Exchange operations on noncrossing spanning trees Csaba D. T oth - - PowerPoint PPT Presentation

exchange operations on noncrossing spanning trees
SMART_READER_LITE
LIVE PREVIEW

Exchange operations on noncrossing spanning trees Csaba D. T oth - - PowerPoint PPT Presentation

Exchange operations on noncrossing spanning trees Csaba D. T oth Cal State Northridge, Los Angeles, CA and Tufts University, Medford, MA Spanning Trees Elementary Operations abstract spanning tree = connected graph on n vertices that


slide-1
SLIDE 1

Exchange operations on noncrossing spanning trees

Csaba D. T´

  • th

Cal State Northridge, Los Angeles, CA and Tufts University, Medford, MA

slide-2
SLIDE 2

Spanning Trees — Elementary Operations

There are nn−2 spanning trees on n labeled vertices [Cayley, 1889] Exchange property for graphic maroids: If T1 = (V, E1) and T2 = (V, E2) are spanning trees, ∀e1 ∈ E1 ∃e2 ∈ E2 : (V, E1 − e1 + e2) is a spanning tree. For n ≥ 4, there exist two edge-disjoint spanning trees. So the diameter of the exchange graph equals n − 1. abstract spanning tree = connected graph on n vertices that does not contain cycles.

slide-3
SLIDE 3

Spanning Trees — Elementary Operations

For |S| = n, Ω(12.54n) ≤ max

|S|=n |T (S)| ≤ O(141.07n).

[Huemer and de Mier, 2015; Hoffmann et al. 2013] S = set of n points in general position in R2, T (S) = set of plane spanning trees on S. plane spanning tree = a straight-line spanning tree

  • n n points in the plane, no two edges cross.
  • The matroid exchange may introduce crossings!
  • We restrict exchanges to plane spanning trees.
slide-4
SLIDE 4

Spanning Trees — Elementary Operations

Let T1 = (S, E1) and T2 = (S, E2) be two trees in T (S). The operation that replaces T1 by T2 is

  • an exchange if there are edges e1 and e2 such that

E1 \ E2 = {e1} and E2 \ E1 = {e2} (i.e., delete an edge e1 from E1 and insert a new edge e2).

  • A compatible exchange is an exchange such that the

graph (S, E1 ∪ E2) is a noncrossing straight-line graph (i.e., e1 and e2 do not cross).

  • A rotation is a compatible exchange such that e1 and

e2 have a common endpoint p = e1 ∩ e2.

  • An empty-triangle rotation is a rotation such that the

edges of neither T1 nor T2 intersect the interior of the triangle ∆(pqr) formed by the vertices of e1 and e2.

  • An edge slide is an empty-triangle rotation such that

qr ∈ E1 ∩ E2.

slide-5
SLIDE 5

Spanning Trees — Elementary Operations

Exchange Compatible Exchange Rotation Empty-Triangle Rotation Edge Slide

slide-6
SLIDE 6

Spanning Trees — Elementary Operations

Operation Single Operation Single Operation Upper Bound Lower Bound Exchange 2n − 4 ⌊ 3n

2 ⌋ − 5 [HHM+99]

Compatible Ex. 2n − 4 ⌊ 3n

2 ⌋ − 5

Rotation 2n − 4 [AF96] ⌊ 3n

2 ⌋ − 4

Empty-Tri. Rot. O(n log n) ⌊ 3n

2 ⌋ − 4

Edge Slide O(n2) [AR07] Ω(n2) [AR07] All five operations define connected transition graphs for every point set in general position. Current upper and lower bounds for the diameter

slide-7
SLIDE 7

Spanning Trees — Simultaneous Operations

Upper and lower bounds for the diameter under simultaneous operations. Convex Position Empty-Tri. Rot. 4 3 Edge Slide O(log n) Ω(log n) Operation Simultaneous Simultaneous Upper Bound Lower Bound Exchange 1 1 Compatible Ex. O(log n) [AAH02] Ω(

log n log log n) [BRU+09]

Rotation O(log n) Ω(

log n log log n)

Empty-Tri. Rot. 8n Ω(log n) Edge Slide O(n2) [AR07] Ω(n)

slide-8
SLIDE 8

Spanning Trees — Exchange Operation

Lower bound construction: It takes ⌊ 3n

2 ⌋ − 5 exchanges to

transform T1 to T2. [Hernando, Hurtado, M´ arquez, Mora, and Noy, 1999]

slide-9
SLIDE 9

Spanning Trees — Exchange Operation

Lower bound construction: It takes ⌊ 3n

2 ⌋ − 5 exchanges to

transform T1 to T2. [Hernando, Hurtado, M´ arquez, Mora, and Noy, 1999] The same consturction gives a lower bound of ⌊ 3n

2 ⌋ − 4 for

rotation operations.

slide-10
SLIDE 10

Spanning Trees — Exchange Operation

n − 2 exchanges can transform any plane graph into a star centered at the convex hull. ⇒ Diameter ≤ 2n − 4 [Avis & Fukuda, 1996]

slide-11
SLIDE 11

Spanning Trees — Exchange Operation

n − 2 exchanges can transform any plane graph into a star centered at the convex hull. ⇒ Diameter ≤ 2n − 4 [Avis & Fukuda, 1996] Let v be a vertex on the convex hull. While T is not a star centered at v,

  • v sees an entire edge ab.
  • Rotate ab to av or bv.

v

slide-12
SLIDE 12

Spanning Trees — Exchange Operation

n − 2 exchanges can transform any plane graph into a star centered at the convex hull. ⇒ Diameter ≤ 2n − 4 [Avis & Fukuda, 1996] Let v be a vertex on the convex hull. While T is not a star centered at v,

  • v sees an entire edge ab.
  • Rotate ab to av or bv.

a b v

slide-13
SLIDE 13

Spanning Trees — Exchange Operation

n − 2 exchanges can transform any plane graph into a star centered at the convex hull. ⇒ Diameter ≤ 2n − 4 [Avis & Fukuda, 1996] Let v be a vertex on the convex hull. While T is not a star centered at v,

  • v sees an entire edge ab.
  • Rotate ab to av or bv.

a b v

slide-14
SLIDE 14

Spanning Trees — Exchange Operation

n − 2 exchanges can transform any plane graph into a star centered at the convex hull. ⇒ Diameter ≤ 2n − 4 [Avis & Fukuda, 1996] Let v be a vertex on the convex hull. While T is not a star centered at v,

  • v sees an entire edge ab.
  • Rotate ab to av or bv.

a b v

slide-15
SLIDE 15

Spanning Trees — Exchange Operation

n − 2 exchanges can transform any plane graph into a star centered at the convex hull. ⇒ Diameter ≤ 2n − 4 [Avis & Fukuda, 1996] Let v be a vertex on the convex hull. While T is not a star centered at v,

  • v sees an entire edge ab.
  • Rotate ab to av or bv.

a b v For n ≥ 3 points in convex position: diameter ≤ 23n

12 − 5.

[Lonner & T., 2018]

slide-16
SLIDE 16

Spanning Trees — Empty-Triangle Rotation

Let ℓ be a halving line. Triangulate T. For every triangle ∆ along ℓ (in stabbing order),

  • If the first edge of ∆

crossed by ℓ is in T, then replace it with another edge of ∆. At most 3n empty-triangle rotations can remove all but one edges between the two halves. f(n) ≤ 3n + 2f(n/2) ⇒ Diameter is O(n log n)

slide-17
SLIDE 17

Spanning Trees — Empty-Triangle Rotation

Let ℓ be a halving line. Triangulate T. For every triangle ∆ along ℓ (in stabbing order),

  • If the first edge of ∆

crossed by ℓ is in T, then replace it with another edge of ∆. At most 3n empty-triangle rotations can remove all but one edges between the two halves. f(n) ≤ 3n + 2f(n/2) ⇒ Diameter is O(n log n) ℓ

slide-18
SLIDE 18

Spanning Trees — Empty-Triangle Rotation

Let ℓ be a halving line. Triangulate T. For every triangle ∆ along ℓ (in stabbing order),

  • If the first edge of ∆

crossed by ℓ is in T, then replace it with another edge of ∆. At most 3n empty-triangle rotations can remove all but one edges between the two halves. f(n) ≤ 3n + 2f(n/2) ⇒ Diameter is O(n log n) ℓ

slide-19
SLIDE 19

Spanning Trees — Empty-Triangle Rotation

Let ℓ be a halving line. Triangulate T. For every triangle ∆ along ℓ (in stabbing order),

  • If the first edge of ∆

crossed by ℓ is in T, then replace it with another edge of ∆. At most 3n empty-triangle rotations can remove all but one edges between the two halves. f(n) ≤ 3n + 2f(n/2) ⇒ Diameter is O(n log n) ℓ

slide-20
SLIDE 20

Spanning Trees — Empty-Triangle Rotation

Let ℓ be a halving line. Triangulate T. For every triangle ∆ along ℓ (in stabbing order),

  • If the first edge of ∆

crossed by ℓ is in T, then replace it with another edge of ∆. At most 3n empty-triangle rotations can remove all but one edges between the two halves. f(n) ≤ 3n + 2f(n/2) ⇒ Diameter is O(n log n) ℓ

slide-21
SLIDE 21

Spanning Trees — Empty-Triangle Rotation

Let ℓ be a halving line. Triangulate T. For every triangle ∆ along ℓ (in stabbing order),

  • If the first edge of ∆

crossed by ℓ is in T, then replace it with another edge of ∆. At most 3n empty-triangle rotations can remove all but one edges between the two halves. f(n) ≤ 3n + 2f(n/2) ⇒ Diameter is O(n log n) ℓ

slide-22
SLIDE 22

Spanning Trees — Empty-Triangle Rotation

Let ℓ be a halving line. Triangulate T. For every triangle ∆ along ℓ (in stabbing order),

  • If the first edge of ∆

crossed by ℓ is in T, then replace it with another edge of ∆. At most 3n empty-triangle rotations can remove all but one edges between the two halves. f(n) ≤ 3n + 2f(n/2) ⇒ Diameter is O(n log n) ℓ

slide-23
SLIDE 23

Spanning Trees — Empty-Triangle Rotation

Let ℓ be a halving line. Triangulate T. For every triangle ∆ along ℓ (in stabbing order),

  • If the first edge of ∆

crossed by ℓ is in T, then replace it with another edge of ∆. At most 3n empty-triangle rotations can remove all but one edges between the two halves. f(n) ≤ 3n + 2f(n/2) ⇒ Diameter is O(n log n) ℓ

slide-24
SLIDE 24

Spanning Trees — Empty-Triangle Rotation

Let ℓ be a halving line. Triangulate T. For every triangle ∆ along ℓ (in stabbing order),

  • If the first edge of ∆

crossed by ℓ is in T, then replace it with another edge of ∆. At most 3n empty-triangle rotations can remove all but one edges between the two halves. f(n) ≤ 3n + 2f(n/2) ⇒ Diameter is O(n log n) ℓ

slide-25
SLIDE 25

Spanning Trees — Empty-Triangle Rotation

Let ℓ be a halving line. Triangulate T. For every triangle ∆ along ℓ (in stabbing order),

  • If the first edge of ∆

crossed by ℓ is in T, then replace it with another edge of ∆. At most 3n empty-triangle rotations can remove all but one edges between the two halves. f(n) ≤ 3n + 2f(n/2) ⇒ Diameter is O(n log n) ℓ

slide-26
SLIDE 26

Spanning Trees — Empty-Triangle Rotation

Let ℓ be a halving line. Triangulate T. For every triangle ∆ along ℓ (in stabbing order),

  • If the first edge of ∆

crossed by ℓ is in T, then replace it with another edge of ∆. At most 3n empty-triangle rotations can remove all but one edges between the two halves. f(n) ≤ 3n + 2f(n/2) ⇒ Diameter is O(n log n) ℓ

slide-27
SLIDE 27

Spanning Trees — Empty-Triangle Rotation

Let ℓ be a halving line. Triangulate T. For every triangle ∆ along ℓ (in stabbing order),

  • If the first edge of ∆

crossed by ℓ is in T, then replace it with another edge of ∆. At most 3n empty-triangle rotations can remove all but one edges between the two halves. f(n) ≤ 3n + 2f(n/2) ⇒ Diameter is O(n log n) ℓ

slide-28
SLIDE 28

Spanning Trees — Empty-Triangle Rotation

Let ℓ be a halving line. Triangulate T. For every triangle ∆ along ℓ (in stabbing order),

  • If the first edge of ∆

crossed by ℓ is in T, then replace it with another edge of ∆. At most 3n empty-triangle rotations can remove all but one edges between the two halves. f(n) ≤ 3n + 2f(n/2) ⇒ Diameter is O(n log n) ℓ

slide-29
SLIDE 29

Spanning Trees — Empty-Triangle Rotation

Let ℓ be a halving line. Triangulate T. For every triangle ∆ along ℓ (in stabbing order),

  • If the first edge of ∆

crossed by ℓ is in T, then replace it with another edge of ∆. At most 3n empty-triangle rotations can remove all but one edges between the two halves. f(n) ≤ 3n + 2f(n/2) ⇒ Diameter is O(n log n) ℓ

slide-30
SLIDE 30

Simultaneous Empty-Triangle Rotation

Let ℓ be a halving line. Triangulate T. For every triangle ∆ along T (in stabbing order),

  • If the first edge of ∆

crossed by ℓ is in T, then replace it with another edge of ∆. At most 3n empty triangle rotations can remove all but one edges between the two halves. f(n) ≤ 3n + f(n/2) ⇒ Diameter is O(n) ℓ

slide-31
SLIDE 31

Simultaneous Empty Triangle Rotations

logn(y)/2

Tree T1 contains a horizontal edge pq. Tree T2 is a star centered at r. p q r Ω(log n) simultaneous empty-triangle rotations are sometimes necessary:

slide-32
SLIDE 32

Simultaneous Empty Triangle Rotations

logn(y)/2

Tree T1 contains a horizontal edge pq. Tree T2 is a star centered at r. p q r Ω(log n) simultaneous empty-triangle rotations are sometimes necessary:

slide-33
SLIDE 33

Simultaneous Empty Triangle Rotations

logn(y)/2

Tree T1 contains a horizontal edge pq. Tree T2 is a star centered at r. p q r Ω(log n) simultaneous empty-triangle rotations are sometimes necessary:

slide-34
SLIDE 34

Simultaneous Empty Triangle Rotations

logn(y)/2

Tree T1 contains a horizontal edge pq. Tree T2 is a star centered at r. p q r Ω(log n) simultaneous empty-triangle rotations are sometimes necessary:

slide-35
SLIDE 35

Simultaneous Empty Triangle Rotations

logn(y)/2

Tree T1 contains a horizontal edge pq. Tree T2 is a star centered at r. p q r Ω(log n) simultaneous empty-triangle rotations are sometimes necessary:

slide-36
SLIDE 36

Simultaneous Empty Triangle Rotations

logn(y)/2

Tree T1 contains a horizontal edge pq. Tree T2 is a star centered at r. p q r Ω(log n) simultaneous empty-triangle rotations are sometimes necessary:

slide-37
SLIDE 37

Simultaneous Empty Triangle Rotations

logn(y)/2

Tree T1 contains a horizontal edge pq. Tree T2 is a star centered at r. p q r Ω(log n) simultaneous empty-triangle rotations are sometimes necessary:

slide-38
SLIDE 38

Simultaneous Empty Triangle Rotations

logn(y)/2

Tree T1 contains a horizontal edge pq. Tree T2 is a star centered at r. p q r Ω(log n) simultaneous empty-triangle rotations are sometimes necessary:

slide-39
SLIDE 39

Simultaneous Empty Triangle Rotations

logn(y)/2

Tree T1 contains a horizontal edge pq. Tree T2 is a star centered at r. p q r Ω(log n) simultaneous empty-triangle rotations are sometimes necessary:

slide-40
SLIDE 40

Simultaneous Empty Triangle Rotations

logn(y)/2

Tree T1 contains a horizontal edge pq. Tree T2 is a star centered at r. p q r Ω(log n) simultaneous empty-triangle rotations are sometimes necessary:

slide-41
SLIDE 41

Simultaneous Empty Triangle Rotations

logn(y)/2

Tree T1 contains a horizontal edge pq. Tree T2 is a star centered at r. p q r Ω(log n) simultaneous empty-triangle rotations are sometimes necessary:

slide-42
SLIDE 42

Simultaneous Empty Triangle Rotations

logn(y)/2

Tree T1 contains a horizontal edge pq. Tree T2 is a star centered at r. p q r Ω(log n) simultaneous empty-triangle rotations are sometimes necessary:

slide-43
SLIDE 43

Spanning Trees — Simultaneous Rotations

O(log n) simultaneous rotations can transform any plane graph into a star centered at the convex hull. ⇒ Diameter is O(log n)

slide-44
SLIDE 44

Spanning Trees — Simultaneous Rotations

O(log n) simultaneous rotations can transform any plane graph into a star centered at the convex hull. ⇒ Diameter is O(log n) p p p p

  • Let p be an extreme point.
  • Assume p = (0, −∞)

by a projective trafo.

  • While T is not a star

centered at p, Apply starify(p)

slide-45
SLIDE 45

Spanning Trees — Simultaneous Rotations

O(log n) simultaneous rotations can transform any plane graph into a star centered at the convex hull. ⇒ Diameter is O(log n) p p p p

  • Let p be an extreme point.
  • Assume p = (0, −∞)

by a projective trafo.

  • While T is not a star

centered at p, Apply starify(p)

slide-46
SLIDE 46

Spanning Trees — Simultaneous Rotations

O(log n) simultaneous rotations can transform any plane graph into a star centered at the convex hull. ⇒ Diameter is O(log n) p p p p

  • Let p be an extreme point.
  • Assume p = (0, −∞)

by a projective trafo.

  • While T is not a star

centered at p, Apply starify(p) e f Pe Pf

slide-47
SLIDE 47

Spanning Trees — Simultaneous Rotations

O(log n) simultaneous rotations can transform any plane graph into a star centered at the convex hull. ⇒ Diameter is O(log n) p p p p

  • Let p be an extreme point.
  • Assume p = (0, −∞)

by a projective trafo.

  • While T is not a star

centered at p, Apply starify(p) e f Pe Pf

slide-48
SLIDE 48

Spanning Trees — Simultaneous Rotations

O(log n) simultaneous rotations can transform any plane graph into a star centered at the convex hull. ⇒ Diameter is O(log n) p p p p

  • Let p be an extreme point.
  • Assume p = (0, −∞)

by a projective trafo.

  • While T is not a star

centered at p, Apply starify(p) e f Pe Pf

slide-49
SLIDE 49

Spanning Trees — Simultaneous Rotations

O(log n) simultaneous rotations can transform any plane graph into a star centered at the convex hull. ⇒ Diameter is O(log n) p p p p

  • Let p be an extreme point.
  • Assume p = (0, −∞)

by a projective trafo.

  • While T is not a star

centered at p, Apply starify(p)

slide-50
SLIDE 50

Spanning Trees — Simultaneous Rotations

O(log n) simultaneous rotations can transform any plane graph into a star centered at the convex hull. ⇒ Diameter is O(log n) p p p p

  • Let p be an extreme point.
  • Assume p = (0, −∞)

by a projective trafo.

  • While T is not a star

centered at p, Apply starify(p) starify(p) maintains a plane spanning tree. The sum of “discretre” horizontal extents all edges decreases by a factor of 1

2.

⇒ Algo. terminates after O(log n) moves.

slide-51
SLIDE 51

Spanning Trees — Simultaneous Rotations

Each iteration of starify(p) can be implemented in at most 4 simultaneous rotations. p e Pe u v

slide-52
SLIDE 52

Spanning Trees — Simultaneous Rotations

Each iteration of starify(p) can be implemented in at most 4 simultaneous rotations. p e Pe u v

slide-53
SLIDE 53

Spanning Trees — Simultaneous Rotations

Each iteration of starify(p) can be implemented in at most 4 simultaneous rotations. p e Pe u v

slide-54
SLIDE 54

Spanning Trees — Simultaneous Rotations

Each iteration of starify(p) can be implemented in at most 4 simultaneous rotations. p e Pe u v

slide-55
SLIDE 55

Spanning Trees — Simultaneous Rotations

Each iteration of starify(p) can be implemented in at most 4 simultaneous rotations. p e Pe u v

slide-56
SLIDE 56

Spanning Trees — Simultaneous Rotations

Each iteration of starify(p) can be implemented in at most 4 simultaneous rotations. p e Pe u v

slide-57
SLIDE 57

Spanning Trees — Simultaneous Rotations

Each iteration of starify(p) can be implemented in at most 4 simultaneous rotations. p e Pe u v

slide-58
SLIDE 58

Spanning Trees — Simultaneous Rotations

Each iteration of starify(p) can be implemented in at most 4 simultaneous rotations. p e Pe u v

slide-59
SLIDE 59

Spanning Trees — Elementary Operations

Operation Single Operation Single Operation Upper Bound Lower Bound Exchange 2n − 4 ⌊ 3n

2 ⌋ − 5 [HHM+99]

Compatible Ex. 2n − 4 ⌊ 3n

2 ⌋ − 5

Rotation 2n − 4 [AF96] ⌊ 3n

2 ⌋ − 4

Empty-Tri. Rot. O(n log n) ⌊ 3n

2 ⌋ − 4

Edge Slide O(n2) [AR07] Ω(n2) [AR07] All five operations define connected transition graphs for every point set in general position. Current upper and lower bounds for the diameter

slide-60
SLIDE 60

Spanning Trees — Simultaneous Operations

Upper and lower bounds for the diameter under simultaneous operations. Convex Position Empty-Tri. Rot. 4 3 Edge Slide O(log n) Ω(log n) Operation Simultaneous Simultaneous Upper Bound Lower Bound Exchange 1 1 Compatible Ex. O(log n) [AAH02] Ω(

log n log log n) [BRU+09]

Rotation O(log n) Ω(

log n log log n)

Empty-Tri. Rot. 8n Ω(log n) Edge Slide O(n2) [AR07] Ω(n)

slide-61
SLIDE 61

Reconstuct Crossings from Plane Spanning Trees

Keller & Perles [2016]: Given the exchange graph on T (S), for some point set S, one can compute the intersection graph

  • f the edges of K(S). In other words, the exchange graph

determines which pairs of edges of K(S) cross. S = set of n points in general position in R2, T (S) = set of plane spanning trees on S. K(S) = complete geometric graph on S.

slide-62
SLIDE 62

Reconstuct Crossings from Plane Spanning Trees

Keller & Perles [2016]: Given the exchange graph on T (S), for some point set S, one can compute the intersection graph

  • f the edges of K(S). In other words, the exchange graph

determines which pairs of edges of K(S) cross. S = set of n points in general position in R2, T (S) = set of plane spanning trees on S. K(S) = complete geometric graph on S. Oropeza & T. [2018]: Given the compatible exchange graph

  • n T (S), for some point set S, one can compute the

intersection graph of the edges of K(S). In other words, the compatoble exchange graph determines which pairs of edges of K(S) cross.

slide-63
SLIDE 63

Open Problems

Improve the diameter bounds for the “tree graphs.”

  • Are ⌊ 3

2n⌋ exchange operations enough to transform a

plane spanning tree to any other plane spanning tree?

  • Is the diameter for empty-triangle rotation O(n)?
  • Is the diameter for simultanous edge slides Θ(n), or

Θ(n2), or something in between? Transformation graphs for other variants:

  • Is the space of plane spanning trees of max degee≤ k

connected under any or all of the five operations?

  • If the edges have unique labels, can these operations

“shuffle” the labels arbitrarily?

slide-64
SLIDE 64

Open Problems

Reconstruction of intersection pattens from “tree graphs.”

  • Does the transition graph of rotations contain enough

informartion to reconstruct the intersection graph of the edges of K(S)?

  • For finding a possible counterexample, we need to

generate finite point sets S1, S2 ⊂ R2 such that |S1| = |S2|, |T (S1)| = |T (S2)|, and the intersection patterns of K(S1) and K(S2) are different.

slide-65
SLIDE 65

Thank you for your attention!