Orthant-Strictly Monotonic Norms, Generalized Top-k and k-Support Norms and the L0 Pseudonorm

TL;DR

This paper introduces orthant-strictly monotonic norms and their connection to generalized top-k and k-support norms, providing a systematic way to express the l0 pseudonorm level sets through norm differences.

Contribution

It defines orthant-strictly monotonic norms and shows they generate strictly increasingly graded norm sequences related to the l0 pseudonorm.

Findings

Orthant-strictly monotonic norms encompass lp norms and others.

Generated norm sequences are strictly increasingly graded.

Level sets of l0 pseudonorm are expressed via differences of norms.

Abstract

The so-called l0 pseudonorm on the Euclidean space Rd counts the number of nonzero components of a vector. We say that a sequence of norms is strictly increasingly graded (with respect to the l0 pseudonorm) if it is nondecreasing and that the sequence of norms of a vector~x becomes stationary exactly at the index l0(x). In this paper, with any (source) norm, we associate sequences of generalized top-k and k-support norms, and we also introduce the new class of orthant-strictly monotonic norms (that encompasses the lp norms, but for the extreme ones). Then, we show that an orthant-strictly monotonic source norm generates a sequence of generalized top-k norms which is strictly increasingly graded. With this, we provide a systematic way to generate sequences of norms with which the level sets of the l0 pseudonorm are expressed by means of the difference of two norms. Our results rely on the study of orthant-strictly monotonic norms.