arXiv subjects

Weinan E

Publications and source records attributed to Weinan E.

At least 163 records · Page 9Linked to original sources

arXiv2015

Functional Frank-Wolfe Boosting for General Loss Functions

Boosting is a generic learning method for classification and regression. Yet, as the number of base hypotheses becomes larger, boosting can lead to a deterioration of test performance. Overfitting is an important and ubiquitous phenomenon, especially in regression settings. To avoid overfitting, we consider using $l_1$ regularization. We propose a novel Frank-Wolfe type boosting algorithm (FWBoost) applied to general loss functions. By using exponential loss, the FWBoost algorithm can be rewritten as a variant of AdaBoost for binary classification. FWBoost algorithms have exactly the same form as existing boosting methods, in terms of making calls to a base learning algorithm with different weights update. This direct connection between boosting and Frank-Wolfe yields a new algorithm that is as practical as existing boosting methods but with new guarantees and rates of convergence. Experimental results show that the test performance of FWBoost is not degraded with larger rounds in boosting, which is consistent with the theoretical analysis.

stat.ML↗

arXiv2015

Multiscale Adaptive Representation of Signals: I. The Basic Framework

We introduce a framework for designing multi-scale, adaptive, shift-invariant frames and bi-frames for representing signals. The new framework, called AdaFrame, improves over dictionary learning-based techniques in terms of computational efficiency at inference time. It improves classical multi-scale basis such as wavelet frames in terms of coding efficiency. It provides an attractive alternative to dictionary learning-based techniques for low level signal processing tasks, such as compression and denoising, as well as high level tasks, such as feature extraction for object recognition. Connections with deep convolutional networks are also discussed. In particular, the proposed framework reveals a drawback in the commonly used approach for visualizing the activations of the intermediate layers in convolutional networks, and suggests a natural alternative.

cs.CV↗

arXiv2014

On the distinguished limits of the Navier slip model of the moving contact line problem

When a droplet spreads on a solid substrate, it is unclear what are the correct boundary conditions to impose at the moving contact line. The classical no-slip condition is generally acknowledged to lead to a non-integrable singularity at the moving contact line, for which a slip condition, associated with a small slip parameter, $λ$, serves to alleviate. In this paper, we discuss what occurs as the slip parameter, $λ$, tends to zero. In particular, we explain how the zero-slip limit should be discussed in consideration of two distinguished:: one where time is held constant $t = O(1)$, and one where time tends to infinity at the rate $t = O(|\log λ|)$. The crucial result is that in the case where time is held constant, the $λ\to 0$ limit converges to the slip-free equation, and contact line slippage occurs as a regular perturbative effect. However, if $λ\to 0$ and $t \to \infty$, then contact line slippage is a leading-order singular effect.

physics.flu-dyn↗

arXiv2013

Exact renormalization group analysis of turbulent transport by the shear flow

The exact renormalization group (RG) method initiated by Wilson and further developed by Polchinski is used to study the shear flow model proposed by Avellaneda and Majda as a simplified model for the diffusive transport of a passive scalar by a turbulent velocity field. It is shown that this exact RG method is capable of recovering all the scaling regimes as the spectral parameters of velocity statistics vary, found by Avellaneda and Majda in their rigorous study of this model. This gives further confidence that the RG method, if implemented in the right way instead of using drastic truncations as in the Yakhot-Orszag's approximate RG scheme, does give the correct prediction for the large scale behaviors of solutions of stochastic partial differential equations (PDE). We also derive the analog of the "large eddy simulation" models when a finite amount of small scales are eliminated from the problem.

math-ph↗

arXiv2013

Efficient iterative method for solving the Dirac-Kohn-Sham density functional theory

We present for the first time an efficient iterative method to directly solve the four-component Dirac-Kohn-Sham (DKS) density functional theory. Due to the existence of the negative energy continuum in the DKS operator, the existing iterative techniques for solving the Kohn-Sham systems cannot be efficiently applied to solve the DKS systems. The key component of our method is a novel filtering step (F) which acts as a preconditioner in the framework of the locally optimal block preconditioned conjugate gradient (LOBPCG) method. The resulting method, dubbed the LOBPCG-F method, is able to compute the desired eigenvalues and eigenvectors in the positive energy band without computing any state in the negative energy band. The LOBPCG-F method introduces mild extra cost compared to the standard LOBPCG method and can be easily implemented. We demonstrate our method in the pseudopotential framework with a planewave basis set which naturally satisfies the kinetic balance prescription. Numerical results for Pt$_{2}$, Au$_{2}$, TlF, and Bi$_{2}$Se$_{3}$ indicate that the LOBPCG-F method is a robust and efficient method for investigating the relativistic effect in systems containing heavy elements.

physics.comp-ph↗

arXiv2012

Modified string method for finding minimum energy path

We present an efficient algorithm for calculating the minimum energy path (MEP) and energy barriers between local minima on a multidimensional potential energy surface (PES). Such paths play a central role in the understanding of transition pathways between metastable states. Our method relies on the original formulation of the string method [Phys. Rev. B ${\bf 66}$, 052301 (2002)], i.e. to evolve a smooth curve along a direction normal to the curve. The algorithm works by performing minimization steps on hyperplanes normal to the curve. Therefore the problem of finding MEP on the PES is remodeled as a set of constrained minimization problems. This provides the flexibility of using minimization algorithms faster than the steepest descent method used in the simplified string method [J. Chem. Phys., ${\bf 126}$(16),164103 (2007)]. At the same time, it provides a more direct analog of the finite temperature string method. The applicability of the algorithm is demonstrated using various examples.

physics.comp-ph↗

arXiv2011

Atomistic simulations of rare events using gentlest ascent dynamics

The dynamics of complex systems often involve thermally activated barrier crossing events that allow these systems to move from one basin of attraction on the high dimensional energy surface to another. Such events are ubiquitous, but challenging to simulate using conventional simulation tools, such as molecular dynamics. Recently, Weinan E et al. [Nonlinearity, 24(6),1831(2011)] proposed a set of dynamic equations, the gentlest ascent dynamics (GAD), to describe the escape of a system from a basin of attraction and proved that solutions of GAD converge to index-1 saddle points of the underlying energy. In this paper, we extend GAD to enable finite temperature simulations in which the system hops between different saddle points on the energy surface. An effective strategy to use GAD to sample an ensemble of low barrier saddle points located in the vicinity of a locally stable configuration on the high dimensional energy surface is proposed. The utility of the method is demonstrated by studying the low barrier saddle points associated with point defect activity on a surface. This is done for two representative systems, namely, (a) a surface vacancy and ad-atom pair and (b) a heptamer island on the (111) surface of copper.

cond-mat.mtrl-sci↗

arXiv2011

Cauchy-Born rule and spin density wave for the spin-polarized Thomas-Fermi-Dirac-von Weizsacker model

The electronic structure (electron charges and spins) of a perfect crystal under external magnetic field is analyzed using the spin-polarized Thomas-Fermi-Dirac-von Weizsacker model. An extension of the classical Cauchy-Born rule for crystal lattices is established for the electronic structure under sharp stability conditions on charge density wave and spin density wave. A Landau-Lifschitz type micromagnetic energy functional is derived.

math-ph↗

arXiv2011

The Gentlest Ascent Dynamics

Dynamical systems that describe the escape from the basins of attraction of stable invariant sets are presented and analyzed. It is shown that the stable fixed points of such dynamical systems are the index-1 saddle points. Generalizations to high index saddle points are discussed. Both gradient and non-gradient systems are considered. Preliminary results on the nature of the dynamical behavior are presented.

math.DS↗

arXiv2010

Effective Maxwell equations from time-dependent density functional theory

The behavior of interacting electrons in a perfect crystal under macroscopic external electric and magnetic fields is studied. Effective Maxwell equations for the macroscopic electric and magnetic fields are derived starting from time-dependent density functional theory. Effective permittivity and permeability coefficients are obtained.

math-ph↗

arXiv2009

Density-gradient-corrected embedded atom method

Through detailed comparisons between Embedded Atom Method (EAM) and first-principles calculations for Al, we find that EAM tends to fail when there are large electron density gradients present. We attribute the observed failures to the violation of the uniform density approximation (UDA) underlying EAM. To remedy the insufficiency of UDA, we propose a gradient-corrected EAM model which introduces gradient corrections to the embedding function in terms of exchange-correlation and kinetic energies. Based on the perturbation theory of "quasiatoms" and density functional theory, the new embedding function captures the essential physics missing in UDA, and paves the way for developing more transferable EAM potentials. With Voter-Chen EAM potential as an example, we show that the gradient corrections can significantly improve the transferability of the potential.

cond-mat.mes-hall↗

arXiv2008

Multipole Representation of the Fermi Operator with Application to the Electronic Structure Analysis of Metallic Systems

We propose a multipole representation of the Fermi-Dirac function and the Fermi operator, and use this representation to develop algorithms for electronic structure analysis of metallic systems. The new algorithm is quite simple and efficient. Its computational cost scales logarithmically with $βΔ\eps$ where $β$ is the inverse temperature, and $Δ\eps$ is the width of the spectrum of the discretized Hamiltonian matrix.

cond-mat.mtrl-sci↗

arXiv2008

Some Critical Issues for the "Equation-Free" Approach to Multiscale Modeling

The "equation-free'' approach has been proposed in recent years as a general framework for developing multiscale methods to efficiently capture the macroscale behavior of a system using only the microscale models. In this paper, we take a close look at some of the algorithms proposed under the "equation-free'' umbrella, the projective integrators and the patch dynamics. We discuss some very simple examples in the context of the "equation-free'' approach. These examples seem to indicate that while its general philosophy is quite attractive and indeed similar to many other approaches in concurrent multiscale modeling, there are severe limitations to the specific implementation proposed by the equation-free approach.

math.NA↗

arXiv2005

Numerical study of metastability due to tunneling: The quantum string method

We generalize the string method, originally designed for the study of thermally activated rare events, to the calculation of quantum tunneling rates. This generalization is based on the analogy between quantum mechanics and statistical mechanics in the path-integral formalism. The quantum string method first locates, in the space of imaginary-time trajectories, the minimal action path (MAP) between two minima of the imaginary-time action. From the MAP, the saddle-point (``bounce'') action associated with the exponential barrier penetration probability is obtained and the pre-exponential factor (the ratio of determinants) for the tunneling rate evaluated using stochastic simulation. The quantum string method is implemented to calculate the zero-temperature escape rates for the metastable zero-voltage states in the current-biased Josephson tunnel junction model. In the regime close to the critical bias current, direct comparison of the numerical and analytical results yields good agreement. Our calculations indicate that for the nanojunctions encountered in many experiments today, the (absolute) escape rates should be measurable at bias current much below the critical current.

cond-mat.stat-mech↗

arXiv2004

Multiscale simulations in simple metals: a density-functional based methodology

We present a formalism for coupling a density functional theory-based quantum simulation to a classical simulation for the treatment of simple metallic systems. The formalism is applicable to multiscale simulations in which the part of the system requiring quantum-mechanical treatment is spatially confined to a small region. Such situations often arise in physical systems where chemical interactions in a small region can affect the macroscopic mechanical properties of a metal. We describe how this coupled treatment can be accomplished efficiently, and we present a coupled simulation for a bulk aluminum system.

cond-mat.mtrl-sci↗

arXiv2003

Noise-driven transition to quasi-deterministic limit cycle dynamics in excitable systems

The effect of small-amplitude noise on excitable systems with large time-scale separation is analyzed. It is found that small random perturbations of the fast excitatory variable result in the onset of a quasi-deterministic limit cycle behavior, absent without noise. The limit cycle is established at a critical value of the amplitude of the noise, and its period is nontrivially determined by the relationship between the noise amplitude and the time scale ratio. It is argued that this effect might provide a mechanism by which the function of biological systems operating in noisy environments can be robustly controlled by the level of the noise.

q-bio.QM↗

arXiv2002

Energy landscapes and rare events

Many problems in physics, material sciences, chemistry and biology can be abstractly formulated as a system that navigates over a complex energy landscape of high or infinite dimensions. Well-known examples include phase transitions of condensed matter, conformational changes of biopolymers, and chemical reactions. The energy landscape typically exhibits multiscale features, giving rise to the multiscale nature of the dynamics. This is one of the main challenges that we face in computational science. In this report, we will review the recent work done by scientists from several disciplines on probing such energy landscapes. Of particular interest is the analysis and computation of transition pathways and transition rates between metastable states. We will then present the string method that has proven to be very effective for some truly complex systems in material science and chemistry.

math.NA↗

arXiv2002

The Heterogeneous Multi-Scale Method

The heterogeneous multi-scale method (HMM) is a general strategy for dealing with problems involving multi-scales, with multi-physics, using multi-grids. It not only unifies several existing multi-scale methods, but also provide a methodology for designing new algorithms for new applications. In this paper, we review the history of multi-scale modeling and simulation that led to the development of HMM, the methodology itself together with some applications, and the mathematical theory of stability and accuracy.

physics.comp-ph↗