Search NASA⌕ Search

SEARCH · Search NASA

Results for “computer”

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 1,711 records · Page 95

A fast and robust computational modeling approach for density and shape predictions in powder metallurgy hot isostatic pressing

Powder metallurgy hot isostatic pressing (PM-HIP) is an advanced manufacturing process that produces near-net-shape parts with high material utilization and uniform microstructures. PM-HIP is frequently used for producing small-scale parts with complicated geometries and is potentially economical for producing large-scale parts. However, excessive post-HIP shape distortions can reduce its effectiveness and economic advantage, especially for larger parts. A PM-HIP computational model can predict and help mitigate these distortions. However, due to complex deformation mechanisms and thermo-mechanical coupling present in PM-HIP processes, these non-linear computational models sometimes become numerically unstable. The numerical instabilities in these models can lead to very slow convergence or no convergence at all, which often translates to slow and unreliable models. These limitations are more pronounced in large models with complicated geometries. Hence, in this work, an alternative modeling approach is presented that improves numerical stability and computational performance. The presented approach achieves these improvements through approximating the fully coupled thermo-mechanical PM-HIP model as a decoupled model and adding inertial damping to the model’s mechanical part. In conclusion, a comparison with the fully coupled model indicated a slight dip in prediction accuracy (<5% error) but significant improvements in numerical stability (>20 times larger time step size) and computational performance (5-10 times speed-up with less computational resource usage) when using the presented approach.

Hot isostatic pressing↗

Automatic computation for optimum height planning of apartment buildings to improve solar access

The objective of this study is to suggest a mathematical model and an optimal algorithm for determining the height of apartment buildings to satisfy the solar rights of survey buildings or survey housing units. The objective is also to develop an automatic computation model for the optimum height of apartment buildings and then to clarify the performance and expected effects. To accomplish the objective of this study, the following procedures were followed: (1) The necessity of the height planning of obstruction buildings to satisfy the solar rights of survey buildings or survey housing units is demonstrated by analyzing through a literature review the recent trend of disputes related to solar rights and to examining the social requirements in terms of solar rights. In addition, the necessity of the automatic computation system for height planning of apartment buildings is demonstrated and a suitable analysis method for this system is chosen by investigating the characteristics of analysis methods for solar rights assessment. (2) A case study on the process of height planning of apartment buildings will be briefly described and the problems occurring in this process will then be examined carefully. (3) To develop an automatic computation model for height planning of apartment buildings, geometrical elements forming apartment buildings are defined by analyzing the geometrical characteristics of apartment buildings. In addition, design factors and regulations required in height planning of apartment buildings are investigated. Based on this knowledge, the methodology and mathematical algorithm to adjust the height of apartment buildings by automatic computation are suggested and probable problems and the ways to resolve these problems are discussed. Finally, the methodology and algorithm for the optimization are suggested. (4) Based on the suggested methodology and mathematical algorithm, the automatic computation model for optimum height of apartment buildings is developed and the developed system is verified through the application of some cases. The effects of the suggested model are then demonstrated quantitatively and qualitatively. (author)

14 SOLAR ENERGY↗

Accelerating Radiation Computations for Dynamical Models With Targeted Machine Learning and Code Optimization

Abstract Atmospheric radiation is the main driver of weather and climate, yet due to a complicated absorption spectrum, the precise treatment of radiative transfer in numerical weather and climate models is computationally unfeasible. Radiation parameterizations need to maximize computational efficiency as well as accuracy, and for predicting the future climate many greenhouse gases need to be included. In this work, neural networks (NNs) were developed to replace the gas optics computations in a modern radiation scheme (RTE+RRTMGP) by using carefully constructed models and training data. The NNs, implemented in Fortran and utilizing BLAS for batched inference, are faster by a factor of 1–6, depending on the software and hardware platforms. We combined the accelerated gas optics with a refactored radiative transfer solver, resulting in clear‐sky longwave (shortwave) fluxes being 3.5 (1.8) faster to compute on an Intel platform. The accuracy, evaluated with benchmark line‐by‐line computations across a large range of atmospheric conditions, is very similar to the original scheme with errors in heating rates and top‐of‐atmosphere radiative forcings typically below 0.1 K day −1 and 0.5 W m −2 , respectively. These results show that targeted machine learning, code restructuring techniques, and the use of numerical libraries can yield material gains in efficiency while retaining accuracy.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Computing molecular excited states on a D-Wave quantum annealer

Abstract The possibility of using quantum computers for electronic structure calculations has opened up a promising avenue for computational chemistry. Towards this direction, numerous algorithmic advances have been made in the last five years. The potential of quantum annealers, which are the prototypes of adiabatic quantum computers, is yet to be fully explored. In this work, we demonstrate the use of a D-Wave quantum annealer for the calculation of excited electronic states of molecular systems. These simulations play an important role in a number of areas, such as photovoltaics, semiconductor technology and nanoscience. The excited states are treated using two methods, time-dependent Hartree–Fock (TDHF) and time-dependent density-functional theory (TDDFT), both within a commonly used Tamm–Dancoff approximation (TDA). The resulting TDA eigenvalue equations are solved on a D-Wave quantum annealer using the Quantum Annealer Eigensolver (QAE), developed previously. The method is shown to reproduce a typical basis set convergence on the example $$\hbox {H}_2$$ H 2 molecule and is also applied to several other molecular species. Characteristic properties such as transition dipole moments and oscillator strengths are computed as well. Three potential energy profiles for excited states are computed for $$\hbox {NH}_3$$ NH 3 as a function of the molecular geometry. Similar to previous studies, the accuracy of the method is dependent on the accuracy of the intermediate meta-heuristic software called qbsolv.

74 ATOMIC AND MOLECULAR PHYSICS↗

Federated Access from DOE Labs to Distributed Storage in the EIC Era of Computing

The Electron Ion Collider (EIC) collaboration and future experiment is a unique scientific ecosystem within Nuclear Physics as the experiment starts right off as a crosscollaboration between Brookhaven National Lab (BNL) & Jefferson Lab (JLab). As a result, this muti-lab computing model tries at best to provide services accessible from anywhere by anyone who is part of the collaboration. While the computing model for the EIC is not finalized, it is anticipated that the computational and storage resources will be made accessible to a wide range of collaborators across the world. The use of federated ID seems to be a critical element to the strategy of providing such services, allowing seamless access to each lab site computing resources. However, providing Federated access to a Federated storage is not a trivial matter and has its share of technical challenges. In this contribution, we focus on the steps we took towards the deployment of a distributed object storage system that integrates with Amazon S3 and Federated ID. We will first cover for and explain the first stage storage solutions provided to the EIC during the detector design phase. Our initial test deployment consisted of Lustre storage using MinIO, hence providing an S3 interface. High Availability load balancers were added later to provide the initial scalability it lacked. Performance of that system will be shown. While this embryonic solution worked well, it had many limitations. Looking ahead, the Ceph object storage is considered a top-of-the-line solution in the storage community - since the Ceph Object Gateway is compatible with the Amazon S3 API out of the box, our next phase will use a native S3 storage. Our Ceph deployment will consist of erasure coded storage nodes to maximize storage potential along with multiple Ceph Object Gateways for redundant access. We will compare performance of our next stage implementations. Finally, we will present how to leverage OpenID Connect with the Ceph Object Gateway’s to enable Federated ID access. We hope this contribution will serve the community needs as we move forward with cross-lab collaborations and the need for Federated ID access to distributed compute facilities.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Modeling Distributed Computing Infrastructures for HEP Applications

Predicting the performance of various infrastructure design options in complex federated infrastructures with computing sites distributed over a wide area network that support a plethora of users and workflows, such as the Worldwide LHC Computing Grid (WLCG), is not trivial. Due to the complexity and size of these infrastructures, it is not feasible to deploy experimental test-beds at large scales merely for the purpose of comparing and evaluating alternate designs. An alternative is to study the behaviours of these systems using simulation. This approach has been used successfully in the past to identify efficient and practical infrastructure designs for High Energy Physics (HEP). A prominent example is the Monarc simulation framework, which was used to study the initial structure of the WLCG. New simulation capabilities are needed to simulate large-scale heterogeneous computing systems with complex networks, data access and caching patterns. A modern tool to simulate HEP workloads that execute on distributed computing infrastructures based on the SimGrid and WRENCH simulation frameworks is outlined. Studies of its accuracy and scalability are presented using HEP as a case-study. Hypothetical adjustments to prevailing computing architectures in HEP are studied providing insights into the dynamics of a part of the WLCG and candidates for improvements.

Horzela, Maximilian↗

Accelerating science: The usage of commercial clouds in ATLAS Distributed Computing

The ATLAS experiment at CERN is one of the largest scientific machines built to date and will have ever growing computing needs as the Large Hadron Collider collects an increasingly larger volume of data over the next 20 years. ATLAS is conducting R&D projects on Amazon Web Services and Google Cloud as complementary resources for distributed computing, focusing on some of the key features of commercial clouds: lightweight operation, elasticity and availability of multiple chip architectures. The proof of concept phases have concluded with the cloud-native, vendoragnostic integration with the experiment’s data and workload management frameworks. Google Cloud has been used to evaluate elastic batch computing, ramping up ephemeral clusters of up to O(100k) cores to process tasks requiring quick turnaround. Amazon Web Services has been exploited for the successful physics validation of the Athena simulation software on ARM processors. We have also set up an interactive facility for physics analysis allowing endusers to spin up private, on-demand clusters for parallel computing with up to 4 000 cores, or run GPU enabled notebooks and jobs for machine learning applications. The success of the proof of concept phases has led to the extension of the Google Cloud project, where ATLAS will study the total cost of ownership of a production cloud site during 15 months with 10k cores on average, fully integrated with distributed grid computing resources and continue the R&D projects.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Further Evidence in Support of the Universal Nilpotent Grammatical Computational Paradigm of Quantum Physics

Further evidence is presented in favour of the computational paradigm, conceived and constructed by Rowlands and Diaz, as detailed in Rowlands' book Zero to Infinity (2007), and in particular the authors' paper 'The Grammatical Universe: the Laws of Thermodynamics and Quantum Entanglement'. The paradigm, which has isomorphic group and algebraic quantum mechanical language interpretations, not only predicts the well-established facts of quantum physics, the periodic table, chemistry / valence and of molecular biology, whose understanding it extends; it also provides an elegant, simple solution to the unresolved quantum measurement problem. In this fundamental paradigm, all the computational constructs / predictions that emerge, follow from the simple fact, that, as in quantum mechanics, the wave function is defined only up to an arbitrary fixed phase. This fixed phase provides a simple physical understanding of the quantum vacuum in quantum field theory, where only relative phases, known to be able to encode 3+1 relativistic space-time geometries, can be measured. It is the arbitrary fixed measurement standard, against which everything that follows is to be measured, even though the standard itself cannot be, since nothing exists against which to measure it. The standard, as an arbitrary fixed reference phase, functions as the holographic basis for a self-organized universal quantum process of emergent novel fermion states of matter where, following each emergence, the arbitrary standard is re-fixed anew so as to provide a complete history / holographic record or hologram of the current fixed past, advancing an unending irreversible evolution, such as is the evidence of our senses. The fermion states, in accord with the Pauli exclusion principle, each correspond to a unique nilpotent symbol in the infinite alphabet (which specifies the grammar in this nilpotent universal computational rewrite system (NUCRS) paradigm); and the alphabet, as Hill and Rowlands hypothesize on substantial evidence [26], includes that of the RNA / DNA genetic code and, as holographic phase encodings / holograms, the 4D geometries of all living systems as self-organised grammatical computational rewrite machines / machinery. Human brains, natural grammatical (written symbol) languages, 4D geometric self-awareness and a totally new emergent property of matter, human consciousness, can thus with some measure of confidence be postulated as further genetic consequences which follow from this self-organizing fundamental rewrite NUCRS construction. For it, like natural language, possesses a semantics and not just a syntax, where the initial symbol, i.e. the arbitrary fixed phase measurement standard, is able to function as the template for the blueprints of the emergent 4D relativistic real and virtual geometries to come, in a 'from the Self Creation to the creation of the human self' computational rewrite process evolution.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Density functional theory study of the interaction of vinyl radical, ethyne, and ethene with benzene, aimed to define an affordable computational level to investigate stability trends in large van der Waals complexes

Our purpose is to identify a computational level sufficiently dependable and affordable to assess trends in the interaction of a variety of radical or closed shell unsaturated hydro-carbons A adsorbed on soot platelet models B. These systems, of environmental interest, would unavoidably have rather large sizes, thus prompting to explore in this paper the performances of relatively low-level computational methods and compare them with higher-level reference results. To this end, the interaction of three complexes between non-polar species, vinyl radical, ethyne, or ethene (A) with benzene (B) is studied, since these species, involved themselves in growth processes of polycyclic aromatic hydrocarbons (PAHs) and soot particles, are small enough to allow high-level reference calculations of the interaction energy ΔE{sub AB}. Counterpoise-corrected interaction energies ΔE{sub AB} are used at all stages. (1) Density Functional Theory (DFT) unconstrained optimizations of the A−B complexes are carried out, using the B3LYP-D, ωB97X-D, and M06-2X functionals, with six basis sets: 6-31G(d), 6-311 (2d,p), and 6-311++G(3df,3pd); aug-cc-pVDZ and aug-cc-pVTZ; N07T. (2) Then, unconstrained optimizations by Møller-Plesset second order Perturbation Theory (MP2), with each basis set, allow subsequent single point Coupled Cluster Singles Doubles and perturbative estimate of the Triples energy computations with the same basis sets [CCSD(T)//MP2]. (3) Based on an additivity assumption of (i) the estimated MP2 energy at the complete basis set limit [E{sub MP2/CBS}] and (ii) the higher-order correlation energy effects in passing from MP2 to CCSD(T) at the aug-cc-pVTZ basis set, ΔE{sub CC-MP}, a CCSD(T)/CBS estimate is obtained and taken as a computational energy reference. At DFT, variations in ΔE{sub AB} with basis set are not large for the title molecules, and the three functionals perform rather satisfactorily even with rather small basis sets [6-31G(d) and N07T], exhibiting deviation from the computational reference of less than 1 kcal mol{sup −1}. The zero-point vibrational energy corrected estimates Δ(E{sub AB}+ZPE), obtained with the three functionals and the 6-31G(d) and N07T basis sets, are compared with experimental D{sub 0} measures, when available. In particular, this comparison is finally extended to the naphthalene and coronene dimers and to three π−π associations of different PAHs (R, made by 10, 16, or 24 C atoms) and P (80 C atoms)

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Spin-transfer torque magnetoresistive random-access memory technologies for normally off computing (invited)

Most parts of present computer systems are made of volatile devices, and the power to supply them to avoid information loss causes huge energy losses. We can eliminate this meaningless energy loss by utilizing the non-volatile function of advanced spin-transfer torque magnetoresistive random-access memory (STT-MRAM) technology and create a new type of computer, i.e., normally off computers. Critical tasks to achieve normally off computers are implementations of STT-MRAM technologies in the main memory and low-level cache memories. STT-MRAM technology for applications to the main memory has been successfully developed by using perpendicular STT-MRAMs, and faster STT-MRAM technologies for applications to the cache memory are now being developed. The present status of STT-MRAMs and challenges that remain for normally off computers are discussed.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Superior model for fault tolerance computation in designing nano-sized circuit systems

As CMOS technology scales nano-metrically, reliability turns out to be a decisive subject in the design methodology of nano-sized circuit systems. As a result, several computational approaches have been developed to compute and evaluate reliability of desired nano-electronic circuits. The process of computing reliability becomes very troublesome and time consuming as the computational complexity build ups with the desired circuit size. Therefore, being able to measure reliability instantly and superiorly is fast becoming necessary in designing modern logic integrated circuits. For this purpose, the paper firstly looks into the development of an automated reliability evaluation tool based on the generalization of Probabilistic Gate Model (PGM) and Boolean Difference-based Error Calculator (BDEC) models. The Matlab-based tool allows users to significantly speed-up the task of reliability analysis for very large number of nano-electronic circuits. Secondly, by using the developed automated tool, the paper explores into a comparative study involving reliability computation and evaluation by PGM and, BDEC models for different implementations of same functionality circuits. Based on the reliability analysis, BDEC gives exact and transparent reliability measures, but as the complexity of the same functionality circuits with respect to gate error increases, reliability measure by BDEC tends to be lower than the reliability measure by PGM. The lesser reliability measure by BDEC is well explained in this paper using distribution of different signal input patterns overtime for same functionality circuits. Simulation results conclude that the reliability measure by BDEC depends not only on faulty gates but it also depends on circuit topology, probability of input signals being one or zero and also probability of error on signal lines.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Computationally efficient characterization of potential energy surfaces based on fingerprint distances

An analysis of the network defined by the potential energy minima of multi-atomic systems and their connectivity via reaction pathways that go through transition states allows us to understand important characteristics like thermodynamic, dynamic, and structural properties. Unfortunately computing the transition states and reaction pathways in addition to the significant energetically low-lying local minima is a computationally demanding task. We here introduce a computationally efficient method that is based on a combination of the minima hopping global optimization method and the insight that uphill barriers tend to increase with increasing structural distances of the educt and product states. This method allows us to replace the exact connectivity information and transition state energies with alternative and approximate concepts. Without adding any significant additional cost to the minima hopping global optimization approach, this method allows us to generate an approximate network of the minima, their connectivity, and a rough measure for the energy needed for their interconversion. This can be used to obtain a first qualitative idea on important physical and chemical properties by means of a disconnectivity graph analysis. Besides the physical insight obtained by such an analysis, the gained knowledge can be used to make a decision if it is worthwhile or not to invest computational resources for an exact computation of the transition states and the reaction pathways. Furthermore it is demonstrated that the here presented method can be used for finding physically reasonable interconversion pathways that are promising input pathways for methods like transition path sampling or discrete path sampling.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Ultra-fast computation of electronic spectra for large systems by tight-binding based simplified Tamm-Dancoff approximation (sTDA-xTB)

The computational bottleneck of the extremely fast simplified Tamm-Dancoff approximated (sTDA) time-dependent density functional theory procedure [S. Grimme, J. Chem. Phys. 138, 244104 (2013)] for the computation of electronic spectra for large systems is the determination of the ground state Kohn-Sham orbitals and eigenvalues. This limits such treatments to single structures with a few hundred atoms and hence, e.g., sampling along molecular dynamics trajectories for flexible systems or the calculation of chromophore aggregates is often not possible. The aim of this work is to solve this problem by a specifically designed semi-empirical tight binding (TB) procedure similar to the well established self-consistent-charge density functional TB scheme. The new special purpose method provides orbitals and orbital energies of hybrid density functional character for a subsequent and basically unmodified sTDA procedure. Compared to many previous semi-empirical excited state methods, an advantage of the ansatz is that a general eigenvalue problem in a non-orthogonal, extended atomic orbital basis is solved and therefore correct occupied/virtual orbital energy splittings as well as Rydberg levels are obtained. A key idea for the success of the new model is that the determination of atomic charges (describing an effective electron-electron interaction) and the one-particle spectrum is decoupled and treated by two differently parametrized Hamiltonians/basis sets. The three-diagonalization-step composite procedure can routinely compute broad range electronic spectra (0-8 eV) within minutes of computation time for systems composed of 500-1000 atoms with an accuracy typical of standard time-dependent density functional theory (0.3-0.5 eV average error). An easily extendable parametrization based on coupled-cluster and density functional computed reference data for the elements H–Zn including transition metals is described. The accuracy of the method termed sTDA-xTB is first benchmarked for vertical excitation energies of open- and closed-shell systems in comparison to other semi-empirical methods and applied to exemplary problems in electronic spectroscopy. As side products of the development, a robust and efficient valence electron TB method for the accurate determination of atomic charges as well as a more accurate calculation scheme of dipole rotatory strengths within the Tamm-Dancoff approximation is proposed.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

The influence of boundary and edge-plasma modeling in computations of axisymmetric vertical displacement

A number of previously published studies apply extended-MHD computations to model macroscopic dynamics of tokamak vertical displacement events (VDEs). The characteristic plasma-wall contact and resistive-wall diffusion imply sensitivity to boundary conditions in a general sense, but which conditions have significant influence depends on how the plasma is modeled. This work examines these dependencies by comparing results of axisymmetric extended-MHD computations with different sets of boundary conditions, plasma thermal-conduction models, and wall resistivity values. The geometry and plasma parameters of the computations represent a modest-sized tokamak. The forced-displacement scenario and computational setup are similar to those used in previous 3D computations. The results show that for a given wall resistivity, the VDE time-scale is most sensitive to variations in the boundary and thermal-conduction parameters that affect electron thermal transport. The electrical conductivity depends on electron temperature, and the dependence on thermal transport stems from its influence on the electrical circuit that includes the open-field halo current. Lastly, conditions that lead to hotter, broader halo regions slow the evolution. Significant sensitivity to the boundary condition on plasma flow-velocity exists when electron thermal conduction is restricted and electron energy loss is convective, which is expected for conditions at the entrance of the magnetic presheath.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

On applications of quantum computing to plasma simulations

Quantum computing is gaining increased attention as a potential way to speed up simulations of physical systems, and it is also of interest to apply it to simulations of classical plasmas. However, quantum information science is traditionally aimed at modeling linear Hamiltonian systems of a particular form that is found in quantum mechanics, so extending the existing results to plasma applications remains a challenge. Here, we report a preliminary exploration of the long-term opportunities and likely obstacles in this area. First, we show that many plasma-wave problems are naturally representable in a quantumlike form and thus are naturally fit for quantum computers. Second, we consider more general plasma problems that include non-Hermitian dynamics (instabilities, irreversible dissipation) and nonlinearities. We show that by extending the configuration space, such systems can also be represented in a quantumlike form and thus can be simulated with quantum computers too, albeit that requires more computational resources compared to the first case. Third, we outline potential applications of hybrid quantum–classical computers, which include analysis of global eigenmodes and also an alternative approach to nonlinear simulations.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

RASPA3

RASPA3, a molecular simulation code for computing adsorption and diffusion in nanoporous materials and thermodynamic and transport properties of fluids. It implements force field based classical Monte Carlo/molecular dynamics in various ensembles. RASPA3 is rewritten from the ground up in C++23 with speed and code readability in mind. Transition-matrix Monte Carlo is added to compute the density of states and free energies. The Monte Carlo code for rigid molecules is based on quaternions, and the atomic positions needed in the energy evaluation are recreated from the center of mass position and quaternion orientation. The expanded ensemble methodology for fractional molecules, with a scaling parameter λ between 0 and 1, now also keeps track of analytic expressions of dU/dλ, allowing independent verification of the chemical potential using thermodynamic integration. The source code is freely available under the MIT license on GitHub.

Dubbeldam, David↗

Quantum graph learning and algorithms applied in quantum computer sciences and image classification

Graph and network theory play a fundamental role in quantum computer sciences, including quantum information and computation. Random graphs and complex network theory are pivotal in predicting novel quantum phenomena, where entangled links are represented by edges. Quantum algorithms have been developed to enhance solutions for various network problems, giving rise to quantum graph computing and quantum graph learning (QGL). Here, in this review, we explore graph theory and graph learning methods as powerful tools for quantum computers to generate efficient solutions to problems beyond the reach of classical systems. We delve into the development of quantum complex network theory and its applications in quantum computation, materials discovery, and research. We also discuss quantum machine learning (QML) methodologies for effective image classification using qubits, quantum gates, and quantum circuits. Additionally, the paper addresses the challenges of QGL and algorithms, emphasizing the steps needed to develop flexible QGL solvers. This review presents a comprehensive overview of the fields of QGL and QML, highlights recent advancements, and identifies opportunities for future research.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Parallelizable adjoint stencil computations using transposed forward-mode algorithmic differentiation

Algorithmic differentiation (AD) is a tool for generating discrete adjoint solvers, which efficiently compute gradients of functions with many inputs, for example for use in gradient-based optimization. AD is often applied to large computations such as stencil operators, which are an important part of most structured-mesh PDE solvers. Stencil computations are often parallelized, for example by using OpenMP, and optimized by using techniques such as cache-blocking and tiling to fully utilize multicore CPUs and many-core accelerators and GPUs. Differentiating these codes with conventional reverse-mode AD results in adjoint codes that cannot be expressed as stencil operations and may not be easily parallelizable. They thus leave most of the compute power of modern architectures unused. We present a method that combines forward-mode AD and loop transformation to generate adjoint solvers that use the same memory access pattern as the original computation that they are derived from and can benefit from the same optimization techniques. The effectiveness of this method is demonstrated by generating a scalable adjoint CFD solver for multicore CPUs and Xeon Phi accelerators.

algorithmic differentiation↗