Publications

Barnard's Loop is a famous arc of Hα emission located in the Orion star-forming region. Here, we provide evidence of a possible formation mechanism for Barnard's Loop and compare our results with recent work suggesting a major feedback event occurred in the region around 6 Myr ago. We present a 3D model of the large-scale Orion region, indicating coherent, radial, 3D expansion of the OBP-Near/Briceño-1 (OBP-B1) cluster in the middle of a large dust cavity. The large-scale gas in the region also appears to be expanding from a central point, originally proposed to be Orion X. OBP-B1 appears to serve as another possible center, and we evaluate whether Orion X or OBP-B1 is more likely to be the cause of the expansion. We find that neither cluster served as the single expansion center, but rather a combination of feedback from both likely propelled the expansion. Recent 3D dust maps are used to characterize the 3D topology of the entire region, which shows Barnard's Loop's correspondence with a large dust cavity around the OPB-B1 cluster. The molecular clouds Orion A, Orion B, and Orion λ reside on the shell of this cavity. Simple estimates of gravitational effects from both stars and gas indicate that the expansion of this asymmetric cavity likely induced anisotropy in the kinematics of OBP-B1. We conclude that feedback from OBP-B1 has affected the structure of the Orion A, Orion B, and Orion λ molecular clouds and may have played a major role in the formation of Barnard's Loop.

Single-molecule force spectroscopy (smFS) is a powerful approach to studying molecular self-organization. However, the coupling of the molecule with the ever-present experimental device introduces artifacts, that complicate the interpretation of these experiments. Performing statistical inference to learn hidden molecular properties is challenging because these measurements produce non-Markovian time series, and even minimal models lead to intractable likelihoods. To overcome these challenges, we developed a computational framework built on novel statistical methods called simulation-based inference (SBI). SBI enabled us to directly estimate the Bayesian posterior, and extract reduced quantitative models from smFS, by encoding a mechanistic model into a simulator in combination with probabilistic deep learning. Using synthetic data, we could systematically disentangle the measurement of hidden molecular properties from experimental artifacts. The integration of physical models with machine-learning density estimation is general, transparent, easy to use, and broadly applicable to other types of biophysical experiments.

Single-molecule force spectroscopy (smFS) is a powerful approach to studying molecular self-organization. However, the coupling of the molecule with the ever-present experimental device introduces artifacts, that complicate the interpretation of these experiments. Performing statistical inference to learn hidden molecular properties is challenging because these measurements produce non-Markovian time series, and even minimal models lead to intractable likelihoods. To overcome these challenges, we developed a computational framework built on novel statistical methods called simulation-based inference (SBI). SBI enabled us to directly estimate the Bayesian posterior, and extract reduced quantitative models from smFS, by encoding a mechanistic model into a simulator in combination with probabilistic deep learning. Using synthetic data, we could systematically disentangle the measurement of hidden molecular properties from experimental artifacts. The integration of physical models with machine-learning density estimation is general, transparent, easy to use, and broadly applicable to other types of biophysical experiments.

Many applications in magnetic confinement fusion require the efficient calculation of surface integrals with singular integrands. The singularity subtraction approaches typically used to handle such singularities are complicated to implement and low-order accurate. In contrast, we demonstrate that the Kapur–Rokhlin quadrature scheme is well-suited for the logarithmically singular integrals encountered for a toroidally axisymmetric confinement system, is easy to implement and is high-order accurate. As an illustration, we show how to apply this quadrature scheme for the efficient and accurate calculation of the normal component of the magnetic field due to the plasma current on the plasma boundary, via the virtual-casing principle.

We propose a hierarchical tensor-network approach for approximating high-dimensional probability density via empirical distribution. This leverages randomized singular value decomposition (SVD) techniques and involves solving linear equations for tensor cores in this tensor network. The complexity of the resulting algorithm scales linearly in the dimension of the high-dimensional density. An analysis of estimation error demonstrates the effectiveness of this method through several numerical experiments.

We develop a density-matrix renormalization group (DMRG) algorithm for the simulation of quantum circuits. This algorithm can be seen as the extension of the time-dependent DMRG from the usual situation of Hermitian Hamiltonian matrices to quantum circuits defined by unitary matrices. For small circuit depths, the technique is exact and equivalent to other matrix product state–based techniques. For larger depths, it becomes approximate in exchange for an exponential speed up in computational time. Like an actual quantum computer, the quality of the DMRG results is characterized by a finite fidelity. However, unlike a quantum computer, the fidelity depends strongly on the quantum circuit considered. For the most difficult possible circuit for this technique, the so-called "quantum supremacy" benchmark of Google LLC [Arute et al., Nature 574, 505 (2019)], we find that the DMRG algorithm can generate bit strings of the same quality as the seminal Google experiment on a single computing core. For a more structured circuit used for combinatorial optimization (quantum approximate optimization algorithm), we find a drastic improvement of the DMRG results with error rates dropping by a factor of 100 compared with random quantum circuits. Our results suggest that the current bottleneck of quantum computers is their fidelities rather than the number of qubits.

A major challenge in ART is to select high-quality oocytes and embryos. The metabolism of oocytes and embryos has long been linked to their viability, suggesting the potential utility of metabolic measurements to aid in selection. Here, we review recent work on noninvasive metabolic imaging of cumulus cells, oocytes, and embryos. We focus our discussion on fluorescence lifetime imaging microscopy (FLIM) of the autofluorescent coenzymes NAD(P)H and flavine adenine dinucleotide (FAD+), which play central roles in many metabolic pathways. FLIM measurements provide quantitative information on NAD(P)H and FAD+ concentrations and engagement with enzymes, leading to a robust means of characterizing the metabolic state of cells. We argue that FLIM is a promising approach to aid in oocyte and embryo selection.

Systems neuroscientists seek a mechanistic and computational understanding of neural activity. In the visual system, the jump from sub-cortical to cortical brings about many important changes in representation and circuitry. By focusing on the important transformation of spatial information between thalamus and cortex, this work provides a better understanding of the computations that underlie visual processing. We performed a series of experiments in anesthetized primates, recording from individual neurons in the lateral geniculate nucleus (LGN) of the thalamus and the primary visual cortex (V1) using single grating stimuli and mixtures of gratings. To make sense of these different recordings, we fit the measured responses to both mechanistic and more computational models.
In the first chapter, we bring together previous accounts - in both the LGN and in V1 - of shifts in spatial frequency tuning with image contrast. We use a common stimulus set comprised of sinusoidal gratings that vary in spatial frequency and contrast. Fitting canonical, mechanistic models which capture our understanding of each area's receptive field structure, we show that the tuning shifts in V1 are larger than those in the LGN. This result suggests that shifts in LGN selectivity are inherited in V1, but further intracortical processing contributes to the more pronounced tuning shifts.

In the second and third chapters, we turn our focus to stimuli of intermediate complexity. We used superimposed mixtures of gratings as well as a more direct masking experiment to measure the tuning of spatial frequency suppression. In the second chapter, we report stronger spatial frequency-dependent suppression in V1 than in LGN, and find that suppression is typically strongest for frequencies at or below the cell's preference. These stimulus sets were also designed to evoke a broad range of responses which help constrain our computational model of spatial frequency selectivity. In the third chapter, we fit this model to the observed neuronal responses. The model implements divisive normalization, a canonical computation in cortical processing that accounts for a wide variety of observed neural activity. In the standard model of normalization, the response of a given neuron is normalized by the activity of nearby neurons that are selective across a wide range of stimulus values and features. We show that introducing a spatial-frequency tuned weighting of the normalization signal can preserve gain control while also better accounting for shifts in spatial frequency tuning and the observed suppression to complex stimuli. The tuning of the normalization was typically found to be stronger for frequencies below the cell’s peak, highlighting the role of low frequency suppression in shaping selectivity.

This paper presents a simple formalism for calculating X-ray absorption (XAS) and resonant inelastic x-ray scattering (RIXS) that has as input orbital-resolved density of states from a single-particle or many-body

Thin films of the solid solution NdLaNiO3 are grown in order to study the expected 0 K phase transitions at a specific composition. We experimentally map out the structural, electronic and magnetic properties as a function of x and a discontinuous, possibly first order, insulator–metal transition is observed at low temperature when x = 0.2. Raman spectroscopy and scanning transmission electron microscopy show that this is not associated with a correspondingly discontinuous global structural change. On the other hand, results from density functional theory (DFT) and combined DFT and dynamical mean field theory calculations produce a 0 K first order transition at around this composition. We further estimate the temperature-dependence of the transition from thermodynamic considerations and find that a discontinuous insulator–metal transition can be reproduced theoretically and implies a narrow insulator–metal phase coexistence with x. Finally, muon spin rotation (µSR) measurements suggest that there are non-static magnetic moments in the system that may be understood in the context of the first order nature of the 0 K transition and its associated phase coexistence regime.