Joint Institute for Computational Sciences

The study presents a novel process to design lightweight, high-performance cooling manifolds for power electronics using generative design. The process begins with a baseline design that defines the constraints of the manifold with regard to the target cooling geometry and flow path. A flow optimization is then performed to optimize flow distribution and maximize convective efficiency. Once a final fluid volume is obtained, a structural optimization is conducted to minimize weight and material usage. The simulation results for the final design demonstrated a 40.1% increase in the average heat transfer coefficient, a 7.5% decrease in average chip temperature, a 76.6% improvement in temperature uniformity, and a 63.3% reduction in weight at the expense of a minimal 5.1% increase in pressure drop compared to the baseline design.

Abstract Cells have evolved to sense a wide range of input combinations and integrate those signals through signaling pathways to produce context-specific responses, such as differentiation, cell-type specification, and patterning. To replicate this information-processing capacity, synthetic biology has developed large-scale circuitry inspired by the fundamental principles of computer science. Within this framework, neuromorphic computing implemented using genetic circuits offers the opportunity to significantly enhance the computational capabilities of single cells. In this work, we establish design principles for implementing neuromorphic computing in living cells by identifying the key feature that enables a chemical reaction network to function as a perceptron: an input-output mapping with a tunable threshold. We demonstrate that four ubiquitous chemical reaction networks, namely molecular sequestration, catalytic degradation, competitive binding, and activation/deactivation cycles, all satisfy this requirement and can be engineered as perceptrons. By layering these perceptrons into multi-layer architectures, we then show how to construct both linear and nonlinear decision boundaries through rational tuning of production rates that encode network weights. As proof of principle, we apply this framework to design neural networks capable of discriminating between healthy and cancer cells based on gene expression data from 19 tissue types. Together, this work formalizes the design principles for engineering genetic circuits as neural networks and establishes a foundation for implementing next-generation cellular computation.

Controlling ionic transport in liquids is anticipated to provide new scientific and technological opportunities, but it requires accurate knowledge of atomic-scale dynamics of ions beyond the hydrodynamic description. Atomic dynamics in liquids is characterized by strong and dynamical correlations among atoms, which render a conventional approach to describing the dynamics in reciprocal space challenging. We propose an alternative approach to describe liquid dynamics in real space and time using the time-delayed two-body correlation function, Van Hove correlation function, and discuss recent results on the local dynamics in water and electrolytes.

We developed a framework merging unsupervised and supervised machine learning to classify lightning radio signals, and applied it to the possible detection of terrestrial gamma-ray flashes (TGFs). Recent studies have established a tight connection between energetic in-cloud pulses (EIPs, >150 kA) and a subset of TGFs, enabling continuous and large-scale ground-based TGF detection. However, even with a high peak current threshold, it is time-consuming to manually search for EIPs in a background of many non-EIP events, and it becomes even more difficult when a lower peak-current threshold is used. Machine learning classifiers are an effective tool. Beginning with unsupervised learning, spectral clustering is performed on the low-dimensional features extracted by an autoencoder from raw radio waveforms, showing that +EIPs naturally constitute a distinct class of waveform and 67% of the total population. The clustering results are used to form a labeled dataset (~10,000 events) to further train supervised convolutional neural network (CNN) that targets for +EIPs. Our CNN models identify on average 95.2% of true +EIPs with accuracy up to 98.7%, representing a powerful tool for +EIP classification. The pretrained CNN classifier is further applied to identify lower peak current EIPs (LEIPs, >50 kA) from a larger dataset (~30,000 events). Among 10 LEIPs coincident with Fermi TGF observations, 2 previously reported TGFs and 2 unreported but suspected TGFs are found, while the majority are not associated with detectable TGFs. In addition, unsupervised clustering is found to reflect characteristics of the ionosphere reflection height and its effect on radio wave propagation.

Gravitational wave high-energy Electromagnetic Counterpart All-sky Monitor (GECAM) is a space-borne instrument dedicated to monitoring high-energy transients, thereinto Terrestrial Gamma-ray Flashes (TGFs) and Terrestrial Electron Beams (TEBs). We propose a TGF/TEB search algorithm, with which 147 bright TGFs and 4 TEBs are identified during an effective observation time of $\sim$ 9 months. We show that, with gamma-ray and charged particle detectors, GECAM can effectively identify and distinguish TGFs and TEBs, and measure their temporal and spectral properties in detail. Moreover, we find an interesting TEB consisting of two pulses with a separation of $\sim$ 150 ms, which is expected to originate from a lightning process near the geomagnetic footprint. We also find that the GECAM TGF’s lightning-association ratio is $\sim$ 80\% in the east Asia region using the GLD360 lightning network, which is significantly higher than previous observations.

Despite C/C++ and Python both being very popular programming languages, each tool possesses unique advantages and disadvantages. Notably, computers can run C/C++ code very quickly, but C/C++ code has to first be compiled and the syntax can be difficult for new programmers to understand. Python code, however, sacrifices speed for an easy-to-understand syntax and can be run interactively. Thankfully, it is possible to combine the benefits of Python and C/C++. For example, NumPy is a popular package of linear algebra operations written in C but used with Python. Such a combination allows programmers to not only utilize the fast speeds of C code but also Python's simple syntax. NumPy's potential, however, is limited by its inability to run on graphics processing units (GPUs), processors specialized for handling computations. On the other hand, a linear algebra library known as Matrix Algebra on GPU and Multicore Architectures (MAGMA) is suited for running its code on GPUs. Coupled with the fact that its code is written in C/C++, MAGMA offers extremely fast computations. To combine MAGMA's speed with Python's easy-to-understand syntax, we researched how to use C++ code with Python. By researching a tool known as Simplified Wrapper and Interface Generator (SWIG), we created PyMAGMA - a library of chosen MAGMA functions which can be imported in Python 3.9 for use.

Understanding the microscopic mechanism of the transition of glass remains one of the most challenging topics in Condensed Matter Physics. What controls the sharp slowing down of molecular motion upon approaching the glass transition temperature Tg, whether there is an underlying thermodynamic transition at some finite temperature below Tg, what the role of cooperativity and heterogeneity are, and many other questions continue to be topics of active discussions. This review focuses on the mechanisms that control the steepness of the temperature dependence of structural relaxation (fragility) in glass-forming liquids. We present a brief overview of the basic theoretical models and their experimental tests, analyzing their predictions for fragility and emphasizing the successes and failures of the models. Special attention is focused on the connection of fast dynamics on picosecond time scales to the behavior of structural relaxation on much longer time scales. A separate section discusses the specific case of polymeric glass-forming liquids, which usually have extremely high fragility. We emphasize the apparent difference between the glass transitions in polymers and small molecules. We also discuss the possible role of quantum effects in the glass transition of light molecules and highlight the recent discovery of the unusually low fragility of water. At the end, we formulate the major challenges and questions remaining in this field.

Based on ground-based Doppler radar data with high spatial and temporal resolution, our study focuses on rapid intensification (RI) of a sheared tropical cyclone (TC). The asymmetric inner-core structure and its impact on RI are presented using the Advanced Regional Prediction System-3DVAR. The time evolution of the TC inner-core characteristics shows that Meranti experiences two stages of RI: the asymmetric dominant period and the axisymmetric dominant period. The comparison between the two RI stages shows that the vortex intensified more in the asymmetric stage than the axisymmetric stage. The wavenumber-1 component introduced by the vertical wind shear dominates the asymmetric structure. In the moderate-to-strong shear environment, deep convection occurs mainly in the downshear-left quadrant, generating substantial diabatic heating in the downshear region. Under a suitable inner-core structure configuration, the downshear vortices are entrained cyclonically inward, resulting in RI. It is the interaction between the axisymmetric inner-core structure of the TC and the mesoscale vortex that controls the intensification process, which can be replicated by the nondivergent barotropic model with a similar initial field. This infrequent case provided observation of RI during a strong shear period, which suggests that the inner core structure is vital for TC RI.