CPU Minutes Research Articles

Near real-time calculation of the critical plane in all surface nodes of large metallic structures under multiaxial fatigue loading by visualization of the solution space

Applications of critical-plane-based fatigue criteria on large FE meshes remain uncommon, primarily due to the absence of a search scheme capable of identifying the critical plane for multiaxial loading scenarios in a way that is both reliable and fast enough. To address this gap, we introduce a three-module optimization scheme based on the Maximum Variance Method (MVM), guided by insights learned from visualizing the solution space. These visualizations not only validate the MVM strategy but also reveal a symmetry axis, which in turn assists in selecting the most critical among reciprocal solutions. Through iterative refinement of optimization parameters on automatically generated loading signals, our algorithm achieves a significant speedup over existing methods, while improving accuracy. The potential that comes with a processing speed of 10k nodes per CPU minute is demonstrated on a design challenge involving an additively manufactured Ti–6Al–4V bearing house subjected to non-proportional loading. A full critical-plane-based fatigue analysis on the 100k surface nodes takes 75 s with C++ PPL across 16 threads, which is fast enough to enable integration into iterative procedures such as topology optimization, promising substantial advancements in both academic research and engineering practices.

Development of a neural network model for peeling–ballooning stability analysis in the KSTAR tokamak pedestals

The neural network model, MISHKA-NN is developed to mitigate the computational burden associated with the linear ideal magnetohydrodynamic (MHD) stability analysis of the pedestal based on the peeling–ballooning (P–B) model. By utilizing both 1D plasma profiles (current density, pressure gradient, and safety factor) and 0D parameters (plasma geometry, total current, and toroidal mode number), the model predicts linear growth rate of edge-localized ideal MHD instability in a given equilibrium state. By enabling the prediction of each instability within a second, the model reduces the time required for plotting a pedestal P–B stability diagram ( j−α diagram) from approximately 100 CPU hours to a few CPU minutes. Notably, even with the utilization of parametric pressure and current profiles and plasma boundary shapes for the training dataset, the model shows a satisfactory level of performance in benchmarking the j−α diagram for the reconstructed equilibrium from a KSTAR tokamak experiment. We anticipate the model to serve as a versatile alternative to 2D linear MHD stability codes, alleviating numerical costs.

Generative Design of Sheet Metal Structures

Sheet Metal (SM) fabrication is perhaps one of the most common metalworking technique. Despite its prevalence, SM design is manual and costly, with rigorous practices that restrict the search space, yielding suboptimal results. In contrast, we present a framework for the first automatic design of SM parts. Focusing on load bearing applications, our novel system generates a high-performing manufacturable SM that adheres to the numerous constraints that SM design entails: The resulting part minimizes manufacturing costs while adhering to structural, spatial, and manufacturing constraints. In other words, the part should be strong enough, not disturb the environment, and adhere to the manufacturing process. These desiderata sum up to an elaborate, sparse, and expensive search space. Our generative approach is a carefully designed exploration process, comprising two steps. In Segment Discovery connections from the input load to attachable regions are accumulated, and during Segment Composition the most performing valid combination is searched for. For Discovery, we define a slim grammar, and sample it for parts using a Markov-Chain Monte Carlo (MCMC) approach, ran in intercommunicating instances (i.e, chains) for diversity. This, followed by a short continuous optimization, enables building a diverse and high-quality library of substructures. During Composition, a valid and minimal cost combination of the curated substructures is selected. To improve compliance significantly without additional manufacturing costs, we reinforce candidate parts onto themselves --- a unique SM capability called self-riveting. we provide our code and data in https://github.com/amir90/AutoSheetMetal. We show our generative approach produces viable parts for numerous scenarios. We compare our system against a human expert and observe improvements in both part quality and design time. We further analyze our pipeline's steps with respect to resulting quality, and have fabricated some results for validation. We hope our system will stretch the field of SM design, replacing costly expert hours with minutes of standard CPU, making this cheap and reliable manufacturing method accessible to anyone.

Learning Optimal Solutions via an LSTM-Optimization Framework

In this study, we present a deep learning-optimization framework to tackle dynamic mixed-integer programs. Specifically, we develop a bidirectional Long Short Term Memory (LSTM) framework that can process information forward and backward in time to learn optimal solutions to sequential decision-making problems. We demonstrate our approach in predicting the optimal decisions for the single-item capacitated lot-sizing problem (CLSP), where a binary variable denotes whether to produce in a period or not. Due to the dynamic nature of the production and inventory levels that must meet the periodic demand, the CLSP can be treated as a sequence labeling task where a recurrent neural network can capture the problem’s temporal dynamics. Computational results show that our LSTM-Optimization (LSTM-Opt) framework significantly reduces the solution time of benchmark CLSP problems without much loss in feasibility and optimality. For example, the predictions at the 85% level reduce the CPLEX solution time by a factor of 9 on average for over 240000 test instances with an optimality gap of less than 0.05% and 0.4% infeasibility in the test set. Also, models trained using shorter planning horizons can successfully predict the optimal solution of the instances with longer planning horizons. For the hardest data set, the LSTM predictions at the 25% level reduce the solution time of 70 CPU hours to less than 2 CPU minutes with an optimality gap of 0.8% and without any infeasibility. The LSTM-Opt framework outperforms classical ML algorithms, such as the logistic regression and random forest, in terms of the solution quality, and exact approaches, such as the ( $$\ell$$ , S) and dynamic programming-based inequalities, with respect to the solution time improvement. Our machine learning approach could be beneficial in tackling sequential decision-making problems similar to CLSP, which need to be solved repetitively, frequently, and in a fast manner.

Analysing the SEDs of protoplanetary disks with machine learning

Context. The analysis of spectral energy distributions (SEDs) of protoplanetary disks to determine their physical properties is known to be highly degenerate. Hence, a full Bayesian analysis is required to obtain parameter uncertainties and degeneracies. The main challenge here is computational speed, as one proper full radiative transfer model requires at least a couple of CPU minutes to compute. Aims. We performed a full Bayesian analysis for 30 well-known protoplanetary disks to determine their physical disk properties, including uncertainties and degeneracies. To circumvent the computational cost problem, we created neural networks (NNs) to emulate the SED generation process. Methods. We created two sets of MCFOST Monte Carlo radiative transfer disk models to train and test two NNs that predict SEDs for continuous and discontinuous disks, with 18 and 26 free model parameters, respectively. A Bayesian analysis was then performed on 30 protoplanetary disks with SED data collected by the FP7-Space DIANA project to determine the posterior distributions of all parameters. We ran this analysis twice, (i) with old distances and additional parameter constraints as used in a previous study, to compare results, and (ii) with updated distances and free choice of parameters to obtain homogeneous and unbiased model parameters. We evaluated the uncertainties in the determination of physical disk parameters from SED analysis, and detected and quantified the strongest degeneracies. Results. The NNs are able to predict SEDs within ~1 ms with uncertainties of about 5% compared to the true SEDs obtained by the radiative transfer code. We find parameter values and uncertainties that are significantly different from previous values obtained by χ2 fitting. Comparing the global evidence for continuous and discontinuous disks, we find that 26 out of 30 objects are better described by disks that have two distinct radial zones. The analysed sample shows a significant trend for massive disks to have small scale heights, which is consistent with lower midplane temperatures in massive disks. We find that the frequently used analytic relationship between disk dust mass and millimetre-flux systematically underestimates the dust mass for high-mass disks (dust mass ≥10−4 M⊙). We determine how well the dust mass can be determined with our method for different numbers of flux measurements. As a byproduct, we created an interactive graphical tool that instantly returns the SED predicted by our NNs for any parameter combination.

Few observation binary orbit solver (fobos) from two (or more) astrometric observations

ABSTRACT We have developed a new, fast method of estimating the orbital properties of a binary or triple system using as few as two epochs of astrometric data. fobos (Few Observation Binary Orbit Solver) uses a flat prior brute force Monte Carlo method to produce probability density functions of the likely orbital parameters. We test the code on fake observations and show that it can (fairly often) constrain the semi-major axis to within a factor of 2–3, and the inclination to within ∼20° from only two astrometric observations. We also show that the 68 and 95 per cent confidence intervals are statistically reliable. Applying this method to triple systems allows the relative inclination of the secondary and tertiary star orbits to be constrained. fobos can usually find a statistically significant number of possible matches in CPU minutes for binary systems, and CPU hours for triple systems.

Performance evaluation of GPU- and cluster-computing for parallelization of compute-intensive tasks

PurposeThis paper aims to evaluate different approaches for the parallelization of compute-intensive tasks. The study compares a Java multi-threaded algorithm, distributed computing solutions with MapReduce (Apache Hadoop) and resilient distributed data set (RDD) (Apache Spark) paradigms and a graphics processing unit (GPU) approach with Numba for compute unified device architecture (CUDA).Design/methodology/approachThe paper uses a simple but computationally intensive puzzle as a case study for experiments. To find all solutions using brute force search, 15! permutations had to be computed and tested against the solution rules. The experimental application comprises a Java multi-threaded algorithm, distributed computing solutions with MapReduce (Apache Hadoop) and RDD (Apache Spark) paradigms and a GPU approach with Numba for CUDA. The implementations were benchmarked on Amazon-EC2 instances for performance and scalability measurements.FindingsThe comparison of the solutions with Apache Hadoop and Apache Spark under Amazon EMR showed that the processing time measured in CPU minutes with Spark was up to 30% lower, while the performance of Spark especially benefits from an increasing number of tasks. With the CUDA implementation, more than 16 times faster execution is achievable for the same price compared to the Spark solution. Apart from the multi-threaded implementation, the processing times of all solutions scale approximately linearly. Finally, several application suggestions for the different parallelization approaches are derived from the insights of this study.Originality/valueThere are numerous studies that have examined the performance of parallelization approaches. Most of these studies deal with processing large amounts of data or mathematical problems. This work, in contrast, compares these technologies on their ability to implement computationally intensive distributed algorithms.

Dalek: A Deep Learning Emulator for TARDIS

Abstract Supernova spectral time series contain a wealth of information about the progenitor and explosion process of these energetic events. The modeling of these data requires the exploration of very high dimensional posterior probabilities with expensive radiative transfer codes. Even modest parameterizations of supernovae contain more than 10 parameters and a detailed exploration demands at least several million function evaluations. Physically realistic models require at least tens of CPU minutes per evaluation putting a detailed reconstruction of the explosion out of reach of traditional methodology. The advent of widely available libraries for the training of neural networks combined with their ability to approximate almost arbitrary functions with high precision allows for a new approach to this problem. Instead of evaluating the radiative transfer model itself, one can build a neural network proxy trained on the simulations but evaluating orders of magnitude faster. Such a framework is called an emulator or surrogate model. In this work, we present an emulator for the tardis supernova radiative transfer code applied to Type Ia supernova spectra. We show that we can train an emulator for this problem given a modest training set of 100,000 spectra (easily calculable on modern supercomputers). The results show an accuracy on the percent level (that are dominated by the Monte Carlo nature of tardis and not the emulator) with a speedup of several orders of magnitude. This method has a much broader set of applications and is not limited to the presented problem.

Recovering Thermodynamics from Spectral Profiles observed by IRIS: A Machine and Deep Learning Approach

Abstract Inversion codes allow the reconstruction of a model atmosphere from observations. With the inclusion of optically thick lines that form in the solar chromosphere, such modeling is computationally very expensive because a non-LTE evaluation of the radiation field is required. In this study, we combine the results provided by these traditional methods with machine and deep learning techniques to obtain similar-quality results in an easy-to-use, much faster way. We have applied these new methods to Mg ii h and k lines observed by the Interface Region Imaging Spectrograph (IRIS). As a result, we are able to reconstruct the thermodynamic state (temperature, line-of-sight velocity, nonthermal velocities, electron density, etc.) in the chromosphere and upper photosphere of an area equivalent to an active region in a few CPU minutes, speeding up the process by a factor of 105 − 106. The open-source code accompanying this Letter will allow the community to use IRIS observations to open a new window to a host of solar phenomena.

Wave propagation in head on bonnet impact: Material and design issues

Head on bonnet impact is becoming more and more important in automotive design as regulations on pedestrian safety become more demanding. Despite the relatively low amount of energy involved, these impacts are truly dynamic phenomena as the event duration is comparable with the traveling time of the different wavefronts generated by the impact. In this paper, we show that we can build up a simplified model for the impact based on wave propagation analysis. Using this model, we can analyze head acceleration on existing bonnets or predict it on new ones. Head acceleration in a bonnet impact can thus be estimated over the whole area of the bonnet with a few minutes of CPU.

Reliability-based design optimization using SORM and SQP

In this work a second order approach for reliability-based design optimization (RBDO) with mixtures of uncorrelated non-Gaussian variables is derived by applying second order reliability methods (SORM) and sequential quadratic programming (SQP). The derivation is performed by introducing intermediate variables defined by the incremental iso-probabilistic transformation at the most probable point (MPP). By using these variables in the Taylor expansions of the constraints, a corresponding general first order reliability method (FORM) based quadratic programming (QP) problem is formulated and solved in the standard normal space. The MPP is found in the physical space in the metric of Hasofer-Lind by using a Newton algorithm, where the efficiency of the Newton method is obtained by introducing an inexact Jacobian and a line-search of Armijo type. The FORM-based SQP approach is then corrected by applying four SORM approaches: Breitung, Hohenbichler, Tvedt and a recent suggested formula. The proposed SORM-based SQP approach for RBDO is accurate, efficient and robust. This is demonstrated by solving several established benchmarks, with values on the target of reliability that are considerable higher than what is commonly used, for mixtures of five different distributions (normal, lognormal, Gumbel, gamma and Weibull). Established benchmarks are also generalized in order to study problems with large number of variables and several constraints. For instance, it is shown that the proposed approach efficiently solves a problem with 300 variables and 240 constraints within less than 20 CPU minutes on a laptop. Finally, a most well-know deterministic benchmark of a welded beam is treated as a RBDO problem using the proposed SORM-based SQP approach.

Spatio-energetic cross talk in photon counting detectors: Detector model and correlated Poisson data generator.

An x-ray photon interacts with photon counting detectors (PCDs) and generates an electron charge cloud or multiple clouds. The clouds (thus, the photon energy) may be split between two adjacent PCD pixels when the interaction occurs near pixel boundaries, producing a count at both of the pixels. This is called double-counting with charge sharing. (A photoelectric effect with K-shell fluorescence x-ray emission would result in double-counting as well). As a result, PCD data are spatially and energetically correlated, although the output of individual PCD pixels is Poisson distributed. Major problems include the lack of a detector noise model for the spatio-energetic cross talk and lack of a computationally efficient simulation tool for generating correlated Poisson data. A Monte Carlo (MC) simulation can accurately simulate these phenomena and produce noisy data; however, it is not computationally efficient. In this study, the authors developed a new detector model and implemented it in an efficient software simulator that uses a Poisson random number generator to produce correlated noisy integer counts. The detector model takes the following effects into account: (1) detection efficiency; (2) incomplete charge collection and ballistic effect; (3) interaction with PCDs via photoelectric effect (with or without K-shell fluorescence x-ray emission, which may escape from the PCDs or be reabsorbed); and (4) electronic noise. The correlation was modeled by using these two simplifying assumptions: energy conservation and mutual exclusiveness. The mutual exclusiveness is that no more than two pixels measure energy from one photon. The effect of model parameters has been studied and results were compared with MC simulations. The agreement, with respect to the spectrum, was evaluated using the reduced χ2 statistics or a weighted sum of squared errors, χred2(≥1), where χred2=1 indicates a perfect fit. The model produced spectra with flat field irradiation that qualitatively agree with previous studies. The spectra generated with different model and geometry parameters allowed for understanding the effect of the parameters on the spectrum and the correlation of data. The agreement between the model and MC data was very strong. The mean spectra with 90 keV and 140 kVp agreed exceptionally well: χred2 values were 1.049 with 90 keV data and 1.007 with 140 kVp data. The degrees of cross talk (in terms of the relative increase from single pixel irradiation to flat field irradiation) were 22% with 90 keV and 19% with 140 kVp for MC simulations, while they were 21% and 17%, respectively, for the model. The covariance was in strong agreement qualitatively, although it was overestimated. The noisy data generation was very efficient, taking less than a CPU minute as opposed to CPU hours for MC simulators. The authors have developed a novel, computationally efficient PCD model that takes into account double-counting and resulting spatio-energetic correlation between PCD pixels. The MC simulation validated the accuracy.

An integer programming approach to optimal basic block instruction scheduling for single-issue processors

We present a novel integer programming formulation for basic block instruction scheduling on single-issue processors. The problem can be considered as a very general sequential task scheduling problem with delayed precedence constraints. Our model is based on the linear ordering problem and has, in contrast to the last IP model proposed, numbers of variables and constraints that are strongly polynomial in the instance size. Combined with improved preprocessing techniques and given a time limit of ten minutes of CPU and system time, our branch-and-cut implementation is capable to solve all but eleven of the 369,861 basic blocks of the SPEC 2000 integer and floating point benchmarks to proven optimality. This is competitive to the current state-of-the art constraint programming approach that has also been evaluated on this test suite.

OPTIMIZED LARGE-SCALE CMB LIKELIHOOD AND QUADRATIC MAXIMUM LIKELIHOOD POWER SPECTRUM ESTIMATION

We revisit the problem of exact CMB likelihood and power spectrum estimation with the goal of minimizing computational cost through linear compression. This idea was originally proposed for CMB purposes by Tegmark et al.\ (1997), and here we develop it into a fully working computational framework for large-scale polarization analysis, adopting \WMAP\ as a worked example. We compare five different linear bases (pixel space, harmonic space, noise covariance eigenvectors, signal-to-noise covariance eigenvectors and signal-plus-noise covariance eigenvectors) in terms of compression efficiency, and find that the computationally most efficient basis is the signal-to-noise eigenvector basis, which is closely related to the Karhunen-Loeve and Principal Component transforms, in agreement with previous suggestions. For this basis, the information in 6836 unmasked \WMAP\ sky map pixels can be compressed into a smaller set of 3102 modes, with a maximum error increase of any single multipole of 3.8\% at $\ell\le32$, and a maximum shift in the mean values of a joint distribution of an amplitude--tilt model of 0.006$\sigma$. This compression reduces the computational cost of a single likelihood evaluation by a factor of 5, from 38 to 7.5 CPU seconds, and it also results in a more robust likelihood by implicitly regularizing nearly degenerate modes. Finally, we use the same compression framework to formulate a numerically stable and computationally efficient variation of the Quadratic Maximum Likelihood implementation that requires less than 3 GB of memory and 2 CPU minutes per iteration for $\ell \le 32$, rendering low-$\ell$ QML CMB power spectrum analysis fully tractable on a standard laptop.

Fast protein loop sampling and structure prediction using distance-guided sequential chain-growth Monte Carlo method.

Loops in proteins are flexible regions connecting regular secondary structures. They are often involved in protein functions through interacting with other molecules. The irregularity and flexibility of loops make their structures difficult to determine experimentally and challenging to model computationally. Conformation sampling and energy evaluation are the two key components in loop modeling. We have developed a new method for loop conformation sampling and prediction based on a chain growth sequential Monte Carlo sampling strategy, called Distance-guided Sequential chain-Growth Monte Carlo (DiSGro). With an energy function designed specifically for loops, our method can efficiently generate high quality loop conformations with low energy that are enriched with near-native loop structures. The average minimum global backbone RMSD for 1,000 conformations of 12-residue loops is Å, with a lowest energy RMSD of Å, and an average ensemble RMSD of Å. A novel geometric criterion is applied to speed up calculations. The computational cost of generating 1,000 conformations for each of the x loops in a benchmark dataset is only about cpu minutes for 12-residue loops, compared to ca cpu minutes using the FALCm method. Test results on benchmark datasets show that DiSGro performs comparably or better than previous successful methods, while requiring far less computing time. DiSGro is especially effective in modeling longer loops (– residues).

A binary branch and bound algorithm to minimize maximum scheduling cost

This paper examines a single machine scheduling problem of minimizing the maximum scheduling cost that is nondecreasing with job completion time. Job release dates and precedence constraints are considered. We assume that each job can be processed exactly once without preemption. This is a classical scheduling problem, and is specifically useful in the scheduling of medical treatments. We develop a simple branch and bound algorithm to solve the scheduling problem optimally. A binary branching technique is developed. We use a preemptive solution approach to locate a lower bound, and design a simple heuristic to find an upper bound. Our algorithm is easy to implement and finds optimal schedules in one CPU minute for almost all instances tested, with up to 1000 jobs.

A heuristic for the inventory management of smart vending machine systems

Purpose: The purpose of this paper is to propose a heuristic for the inventory management of smart vending machine systems with product substitution under the replenishment point, order-up-to level policy and to evaluate its performance. Design/methodology/approach: The heuristic is developed on the basis of the decoupled approach. An integer linear mathematical model is built to determine the number of product storage compartments and replenishment threshold for each smart vending machine in the system and the Clarke and Wright’s savings algorithm is applied to route vehicles for inventory replenishments of smart vending machines that share the same delivery days. Computational experiments are conducted on several small-size test problems to compare the proposed heuristic with the integrated optimization mathematical model with respect to system profit. Furthermore, a sensitivity analysis is carried out on a medium-size test problem to evaluate the effect of the customer service level on system profit using a computer simulation. Findings: The results show that the proposed heuristic yielded pretty good solutions with 5.7% error rate on average compared to the optimal solutions. The proposed heuristic took about 3 CPU minutes on average in the test problems being consisted of 10 five-product smart vending machines. It was confirmed that the system profit is significantly affected by the customer service level.

A Molecular Mechanics Approach to Modeling Protein–Ligand Interactions: Relative Binding Affinities in Congeneric Series

We introduce the "Prime-ligand" method for ranking ligands in congeneric series. The method employs a single scoring function, the OPLS-AA/GBSA molecular mechanics/implicit solvent model, for all stages of sampling and scoring. We evaluate the method using 12 test sets of congeneric series for which experimental binding data is available in the literature, as well as the structure of one member of the series bound to the protein. Ligands are "docked" by superimposing a common stem fragment among the compounds in the series using a crystal complex from the Protein Data Bank and sampling the conformational space of the variable region. Our results show good correlation between our predicted rankings and the experimental data for cases in which binding affinities differ by at least 1 order of magnitude. For 11 out of 12 cases, >90% of such ligand pairs could be correctly ranked, while for the remaining case, Factor Xa, 76% of such pairs were correctly ranked. A small number of compounds could not be docked using the current protocol because of the large size of functional groups that could not be accommodated by a rigid receptor. CPU requirements for the method, involving CPU minutes per ligand, are modest compared with more rigorous methods that use similar force fields, such as free energy perturbation. We also benchmark the scoring function using series of ligands bound to the same protein within the CSAR data set. We demonstrate that energy minimization of ligands in the crystal structures is critical to obtain any correlation with experimentally determined binding affinities.

TH-D-BRC-03: A Fast Scatter-Correction Algorithm for KeV CBCT

Purpose: To develop a fast Monte‐Carlo‐based scatter‐correction algorithm for clinical keV cone‐beam CT(CBCT)images.Method and Materials: Estimates of the scatter in the projection‐views of a CBCT scan were obtained by an iterative process, each step consisting of: (1) a coarse CBCT reconstruction; (2) simulation of photon histories for projections using a purpose‐written Monte Carlo code; (3) scoring scatter contributions to fixed points on the detector (a “forced detection” technique); and (4) subtraction of scatter‐estimates from the measured pixel‐values. The scatter signal at each pixel was estimated using linear interpolationspatially between the values calculated at the fixed points and angularly between projection angles. Following convergence to a set of scatter‐corrected profiles, a final full‐resolution scatter‐corrected reconstruction was performed. All CBCT reconstructions were performed using software developed in‐house. The x‐ray tube spectrum and the energy‐response of the detector were both modeled. To validate the technique, projection measurements (120 kV and 0.4 mAs per projection) of a Catphan quality‐assurance phantom (The Phantom Laboratory) were obtained using a Synergy XVI CBCT unit (Elekta Limited). Results: Typically the algorithm took less than 2 min to complete 4 iterations on a desktop PC, after which convergence was obtained. Qualitatively, the algorithm resulted in an improved image with the characteristic ‘cupping’ artifacts, due to scatter, disappearing. Quantitatively, non‐uniformity was decreased after correction from about 15% to 1% or less at a cost of an increase in image noise from 3.7% to 5.1%. CT number accuracy was also markedly improved. Conclusion: It was shown Monte‐Carlo‐based scatter‐correction of clinical keV CBCTimages does not have to be prohibitively slow. Such a scatter‐correction can be successfully performed in a few CPU minutes.

TU‐EE‐A1‐01: Validation of a Prototype Deterministic Solver for Photon Beam Dose Calculations On Acquired CT Data in the Presence of Narrow Beams and Heterogeneities

Purpose: To evaluate a deterministic method for solving the neutral and charged particle transport equations in cases where electron disequilibrium is significant. Methods and Materials: A prototype deterministic solver, Acuros®, has been developed and validated for photon beam radiotherapy. Acuros is based on the Attila® radiation transport code, which solves the differential form of the linear Boltzmann transport equation for neutral particles, and the linear Boltzmann‐Fokker‐Plank transport equation for charged particles. The angular and energy dependent photon and electron flux is solved at every spatial unknown in the computational domain, and quantities such as dose‐to‐medium (DM) are obtained by multiplying the energy dependent particle flux by the energy dependent energy deposition cross section for the associated image pixel material. Comparisons were made with the Monte Carlo code EGSnrc (DOSXYZnrc) for a head‐and‐neck case having eight 1.5×1.5 cm2 photon beams with a 6 MV photon spectrum. The Acuros calculation used 123,000 elements. Since four spatial unknowns were solved in each element, this equated to 492,000 spatial degrees of freedom.The DOSXYZnrc calculation used 2.5×2.5×2.5 mm3 voxels. Results: Results were compared at 95,183 image pixels where dose > 5% of Dmax. 99.02% of pixels satisfied the 3%/3mm criteria (94,251 out of 95,183 pixels). Computational times for the Acuros and DOSXYZnrc calculations were approximately 19.6 CPU minutes (2.2 GHz Opteron processor)and 2,890 CPU minutes (0.4% avg. uncertainty in voxels > Dmax/2), respectively. Through a rewritten, radiotherapy specific solver, Acuros computational times were further reduced to 7.45 CPU minutes without affecting accuracy. When only doses in regions greater than 20% of Dmax are of interest, computational times are further reduced to 3.08 CPU minutes. Conclusions: A clinically viable combination of dose calculation speed and accuracy has been achieved using a radiotherapy specific deterministic solver.Research funded in part by NIH grant 1R43 CA105806‐01A1.