arXiv subjects

Divakar Viswanath

Publications and source records attributed to Divakar Viswanath.

16 recordsLinked to original sources

arXiv2018

Prediction of Dynamical time Series Using Kernel Based Regression and Smooth Splines

Prediction of dynamical time series with additive noise using support vector machines or kernel based regression has been proved to be consistent for certain classes of discrete dynamical systems. Consistency implies that these methods are effective at computing the expected value of a point at a future time given the present coordinates. However, the present coordinates themselves are noisy, and therefore, these methods are not necessarily effective at removing noise. In this article, we consider denoising and prediction as separate problems for flows, as opposed to discrete time dynamical systems, and show that the use of smooth splines is more effective at removing noise. Combination of smooth splines and kernel based regression yields predictors that are more accurate on benchmarks typically by a factor of 2 or more. We prove that kernel based regression in combination with smooth splines converges to the exact predictor for time series extracted from any compact invariant set of any sufficiently smooth flow. As a consequence of convergence, one can find examples where the combination of kernel based regression with smooth splines is superior by even a factor of $100$. The predictors that we compute operate on delay coordinate data and not the full state vector, which is typically not observable.

stat.ML↗

arXiv2018

Delay Embedding of Periodic Orbits Using a Fixed Observation Function

Delay coordinates are a widely used technique to pass from observations of a dynamical system to a representation of the dynamical system as an embedding in Euclidean space. Current proofs show that delay coordinates of a given dynamical system result in embeddings generically over a space of observations (Sauer, Yorke, Casdagli, J. Stat. Phys., vol. 65 (1991), p. 579-616). Motivated by applications of the embedding theory, we consider the situation where the observation function is fixed. For example, the observation function may simply be some fixed coordinate of the state vector. For a fixed observation function (any nonzero linear combination of coordinates) and for the special case of periodic solutions, we prove that delay coordinates result in an embedding generically over the space of flows in the $C^{r}$ topology with $r\geq2$.

math.DS↗

arXiv2018

Prevalence of Delay Embeddings with a Fixed Observation Function

Let $x_{j+1}=ϕ(x_{j})$, $x_{j}\in\mathbb{R}^{d}$, be a dynamical system with $ϕ$ being a diffeomorphism. Although the state vector $x_{j}$ is often unobservable, the dynamics can be recovered from the delay vector $\left(o(x_{1}),\ldots,o(x_{D})\right)$, where $o$ is the scalar-valued observation function and $D$ is the embedding dimension. The delay map is an embedding for generic $o$, and more strongly, the embedding property is prevalent. We consider the situation where the observation function is fixed at $o=π_{1}$, with $π_{1}$ being the projection to the first coordinate. However, we allow polynomial perturbations to be applied directly to the diffeomorphism $ϕ$, thus mimicking the way dynamical systems are parametrized. We prove that the delay map is an embedding with probability one with respect to the perturbations. Our proof introduces a new technique for proving prevalence using the concept of Lebesgue points.

math.DS↗

arXiv2015

Navier-Stokes solver using Green's functions II: spectral integration of channel flow and plane Couette flow

The Kleiser-Schumann algorithm has been widely used for the direct numerical simulation of turbulence in rectangular geometries. At the heart of the algorithm is the solution of linear systems which are tridiagonal except for one row. This note shows how to solve the Kleiser-Schumann problem using perfectly triangular matrices. An advantage is the ability to use functions in the LAPACK library. The method is used to simulate turbulence in channel flow at $Re=80,000$ (and $Re_τ=2400$) using $10^{9}$ grid points. An assessment of the length of time necessary to eliminate transient effects in the initial state is included.

physics.flu-dyn↗

arXiv2015

Intermittency at Fine Scales and Complex Singularities of Turbulent Couette Flow

Fine scales of turbulent velocity fields, beyond the inertial range and well into the dissipative range, are highly intermittent. It has been hypothesized that complex plane singularities are the principal mechanism behind fine scale intermittency. In this article, we view the velocity field of a turbulent flow as an analytic function of time. Although the function is only available for real values of time, we present a numerical technique to analytically continue the function to complex values of time, and with sufficient fidelity to locate and visualize the singularity closest to the real axis. Using this technique, we demonstrate a robust connection between temporal intermittency and the location of singularities in the complex plane.

physics.flu-dyn↗

arXiv2015

Error Analysis of Finite Differences and the Mapping Parameter in Spectral Differentiation

The Chebyshev points are commonly used for spectral differentiation in non-periodic domains. The rounding error in the Chebyshev approximation to the $n$-the derivative increases at a rate greater than $n^{2m}$ for the $m$-th derivative. The mapping technique of Kosloff and Tal-Ezer (\emph{J. Comp. Physics}, vol. 104 (1993), p. 457-469) ameliorates this increase in rounding error. We show that the argument used to justify the choice of the mapping parameter is substantially incomplete. We analyze rounding error as well as discretization error and give a more complete argument for the choice of the mapping parameter. If the discrete cosine transform is used to compute derivatives, we show that a different choice of the mapping parameter yields greater accuracy.

math.NA↗

arXiv2014

Metric Entropy and the Optimal Prediction of Chaotic Signals

Suppose we are given a time series or a signal $x(t)$ for $0\leq t\leq T$. We consider the problem of predicting the signal in the interval $T<t\leq T+t_{f}$ from a knowledge of its history and nothing more. We ask the following question: what is the largest value of $t_{f}$ for which a prediction can be made? We show that the answer to this question is contained in a fundamental result of information theory due to Wyner, Ziv, Ornstein, and Weiss. In particular, for the class of chaotic signals, the upper bound is $t_{f}\leq\log_{2}T/H$ in the limit $T\rightarrow\infty$, with $H$ being entropy in a sense that is explained in the text. If $\bigl|x(T-s)-x(t^{\ast}-s)\bigr|$ is small for $0\leq s\leqτ$, where $τ$ is of the order of a characteristic time scale, the pattern of events leading up to $t=T$ is similar to the pattern of events leading up to $t=t^{\ast}$. It is reasonable to expect $x(t^{\ast}+t_{f})$ to be a good predictor of $x(T+t_{f}).$ All existing methods for prediction use this idea in some way or the other. Unfortunately, this intuitively reasonable idea is fundamentally deficient and all existing methods fall well short of the Wyner-Ziv entropy bound on $t_{f}$. An optimal predictor should decompose the distance between the pattern of events leading up to $t=T$ and the pattern leading up to $t=t^{\ast}$ into stable and unstable components. A good match should have suitably small unstable components but will in general allow stable components which are as large as the tolerance for correct prediction. For the special case of hyperbolic toral automorphisms, we derive an optimal predictor using Pade approximation.

nlin.CD↗

arXiv2014

Navier-Stokes solver using Green's functions I: channel flow and plane Couette flow

Numerical solvers of the incompressible Navier-Stokes equations have reproduced turbulence phenomena such as the law of the wall, the dependence of turbulence intensities on the Reynolds number, and experimentally observed properties of turbulence energy production. In this article, we begin a sequence of investigations whose eventual aim is to derive and implement numerical solvers that can reach higher Reynolds numbers than is currently possible. Every time step of a Navier-Stokes solver in effect solves a linear boundary value problem. The use of Green's functions leads to numerical solvers which are highly accurate in resolving the boundary layer, which is a source of delicate but exceedingly important physical effects at high Reynolds numbers. The use of Green's functions brings with it a need for careful quadrature rules and a reconsideration of time steppers. We derive and implement Green's function based solvers for the channel flow and plane Couette flow geometries. The solvers are validated by reproducing turbulent signals which are in good qualitative and quantitative agreement with experiment.

physics.flu-dyn↗

arXiv2014

Accuracy and stability of inversion of power series

This article considers the numerical inversion of the power series $p(x)=1+b_{1}x+b_{2}x^{2}+\cdots$ to compute the inverse series $q(x)$ satisfying $p(x)q(x)=1$. Numerical inversion is a special case of triangular back-substitution, which has been known for its beguiling numerical stability since the classic work of Wilkinson (1961). We prove the numerical stability of inversion of power series and obtain bounds on numerical error. A range of examples show these bounds to be quite good. When $p(x)$ is a polynomial and $x=a$ is a root with $p(a)=0$, we show that root deflation via the simple division $p(x)/(x-a)$ can trigger instabilities relevant to polynomial root finding and computation of finite-difference weights. When $p(x)$ is a polynomial, the accuracy of the computed inverse $q(x)$ is connected to the pseudozeros of $p(x)$.

math.NA↗

arXiv2012

Spectral integration of linear boundary value problems

Spectral integration was deployed by Orszag and co-workers (1977, 1980, 1981) to obtain stable and efficient solvers for the incompressible Navier-Stokes equation in rectangular geometries. Two methods in current use for channel flow and plane Couette flow, namely, Kleiser-Schumann (1980) and Kim-Moin-Moser (1977), rely on the same technique. In its current form, the technique of spectral integration, as applied to the Navier-Stokes equations, is dominated by rounding errors at higher Reynolds numbers which would otherwise be within reach. In this article, we derive a number of versions of spectral integration and explicate their properties, with a view to extending the Kleiser-Schumann and Kim-Moin-Moser algorithms to higher Reynolds numbers. More specifically, we show how spectral integration matrices that are banded, but bordered by dense rows, can be reduced to purely banded matrices. Key properties, such as the accuracy of spectral integration even when Green's functions are not resolved by the underlying grid, the accuracy of spectral integration in spite of ill-conditioning of underlying linear systems, and the accuracy of derivatives, are thoroughly explained.

math.NA↗

arXiv2011

Barycentric Hermite Interpolation

Let $z_{1},\ldots,z_{K}$ be distinct grid points. If $f_{k,0}$ is the prescribed value of a function at the grid point $z_{k}$, and $f_{k,r}$ the prescribed value of the $r$\foreignlanguage{american}{-th} derivative, for $1\leq r\leq n_{k}-1$, the Hermite interpolant is the unique polynomial of degree $N-1$ ($N=n_{1}+\cdots+n_{K}$) which interpolates the prescribed function values and function derivatives. We obtain another derivation of a method for Hermite interpolation recently proposed by Butcher et al. {[}\emph{Numerical Algorithms, vol. 56 (2011), p. 319-347}{]}. One advantage of our derivation is that it leads to an efficient method for updating the barycentric weights. If an additional derivative is prescribed at one of the interpolation points, we show how to update the barycentric coefficients using only $\mathcal{O}\left(N\right)$ operations. Even in the context of confluent Newton series, a comparably efficient and general method to update the coefficients appears not to be known. If the method is properly implemented, it computes the barycentric weights with fewer operations than other methods and has very good numerical stability even when derivatives of high order are involved. We give a partial explanation of its numerical stability.

math.NA↗

arXiv2011

Finite Difference Weights, Spectral Differentiation, and Superconvergence

Let $z_{1},z_{2},...,z_{N}$ be a sequence of distinct grid points. A finite difference formula approximates the $m$-th derivative $f^{(m)}(0)$ as $\sum w_{k}f(z_{k})$, with $w_{k}$ being the weights. We derive an algorithm for finding the weights $w_{k}$ which is an improvement of an algorithm of Fornberg (\emph{Mathematics of Computation}, vol. 51 (1988), p. 699-706). This algorithm uses fewer arithmetic operations than that of Fornberg by a factor of $4/(5m+5)$ while being equally accurate. The algorithm that we derive computes finite difference weights accurately even when $m$, the order of the derivative, is as high as 16. In addition, the algorithm generalizes easily to the efficient computation of spectral differentiation matrices. The order of accuracy of the finite difference formula for $f^{(m)}(0)$ with grid points $hz_{k}$, $1\leq k\leq N$, is typically $\mathcal{O}(h^{N-m})$. However, the most commonly used finite difference formulas have an order of accuracy that is higher than the typical. For instance, the centered difference approximation $(f(h)-2f(0)+f(-h))/h^{2}$ to $f"(0)$ has an order of accuracy equal to 2 not 1. Even unsymmetric finite difference formulas can exhibit such superconvergence or boosted order of accuracy, as shown by the explicit algebraic condition that we derive. If the grid points are real, we prove a basic result stating that the order of accuracy can never be boosted by more than 1.

math.NA↗

arXiv2009

Complex Singularities and the Lorenz Attractor

The Lorenz attractor is one of the best known examples of applied mathematics. However, much of what is known about it is a result of numerical calculations and not of mathematical analysis. As a step toward mathematical analysis, we allow the time variable in the three dimensional Lorenz system to be complex, hoping that solutions that have resisted analysis on the real line will give up their secrets in the complex plane. Knowledge of singularities being fundamental to any investigation in the complex plane, we build upon earlier work and give a complete and consistent formal development of complex singularities of the Lorenz system using {\it psi series}. The psi series contain two undetermined constants. In addition, the location of the singularity is undetermined as a consequence of the autonomous nature of the Lorenz system. We prove that the psi series converge, using a technique that is simpler and more powerful than that of Hille, thus implying a two parameter family of singular solutions of the Lorenz system. We pose three questions, answers to which may bring us closer to understanding the connection of complex singularities to Lorenz dynamics.

math.DS↗

arXiv2009

Exact and Asymptotic Conditions on Traveling Wave Solutions of the Navier-Stokes Equations

We derive necessary conditions that traveling wave solutions of the Navier-Stokes equations must satisfy in the pipe, Couette, and channel flow geometries. Some conditions are exact and must hold for any traveling wave solution irrespective of the Reynolds number ($Re$). Other conditions are asymptotic in the limit $Re\to\infty$. The exact conditions are likely to be useful tools in the study of transitional structures. For the pipe flow geometry, we give computations up to $Re=100000$ showing the connection of our asymptotic conditions to critical layers that accompany vortex structures at high $Re$.

physics.flu-dyn↗

arXiv2008

Heteroclinic connections in plane Couette flow

Plane Couette flow transitions to turbulence for Re~325 even though the laminar solution with a linear profile is linearly stable for all Re (Reynolds number). One starting point for understanding this subcritical transition is the existence of invariant sets in the state space of the Navier Stokes equation, such as upper and lower branch equilibria and periodic and relative periodic solutions, that are quite distinct from the laminar solution. This article reports several heteroclinic connections between such objects and briefly describes a numerical method for locating heteroclinic connections. Computing such connections is essential for understanding the global dynamics of spatially localized structures that occur in transitional plane Couette flow. We show that the nature of streaks and streamwise rolls can change significantly along a heteroclinic connection.

physics.flu-dyn↗

arXiv1999

Convergence of numerical ODE solvers and Lyapunov's theory of stability

For the ordinary differential equation (ODE) $\dot{x}(t) = f(t,x)$, $x(0) = x_0$, $t\geq 0$, $x\in R^d$, assume $f$ to be at least continuous in $t$ and locally Lipshitz in $x$, and if necessary, several times continuously differentiable in $t$ and $x$. We associate a conditioning function $E(t)$ with each solution $x(t)$ which captures the accumulation of global error in a numerical approximation in the following sense: if $\tilde{x}(t;h)$ is an approximation derived from a single step method of time step $h$ and order $r$ then $\norm{\tilde{x}(t;h) - x(t)} < K(E(t)+ε)h^r$ for $0\leq t\leq T$, any $ε> 0$, sufficiently small $h$, and a constant $K>0$. Using techniques from the stability theory of differential equations, this paper gives conditions on $x(t)$ for $E(t)$ to be upper bounded linearly or by a constant for $t\geq 0$. More concretely, these techniques give constant or linear bounds on $E(t)$ when $x(t)$ is a trajectory of a dynamical system which falls into a stable, hyperbolic fixed point; or into a stable, hyperbolic cycle; or into a normally hyperbolic and contracting manifold with quasiperiodic flow on the manifold.

math.DS↗