Engineering Papers⌕ Search

Engineering topics

Michaud, Isaac James

Publications and source records attributed to Michaud, Isaac James.

Implications of new Reasoning Capabilities for Science and Security: Results from a Quick Initial Study

On Thursday, September 12 OpenAI released “a new series of models designed to spend more time thinking… they can reason through complex tasks and solve harder problems than previous models in science, coding, and math.” These models are referred to as o1-preview and o1-mini and appear to be first results of what had been a closely held project called Strawberry within OpenAI. The models are not described as successors in the earlier GPT series because they provide a qualitatively different type of capability, especially step-by-step reasoning.

97 MATHEMATICS AND COMPUTING↗

Sequential Bayesian Methods for Analyzing Computer Models

Efficient analysis of computer models is essential for the validation and uncertainty quantification of those models. Surrogate models, and Gaussian processes in particular, are a common and powerful approach to analyzing computer models that treat computer models as a black-box function. Gaussian processes form a Bayesian model over a space of functions that gives a measure of uncertainty about the computer model output at unobserved locations and a framework for sequential sampling. This tutorial will show how to fit Gaussian processes on a series of test functions and apply sequential design techniques to estimate extrema, level sets, and reliabilities of those test functions.

97 MATHEMATICS AND COMPUTING↗

EUCLID Experiment: Design, Execution, and Early Results [Slides]

This presentation covers design, execution, and early results for the EUCLID experiment. Unresolved compensating errors for Pu-239 and the Joint uncertainty problems are discussed. Various tools and design implementations are discussed along with upcoming work. Additional slides compare EUCLID calculations to experiment for several nuclear data libraries. ENDF/B-VIII.1β2 is named e81b2.

97 MATHEMATICS AND COMPUTING↗

Uncovering Where Compensating Errors Could Hide in ENDF/B-VIII.0

Unconstrained physics spaces between two or more nuclear data observables in a library occur when their values can be simultaneously adjusted without violating the uncertainties in either differential information or simulations of relevant integral experiments. Differential data are often too imprecise to fully bound all nuclear data observables of interest for application simulations. Integral data are simulated with combinations of nuclear data so that an error in one observable may be hidden by a counterbalancing error in another. In this manner compensating errors may lurk within nuclear data libraries and these errors have the potential to undermine the predictive power of neutron transport simulations, particularly in situations where there is no conclusive validation experiment that resembles the application of interest. The EUCLID project (Experiments Underpinned by Computational Learning for Improvements in Nuclear Data) developed a preliminary workflow to identify these unconstrained physics spaces by bringing together results from a large collection of integral experiments with their simulated counter-parts as well as differential information that have a one-to-one correspondence to nuclear data. This wealth of information is processed by machine learning tools for subsequent refinement by human experts. Here, we show how the EUCLID work-flow is executed by applying it first to 239 Pu and then to 9 Be nuclear data in ENDF/B-VIII.0.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Understanding the impact of nuclear-data covariances on various integral responses using adjustment

The EUCLID (Experiments Underpinned by Computational Learning for Improvements in Nuclear Data) project created a library of sensitivities for nine different integral responses with respect to nuclear data. These integral responses were obtained from measurements at LLNL (Lawrence Livermore National Laboratory) pulsed spheres, critical and sub-critical assemblies. At the same time, covariances for ENDF/B-VIII.0 were processed at LANL (Los Alamos National Laboratory). The combination of these data allow us to study the impact of nuclear-data covariances on various integral responses, either by forward-propagating covariances via sensitivities, or by using nuclear data, integral responses, and sensitivities for adjustment. Here, we will present: the impact of 1 H, 9 Be, 12 C, 27 Al, 56 Fe, 235,238 U, and 239,240 Pu ENDF/B-VIII.0 covariances on simulated bounds of the following integral responses: LLNL pulsed-spheres neutron-leakage spectra, the effective neutron multiplication factor, reaction rates, and reactivity coefficients of ICSBEP critical assemblies. Also, adjustment results with the same nuclear-data covariances and responses will be discussed.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Nuclear data covariances are critical input to determine upper sub-critical limits and to design experiments to increase it [Slides]

This presentation discusses how Upper Subcritical Limits (USL) are key parameters to determine operational limits in nuclear criticality safety evaluations. It also discusses an example of plutonium casting operation using tantalum at LANL PF-4. The Whisper tool at Los Alamos relies on many inputs, including covariance data, leading the presentation to ask if an existing benchmark data be used in Whisper to adjust nuclear data and covariances to justify a higher USL. If not, Whisper can be used to help design an optimal new benchmark experiment. The presentation also seeks to determine what the possible impacts are on USL and operational limits for plutonium casting. In conclusion, nuclear data covariances are used for by Whisper for: GSSL adjustment of nuclear data and covariances, identification of most similar existing benchmark experiments to application, simulation of Upper Subcritical Limit of application, and input to optimization techniques for designing most appropriate new benchmark experiment(s) to meet requirements. This requires a complete set of nuclear data covariances, benchmarks and k-effective sensitivity profiles (for both benchmarks and applications). The presentation concludes by asking if end users should trust results that depend on current covariance data.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Efficient Data Query for Gaussian Process Compressed Data through Value Range Estimation [Slides]

When the resolution of the data increases, data reduction methods are applied to simulation output, including Gaussian process, neural representation and compression algorithms. Lots of data analysis/visualization techniques requires data query, but data query from reduced representation is still challenging. This report will provide examples and provide possible answers to why data query from reduced representation is still challenging.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Uncovering Where Compensating Errors Could Hide in ENDF/B-VIII.0

Unconstrained physics spaces between two or more nuclear data observables in a library occur when their values can be simultaneously adjusted without violating the uncertainties in either differential information or simulations of relevant integral experiments. Differential data are often too imprecise to fully bound all nuclear data observables of interest for application simulations. Integral data are simulated with combinations of nuclear data so that an error in one observable may be hidden by a counterbalancing error in another. In this manner compensating errors may lurk within nuclear data libraries and these errors have the potential to undermine the predictive power of neutron transport simulations, particularly in situations where there is no conclusive validation experiment that resembles the application of interest. The EUCLID project (Experiments Underpinned by Computational Learning for Improvements in Nuclear Data) developed a preliminary workflow to identify these unconstrained physics spaces by bringing together results from a large collection of integral experiments with their simulated counter-parts as well as differential information that have a one-to-one correspondence to nuclear data. This wealth of information is processed by machine learning tools for subsequent refinement by human experts. Here, we show how the EUCLID work-flow is executed by applying it first to 239 Pu and then to 9 Be nuclear data in ENDF/B-VIII.0.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

EUCLID Sensitivity Database

This report documents the EUCLID sensitivity database along with its several use-cases. EUCLID computed sensitivities for the following integral responses: Criticality of ICSBEP critical assemblies, LLNL pulsed-sphere neutron-leakage spectra, reaction rates in selected ICBSEP critical assemblies, delayed neutron factions of selected ICSBEP critical assemblies, reactivity coefficients in two ICSBEP critical assemblies, sub-critical assembly responses and Rossi-alpha of selected critical assemblies. It is described for each response what the reported observable constitutes, the method we used to obtain the sensitivities, and which integral experiments were studied. It is also documented briefly in what format these sensitivities are stored. These sensitivities were used for many aspects of the EUCLID project, like ML-supported large-scale nuclear-data validation, or optimization of integral experiments. But these sensitivities can also be applied for more established processes in the nuclear-data application field such as adjustment or assessing the upper sub-critical limit.

Delayed Neutron Fraction↗

EUCLID: Experiments Underpinned by Computational Learning for Improvements in Nuclear Data [Slides]

EUCLID will design validation experiments optimized to resolve compensating errors and adjust nuclear data to experiments. A big part of the success of the EUCLID proposal was due to previous work supported by NCSP (MCNP, nuclear data, and NCERC capabilities). The work performed under EUCLID will similarly benefit the NCSP mission. It will lead to new MCNP capabilities, improved nuclear data and nuclear data capabilities, and new methodology and tools that will have large impact on future NCERC experiments.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

How New Mexico Leveraged a COVID-19 Case Forecasting Model to Preemptively Address the Healthcare Needs of the State: Quantitative Analysis

Prior to the COVID-19 pandemic, US hospitals relied on static projections of future trends for long-term planning and were only beginning to consider forecasting methods for short-term planning of staffing and other resources. With the overwhelming burden imposed by COVID-19 on the health care system, an emergent need exists to accurately forecast hospitalization needs within an actionable timeframe. Our goal was to leverage an existing COVID-19 case and death forecasting tool to generate the expected number of concurrent hospitalizations, occupied intensive care unit (ICU) beds, and in-use ventilators 1 day to 4 weeks in the future for New Mexico and each of its five health regions. We developed a probabilistic model that took as input the number of new COVID-19 cases for New Mexico from Los Alamos National Laboratory’s COVID-19 Forecasts Using Fast Evaluations and Estimation tool, and we used the model to estimate the number of new daily hospital admissions 4 weeks into the future based on current statewide hospitalization rates. The model estimated the number of new admissions that would require an ICU bed or use of a ventilator and then projected the individual lengths of hospital stays based on the resource need. By tracking the lengths of stay through time, we captured the projected simultaneous need for inpatient beds, ICU beds, and ventilators. We used a postprocessing method to adjust the forecasts based on the differences between prior forecasts and the subsequent observed data. Thus, we ensured that our forecasts could reflect a dynamically changing situation on the ground. Forecasts made between September 1 and December 9, 2020, showed variable accuracy across time, health care resource needs, and forecast horizon. Forecasts made in October, when new COVID-19 cases were steadily increasing, had an average accuracy error of 20.0%, while the error in forecasts made in September, a month with low COVID-19 activity, was 39.7%. Across health care use categories, state-level forecasts were more accurate than those at the regional level. Although the accuracy declined as the forecast was projected further into the future, the stated uncertainty of the prediction improved. Forecasts were within 5% of their stated uncertainty at the 50% and 90% prediction intervals at the 3- to 4-week forecast horizon for state-level inpatient and ICU needs. However, uncertainty intervals were too narrow for forecasts of state-level ventilator need and all regional health care resource needs. Real-time forecasting of the burden imposed by a spreading infectious disease is a crucial component of decision support during a public health emergency. Our proposed methodology demonstrated utility in providing near-term forecasts, particularly at the state level. This tool can aid other stakeholders as they face COVID-19 population impacts now and in the future.

60 APPLIED LIFE SCIENCES↗

Gaussian Process Optimization of Sensitivity-Based Similarity Metrics between New Nuclear Applications and New/Existing Benchmarks [Slides]

This presentation discusses Nuclear Criticality Safety (NCS) and how designing safe, new nuclear criticality experiments requires expert judgement, which could take years of experience. Sensitivity/uncertainty (S/U) analysis can be utilized by less experienced individuals to conservatively estimate uncertainties in important parameters, such as k eff , in newly proposed nuclear experiments. The presentation poses the question of how this analysis can be performed and states that the answer lies in matching new nuclear experiments with existing benchmark experiments using similarity metrics. By increasing the criticality safety of the application in this work, higher mass limits could be used in PF-4 operations. Additionally, the presentation discusses MCNP6.2®, Whisper-1.1, the software that can be used in this analysis. Also discussed is the fact that International Criticality Safety Benchmark Evaluation Project (ICSBEP) benchmarks rarely match new nuclear applications and that there are significant differences in given set of materials and/or geometry. If there are no benchmarks that match the application, the presentation discusses the possibility of creating new benchmarks. In conclusion, this work presents a Gaussian process (GP) optimization scheme that was used to generate new benchmarks with the highest sensitivity-based similarity metrics to user-defined nuclear applications. The Gaussian process optimization successfully designed 3 new experimental benchmarks that were highly correlated to the application of interest and had k eff values near critical. Optimization over c k,i-r has shown that investigating specific isotope reactions for different applications is crucial to designing benchmark experiments. Partial contribution from Pu dominates c k similarity metric. Future work includes testing new stand-alone similarity metrics or new combinations of similarity metrics as the design criterion of this optimization – design criterion is application dependent.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Designing Critical Experiments using Gaussian Process Optimization [Slides]

When designing new nuclear applications, there may be a lack of similarity between the application and existing ICSBEP benchmarks. Without appropriate benchmark data, predictive simulations will have high uncertainty in parameters of interest, such as K eff . The goal of the project is to create an algorithm to optimally design critical experiments that match cross-section sensitivities of an application model while minimizing the number of simulation runs in the optimization.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Estimating Mutual Information for High-to-Low Calibration

Presentation shows that KSG 2 is superior to KSG 1 because it scales locally automatically; KSG estimators are limited to a maximum MI due to sample size; LNC extends the capability of KSG without onerous assumptions; iLNC allows LNC to estimate information gain.

97 MATHEMATICS AND COMPUTING↗