnetic materials.

42 ENGINEERING↗