We will first introduce basic ideas of Compressive Sensing (CS), which
is an emerging
(some would say revolutionary) methodology in signal, image and data
processing.
The theory for CS has so far been built largely on a notion called
Restricted Isometry
Property or RIP. We will point out drawbacks of RIP-based analyses and
introduce
results from a non-RIP analysis including some new extensions. We will
also discuss
some related optimization algorithms.
An old result of Kac-Hammersley says that the complex zeros of a Gaussian
random polynomial \sum_{j = 0}^N a_j z^j with i.i.d. normal coefficients a_j, concentrate
on the unit circle. This seems counter intuitive at first, since the zeros could be anywhere.
We will explain this paradox and show that there is a very general result that empirical measures
of complex zeros tend to `equilibrium measures'. We then give a large deviations principle showing
that the probability of deviation from equilibrium measure is exponentially small.
A random walk is called recurrent if it is sure to return to its starting point and transient otherwise. A famous result of Polya is that simple symmetric random walk on the integer lattice is recurrent in dimensions 1 and 2, and transient in higher dimensions. We study random walks that are small perturbations of simple random walk. Our main result is that if the dimension is high enough then these random walks are transient.
We describe some recent developments involving particle methods in
electrostatics and fluid dynamics.
First we describe a Cartesian treecode for charged particle systems
undergoing screened Coulomb interactions in three dimensions. The
treecode reduces the operation count for computing potentials and
forces from $O(N^2)$ to $O(N\log N)$, where $N$ is the number of
particles in the system. The algorithm is especially well suited for
particles distributed on a surface. This is a step towards a treecode-
accelerated boundary integral solver for implicit solvent models in
biomolecular dynamics. (joint work with Peijun Li, Hans Johnston, and
Weihua Geng)
Second we describe a Lagrangian panel method for computing vortex
sheet motion in three dimensional fluid flow. Here too a treecode is
used, to evaluate the regularized Biot-Savart integral for the
sheet's self-induced velocity. The method is applied to compute the
azimuthal instability of a vortex ring. Details of the core dynamics
are clarified by tracking material lines on the sheet surface.
Results show the collapse of the vortex core in each wavelength and
the radial ejection of ringlets. These events are correlated with
local axial flow in the core of the ring. (joint work with Hualong
Feng and Leon Kaganovskiy)
Various localized patterns have been observed both experimentally and numerically in many reaction-diffusion systems, including stable spots, traveling spots, breathing spot and splitting spot etc. A well-known example is the ferrocyanide-iodate-sulphite reaction that has reproduced similar spot-replication behavior shown by computing Gray-Scott model. Understanding these phenomena could have potential applications in chemical reactions, biological morphogenesis and medical research.
In this talk, I concentrate on analyzing dynamics and instabilities of spike patterns (1D) and spot patterns (2D) in Gray-Scott model. In a specific parameter regime of 1D problem, oscillatory profile and drift instabilities are analyzed through a Stefan type problem. In 2D model, we study the mechanism of spot-replication, for which the splitting criterion is identified. A DAE system is derived to describe the spot dynamics. Competition instability and oscillatory profile instability are also investigated. On top of these results, phase diagrams are plotted in different parameter spaces. The theories are illustrated for infinite domain, unit circle and square domain, and are compared with full numerical simulations.