binary factorizations of the matrix of all ones
play

Binary Factorizations of the Matrix of All Ones Maguy Trefois Paul - PowerPoint PPT Presentation

Binary Factorizations of the Matrix of All Ones Maguy Trefois Paul Van Dooren Jean-Charles Delvenne Universit catholique de Louvain, Belgium ILAS, June 2013 1 Motivation : the finite-time average consensus problem We have: n


  1. Binary Factorizations of the Matrix of All Ones Maguy Trefois Paul Van Dooren Jean-Charles Delvenne Université catholique de Louvain, Belgium ILAS, June 2013 1

  2. Motivation : the finite-time average consensus problem We have: • n communicating agents with an initial position • a communication topology At each time step: • each agent sends its current position to some other agents according to the communication pattern • with the received information, each agent changes its position The goal: after a finite time, all the agents meet at the average of their initial positions 2

  3. Vector of initial positions: � x ( 0 ) Dynamics: x ( t + 1 ) = A t + 1 .� � x ( 0 ) The matrix A respects the communication topology:   0 ? 0 0 0 0 0 ?   A =   0 ? 0 0   ? 0 ? 0 3

  4. x ( t + 1 ) = A t + 1 .� � x ( 0 ) The matrix A is a solution to the consensus if:   0 ? 0 0 ? 0 0 0   • A is of the form   0 ? 0 0   ? ? 0 0   1 . . . 1 . . . • After a finite time m , A m = 1 . . . 4 .   . . .   . . . 1 1 4

  5. x ( t + 1 ) = A t + 1 .� � x ( 0 ) The matrix A is a solution to the consensus if:   0 ? 0 0 ? 0 0 0   • A is of the form   0 ? 0 0   ? ? 0 0   1 . . . 1 . . . • After a finite time m , A m = 1 . . . 4 .   . . .   . . . 1 1 Question : for which communication patterns is it possible to reach the consensus ? 4

  6. We should study the solutions to the equation:   1 . . . 1 A m = 1 . . . . . . n .  ,   . . .  1 . . . 1 where A ∈ R n × n . 5

  7. We should study the solutions to the equation:   1 . . . 1 A m = 1 . . . . . . n .  ,   . . .  1 . . . 1 where A ∈ R n × n . ⇒ difficult to tackle directly 5

  8. We should study the solutions to the equation:   1 . . . 1 A m = 1 . . . . . . n .  ,   . . .  1 . . . 1 where A ∈ R n × n . ⇒ difficult to tackle directly Simpler problem: study the solutions to:   1 . . . 1 A m = . . . . . .  ,   . . .  1 . . . 1 where A ∈ { 0 , 1 } n × n . 5

  9. Outline Factorization problem The De Bruijn matrices Factorizations into commuting factors General form of a root of I n with minimum rank A root class of I n Conclusion 6

  10. Factorization problem The De Bruijn matrices Factorizations into commuting factors General form of a root of I n with minimum rank A root class of I n Conclusion 7

  11. We are looking for the solutions to m � A i = A 1 A 2 ... A m = I n , i = 1 where - I n is the n × n matrix with all ones - each factor A i is an n × n binary matrix. 8

  12. We are looking for the solutions to m � A i = A 1 A 2 ... A m = I n , i = 1 where - I n is the n × n matrix with all ones - each factor A i is an n × n binary matrix. In particular, we are investigating the solutions to: A m = I n , where A is a binary matrix. 8

  13. Factorization problem The De Bruijn matrices Factorizations into commuting factors General form of a root of I n with minimum rank A root class of I n Conclusion 9

  14. Lemma If A ∈ { 0 , 1 } n × n is such that A m = I n , then - A is p-regular, i.e A . 1 = p . 1 and 1 T . A = p . 1 T - n = p m 10

  15. Lemma If A ∈ { 0 , 1 } n × n is such that A m = I n , then - A is p-regular, i.e A . 1 = p . 1 and 1 T . A = p . 1 T - n = p m Definition The De Bruijn matrix of order p and dimension n is a matrix of the form: D ( p , n ) := 1 p ⊗ I n / p ⊗ 1 T p , where - I n / p is the identity matrix of dimension n / p - 1 p is the p × 1 vector with all ones - ⊗ denotes the Kronecker product Moreover, it is imposed that n = p m , for some integer m. 10

  16. 1 1 0 0 0 0 0 0   0 0 1 1 0 0 0 0     0 0 0 0 1 1 0 0     0 0 0 0 0 0 1 1   D ( 2 , 8 ) =   1 1 0 0 0 0 0 0     0 0 1 1 0 0 0 0     0 0 0 0 1 1 0 0   0 0 0 0 0 0 1 1 11

  17. 1 1 0 0 0 0 0 0   0 0 1 1 0 0 0 0     0 0 0 0 1 1 0 0     0 0 0 0 0 0 1 1   D ( 2 , 8 ) =   1 1 0 0 0 0 0 0     0 0 1 1 0 0 0 0     0 0 0 0 1 1 0 0   0 0 0 0 0 0 1 1 Proposition The De Bruijn matrix D ( p , n ) with n = p m is such that D ( p , n ) m = I n . 11

  18. 1 1 0 0 0 0 0 0   0 0 1 1 0 0 0 0     0 0 0 0 1 1 0 0     0 0 0 0 0 0 1 1   D ( 2 , 8 ) =   1 1 0 0 0 0 0 0     0 0 1 1 0 0 0 0     0 0 0 0 1 1 0 0   0 0 0 0 0 0 1 1 Proposition The De Bruijn matrix D ( p , n ) with n = p m is such that D ( p , n ) m = I n . Question : Can we characterize all the roots from the De Bruijn matrices ? 11

  19. Factorization problem The De Bruijn matrices Factorizations into commuting factors General form of a root of I n with minimum rank A root class of I n Conclusion 12

  20. Factorization into commuting factors : Looking for the solutions to: AB = BA = I n , where - A and B are binary matrices - A is p -regular - B is l -regular 13

  21. Factorization problem: AB = BA = I n , where A is p -regular and B is l -regular. Theorem If A and B are commuting factors, then • p . l = n • rank ( A ) ≥ n / p and rank ( B ) ≥ n / l • if rank ( A ) = n / p (resp. rank ( B ) = n / l), then there exist permutation matrices P 1 , P 2 such that P 1 AP T ( resp. P 2 BP T 2 = D ( p , n ) 1 = D ( l , n )) . 14

  22. Question : Is it possible that rank ( A ) > n / p ?  1 0 1 0   1 1 0 0  0 1 0 1 1 1 0 0     A = B =     1 0 0 1 0 0 1 1     0 1 1 0 0 0 1 1 - A and B are 2-regular - AB = BA = I 4 - BUT, rank ( A ) = 3 > 4 / 2 15

  23. Question : Can we choose P 1 = P 2 ?  0 1 0 1 0 1   1 1 0 0 0 0  1 0 1 0 1 0 0 0 1 1 0 0         0 1 0 1 0 1 0 0 0 0 1 1     A = B =     1 0 1 0 1 0 1 1 0 0 0 0         0 1 0 1 0 1 0 0 1 1 0 0     1 0 1 0 1 0 0 0 0 0 1 1 - A is 3-regular, B is 2-regular and AB = BA = I 6 - rank ( A ) = 6 / 3, rank ( B ) = 6 / 2 - BUT, A is not isomorphic to D ( 3 , 6 ) since  3 0 3 0 3 0   2 2 2 1 1 1  0 3 0 3 0 3 1 1 1 2 2 2         3 0 3 0 3 0 2 2 2 1 1 1 A 2 = , D ( 3 , 6 ) 2 =         0 3 0 3 0 3 1 1 1 2 2 2         3 0 3 0 3 0 2 2 2 1 1 1     0 3 0 3 0 3 1 1 1 2 2 2 16

  24. Corollary Let A be a binary matrix satisfying A m = I n . Then, - A is p-regular - if rank ( A ) = n / p, then there are permutation matrices P 1 , P 2 such that P 1 AP T 2 = D ( p , n ) . As previously, - A may have a rank greater than n / p - A may not be isomorphic to D ( p , n ) even though rank ( A ) = n / p 17

  25. Factorization problem The De Bruijn matrices Factorizations into commuting factors General form of a root of I n with minimum rank A root class of I n Conclusion 18

  26. Theorem Let A ∈ { 0 , 1 } n × n such that A m = I n , A is p-regular and p m = n. If rank ( A ) = n / p, then A is isomorphic to a matrix P 1 D ( p , n ) , where P 1 = diag ( Q 1 , ..., Q p ) with each Q i ∈ { 0 , 1 } n / p × n / p is a permutation matrix. 19

  27. Theorem Let A ∈ { 0 , 1 } n × n such that A m = I n , A is p-regular and p m = n. If rank ( A ) = n / p, then A is isomorphic to a matrix P 1 D ( p , n ) , where P 1 = diag ( Q 1 , ..., Q p ) with each Q i ∈ { 0 , 1 } n / p × n / p is a permutation matrix. Not all the matrices of that form are solutions. Indeed, consider 1 1 0 0 0 0 0 0   0 0 1 1 0 0 0 0     0 0 0 0 1 1 0 0     0 0 0 0 0 0 1 1   A =   1 1 0 0 0 0 0 0     0 0 0 0 0 0 1 1     0 0 0 0 1 1 0 0   0 0 1 1 0 0 0 0 19

  28. Factorization problem The De Bruijn matrices Factorizations into commuting factors General form of a root of I n with minimum rank A root class of I n Conclusion 20

  29. A 2-circulant matrix:  0 1 0 1 0  1 0 0 1 0     1 0 1 0 0     0 0 1 0 1   0 1 0 0 1 Theorem (Wu, 2002) Let A ∈ { 0 , 1 } n × n be g-circulant and such that A m = I n . If - g m ≡ 0 mod n - A is p-regular, then A is isomorphic to D ( p , n ) . 21

  30. Definition A nice permutation matrix is built as follows: start with a p × p permutation matrix. Then, replace all the zeros by a zero p × p matrix and each one by a p × p permutation matrix. Repeat this m times. You obtain a permutation matrix of dimension p m . 22

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend


More recommend