matroids & representative sets daniel lokshtanov

44
Matroids & Representative Sets Daniel Lokshtanov

Upload: anis-gardner

Post on 22-Dec-2015

221 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Matroids & Representative Sets Daniel Lokshtanov

Matroids & Representative Sets

Daniel Lokshtanov

Page 2: Matroids & Representative Sets Daniel Lokshtanov

Alice vs Bob

F = {{a,b,c}, {a,c,d}, {b,c,e}}

{b, e}

{a,c,d}

{a, c}

Page 3: Matroids & Representative Sets Daniel Lokshtanov

Rules of the game

Board: universe of size nAll Alice’s sets have size pBob a picks set B of size qAlice wins if she has a set disjoint from B

Page 4: Matroids & Representative Sets Daniel Lokshtanov

Lazy Alice

Alice does not like remembering all those sets.Alice hates losing to Bob.

Can she forget a set A from F, and be sure this will not make the difference

between winning and losing?

Page 5: Matroids & Representative Sets Daniel Lokshtanov

(Ir)relevant Sets

A F is irrelevant if: for every set B of size q such that A B = , there is a set A’ F such that A’ B = .

Alice may forget exactly the irrelevant sets

Page 6: Matroids & Representative Sets Daniel Lokshtanov

Only relevant sets?

B1 B2 B3 Bm

F = { A2A1 A3 Am }…

Page 7: Matroids & Representative Sets Daniel Lokshtanov

Bollobás’ Lemma [1966]

Let A1, A2, … , Am be sets of size p and

B1, B2, … , Bm be sets of size q s.t:

then .

No dependence on universe size n at all!

Page 8: Matroids & Representative Sets Daniel Lokshtanov

Bollobás’ helps Alice

Bollobás’ lemma immediately implies that Alice only needs to remember at most sets.

yay!

Page 9: Matroids & Representative Sets Daniel Lokshtanov

Proof of Bollobás’ Lemma

Consider a random permutation of the universe.

The events «all of Ai before all of Bi» and «All of Aj before all of Bj» are disjoint!

So and hence .

P[all of Ai is before all of Bi] = .Ai BiBj Aj

Page 10: Matroids & Representative Sets Daniel Lokshtanov

Representative Sets

Let F be a family of p-sets. Then q-represents F if for every B of size q such that there exists an with there exists an with .

Corollary of Bollobás: For every F there is an of size at most that q-represents F.

Page 11: Matroids & Representative Sets Daniel Lokshtanov

Computational Problem

Given a family F of p-sets and an integer q, compute a family of size at most that q-represents F.

Page 12: Matroids & Representative Sets Daniel Lokshtanov

Computing Representative Sets

Will show: we can compute representative sets in time essentially where is the matrix multiplication constant < 2.38.

But first – an easy application

Page 13: Matroids & Representative Sets Daniel Lokshtanov

d-Hitting Set

Input: Family F = {S1,…,Sm} of sets of size d over universe U = {v1, …, vn}, integer k.

Question: Does there exist a set X U of size at most k such that for every Si F, Si X ?

Easy branching in time dk

Next: kernel with O(kd) sets and elements

Page 14: Matroids & Representative Sets Daniel Lokshtanov

d-Hitting Set as a Game

F = {{a,b,c}, {a,c,d}, {b,c,e}} Is {b, e} a hitting set?

No, since

{a,c,d}

Page 15: Matroids & Representative Sets Daniel Lokshtanov

Kernel for d-Hitting Set

Compute a k-representative subfamily F’ F of size at most .

Remove all elements not in F’ (at most dkd)

Output the instance F’, k.

Page 16: Matroids & Representative Sets Daniel Lokshtanov

Why is the kernel correct?

May not change a YES instance into a NO instance.

Can a NO instance change into a YES instance?NO instance = Alice always wins

YES instance = Bob can win

We did not forget any sets that made the difference between Alice winning and losing!

Page 17: Matroids & Representative Sets Daniel Lokshtanov

Playing on a matroid

Suppose now that the universe is the edge set of a matroid.

A set A fits a set B if - A and B are disjoint and- A B is independent in the matroid.

Page 18: Matroids & Representative Sets Daniel Lokshtanov

Alice vs Bob on a matroid

F = {{a,b,c}, {a,c,d}, {b,c,e}}

Do you have a set

that fits {b, e}?

&%¤&!!

Note: this game on a uniform matroid of rank p+q is exactly the old game.

Page 19: Matroids & Representative Sets Daniel Lokshtanov

Representative Sets

Let F be a family of p-sets (in a matroid M). Then q-represents F if:

for every B of size q such thatthere exists an that fits B there exists an that also fits B.

Note: representation in a uniform matroid of rank p+q is exactly the old representation.

Page 20: Matroids & Representative Sets Daniel Lokshtanov

Computing Representative Sets

Input: Family F of p-sets over a matroid, integer q, matrix M representing the matroid.

Task: Compute a q-representative subfamily of size at most .

Page 21: Matroids & Representative Sets Daniel Lokshtanov

Playing on a matroidp=4, q=2

3049580290385029238402938502309058010120958303215203852923023023109580420935820230395283032023350203220225822202302350203209802104+4267429810983502239582820320502340958683040938323035802092309532029385308209821522998208357298739829872398253982359823987235239729019380205230958203958293958203958203958203958522938572938575292

M = p+q

F ?!

232401018920320110848338053002

Det

Page 22: Matroids & Representative Sets Daniel Lokshtanov

Fit vs Determinant

If Alices set A and Bob’s set B overlap, then the same column is used twice determinant is 0!

Determinant is nonzero if and only if A fits B.

Page 23: Matroids & Representative Sets Daniel Lokshtanov

Matrix game

ab

cd p+q

p

p+q

q

c

Page 24: Matroids & Representative Sets Daniel Lokshtanov

Generalized Laplace Expansionalmost correct

MBMA p+q

p q

To compute Det

Compute the determinants of all

p p submatrices of MA

Compute the determinants of all

q q submatrices of MB

dimensional vector vA

dimensional vector vB

dot product!*

Page 25: Matroids & Representative Sets Daniel Lokshtanov

Giant Vector game

a (𝑝+𝑞𝑝 )b c

d

c

(𝑝+𝑞𝑝 )

⋅ 0 ?

Page 26: Matroids & Representative Sets Daniel Lokshtanov

Basis

If Alice keeps vectors v1,v2,v3 and v3 = v1 + v2

and v3 fits Bob’s vector vB

Then either v1 or v2 fits vB Alice only needs to keep linearly independent vectors!

At most of them, sincevectors are - dimensional

Finding the basis takes time.

Page 27: Matroids & Representative Sets Daniel Lokshtanov

Wrap up

Alice has a family of p-sets, family of p (p+q) matricesfamily of - dimensonal vectors.

Keep linearly independent vectors, keep the corresponding sets!

Page 28: Matroids & Representative Sets Daniel Lokshtanov

Computing Representative Sets

Theorem: we can compute representative sets of size in time essentially where is the matrix multiplication constant < 2.38.

Page 29: Matroids & Representative Sets Daniel Lokshtanov

Application - Treewidth DP

Have seen several approaches for single exponential algorithms for connectivity problems parameterized by treewidth.

Representative sets gives yet another one

Page 30: Matroids & Representative Sets Daniel Lokshtanov

Hamiltonian Path

Page 31: Matroids & Representative Sets Daniel Lokshtanov

Representative Sets for Matroid Classes

Is it possible to compute representative sets for uniform matroids, graphic matroids or transversal matroids faster than for linear matroids in general?

For uniform matroids, the answer is yes(but proof is sort of complicated)

Page 32: Matroids & Representative Sets Daniel Lokshtanov

Application – k-Path

Input: (directed) graph G, integer k.Question: Is there a simple directed path on k vertices?

Theorem: There is a deterministic time algorithm for k-Path.

Page 33: Matroids & Representative Sets Daniel Lokshtanov

k-Path

Fix a source vertex u.

For vertex v and integer p, define P[v,p] to be the set of (vertex sets of) paths on exactly p vertices from u to v.

Goal: for every v and p k compute a set P’[v,p] that (k-p)-represents P[v,p].

Page 34: Matroids & Representative Sets Daniel Lokshtanov

k-PathGoal: for every v, p k compute a set P’[v,p] that (k-p)-represents P[v,p].

X [v , p ]=¿wv∈ E(G)P′ [w , p−1 ] ⋅ \{v }P′ [u ,1 ]= {{u } }

P′ [v , p ]=reduce(X [v , p ] , k − p)

Need to prove: that (k-p)-represents P[v,p] assuming (k-p+1)-represents P[w,p-1]

Extend all paths that can beextended by v

Page 35: Matroids & Representative Sets Daniel Lokshtanov

Need to prove: that (k-p)-represents P[v,p] assuming (k-p+1)-represents P[w,p-1]

BBu

wv

Size q+1

In

In

Page 36: Matroids & Representative Sets Daniel Lokshtanov

k-Path

Size of family to reduce:

Time to reduce:

Total time:

Also works for weighted k-Path.

( 𝑘𝑝−1)2𝑜(𝑘) ⋅n

Page 37: Matroids & Representative Sets Daniel Lokshtanov

Application – k-Cycle

Input: (directed) graph G, integer k.Question: Is there a simple directed cycle on at least k vertices?

Theorem: 8kpoly(n) algorithm.

Page 38: Matroids & Representative Sets Daniel Lokshtanov

k-Cycle – main lemma

In a shortest cycle C on at least k vertices, we can replace any subpath on k vertices by any other path on k vertices, which is disjoint from the k vertices after it on C.

Page 39: Matroids & Representative Sets Daniel Lokshtanov

k-Cycle – main lemma proof

Puv

Pvw

Pwv

v

u

w

Page 40: Matroids & Representative Sets Daniel Lokshtanov

k-Cycle – algorithm

Guess a vertex u that a shortest cycle C of length at least k passes through.

For every vertex v and integer p, define P[u, p] to be the set of (vertex sets of) paths on exactly p vertices from u to v.

For every vertex v compute a set P’[v] that k-represents P[v,k] using the method from the k-path algorithm.

Size:

Page 41: Matroids & Representative Sets Daniel Lokshtanov

k-Cycle – algorithm

Guess the k’th vertex v on the cycle C.

Main lemma there is a k-cycle containing a path from P’[v] as a subpath!

Check whether there is a path Q in P’[v] such that there is a path back from v to u in G\Q.

Time 8kpoly(n).

Page 42: Matroids & Representative Sets Daniel Lokshtanov

Speeding up

Using similar methods, but trading off space for time one can speed up k-Path to 2.619k and k-Cycle to 6.75k.

Page 43: Matroids & Representative Sets Daniel Lokshtanov

Exercises

Book: 12.9, 12.11, 12.13, 5.9

Page 44: Matroids & Representative Sets Daniel Lokshtanov

Thank You!