Random Walks, Faber Polynomials and Accelerated Power Methods
By: Peter Cowal, Nicholas F. Marshall, Sara Pollock
Potential Business Impact:
Makes computers solve math problems much faster.
In this paper, we construct families of polynomials defined by recurrence relations related to mean-zero random walks. We show these families of polynomials can be used to approximate $z^n$ by a polynomial of degree $\sim \sqrt{n}$ in associated radially convex domains in the complex plane. Moreover, we show that the constructed families of polynomials have a useful rapid growth property and a connection to Faber polynomials. Applications to iterative linear algebra are presented, including the development of arbitrary-order dynamic momentum power iteration methods suitable for classes of non-symmetric matrices.
Similar Papers
Random Walks, Faber Polynomials and Accelerated Power Methods
Numerical Analysis
Makes computers solve hard math problems faster.
On the complex zeros and the computational complexity of approximating the reliability polynomial
Combinatorics
Finds hidden patterns in computer problems.
Approximating functions on ${\mathbb R}^+$ by exponential sums
Numerical Analysis
Makes math problems with curves easier to solve.