Abstract:We call matrix algorithms superfast if they use much fewer flops and memory cells than the input matrix has entries. Using such algorithms is indispensable for Big Data Mining and Analysis, where the input matrices are so immense that one can only access a small fraction of all their entries. A natural remedy is Low Rank Approximation (LRA) of these matrices, which is routinely computed by means of Cross-Approximation iterations for more than a decade of worldwide application in computational practice. We point out and extensively test an important application of superfast LRA to significant acceleration of the celebrated Fast Multipole Method, which turns it into Superfast Multipole Method.

What problem does this paper attempt to address?

The core problem that this paper attempts to solve is: **How to significantly accelerate the Fast Multipole Method (FMM) through Low - Rank Approximation (LRA), thereby achieving the Superfast Multipole Method**. ### Detailed Explanation 1. **Background and Problem Description** - In big data mining and analysis, the input matrices are usually so large that it is impossible to access all entries. Therefore, a method is needed to reduce the computational complexity and memory usage. - **Low - Rank Approximation (LRA)** is an effective solution. It can approximate large matrices with fewer parameters, thus significantly reducing the amount of computation. 2. **Specific Problem** - The Fast Multipole Method (FMM) is an algorithm for efficiently solving large - scale linear systems, especially performing well when dealing with matrices with special structures. However, in some important applications, the low - rank generators required by FMM are unavailable, which limits its efficiency. - The paper proposes an ultra - fast low - rank approximation algorithm based on **Cross - Approximation (CA)**, which can effectively generate these low - rank generators, thus transforming FMM into the Superfast Multipole Method. 3. **Solution** - The paper shows how to use the ultra - fast LRA algorithm to accelerate the key steps of FMM, especially generating the low - rank generators of the off - diagonal blocks of HSS matrices. - Specifically, by applying the ultra - fast LRA algorithm, the calculation of these generators can be completed in a shorter time, thus significantly improving the overall performance of FMM. 4. **Experimental Verification** - The paper verifies the effectiveness of the proposed ultra - fast LRA algorithm through a series of numerical experiments. The experimental results show that even the first cross - approximation cycle can provide reasonable approximation results, and the accuracy is further improved after multiple iterations. 5. **Conclusion** - By introducing the ultra - fast LRA algorithm, the calculation process of FMM can be significantly accelerated, making it more efficient and practical when dealing with large - scale matrices. ### Formula Summary The formulas involved in the paper mainly include: - **Definition of Low - Rank Approximation** \[ M = AB+E,\quad\frac{\|E\|}{\|M\|}\leq\epsilon \] where \(A\in\mathbb{C}^{m\times r}\), \(B\in\mathbb{C}^{r\times n}\), \(\|\cdot\|\) represents a certain matrix norm, and \(\epsilon\) is a small tolerance. - **Definition of HSS Matrix** - For an \(m\times n\) block matrix \(M\), if all diagonal blocks consist of \(O((m + n)r)\) entries, and \(r\) is the maximum rank of its "deflated" block columns, then \(M\) is called an \(r\)-HSS matrix. - **Generator Representation** - Each "deflated" block column \(N\) can be represented as the product of two generator matrices \(F\) and \(H\): \[ N = FH \] where \(F\) is an \(h\times r\) matrix and \(H\) is an \(r\times k\) matrix. Through these formulas and methods, the paper successfully solves the problem of low efficiency of FMM in some application scenarios.

QRP Variation of Cross--Approximation Iterations for Low Rank Approximation

More on low rank approximation of a matrix

CUR Low Rank Approximation of a Matrix at Sublinear Cost

Efficient Randomized Algorithms for the Fixed-Precision Low-Rank Matrix Approximation

Accurate Low Rank Approximation at a Low Computational Cost

Low-Rank Matrix Approximations with Flip-Flop Spectrum-Revealing QR Factorization

An Efficient Gauss--Newton Algorithm for Symmetric Low-Rank Product Matrix Approximations

Krylov Methods are (nearly) Optimal for Low-Rank Approximation

Computing Low-Rank Approximation of a Dense Matrix on Multicore CPUs with a GPU and Its Application to Solving a Hierarchically Semiseparable Linear System of Equations

Nearly Optimal Approximation of Matrix Functions by the Lanczos Method

Fast Convex Optimization Algorithms for Exact Recovery of a Corrupted Low-Rank Matrix

Fast Algorithms for Recovering a Corrupted Low-Rank Matrix

Fast Convex Optimization Algorithms for the Recovery of a Corrupted Low-Rank Matrix

A Multilevel Low-Rank Newton Method with Super-linear Convergence Rate and its Application to Non-convex Problems

Fast $(1+\varepsilon)$-Approximation Algorithms for Binary Matrix Factorization

Low-Rank Approximations with Sparse Factors I: Basic Algorithms and Error Analysis

Faster Algorithms for Schatten-p Low Rank Approximation

Accelerated Gradient Method for A Class of Nonconvex Low Rank Problem: Essentially Matching the Optimal Convex Convergence Rate

Krylov Subspace Recycling for Fast Iterative Least-Squares in Machine Learning

An algorithm for low-rank matrix factorization and its applications

Efficient Algorithms and Error Analysis for the Modified Nystrom Method.