arXiv subjects

Tianyu Zhu

Publications and source records attributed to Tianyu Zhu.

47 records · Page 3Linked to original sources

arXiv2022

Periodic Coupled-Cluster Green's Function for Photoemission Spectra of Realistic Solids

We present an efficient implementation of coupled-cluster Green's function (CCGF) method for simulating photoemission spectra of periodic systems. We formulate the periodic CCGF approach with Brillouin zone sampling in Gaussian basis at the coupled-cluster singles and doubles (CCSD) level. To enable CCGF calculations of realistic solids, we propose an active-space self-energy correction scheme by combining CCGF with cheaper many-body perturbation theory (GW) and implement the model order reduction (MOR) frequency interpolation technique. We find that the active-space self-energy correction and MOR techniques significantly reduce the computational cost of CCGF while maintaining the high accuracy. We apply the developed CCGF approaches to compute spectral properties and band structure of silicon (Si) and zinc oxide (ZnO) crystals using triple-$ζ$ Gaussian basis and medium-size k-point sampling, and find good agreement with experimental measurements.

physics.chem-ph↗

arXiv2022

Learning Instance and Task-Aware Dynamic Kernels for Few Shot Learning

Learning and generalizing to novel concepts with few samples (Few-Shot Learning) is still an essential challenge to real-world applications. A principle way of achieving few-shot learning is to realize a model that can rapidly adapt to the context of a given task. Dynamic networks have been shown capable of learning content-adaptive parameters efficiently, making them suitable for few-shot learning. In this paper, we propose to learn the dynamic kernels of a convolution network as a function of the task at hand, enabling faster generalization. To this end, we obtain our dynamic kernels based on the entire task and each sample and develop a mechanism further conditioning on each individual channel and position independently. This results in dynamic kernels that simultaneously attend to the global information whilst also considering minuscule details available. We empirically show that our model improves performance on few-shot classification and detection tasks, achieving a tangible improvement over several baseline models. This includes state-of-the-art results on 4 few-shot classification benchmarks: mini-ImageNet, tiered-ImageNet, CUB and FC100 and competitive results on a few-shot detection dataset: MS COCO-PASCAL-VOC.

cs.CV↗

arXiv2021

Learning Online for Unified Segmentation and Tracking Models

Tracking requires building a discriminative model for the target in the inference stage. An effective way to achieve this is online learning, which can comfortably outperform models that are only trained offline. Recent research shows that visual tracking benefits significantly from the unification of visual tracking and segmentation due to its pixel-level discrimination. However, it imposes a great challenge to perform online learning for such a unified model. A segmentation model cannot easily learn from prior information given in the visual tracking scenario. In this paper, we propose TrackMLP: a novel meta-learning method optimized to learn from only partial information to resolve the imposed challenge. Our model is capable of extensively exploiting limited prior information hence possesses much stronger target-background discriminability than other online learning methods. Empirically, we show that our model achieves state-of-the-art performance and tangible improvement over competing models. Our model achieves improved average overlaps of66.0%,67.1%, and68.5% in VOT2019, VOT2018, and VOT2016 datasets, which are 6.4%,7.3%, and6.4% higher than our baseline. Code will be made publicly available.

cs.CV↗

arXiv2021

Learn to Predict Sets Using Feed-Forward Neural Networks

This paper addresses the task of set prediction using deep feed-forward neural networks. A set is a collection of elements which is invariant under permutation and the size of a set is not fixed in advance. Many real-world problems, such as image tagging and object detection, have outputs that are naturally expressed as sets of entities. This creates a challenge for traditional deep neural networks which naturally deal with structured outputs such as vectors, matrices or tensors. We present a novel approach for learning to predict sets with unknown permutation and cardinality using deep neural networks. In our formulation we define a likelihood for a set distribution represented by a) two discrete distributions defining the set cardinally and permutation variables, and b) a joint distribution over set elements with a fixed cardinality. Depending on the problem under consideration, we define different training models for set prediction using deep neural networks. We demonstrate the validity of our set formulations on relevant vision problems such as: 1) multi-label image classification where we outperform the other competing methods on the PASCAL VOC and MS COCO datasets, 2) object detection, for which our formulation outperforms popular state-of-the-art detectors, and 3) a complex CAPTCHA test, where we observe that, surprisingly, our set-based network acquired the ability of mimicking arithmetics without any rules being coded.

cs.CV↗

arXiv2020

All-electron Gaussian-based $G_0W_0$ for Valence and Core Excitation Energies of Periodic Systems

We describe an all-electron $G_0W_0$ implementation for periodic systems with $k$-point sampling implemented in a crystalline Gaussian basis. Our full-frequency $G_0W_0$ method relies on efficient Gaussian density fitting integrals and includes both analytic continuation and contour deformation schemes. Due to the compactness of Gaussian bases, no virtual state truncation is required as is seen in many plane-wave formulations. Finite size corrections are included by taking the $q \to 0$ limit of the Coulomb divergence. Using our implementation, we study quasiparticle energies and band structures across a range of systems including molecules, semiconductors, rare gas solids, and metals. We find that the $G_0W_0$ band gaps of traditional semiconductors converge rapidly with respect to the basis size, even for the conventionally challenging case of ZnO. Using correlation-consistent bases of polarized triple-zeta quality, we find the mean absolute relative error of the extrapolated $G_0W_0$@PBE band gaps to be only 5.2% when compared to experimental values. For core excitation binding energies (CEBEs), we find that $G_0W_0$ predictions improve significantly over those from DFT if the $G_0W_0$ calculations are started from hybrid functionals with a high percentage of exact exchange.

cond-mat.mtrl-sci↗

arXiv2020

Ab Initio Full Cell GW+DMFT for Correlated Materials

Quantitative prediction of electronic properties in correlated materials requires simulations without empirical truncations and parameters. We present a method to achieve this goal through a new ab initio formulation of dynamical mean-field theory (DMFT). Instead of using small impurities defined in a low-energy subspace, which require complicated downfolded interactions which are often approximated, we describe a full cell $GW$+DMFT approach, where the impurities comprise all atoms in a unit cell or supercell of the crystal. Our formulation results in large impurity problems, which we treat here with efficient quantum chemistry impurity solvers that work on the real-frequency axis, combined with a one-shot $G_0W_0$ treatment of long-range interactions. We apply our full cell approach to bulk Si, two antiferromagnetic correlated insulators NiO and $α$-Fe$_2$O$_3$, and the paramagnetic correlated metal SrMoO$_3$, with impurities containing up to 10 atoms and 124 orbitals. We find that spectral properties, magnetic moments, and two-particle spin correlation functions are obtained in good agreement with experiments. In addition, in the metal oxide insulators, the balanced treatment of correlations involving all orbitals in the cell leads to new insights into the orbital character around the insulating gap.

cond-mat.str-el↗

arXiv2020

Recent developments in the PySCF program package

PYSCF is a Python-based general-purpose electronic structure platform that both supports first-principles simulations of molecules and solids, as well as accelerates the development of new methodology and complex computational workflows. The present paper explains the design and philosophy behind PYSCF that enables it to meet these twin objectives. With several case studies, we show how users can easily implement their own methods using PYSCF as a development environment. We then summarize the capabilities of PYSCF for molecular and solid-state simulations. Finally, we describe the growing ecosystem of projects that use PYSCF across the domains of quantum chemistry, materials science, machine learning and quantum information science.

physics.chem-ph↗

arXiv2020

Efficient Implementation of Ab Initio Quantum Embedding in Periodic Systems: Density Matrix Embedding Theory

We describe an efficient quantum embedding framework for realistic ab initio density matrix embedding (DMET) calculations in solids. We discuss in detail the choice of orbitals and mapping to a lattice, treatment of the virtual space and bath truncation, and the lattice-to embedded integral transformation. We apply DMET in this ab initio framework to a hexagonal boron nitride monolayer, crystalline silicon, and nickel monoxide in the antiferromagnetic phase, using large embedded clusters with up to 300 embedding orbitals. We demonstrate our formulation of ab initio DMET in the computation of ground-state properties such as the total energy, equation of state, magnetic moment and correlation functions.

cond-mat.str-el↗

arXiv2019

Efficient Formulation of Ab Initio Quantum Embedding in Periodic Systems: Dynamical Mean-Field Theory

We present an efficient ab initio dynamical mean-field theory (DMFT) implementation for quantitative simulations in solids. Our DMFT scheme employs ab initio Hamiltonians defined for impurities comprising the full unit cell or a supercell of atoms and for realistic quantum chemical basis sets. We avoid double counting errors by using Hartree-Fock as the low-level theory. Intrinsic and projected atomic orbitals (IAO+PAO) are chosen as the local embedding basis, facilitating numerical bath truncation. Using an efficient integral transformation and coupled-cluster Green's function (CCGF) impurity solvers, we are able to handle embedded impurity problems with several hundred orbitals. We apply our ab initio DMFT approach to study a hexagonal boron nitride monolayer, crystalline silicon, and nickel oxide in the antiferromagnetic phase, with up to 104 and 78 impurity orbitals in spin-restricted and unrestricted cluster DMFT calculations and over 100 bath orbitals. We show that our scheme produces accurate spectral functions compared to both benchmark periodic coupled-cluster computations and experimental spectra.

cond-mat.str-el↗

arXiv2019

Coupled-cluster impurity solvers for dynamical mean-field theory

We describe the use of coupled-cluster theory as an impurity solver in dynamical mean-field theory (DMFT) and its cluster extensions. We present numerical results at the level of coupled-cluster theory with single and double excitations (CCSD) for the density of states and self-energies of cluster impurity problems in the one- and two-dimensional Hubbard models. Comparison to exact diagonalization shows that CCSD produces accurate density of states and self-energies at a variety of values of $U/t$ and filling fractions. However, the low cost allows for the use of many bath sites, which we define by a discretization of the hybridization directly on the real frequency axis. We observe convergence of dynamical quantities using approximately 30 bath sites per impurity site, with our largest 4-site cluster DMFT calculation using 120 bath sites. We suggest coupled cluster impurity solvers will be attractive in ab initio formulations of dynamical mean-field theory.

cond-mat.str-el↗

arXiv2018

Simultaneous measurement of the light and charge response of liquid xenon to low-energy nuclear recoils at multiple electric fields

Dual-phase liquid xenon (LXe) detectors lead the direct search for particle dark matter. Understanding the signal production process of nuclear recoils in LXe is essential for the interpretation of LXe based dark matter searches. Up to now, only two experiments have simultaneously measured both the light and charge yield at different electric fields, neither of which attempted to evaluate the processes leading to light and charge production. In this letter, results from a neutron calibration of liquid xenon with simultaneous light and charge detection are presented for energies from 3-74 keV, at electric fields of 0.19, 0.49, and 1.02 kV/cm. No significant field dependence of the yields is observed.

physics.ins-det↗