Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “bayesian optimization”

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 865 records · Page 48

Probabilistic models for feedback systems.

In previous work, we developed a Bayesian-based methodology to analyze the reliability of hierarchical systems. The output of the procedure is a statistical distribution of the reliability, thus allowing many questions to be answered. The principal advantage of the approach is that along with an estimate of the reliability, we also can provide statements of confidence in the results. The model is quite general in that it allows general representations of all of the distributions involved, it incorporates prior knowledge into the models, it allows errors in the 'engineered' nodes of a system to be determined by the data, and leads to the ability to determine optimal testing strategies. In this report, we provide the preliminary steps necessary to extend this approach to systems with feedback. Feedback is an essential component of 'complexity' and provides interesting challenges in modeling the time-dependent action of a feedback loop. We provide a mechanism for doing this and analyze a simple case. We then consider some extensions to more interesting examples with local control affecting the entire system. Finally, a discussion of the status of the research is also included.

99 GENERAL AND MISCELLANEOUS↗

Improving Site-Specific Radiological Performance Assessments - 13431

An improved approach is presented for conducting complete and defensible radiological site-specific performance assessments (PAs) to support radioactive waste disposal decisions. The basic tenets of PA were initiated some thirty years ago, focusing on geologic disposals and evaluating compliance with regulations. Some of these regulations were inherently probabilistic (i.e., addressing uncertainty in a quantitative fashion), such as the containment requirements of the U.S. Environmental Protection Agency's (EPA's) 40 CFR 191, Environmental Radiation Protection Standards for Management and Disposal of Spent Nuclear Fuel, High-Level and Transuranic Radioactive Wastes, Chap. 191.13 [1]. Methods of analysis were developed to meet those requirements, but at their core early PAs used 'conservative' parameter values and modeling approaches. This limited the utility of such PAs to compliance evaluation, and did little to inform decisions about optimizing disposal, closure and long-term monitoring and maintenance, or, in general, maintaining doses 'as low as reasonably achievable' (ALARA). This basic approach to PA development in the United States was employed essentially unchanged through the end of the 20. century, principally by the U.S. Department of Energy (DOE). Performance assessments developed in support of private radioactive waste disposal operations, regulated by the U.S. Nuclear Regulatory Commission (NRC) and its agreement states, were typically not as sophisticated. Discussion of new approaches to PA is timely, since at the time of this writing, the DOE is in the midst of revising its Order 435.1, Radioactive Waste Management [2], and the NRC is revising 10 CFR 61, Licensing Requirements for Land Disposal of Radioactive Waste [3]. Over the previous decade, theoretical developments and improved computational technology have provided the foundation for integrating decision analysis (DA) concepts and objective-focused thinking, plus a Bayesian approach to probabilistic modeling and risk analysis, to guide improvements in PA. This decision-making approach, [4, 5, 6] provides a transparent formal framework for using a value- or objective-focused approach to decision-making. DA, as an analytical means to implement structured decision making, provides a context for both understanding how uncertainty affects decisions and for targeting uncertainty reduction. The proposed DA approach improves defensibility and transparency of decision-making. The DA approach is fully consistent with the need to perform realistic modeling (rather than conservative modeling), including evaluation of site-specific factors. Instead of using generic stylized scenarios for radionuclide fate and transport and for human exposures to radionuclides, site-specific scenarios better represent the advantages and disadvantages of alternative disposal sites or engineered designs, thus clarifying their differences as well as providing a sound basis for evaluation of site performance. The full DA approach to PA is described, from explicitly incorporating societal values through stakeholder involvement to model building. Model building involves scoping by considering features, events, processes, and exposure scenarios (FEPSs), development of a conceptual site model (CSM), translation into numerical models and subsequent computation, and model evaluation. These are implemented in a cycle of uncertainty analysis, sensitivity analysis and value of information analysis so that uncertainty can be reduced until sufficient confidence is gained in the decisions to be made. This includes the traditional focus on hydrogeological processes, but also places emphasis on other FEPSs such as biotically-induced transport and human exposure phenomena. The significance of human exposure scenarios is emphasized by modifying the traditional acronym 'FEPs' to include them, hence 'FEPSs'. The radioactive waste community is also recognizing that disposal sites are to be considered a national (or even global) resource. As such, there is a pressing need to optimize their utility within the constraints of protecting human health and the environment. Failing to do so will result in the need for additional sites or options for storing radioactive waste temporarily, assuming a continued need for radioactive waste disposal. Optimization should be performed using DA, including economic analysis, invoked if necessary through the ALARA process. The economic analysis must recognize the cost of implementation (disposal design, closure, maintenance, etc.), and intra- and inter-generational equity in order to ensure that the best possible radioactive waste management decisions are made for the protection of both current and future generations. In most cases this requires consideration of population or collective risk. (authors)

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

Machine learning based approach to predict ductile damage model parameters for polycrystalline metals

Damage models for ductile materials typically need to be parameterized, often with the appropriate parameters changing for a given material depending on the loading conditions. This can make parameterizing these models computationally expensive, since an inverse problem must be solved for each loading condition. Using standard inverse modeling techniques typically requires hundreds or thousands of high-fidelity computer simulations to estimate the optimal parameters. Additionally, the time of a human expert is required to set up the inverse model. Machine learning has recently emerged as an alternative approach to inverse modeling in these settings, where the machine learning model is trained in an offline manner and new parameters can be quickly generated on the fly, after training is complete. Here, this work utilizes such a workflow to enable the rapid parameterization of a ductile damage model called TEPLA with a machine learning inverse model. The machine learning model can efficiently estimate the model parameters much faster, as compared to previously employed methods, such as Bayesian calibration. The results demonstrate good accuracy on a synthetic test dataset and is validated against experimental data.

36 MATERIALS SCIENCE↗

KEPLER INPUT CATALOG: PHOTOMETRIC CALIBRATION AND STELLAR CLASSIFICATION

We describe the photometric calibration and stellar classification methods used by the Stellar Classification Project to produce the Kepler Input Catalog (KIC). The KIC is a catalog containing photometric and physical data for sources in the Kepler mission field of view; it is used by the mission to select optimal targets. Four of the visible-light (g, r, i, z) magnitudes used in the KIC are tied to Sloan Digital Sky Survey magnitudes; the fifth (D51) is an AB magnitude calibrated to be consistent with Castelli and Kurucz (CK) model atmosphere fluxes. We derived atmospheric extinction corrections from hourly observations of secondary standard fields within the Kepler field of view. For these filters and extinction estimates, repeatability of absolute photometry for stars brighter than magnitude 15 is typically 2%. We estimated stellar parameters {l_brace}T{sub eff}, log (g), log (Z), E{sub B-V}{r_brace} using Bayesian posterior probability maximization to match observed colors to CK stellar atmosphere models. We applied Bayesian priors describing the distribution of solar-neighborhood stars in the color-magnitude diagram, in log (Z), and in height above the galactic plane. Several comparisons with samples of stars classified by other means indicate that for 4500 K {<=}T{sub eff} {<=} 6500 K, our classifications are reliable within about {+-}200 K and 0.4 dex in log (g) for dwarfs, with somewhat larger log (g) uncertainties for giants. It is difficult to assess the reliability of our log (Z) estimates, but there is reason to suspect that it is poor, particularly at extreme T{sub eff}. Comparisons between the CK models and observed colors are generally satisfactory with some exceptions, notably for stars cooler than 4500 K. Of great importance for the Kepler mission, for T{sub eff} {<=} 5400 K, comparison with asteroseismic results shows that the distinction between main-sequence stars and giants is reliable with about 98% confidence. Larger errors in log (g) occur for warmer stars, for which our filter set provides inadequate gravity diagnostics. The KIC is available through the MAST data archive.

79 ASTRONOMY AND ASTROPHYSICS↗

Active learning emulators for nuclear two-body scattering in momentum space

In this work we extend the active learning emulators for two-body scattering in coordinate space with error estimation, recently developed by Maldonado et al. [Phys. Rev. C 112, 024002], to coupled-channel scattering in momentum space. Our full-order model (FOM) solver is based on the Lippmann-Schwinger integral equation for the scattering t-matrix as opposed to the radial Schrödinger equation. We use (Petrov-)Galerkin projections and high-fidelity calculations at a few snapshots across the parameter space of the interaction to construct efficient reduced-order models (ROMs), trained by a greedy algorithm for locally optimal snapshot selection. Both the FOM solver and the corresponding ROMs are implemented efficiently in Python using Google's JAX library. We present results for emulating scattering phase shifts in coupled and uncoupled channels and cross sections, and assess the accuracy of the developed ROMs and their computational speedup factors. We also develop emulator error estimation for both the t-matrix and the total cross section. The software framework for reproducing and extending our results is publicly available. Together with our recent advances in developing active-learning emulators for three-body scattering, these emulator frameworks set the stage for full Bayesian calibrations of chiral nuclear interactions and optical models against scattering data with quantified emulator errors.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

CORRECTING VELOCITY DISPERSIONS OF DWARF SPHEROIDAL GALAXIES FOR BINARY ORBITAL MOTION

We show that the measured velocity dispersions of dwarf spheroidal galaxies from about 4 to 10 km s{sup -1} are unlikely to be inflated by more than 30% due to the orbital motion of binary stars and demonstrate that the intrinsic velocity dispersions can be determined to within a few percent accuracy using two-epoch observations with 1-2 yr as the optimal time interval. The crucial observable is the threshold fraction-the fraction of stars that show velocity changes larger than a given threshold between measurements. The threshold fraction is tightly correlated with the dispersion introduced by binaries, independent of the underlying binary fraction and distribution of orbital parameters. We outline a simple procedure to correct the velocity dispersion to within a few percent accuracy by using the threshold fraction and provide fitting functions for this method. We also develop a methodology for constraining properties of binary populations from both single- and two-epoch velocity measurements by including the binary velocity distribution in a Bayesian analysis.

79 ASTRONOMY AND ASTROPHYSICS↗

Support for Reactor Operators in Case of Cyber-Security Threats (NEUP Final Report)

Increasing use of digital technology in nuclear power plants (NPPs) poses cyber-security as a crucial threat to public safety and to continuous energy production. Cyber-security risks are comprised of complex known and unknown interactions between various entities, system vulnerabilities, network protocols, human users and malicious attacks. There is little understanding or research geared towards plant operators’ response under cyber-security threats and operational procedures to cope with such threats. This is particularly critical when a cyber-security event masquerades as a safety incident or an evolving accident. Instead of leading operators to the remediation of the accident, the masked cyber event may lead them to circumvent the safety mechanisms of the plant. Given these challenges, the objective of this project is to develop a tool to characterize abnormal NPP events as “cyber” or “safety” incidents, and to further develop this tool as a real-time operator aid to assist in response to the incident. This NEUP Final Report summarizes the research efforts towards the above goals and the results in the following three aspects. First, a methodology for classifying abnormal NPP events as cyber-attacks or safety incidents was developed. This provides the basis for responding to potential cyber-attacks. In this methodology, the causal relationships between various variables (i.e., physical variables, network variables, and control variables) for a nuclear system under consideration are modeled using dynamic Bayesian networks. Based on observations of certain variables (i.e., the evidence), inference is performed to identify the state of hidden variables, which helps indicate the occurrence of cyber-attacks or safety incidents. The proposed methodology was applied to the full scope NPP simulator system at The Ohio State University as a case study. The result shows that the proposed methodology is capable of distinguishing between cyber-attacks and safety incidents. Second, a game-theoretic approach for responding to cyber-attacks on nuclear power plants was developed. This is the second major goal of the project, and the method is used to recommend optimal responses to reactor operators in the case of cyber-attacks. We consider two cases in this research. In the first case, the interaction between the defender (i.e., the reactor operator) and the attacker is modeled as a two-player, nonzero-sum, Markov game with an infinite time horizon. In the second case, we consider a semi-Markov game between the defender and the attacker within a finite time horizon. The model in the second case enables us to consider the time sensitive response of the defender. In each case, the method for deriving the Nash equilibrium is developed. The Nash Equilibrium of the game provides the valid prediction of both players’ actions because no single player can benefit from unilaterally deviating from the equilibrium policy if the other player adheres to his/hers, hence it provides the best response of the defender to cyber-attacks. The proposed method in each case is demonstrated through a simplified benchmark digital feedwater control system. Third, a prototype display is developed to integrate the above information. The prototype display depicts the communication information to the operator and incorporates the cyber-security tool developed by the team into the interface. It consists of the regular display in current control rooms, as well as a panel used to indicate the occurrence of cyber-attacks or safety events and a panel used to guide the operator’s response to detected cyber-attacks.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

A Bayesian statistical approach to inferring particle dynamics from in-situ pump POC and chloropigment data from the Mediterranean Sea

Concentrations of chloropigments and particulate organic carbon (POC) in large-volume in-situ pump samples from the Mediterranean Sea were used to estimate rate constants of processes that control the fate of particles, and specifically chloropigments, in the water column. In this paper we introduce a Bayesian statistical inversion method that combines the data with a new box model and has the capacity to infer rate constants for POC respiration/dissolution, chlorophyll and pheopigment degradation, and particle aggregation and disaggregation. We use first-order kinetics to model disaggregation, and use both first-order and second-order kinetics to model aggregation. Using these methods, the estimated small-particle (1–70 μm) POC respiration rate constant was 2.44 -1.00 +1.69 yr -1 (0.41 yr). The estimated disaggregation and second-order aggregation rate constants were 85.6 -36.4 +63.4 yr -1 (1.17×10 -2 yr) and 2.78 -1.17 +2.01 μM -1 yr -1 , respectively. Using the optimal rate constants and the corresponding particle concentrations, disaggregation is ~4.2 times faster than the small-size POC dissolution rate, which indicates that disaggregation is a dominant processes at the time of sampling. More importantly, by comparing our results with those of previous studies, we conclude that sampling methods have less influence than tracers themselves on inferring particle dynamic rate constants. We previously introduced a somewhat similar approach to modeling SV sediment trap data, but large volume pumps are a much more common sample collection method in oceanographic surveys than SV sediment traps, and thus our new model should have a wider applicability.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

A Multirange Vehicle Speed Prediction With Application to Model Predictive Control-Based Integrated Power and Thermal Management of Connected Hybrid Electric Vehicles

Abstract Connectivity and automated driving technologies have opened up new research directions in the energy management of vehicles which exploit look-ahead preview and enhance the situational awareness. Despite this advancement, the vehicle speed preview that can be obtained from vehicle-to-vehicle/infrastructure (V2V/I) communications is often limited to a relatively short time-horizon. The vehicular energy systems, specifically those of the electrified vehicles, consist of multiple interacting power and thermal subsystems that respond over different time-scales. Consequently, their optimal energy management can greatly benefit from long-term speed prediction beyond that available through V2V/I communications. Accurately extending the look-ahead preview, on the other hand, is fundamentally challenging due to the dynamic nature of the traffic environment. To address this challenge, we propose a data-driven multirange vehicle speed prediction strategy for arterial corridors with signalized intersections, providing the vehicle speed preview for three different ranges, i.e., short-, medium-, and long-range. The short-range preview is obtained by V2V/I communications. The medium-range preview is realized using a neural network (NN), while the long-range preview is predicted based on a Bayesian network (BN). The predictions are updated in real-time based on the current state of traffic and incorporated into a multihorizon model predictive control (MH-MPC) for integrated power and thermal management (iPTM) of connected vehicles. The results of design and evaluation of the performance of the proposed data-informed MH-MPC for iPTM of connected hybrid electric vehicles (HEVs) using traffic data for real-world city driving are reported.

Automation & Control Systems↗

Calibrating hypersonic turbulence flow models with the HIFiRE-1 experiment using data-driven machine-learned models.

In this paper we study the efficacy of combining machine-learning methods with projection-based model reduction techniques for creating data-driven surrogate models of computationally expensive, high-fidelity physics models. Such surrogate models are essential for many-query applications e.g., engineering design optimization and parameter estimation, where it is necessary to invoke the high-fidelity model sequentially, many times. Surrogate models are usually constructed for individual scalar quantities. However there are scenarios where a spatially varying field needs to be modeled as a function of the model’s input parameters. Here we develop a method to do so, using projections to represent spatial variability while a machine-learned model captures the dependence of the model’s response on the inputs. The method is demonstrated on modeling the heat flux and pressure on the surface of the HIFiRE-1 geometry in a Mach 7.16 turbulent flow. The surrogate model is then used to perform Bayesian estimation of freestream conditions and parameters of the SST (Shear Stress Transport) turbulence model embedded in the high-fidelity (Reynolds-Averaged Navier–Stokes) flow simulator, using shock-tunnel data. The paper provides the first-ever Bayesian calibration of a turbulence model for complex hypersonic turbulent flows. We find that the primary issues in estimating the SST model parameters are the limited information content of the heat flux and pressure measurements and the large model-form error encountered in a certain part of the flow.

42 ENGINEERING↗

An ensemble model of QSAR tools for regulatory risk assessment

Quantitative structure activity relationships (QSARs) are theoretical models that relate a quantitative measure of chemical structure to a physical property or a biological effect. QSAR predictions can be used for chemical risk assessment for protection of human and environmental health, which makes them interesting to regulators, especially in the absence of experimental data. For compatibility with regulatory use, QSAR models should be transparent, reproducible and optimized to minimize the number of false negatives. In silico QSAR tools are gaining wide acceptance as a faster alternative to otherwise time-consuming clinical and animal testing methods. However, different QSAR tools often make conflicting predictions for a given chemical and may also vary in their predictive performance across different chemical datasets. In a regulatory context, conflicting predictions raise interpretation, validation and adequacy concerns. To address these concerns, ensemble learning techniques in the machine learning paradigm can be used to integrate predictions from multiple tools. By leveraging various underlying QSAR algorithms and training datasets, the resulting consensus prediction should yield better overall predictive ability. We present a novel ensemble QSAR model using Bayesian classification. The model allows for varying a cut-off parameter that allows for a selection in the desirable trade-off between model sensitivity and specificity. The predictive performance of the ensemble model is compared with four in silico tools (Toxtree, Lazar, OECD Toolbox, and Danish QSAR) to predict carcinogenicity for a dataset of air toxins (332 chemicals) and a subset of the gold carcinogenic potency database (480 chemicals). Leave-one-out cross validation results show that the ensemble model achieves the best trade-off between sensitivity and specificity (accuracy: 83.8 % and 80.4 %, and balanced accuracy: 80.6 % and 80.8 %) and highest inter-rater agreement [kappa (κ): 0.63 and 0.62] for both the datasets. The ROC curves demonstrate the utility of the cut-off feature in the predictive ability of the ensemble model. In conclusion, this feature provides an additional control to the regulators in grading a chemical based on the severity of the toxic endpoint under study.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Multiple objective optimization in reliability demonstration test

Reliability demonstration tests are usually performed in product design or validation processes to demonstrate whether a product meets specified requirements on reliability. For binomial demonstration tests, the zero-failure test has been most commonly used due to its simplicity and use of minimum sample size to achieve an acceptable consumer’s risk level. However, this test can often result in unacceptably high risk for producers as well as a low probability of passing the test even when the product has good reliability. This paper explicitly explores the interrelationship between multiple objectives that are commonly of interest when planning a demonstration test and proposes structured decision-making procedures using a Pareto front approach for selecting an optimal test plan based on simultaneously balancing multiple criteria. Different strategies are suggested for scenarios with different user priorities and graphical tools are developed to help quantify the trade-offs between choices and to facilitate informed decision making. As a result, potential impacts of some subjective user inputs on the final decision are studied to offer insights and useful guidance for general applications.

97 MATHEMATICS AND COMPUTING↗

Bayesian, Multifidelity Operator Learning for Complex Engineering Systems–A Position Paper

Abstract Deep learning has significantly improved the state-of-the-art in computer vision and natural language processing, and holds great potential to design effective tools for predicting and simulating complex engineering systems. In particular, scientific machine learning seeks to apply the power of deep learning to scientific and engineering tasks, with operator learning (OL) emerging as a particularly effective tool. OL can approximate nonlinear operators arising in complex engineering systems, making it useful for simulating, designing, and controlling those systems. In this position paper, we provide a comprehensive overview of OL, including its potential applications to complex engineering domains. We cover three variations of OL approaches: deterministic OL for modeling nonautonomous systems, OL with uncertainty quantification (UQ) capabilities, and multifidelity OL. For each variation, we discuss drawbacks and potential applications to engineering, in addition to providing a detailed explanation. We also highlight how multifidelity OL approaches with UQ capabilities can be used to design, optimize, and control engineering systems. Finally, we outline some potential challenges for OL within the engineering domain.

Computer Science↗

Uncertainty quantification of a deep learning fuel property prediction model

Deep learning models are being widely used in the field of combustion. Given the black-box nature of typical neural network based models, uncertainty quantification (UQ) is critical to ensure the reliability of predictions as well as the training datasets, and for a principled quantification of noise and its various sources. Deep learning surrogate models for predicting properties of chemical compounds and mixtures have been recently shown to be promising for enabling data-driven fuel design and optimization, with the ultimate goal of improving efficiency and lowering emissions from combustion engines. In this study, UQ is performed for a multi-task deep learning model that simultaneously predicts the research octane number (RON), Motor Octane Number (MON), and Yield Sooting Index (YSI) of pure components and multicomponent blends. The deep learning model is comprised of three smaller networks: Extractor 1, Extractor 2, and Predictor, and a mixing operator. The molecular fingerprints of individual components are encoded via Extractor 1 and Extractor 2, the mixing operator generates fingerprints for mixtures/blends based on linear mixing operation, and the predictor maps the fingerprint to the target properties. Two different classes of UQ methods, Monte Carlo ensemble methods and Bayesian neural networks (BNNs), are employed for quantifying the epistemic uncertainty. Combinations of Bernoulli and Gaussian distributions with DropConnect and DropOut techniques are explored as ensemble methods. All the DropConnect, DropOut and Bayesian layers are applied to the predictor network. Aleatoric uncertainty is modeled by assuming that each data point has an independent uncertainty associated with it. The results of the UQ study are further analyzed to compare the performance of BNN and ensemble methods. Although this study is confined to UQ of fuel property prediction, the methodologies are applicable to other deep learning frameworks that are being widely used in the combustion community.

33 ADVANCED PROPULSION SYSTEMS↗

A Bayesian approach to real-time 3D tumor localization via monoscopic x-ray imaging during treatment delivery

Purpose: Monoscopic x-ray imaging with on-board kV devices is an attractive approach for real-time image guidance in modern radiation therapy such as VMAT or IMRT, but it falls short in providing reliable information along the direction of imaging x-ray. By effectively taking consideration of projection data at prior times and/or angles through a Bayesian formalism, the authors develop an algorithm for real-time and full 3D tumor localization with a single x-ray imager during treatment delivery. Methods: First, a prior probability density function is constructed using the 2D tumor locations on the projection images acquired during patient setup. Whenever an x-ray image is acquired during the treatment delivery, the corresponding 2D tumor location on the imager is used to update the likelihood function. The unresolved third dimension is obtained by maximizing the posterior probability distribution. The algorithm can also be used in a retrospective fashion when all the projection images during the treatment delivery are used for 3D localization purposes. The algorithm does not involve complex optimization of any model parameter and therefore can be used in a ''plug-and-play'' fashion. The authors validated the algorithm using (1) simulated 3D linear and elliptic motion and (2) 3D tumor motion trajectories of a lung and a pancreas patient reproduced by a physical phantom. Continuous kV images were acquired over a full gantry rotation with the Varian TrueBeam on-board imaging system. Three scenarios were considered: fluoroscopic setup, cone beam CT setup, and retrospective analysis. Results: For the simulation study, the RMS 3D localization error is 1.2 and 2.4 mm for the linear and elliptic motions, respectively. For the phantom experiments, the 3D localization error is < 1 mm on average and < 1.5 mm at 95th percentile in the lung and pancreas cases for all three scenarios. The difference in 3D localization error for different scenarios is small and is not statistically significant. Conclusions: The proposed algorithm eliminates the need for any population based model parameters in monoscopic image guided radiotherapy and allows accurate and real-time 3D tumor localization on current standard LINACs with a single x-ray imager.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

BEYONDPLANCK III. Commander3

We describe the computational infrastructure for end-to-end Bayesian cosmic microwave background (CMB) analysis implemented by the BeyondPlanck Collaboration. The code is called Commander3. It provides a statistically consistent framework for global analysis of CMB and microwave observations and may be useful for a wide range of legacy, current, and future experiments. The paper has three main goals. Firstly, we provide a high-level overview of the existing code base, aiming to guide readers who wish to extend and adapt the code according to their own needs or re-implement it from scratch in a different programming language. Secondly, we discuss some critical computational challenges that arise within any global CMB analysis framework, for instance in-memory compression of time-ordered data, fast Fourier transform optimization, and parallelization and load-balancing. Thirdly, we quantify the CPU and RAM requirements for the current BEYONDPLANCK analysis, finding that a total of 1.5 TB of RAM is required for efficient analysis and that the total cost of a full Gibbs sample for LFI is 170 CPU-hrs, including both low-level processing and high-level component separation, which is well within the capabilities of current low-cost computing facilities. The existing code base is made publicly available under a GNU General Public Library (GPL) license.

79 ASTRONOMY AND ASTROPHYSICS↗

The cut-sky cosmic microwave background is not anomalous

The observed angular correlation function of the cosmic microwave background has previously been reported to be anomalous, particularly when measured in regions of the sky uncontaminated by Galactic emission. Recent work by Efstathiou et al. presents a Bayesian comparison of isotropic theories, casting doubt on the significance of the purported anomaly. We extend this analysis to all anisotropic Gaussian theories with vanishing mean (<{delta}T>=0), using the much wider class of models to confirm that the anomaly is not likely to point to new physics. On the other hand if there is any new physics to be gleaned, it results from low-l alignments which will be better quantified by a full-sky statistic. We also consider quadratic maximum likelihood power spectrum estimators that are constructed assuming isotropy. The underlying assumptions are therefore false if the ensemble is anisotropic. Nonetheless we demonstrate that, for theories compatible with the observed sky, these estimators (while no longer optimal) remain statistically superior to pseudo-C{sub l} power spectrum estimators.

79 ASTRONOMY AND ASTROPHYSICS↗

Kernel learning backward SDE filter for data assimilation

In this paper, we develop a kernel learning backward SDE filter method to estimate the state of a stochastic dynamical system based on its partial noisy observations. A system of forward backward stochastic differential equations is used to propagate the state of the target dynamical model, and Bayesian inference is applied to incorporate the observational information. Further, to characterize the dynamical model in the entire state space, we introduce a kernel learning method to learn a continuous global approximation for the conditional probability density function of the target state by using discrete approximated density values as training data. Numerical experiments demonstrate that the kernel learning backward SDE is highly effective.

97 MATHEMATICS AND COMPUTING↗