Publications

The observation of gravitational-waves from merging supermassive black holes will be transformative: the detection of a low-frequency gravitational-wave background can tell us if and how supermassive black holes merge, inform our knowledge of galaxy merger rates and supermassive black hole masses, and enable the possibility of detecting new physics at nanohertz frequencies. All we have to do is time pulsars.

We study the chaotic-like behavior of cosmological simulations by quantifying how minute perturbations grow over time and manifest as macroscopic differences in galaxy properties. When we run pairs of 'shadow' simulations that are identical except for random minute initial displacements to particle positions (e.g. of order 1e-7pc), the results diverge from each other at the individual galaxy level (while the statistical properties of the ensemble of galaxies are unchanged). After cosmological times, the global properties of pairs of 'shadow' galaxies that are matched between the simulations differ from each other generally at a level of ~2-25%, depending on the considered physical quantity. We perform these experiments using cosmological volumes of (25-50Mpc/h)^3 evolved either purely with dark matter, or with baryons and star-formation but no feedback, or using the full feedback model of the IllustrisTNG project. The runs cover four resolution levels spanning a factor of 512 in mass. We find that without feedback the differences between shadow galaxies generally become smaller as the resolution increases, but with the IllustrisTNG model the results are mostly converging towards a 'floor'. This hints at the role of feedback in setting the chaotic properties of galaxy formation. Importantly, we compare the macroscopic differences between shadow galaxies to the overall scatter in various galaxy scaling relations, and conclude that for the star formation-mass and the Tully-Fisher relations the butterfly effect in our simulations contributes significantly to the overall scatter. We find that our results are robust to whether random numbers are used in the sub-grid models or not. We discuss the implications for galaxy formation theory in general and for cosmological simulations in particular.

Motivated by recent experiments, we use the +U extension of the generalized gradient approximation to density functional theory to study superlattices composed of alternating layers of LaNiO3 and LaMnO3. For comparison we also study a rocksalt [(111) double perovskite] structure and bulk LaNiO3 and LaMnO3. A Wannier function analysis indicates that band parameters are transferable from bulk to superlattice situations with the exception of the transition-metal d-level energy, which has a contribution from the change in d-shell occupancy. The charge transfer from Mn to Ni is found to be moderate in the superlattice, indicating metallic behavior, in contrast to the insulating behavior found in recent experiments, while the rocksalt structure is found to be insulating with a large Mn-Ni charge transfer. We suggest a high density of cation antisite defects may account for the insulating behavior experimentally observed in short-period superlattices.

There is currently great interest in the strong coupling between the quantized photon field of a cavity and electronic or other degrees of freedom in materials. A major goal is the creation of novel collective states entangling photons with those degrees of freedom. Here we show that the cooperative effect between strong electron correlations in quantum materials and the long-range interactions induced by the photon field leads to the stabilization of coherent phases of light and matter. By studying a two-band model of interacting electrons coupled to a cavity field, we show that a phase characterized by the simultaneous condensation of excitons and photon superradiance can be realized, hence stabilizing and intertwining two collective phenomena which are rather elusive in the absence of this cooperative effect.

We describe an approach for many-body calculations with a finite-temperature, grand canonical ensemble formalism using auxiliary-field quantum Monte Carlo (AFQMC) with a self-consistent constraint to control the sign problem. The usual AFQMC formalism of Blankenbecler, Scalapino, and Sugar suffers from the sign problem with most physical Hamiltonians, as is well known. Building on earlier ideas to constrain the paths in auxiliary-field space [Phys. Rev. Lett. \textbf{83}, 2777 (1999)] and incorporating recent developments in zero-temperature, canonical-ensemble methods, we discuss how a self-consistent constraint can be introduced in the finite-temperature, grand-canonical-ensemble framework. This together with several other algorithmic improvements discussed here leads to a more accurate, more efficient, and numerically more stable approach for finite-temperature calculations. We carry out a systematic benchmark study in the two-dimensional repulsive Hubbard model at 1/8 doping. Temperatures as low as T=1/80 (in units of hopping) are reached. The finite-temperature method is exact at very high temperatures, and approaches the result of the zero-temperature constrained-path AFQMC as temperature is lowered. The benchmark shows that systematically accurate results are obtained for thermodynamic properties.

Extreme-precision radial velocity (EPRV) exoplanet surveys face considerable data analysis challenges in extracting maximally precise RVs from spectra. Chief among these, particularly for the upcoming generation of red-optimized spectrographs targeting M dwarfs, is the presence of telluric absorption features which are not perfectly known. Another major limitation on the achievable RV precision is the need to adopt an imperfect stellar template against which to cross-correlate or otherwise match the observations. In both cases, precision-limiting reliance on external information can be sidestepped using the data directly. Here we propose a data-driven method to simultaneously extract precise RVs and infer the underlying stellar and telluric spectra using a linear model (in the log of flux). The model employs a convex objective and convex regularization to keep the optimization of the spectral components fast. We implement this method in wobble, an open-source python package which uses TensorFlow in one of its first non-neural-network applications to astronomical data. In this work, we demonstrate the performance of wobble on archival HARPS spectra. We recover the canonical exoplanet 51 Pegasi b, detect the secular RV evolution of the M dwarf Barnard's Star, and retrieve the Rossiter-McLaughlin effect for the Hot Jupiter HD 189733b. The method additionally produces extremely high-S/N composite stellar spectra and detailed time-variable telluric spectra, which we also present here.

The dynamics and deformations of immersed flexible fibers are at the heart of important industrial and biological processes, induce peculiar mechanical and transport properties in the fluids that contain them, and are the basis for novel methods of flow control. Here we focus on the low–Reynolds number regime where advances in studying these fiber–fluid systems have been especially rapid. On the experimental side, this is due to new methods of fiber synthesis, microfluidic flow control, and microscope-based tracking measurement techniques. Likewise, there have been continuous improvements in the specialized mathematical modeling and numerical methods needed to capture the interactions of slender flexible fibers with flows, boundaries, and each other.

Current mechanisms for knowledge transfer in deep networks tend to either share the lower layers between tasks, or build upon representations trained on other tasks. However, existing work in non-deep multi-task and lifelong learning has shown success with using factorized representations of the model parameter space for transfer, permitting more flexible construction of task models. Inspired by this idea, we introduce a novel architecture for sharing latent factorized representations in convolutional neural networks (CNNs). The proposed approach, called a deconvolutional factorized CNN, uses a combination of deconvolutional factorization and tensor contraction to perform flexible transfer between tasks. Experiments on two computer vision data sets show that the DF-CNN achieves superior performance in challenging lifelong learning settings, resists catastrophic forgetting, and exhibits reverse transfer to improve previously learned tasks from subsequent experience without retraining.

We express the recently introduced real-time diagrammatic Quantum Monte Carlo, Phys. Rev. B 91, 245154 (2015), in the Larkin-Ovchinnikov basis in Keldysh space. Based on a perturbation expansion in the local interaction U, the special form of the interaction vertex allows to write diagrammatic rules in which vacuum Feynman diagrams directly vanish. This reproduces the main property of the previous algorithm, without the cost of the exponential sum over Keldysh indices. In an importance sampling procedure, this implies that only interaction times in the vicinity of the measurement time contribute. Such an algorithm can then directly address the long-time limit needed in the study of steady states in out-of-equilibrium systems. We then implement and discuss different variants of Monte Carlo algorithms in the Larkin-Ovchinnikov basis. A sign problem reappears, showing that the cancellation of vacuum diagrams has no direct impact on it.

The GW approximation is based on the neglect of vertex corrections, which appear in the exact self-energy and the exact polarizability. Here, we investigate the importance of vertex corrections in the polarizability only. We calculate the polarizability with equation-of-motion coupled-cluster theory with single and double excitations (EOM-CCSD), which rigorously includes a large class of diagrammatically-defined vertex corrections beyond the random phase approximation (RPA). As is well-known, the frequency-dependent polarizability predicted by EOM-CCSD is quite different and generally more accurate than that predicted by the RPA. We evaluate the effect of these vertex corrections on a test set of 20 atoms and molecules. When using a Hartree-Fock reference, ionization potentials predicted by the GW approximation with the RPA polarizability are typically overestimated with a mean absolute error of 0.3 eV. However, those predicted with a vertex-corrected polarizability are typically underestimated with an increased mean absolute error of 0.5 eV. This result suggests that vertex corrections in the self-energy cannot be neglected, at least for molecules. We also assess the behavior of eigenvalue self-consistency in vertex-corrected GW calculations, finding a further worsening of the predicted ionization potentials.