4.6 Article

Revisiting Finite-Time Distributed Algorithms via Successive Nulling of Eigenvalues

Journal

IEEE SIGNAL PROCESSING LETTERS
Volume 22, Issue 1, Pages 54-57

Publisher

IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/LSP.2014.2346657

Keywords

Average-consensus; distributed algorithms; Eigenvalue multiplicity; finite-time convergence; subspaces

Funding

  1. NSF [CCF 1350264]
  2. Division of Computing and Communication Foundations
  3. Direct For Computer & Info Scie & Enginr [1350264] Funding Source: National Science Foundation

Ask authors/readers for more resources

In this letter, we characterize the finite-time behavior on arbitrary undirected graphs. In particular, we derive distributed iterations that are a function of a linear operator on the underlying graph and show that any arbitrary initial condition can be forced to lie on a particular subspace in a finite time. This subspace can be chosen to have the same dimension as the algebraic multiplicity of any (arbitrarily chosen) eigenvalue of the underlying linear operator and is spanned by the eigenvectors corresponding to the chosen eigenvalue. In other words, finite-time behavior is completely characterized by the algebraic multiplicity of the eigenvalues and the corresponding eigenvectors of the underlying linear operator. We show that finite-time average-consensus can be cast naturally in this setup for which we further develop the necessary and sufficient conditions.

Authors

I am an author on this paper
Click your name to claim this paper and add it to your profile.

Reviews

Primary Rating

4.6
Not enough ratings

Secondary Ratings

Novelty
-
Significance
-
Scientific rigor
-
Rate this paper

Recommended

No Data Available
No Data Available