presentations and extensions of transversal matroidshome.gwu.edu/~jbonin/presentations.pdflet r =...
TRANSCRIPT
Presentations and Extensions
of Transversal Matroids
Joseph E. Bonin
The George Washington University
This includes joint work withAnna de Mier (Universitat Politecnica de Catalunya)
The motivation, by analogy
A matroid M can have inequivalent representations, so a matrixrepresentation A can have extraneous information that limits whichextensions of M can be represented by adding a column to A.
aa′
bb′
cc′
aa′
bb′
cc′
.
The motivation, by analogy
A matroid M can have inequivalent representations, so a matrixrepresentation A can have extraneous information that limits whichextensions of M can be represented by adding a column to A.
aa′
bb′
cc′
aa′
bb′
cc′
.
Transversals matroid often have many presentations.
How do the presentations relate to (e.g., limit) the transversalextensions?
A brief review of transversal matroids
A set system, A = (Ai : i ∈ [r ]), is a multiset of sets.E.g., A1 = {a, b, d , e, f , g}, A2 = {b, c , f , g}, A3 = {d , e, f , g , }.
A1 = {a, b, d, e, f, g}
A2 = {b, c, f, g}
A3 = {d, e, f, g, }
a
b
c
d
e
f
g
A brief review of transversal matroids
A set system, A = (Ai : i ∈ [r ]), is a multiset of sets.E.g., A1 = {a, b, d , e, f , g}, A2 = {b, c , f , g}, A3 = {d , e, f , g , }.
A1 = {a, b, d, e, f, g}
A2 = {b, c, f, g}
A3 = {d, e, f, g, }
a
b
c
d
e
f
g
A1 = {a, b, d, e, f, g}
A2 = {b, c, f, g}
A3 = {d, e, f, g}
a
b
c
d
e
f
g
The independent sets of the transversal matroid M[A] are thepartial transversals of A.
The set system A is a presentation of M[A].
A brief review of transversal matroids
A1 = {a, b, d , e, f , g}, A2 = {b, c , f , g}, A3 = {d , e, f , g , }
A1
A2
A3
a b c d e f g
∗ ∗ 0 ∗ ∗ ∗ ∗0 ∗ ∗ 0 0 ∗ ∗0 0 0 ∗ ∗ ∗ ∗
A brief review of transversal matroids
A1 = {a, b, d , e, f , g}, A2 = {b, c , f , g}, A3 = {d , e, f , g , }
A1
A2
A3
a b c d e f g
∗ ∗ 0 ∗ ∗ ∗ ∗0 ∗ ∗ 0 0 ∗ ∗0 0 0 ∗ ∗ ∗ ∗
A1 = {a, b, d, e, f, g}
A2 = {b, c, f, g} A3 = {d, e, f, g, }
ce
a
bd
f g
We can replace ∗s with positive reals so that (i) determinants thataren’t forced to be zero for generic reasons are nonzero and(ii) each non-zero column sums to 1.
The multitude of presentations
4 of this type 1 of this type 8 of this type
4 of this type 2 of this type
1 of this type
The order on presentations
ce
a
fg
bd
A′1= {a, b, d, e, g}
A2 = {b, c, f, g} A3 = {d, e, f, g}
ce
a
fg
bd
{a, b, d, e, f, g}
{b, c, f, g} {d, e, f, g}
c e
a
fg
b d
{a, b, c, d, e, g}
{b, c, f, g} {d, e, f, g}
c e
a
f
gb d
A1 = {a, b, c, d, e, f, g}
A2 = {b, c, f, g} A3 = {d, e, f, g}
A partial order on thepresentations of M:
(A1, . . . ,Ar ) ≤ (A′1, . . . ,A
′r )
if, up to re-indexing the sets,Ai ⊆ A′
i for all i ∈ [r ].
The order on presentations
ce
a
fg
bd
A′1= {a, b, d, e, g}
A2 = {b, c, f, g} A3 = {d, e, f, g}
ce
a
fg
bd
{a, b, d, e, f, g}
{b, c, f, g} {d, e, f, g}
c e
a
fg
b d
{a, b, c, d, e, g}
{b, c, f, g} {d, e, f, g}
c e
a
f
gb d
A1 = {a, b, c, d, e, f, g}
A2 = {b, c, f, g} A3 = {d, e, f, g}
A partial order on thepresentations of M:
(A1, . . . ,Ar ) ≤ (A′1, . . . ,A
′r )
if, up to re-indexing the sets,Ai ⊆ A′
i for all i ∈ [r ].
A transversal matroid has aunique maximal presentation.
(Mason, 1971.)
Typically there are manyminimal presentations.
Extensions of a presentation of a transversal matroid
We consider only(a) presentations of M with r = r(M) sets, and(b) single-element transversal extensions of M of rank r .
Let A = (Ai : i ∈ [r ]) be a presentation of M.
For x 6∈ E (M) and J ⊆ [r ], set AJ = (AJi : i ∈ [r ]) where
AJi =
{
Ai ∪ x , if i ∈ J,Ai , otherwise.
The extension AJ of A yields the extension M[AJ ] of M.
Examples of extensions
A1
A2 A3 M [A∅]
x a loop
M [A{1}] M [A{1,2}]
M [A{1,3}] M [A{1,2,3}]
M[A{2}] = M[A{1,2}]
M[A{3}] = M[A{1,3}]
M[A{2,3}] = M[A{1,2,3}]
New results on the number of extensions of M from one presentation
8 extensions 8 extensions 8 extensions
6 extensions 5 extensions
5 extensions Let TA be the set oftransversal extensions of Mobtained by extending A.
TheoremLet r = r(M).
A is minimal iff |TA| = 2r ;otherwise |TA| ≤
34 · 2r .
If A neither is minimal norcovers a minimalpresentation, then|TA| ≤
58 · 2r . (2014)
A lattice
A1
A2 A3 M [A∅]
x a loop
M [A{1}] M [A{1,2}]
M [A{1,3}] M [A{1,2,3}]
∅
{1}
{1, 2} {1, 3}
{1, 2, 3}
M[A{2}] = M[A{1,2}]
M[A{3}] = M[A{1,3}]
M[A{2,3}] = M[A{1,2,3}]
A second example of such a lattice
a
b
c
f
d
e
gh
iA1 = {a, b, c} A4 = {g, h, i}
A3 = {d, e, f, g, h, i}
A2 = {b, c, d, e, f}
∅
{1} {2} {3}
{1, 2} {1, 3} {2, 3} {3, 4}
{1, 2, 3} {1, 3, 4} {2, 3, 4}
{1, 2, 3, 4}
Closure operators
A closure operator on a set S is a map σ : 2S → 2S for which
1. X ⊆ σ(X ) for all X ⊆ S ,
2. if X ⊆ Y ⊆ S , then σ(X ) ⊆ σ(Y ), and
3. σ(σ(X )) = σ(X ) for all X ⊆ S .
A σ-closed set is a subset X of S with σ(X ) = X .
The σ-closed sets, ordered by inclusion, form a lattice;the lattice operations are X ∨ Y = σ(X ∪ Y ) and X ∧ Y = X ∩ Y .
A closure operator related to extensions
Let A = (Ai : i ∈ [r ]) be a presentation of M, where r = r(M).
LemmaFor e ∈ E (M)− Ai , (A1,A2, . . . ,Ai−1,Ai ∪ e,Ai+1, . . . ,Ar )is a presentation of M iff e is a coloop of M\Ai . (Bondy & Welsh, 1971)
A closure operator related to extensions
Let A = (Ai : i ∈ [r ]) be a presentation of M, where r = r(M).
LemmaFor e ∈ E (M)− Ai , (A1,A2, . . . ,Ai−1,Ai ∪ e,Ai+1, . . . ,Ar )is a presentation of M iff e is a coloop of M\Ai . (Bondy & Welsh, 1971)
So, for J ⊆ [r ], there is a maximum K ⊆ [r ] with M[AJ ] = M[AK ].
Define σA : 2[r ] → 2[r ] by σA(J) = K .
A closure operator related to extensions
Let A = (Ai : i ∈ [r ]) be a presentation of M, where r = r(M).
LemmaFor e ∈ E (M)− Ai , (A1,A2, . . . ,Ai−1,Ai ∪ e,Ai+1, . . . ,Ar )is a presentation of M iff e is a coloop of M\Ai . (Bondy & Welsh, 1971)
So, for J ⊆ [r ], there is a maximum K ⊆ [r ] with M[AJ ] = M[AK ].
Define σA : 2[r ] → 2[r ] by σA(J) = K .
TheoremThe map σA is a closure operator on [r ]. The join of I and J inthe lattice LA of σA-closed sets is I ∪ J, so LA is distributive.If B = (Bi : i ∈ [r ]) is a presentation of M with Ai ⊆ Bi for all i ,then LB is a sublattice of LA. (2014)
LA and TA are isomorphic lattices
Order TA, the set of extensions of M obtained by extending A, bythe weak order.
For matroids M and N on E , we have M ≤w N in theweak order if rM(X ) ≤ rN(X ) for all X ⊆ E ;equivalently, every independent set of M is independent in N.
TheoremThe map J 7→ M[AJ ] is an isomorphism from LA onto TA,so TA is a distributive lattice. (2014)
Minimal presentations are most important for extensions
TheoremIf A ≤ B, then TB ⊆ TA.
If Ai ⊆ Bi for all i ∈ [r ], and if J ⊆ [r ], then M[AσB(J)] = M[BJ ].(2014)
Corollary
For any single-element transversal extension M ′ of M, there is aminimal presentation A of M and a set J with M ′ = M[AJ ]. (2013)
A characterization of minimal presentations
TheoremLet r = r(M). A presentation A of M is minimal iff |TA| = 2r .
(2013)
A characterization of minimal presentations
TheoremLet r = r(M). A presentation A of M is minimal iff |TA| = 2r .
(2013)
LemmaThe presentation A is minimal iff each set in A is a cocircuit of M.
(Las Vergnas, 1971; Bondy and Welsh, 1971)
A characterization of minimal presentations
TheoremLet r = r(M). A presentation A of M is minimal iff |TA| = 2r .
(2013)
LemmaThe presentation A is minimal iff each set in A is a cocircuit of M.
(Las Vergnas, 1971; Bondy and Welsh, 1971)
If A is not minimal, then r(M\Ai ) < r − 1 for some Ai ∈ A.
Set J = [r ]− i .
So x is a coloop of M[AJ ]\Ai , so M[AJ ] = M[AJ∪i ], so |TA| < 2r .
A characterization of minimal presentations
TheoremLet r = r(M). A presentation A of M is minimal iff |TA| = 2r .
(2013)
LemmaThe presentation A is minimal iff each set in A is a cocircuit of M.
(Las Vergnas, 1971; Bondy and Welsh, 1971)
If A is minimal and i ∈ [r ], then r(M\Ai ) = r − 1.
If J ⊆ [r ] and i 6∈ J, then x is not a coloop of M[AJ ]\Ai
since M[AJ ]\Ai extends M\Ai , so J is closed. Thus, LA = 2[r ].
Non-minimal presentations lose at least a quarter of the extensions
TheoremIf A is not minimal, then |TA| = |LA| ≤
34 · 2r . (2014)
This follows by finding the maximal proper sublattices of 2[r ] thatcontain ∅ and [r ], which are
Lij ={
X ⊆ [r ] : X ∩ {i , j} 6= {i}}
for any distinct i , j in [r ]; also, |Lij | =34 · 2
r .
Non-minimal presentations lose at least a quarter of the extensions
TheoremIf A is not minimal, then |TA| = |LA| ≤
34 · 2r . (2014)
This follows by finding the maximal proper sublattices of 2[r ] thatcontain ∅ and [r ], which are
Lij ={
X ⊆ [r ] : X ∩ {i , j} 6= {i}}
for any distinct i , j in [r ]; also, |Lij | =34 · 2
r .
Finding the next-largest sublattices (two types), and showing thatthe number of extensions goes down, gives:
TheoremIf A neither is minimal nor covers a minimal presentation, then|TA| = |LA| ≤
58 · 2
r . (2014)
All sublattices of 2[r ] can be LA
TheoremLet L be a sublattice of 2[r ] (so I ∩ J ∈ L and I ∪ J ∈ L for allI , J ∈ L) with both ∅ and [r ] in L.
There is a transversal matroid M of rank r with L = LA, where Ais the maximal presentation of M.
For each integer n ≥ r , there is a presentation A of the uniformmatroid Ur ,n on [n] with L = LA. (2014)
How many extensions can be common to two presentations?
TheoremIf A and B are different presentations of M, then
2 ≤ |TA ∩ TB| ≤3
4· 2r . (2014)
How many extensions can be common to two presentations?
TheoremIf A and B are different presentations of M, then
2 ≤ |TA ∩ TB| ≤3
4· 2r . (2014)
TheoremThe set LA,B = {J ∈ LA : M[AJ ] = M[BK ] for some K ∈ LB}is a sublattice of LA. (2014)
How many extensions can be common to two presentations?
TheoremIf A and B are different presentations of M, then
2 ≤ |TA ∩ TB| ≤3
4· 2r . (2014)
TheoremThe set LA,B = {J ∈ LA : M[AJ ] = M[BK ] for some K ∈ LB}is a sublattice of LA. (2014)
Closure under unions follows from:
TheoremFor all subsets J and K of [r ], the join of M[AJ ] and M[AK ] in thelattice of all extensions of M is transversal and is M[AJ∪K ]. (2013)
The corresponding statement for meets is false.
A question
Under the weak order, the set of all single-element extensions of amatroid M is a lattice.
Question
Is the set of all single-element transversal extensions of M, orderedby the weak order, a lattice?
As we saw, TA is a lattice, but this question concerns the union ofTA over all presentations A of M.
a a′
b
b′ c
c′
M
.
a a′
b
b′ c
c′
M1
a a′
b
b′ c
c′
M2
a
a′b
b′ c
c′
extension meeta a′
b
b′ c
c′
transversal meet
loop
A question
Under the weak order, the set of all single-element extensions of amatroid M is a lattice.
Question
Is the set of all single-element transversal extensions of M, orderedby the weak order, a lattice?
As we saw, TA is a lattice, but this question concerns the union ofTA over all presentations A of M.
M
a a′
c
b b′
d
Mc
a a′
cx
b b′
d
Md
a a′
c
b b′
d x
extension join
a a′
c
b b′
dx
transversal join
a a′
c
b b′
d
x
Thank you for listening.