The talk will discuss the mean value theorem and Wooley's breakthrough with his "efficent congruencing" method.
Many numerical solvers of ordinary differential equations require problems to be posed as a system of first order differential equations. This means that if one wishes to solve higher order problems, the system have to be rewritten, which is a cumbersome and error-prone process. This talk presents a technique for automatically doing such reformulations.
Choreographies are periodic solutions of the n-body problem in which all of the bodies have unit masses, share a common orbit and are uniformly spread along it. In this talk, I will present an algorithm for numerical computation and stability analysis of choreographies. It is based on approximations by trigonometric polynomials, minimization of the action functional using a closed-form expression of the gradient, quasi-Newton methods, automatic differentiation and Floquet stability analysis.
A year ago I gave a talk raising questions about Faraday shielding which stimulated discussion with John Ockendon and others and led to a collaboration with Jon Chapman and Dave Hewett. The problem is one of harmonic functions subject to constant-potential boundary conditions. A year later, we are happy with the solution we have found, and the paper will appear in SIAM Review. Though many assume as we originally did that Faraday shielding must be exponentially effective, and Feynman even argues this explicitly in his Lectures, we have found that in fact, the shielding is only linear. Along the way to explaining this we make use of Mikhlin's numerical method of series expansion, homogenization by multiple scales analysis, conformal mapping, a phase transition, Brownian motion, some ideas recollected from high school about electrostatic induction, and a constrained quadratic optimization problem solvable via a block 2x2 KKT matrix.
All-at-once schemes aim to solve all time-steps of parabolic PDE-constrained optimization problems in one coupled computation, leading to exceedingly large linear systems requiring efficient iterative methods. We present a new block diagonal preconditioner which is both optimal with respect to the mesh parameter and parallelizable over time, thus can provide significant speed- up. We will present numerical results to demonstrate the effectiveness of this preconditioner.
There is a beautiful problem resulting from arithmetic number theory where a continuous and compactly supported function's 3-fold autoconvolution is constant. In this talk, we optimize the coefficients of a Chebyshev series multiplied by an endpoint singularity to obtain a highly accurate approximation to this constant. Convolving functions with endpoint singularities turns out to be a challenge for standard quadrature routines. However, variable transformations inducing double exponential endpoint decay are used to effectively annihilate the singularities in a way that keeps accuracy high and complexity low.
Flow thought a porous media is usually described by assuming the superficial velocity can be expressed in terms of a constant permeability and a pressure gradient. In poroelastic flows the underlying elastic matrix responds to changes in the fluid pressure. When the elastic deformation is allowed to influence the permeability through the elastic strain, it becomes possible for increased fluid pressure gradient not to result in increased flow, but to decrease the permeability and potentially this may close off or choke the flow. I will talk about a simple model problem for a number of different elastic constitutive models and a number of different permeability-strain models and examine whether there is a general criterion that can be derived to show when, or indeed if, choking can occur for different elasticity-permeability combinations.
We present an algorithm, Parallel-$\ell_0$, for {\em combinatorial compressed sensing} (CCS), where the sensing matrix $A \in \mathbb{R}^{m\times n}$ is the adjacency matrix of an expander graph. The information preserving nature of expander graphs allow the proposed algorithm to provably recover a $k$-sparse vector $x\in\mathbb{R}^n$ from $m = \mathcal{O}(k \log (n/k))$ measurements $y = Ax$ via $\mathcal{O}(\log k)$ simple and parallelizable iterations when the non-zeros in the support of the signal form a dissociated set, meaning that all of the $2^k$ subset sums of the support of $x$ are pairwise different. In addition to the low computational cost, Parallel-$\ell_0$ is observed to be able to recover vectors with $k$ substantially larger than previous CCS algorithms, and even higher than $\ell_1$-regularization when the number of measurements is significantly smaller than the vector length.
This talk concerns the numerical solution of elliptic partial differential equations posed on general smooth surfaces by the Closest Point Method. Based on the closest point representation of the surface, we formulate an embedding equation in a narrow band surrounding the surface, then discretize it using standard finite differences and interpolation schemes. Numerical convergence of the method will be discussed. In order to solve the resulting large sparse linear systems, we propose a specific geometric multigrid method which makes use of the closest point representation of the surface.