Uncertainty quantification (UQ) is a key component for predictive computational science. Irregardless of the specific method employed for this task, several computational simulations are needed to perform satisfactory UQ studies. Multilevel and multi-fidelity strategies have gained popularity because they can lower the UQ computational requirements, which is especially important whenever the computational simulations require high-fidelity and expensive-to-run computer codes. In the context of sampling-based UQ approaches, it is paramount to reduce the variance of estimators to reach the required precision in the statistics of interest. Multilevel sampling methods, such as multilevel Monte Carlo (MLMC) and multilevel quasi-Monte Carlo (MLQMC), enable provably lower variance than single fidelity Monte Carlo estimators by leveraging a hierarchy of model fidelities to reduce the cost of approximating expectations of a quantity of interest for a given estimator precision. Existing MLQMC methods often rely on multiple independent randomizations of a low-discrepancy (LD) sequence to estimate statistical error contributions on each level. While this approach is standard, these replications take up a significant part of the available computational budget, which could otherwise be spent on increasing the number of points from a single LD sequence and further reducing the variance. However, a single LD sequence does not provide the level-wise statistical error estimates needed to solve the sample allocation problem. We propose to recast MLQMC in a Bayesian cubature framework, which uses a single LD sequence and quantifies statistical error through the posterior variance of a Gaussian process (GP) model. When paired with certain types of LD sequences, GP regression and hyperparameter optimization can be carried out with only O(nlogn) computations and O(n) storage, where n is the number of samples. Building on the adaptive sample allocation used in traditional MLQMC, where the number of samples is doubled on the level with the greatest expected benefit, we introduce a new Bayesian utility function that balances the computational cost of doubling against the anticipated reduction in posterior uncertainty. We also propose a new digitally-shift-invariant (DSI) kernel of adaptive smoothness, which combines multiple higher-order DSI kernels through a weighted sum of smoothness parameters, for use with fast digital net GPs. A series of numerical experiments with standard UQ benchmark problems illustrate the performance of our fast Bayesian MLQMC method and associated error estimates for both single-level and multilevel problems with a fixed number of levels. The Bayesian error estimates obtained using digital nets are found to be reliable, although, in some cases, mildly conservative.
Fast charging of lithium-ion batteries at subzero temperatures is limited by increased polarization, reduced charge acceptance, and accelerated degradation. This study develops and validates a combined preheating framework for commercial 21700 NMC811/graphite cells by integrating adaptive internal pulse-current heating with external phase-change composite (PCC)-based heating. Experiments were conducted at initial SOC levels of 30%, 50%, and 80% under −20°C ambient conditions. A control-oriented equivalent-circuit model, parameterized using hybrid pulse power characterization and electrochemical impedance spectroscopy data, was coupled with a three-dimensional thermal model to predict temperature rise, thermal non-uniformity, and energy demand. A voltage-constrained adaptive pulse-current strategy maximized internal heat generation while limiting the absolute SOC drift to within 0.43%. Combined preheating outperformed internal-only and external-only approaches across all SOC levels. At 30% SOC, the strategy raised the cell above 35°C in about 15 min, achieved a heating rate of 3.61°C/min, reduced preheating time by about 30%, and decreased the temperature difference from 2.5°C to 0.9°C. After 200 equivalent full cycles (EFC) at −20°C, the combined-preheated cell retained 86% state of health, compared with 50% for the unheated cell.
A dynamical system is considered safe if its state trajectory remains within a predetermined safe set for all time. Control barrier functions (CBFs) can be used to enforce set invariance, while adaptive control methods maintain robustness and reference-model tracking performance of the closed-loop system in the presence of uncertainties. However, slow actuator dynamics can degrade tracking performance and safety. This study proposes a constructive, safety-critical model-reference-adaptive control (MRAC) framework, in which safe reference commands are generated via a CBF-based quadratic program (QP). MRAC ensures that the tracking error norm remains within a user-defined bound, guarantees exact parameter convergence with a restricted potential function, and incorporates concurrent learning terms into the projection operator-based adaptive update law. A hedging method is used that modifies the ideal reference model dynamics to account for actuator dynamics to prevent degradation of the adaptation mechanism and preserve correct parameter updates in the presence of actuator bandwidth limitations. Tracking and estimation error bounds are identified through Lyapunov stability analysis, and a linear matrix inequalities-based framework is used for computing the minimum allowable actuator bandwidth. These enable the calculation of a decaying safety bound, which is subsequently used in a QP constrained by CBFs to provide performance-predictable safe reference commands that guarantee safety for the uncertain system with actuator dynamics. The effectiveness of the method is demonstrated through simulation-based studies.
Junction Field Effect Transistors (JFETs) are often now considered niche when compared with Metal Oxide Semiconductor Field Effect Transistor (MOSFET) and Complementary Metal Oxide Semiconductor (CMOS) transistor technologies. However, they are still favoured and frequently selected for very low noise analogue Charge Sensitive Preamplifiers (CSPs). The NJ26/2N4416/2N4416A series has proved to be a popular choice of JFET at the front end of feedback resistor-less CSPs for wide-bandgap compound semiconductor radiation spectrometers. However, the design of a feedback resistor-less CSP requires the input JFET to be forward biased, which is beyond the bias regime for which they were designed. Datasheets supplied by manufacturers offer little insight into the input capacitances in this forward biased regime. However, it is the common source input capacitance of the JFET that contributes directly to the 1/f and the white series electronic noise of Xray and electron spectrometers. In this work, the common source reverse feedback and common source input capacitances of four 2N4416 and four 2N4416A JFETs supplied by Vishay Siliconix and InterFET Corporation have been measured as functions of applied bias (-10 V < V < 0.5 V) and operating temperature (-20 degrees C > T > 100 degrees C). Inter alia, it is shown that at the input of the CSP, the Vishay Siliconix devices contribute less capacitance (and consequently less 1/f and white series noise) than the InterFET Corporation devices.
Remote and hybrid work have expanded employees’ ability to adopt work-related tools outside formal organizational control, intensifying concerns about Shadow IT and its security implications. This study examines Shadow IT as a context-sensitive security behavior shaped by socio-cognitive evaluations and perceived situational strain. Using the Theory of Planned Behavior (TPB) as a baseline and extending it with General Strain Theory (GST), the study examines whether situational strain influences Shadow IT directly and indirectly through negative affect. Survey data from 674 remote and hybrid workers were analyzed using PLS-SEM. Results show that situational strain has a positive direct association with Shadow IT use, suggesting that strained conditions can increase the appeal of unsanctioned tools as practical workarounds. However, strain also increases negative affect, which is associated with lower Shadow IT use, indicating a competitive partial mediation. These findings reveal competing strain mechanisms: strain can encourage workaround behavior while generating affective responses that inhibit it.
