phase&transi+ons&in&random&dyadic&tilings& … ·...
TRANSCRIPT
![Page 1: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/1.jpg)
Phase Transi+ons in Random Dyadic Tilings and Rectangular Dissec+ons
Sarah Cannon, Sarah Miracle and Dana Randall
Georgia Ins4tute of Technology
![Page 2: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/2.jpg)
Rectangular Dissec+ons
Rectangular dissec4ons arise in: • VLSI layout • Mapping graphs for floor layouts • Rou4ngs and placements • Combinatorics
Rectangular Dissec4on: A par44on of a laDce region into rectangles whose corners lie on laDce points.
![Page 3: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/3.jpg)
Rectangular Dissec+ons
Par44on n x n laDce region into rectangles such that:
1. There are n rectangles each with area n 2. The corners of rectangles lie on laDce points 3. n = 2k for an even integer k
The Edge-‐Flip Chain Repeat: 1. Pick an random edge e, 2. If e is flippable, flip edge e
with probability ½ �1 �2
)
Open Ques4on: Does the edge-‐flip chain mix rapidly?
![Page 4: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/4.jpg)
Talk Outline
1. Background and Previous Work 2. Our Results 3. Proof Ideas
![Page 5: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/5.jpg)
Related Work: Triangula+ons
• Triangula4ons of general point sets: Open
• Triangula4ons of point sets in convex posi4on: Fast [McShine, Tetali ‘98], [Molloy, Reed, Steiger ‘98]
• Triangula4ons on subsets of Z2: Open
• Weighted Triangula4ons on subsets of Z2 [Caputo, Mar4nelli, Sinclair, Stauffer ‘13]
The edge-‐flip chain: )
�1 �2
![Page 6: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/6.jpg)
Related Work: Weighted Triangula+ons
Weight (σ) = λ(total length of edges)
[Pietro Caputo, Fabio Martinelli, Alistair Sinclair and Alexandre Stauffer. “Random Lattice Triangulations: Structure and Algorithms.” 45th Symposium on Theory of Computing (STOC), 2013.]
[Caputo, Mar4nelli, Sinclair, Stauffer ‘13]
E.g., for λ > 1,
larger weight smaller weight
![Page 7: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/7.jpg)
Related Work: Weighted Triangula+ons
The edge-‐flip chain:
Weight (σ) = λ(total length of edges)
Results [CMSS]:
[Pietro Caputo, Fabio Martinelli, Alistair Sinclair and Alexandre Stauffer. “Random Lattice Triangulations: Structure and Algorithms.” 45th Symposium on Theory of Computing (STOC), 2013.]
)
�1 �2
[Caputo, Mar4nelli, Sinclair, Stauffer ‘13]
�c = 1
Polynomial Mixing Exponential Mixing
�:�0
Conjectured
Phase Transition
![Page 8: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/8.jpg)
Previous Work: Rectangular Dissec+ons
Special Cases:
1. Domino Tilings
2. Dyadic Tilings
The edge-‐flip chain:
�1 �2
)
Fast: [Luby, Randall, Sinclair ‘01], [Randall, Tetali ‘00]
![Page 9: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/9.jpg)
Special Cases: 2. Dyadic Tilings
A dyadic rectangle is a region R with dimensions
where a, b, s and t are nonnega4ve integers.
A dyadic 4ling of the 2k x 2k square is a set of 2k dyadic rectangles, each with area 2k (whose union is the full square).
Dyadic Not dyadic ,
Not a dyadic +ling A dyadic +ling
![Page 10: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/10.jpg)
Previous Results – Dyadic Tilings
The edge-‐flip chain connects the set of dyadic 4lings. [Janson, Randall, Spencer ‘02]
�1 �2
)The edge-‐flip chain:
Open Ques4on: Does the edge-‐flip chain converge quickly?
There is a different Markov chain that converges quickly. [JRS]
![Page 11: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/11.jpg)
Talk Outline
1. Background and Previous Work 2. Our Results 3. Proof Ideas
![Page 12: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/12.jpg)
Our Results: Connec+vity
The Edge-‐Flip Chain Repeat: 1. Pick an random edge e, 2. If e is flippable, flip edge e
with probability ½ �1 �2
)
Theorem 1: The Edge-‐Flip Chain connects the set of all dissec4ons of the n x n laDce region into n rectangles of size n.
![Page 13: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/13.jpg)
Weighted Rectangular Dissec+ons
Given an input parameter λ > 0,
Weight (σ) = λ(total length of edges).
smaller weight larger weight
For λ < 1,
smaller weight larger weightsmaller weight larger weight
For λ > 1,
smaller weight larger weight
![Page 14: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/14.jpg)
�1 �2
)
Weighted Rectangular Dissec+ons
Given an input parameter λ > 0,
Weight (σ) = λ(total length of edges).
The Weighted Edge-‐Flip Chain Repeat: 1. Pick a random edge e and p ∈ 𝑢 (0,1) 2. If e is flippable, let e’ be the new edge it can be flipped to. 3. Flip edge e with probability ½ if p < λ(|e’| - |e|).
�1�2
)
![Page 15: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/15.jpg)
The Mixing Time Defini4on: The total varia4on distance is
||Pt,π|| = max __ ∑ |Pt(x,y) – π(y)|. x∈ Ω y∈ Ω 2
1
A Markov chain is polynomial mixing if τ(ε) is poly(n, log(ε-‐1)). (n is the number of rectangles)
Defini4on: Given ε, the mixing 4me is
τ(ε) = min {t: ||Pt’,π|| < ε, t’ ≥ t}. A
A Markov chain is exponen4al mixing if τ(ε) is at least exp(n).
![Page 16: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/16.jpg)
Our Results: Dyadic Tilings
λ = 0.80 λ = 1.00 λ = 1.03
Rigorous proofs all the way to the cri4cal point λc= 1 !
?
�c = 1
Polynomial Mixing
�:Exponential Mixing
![Page 17: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/17.jpg)
Our Results: Rectangular Dissec+ons
λ = 0.80 λ = 1.00 λ = 1.03
Exponen4al mixing for very different reasons
?
� = 1
Exponential Mixing
�:Exponential Mixing
![Page 18: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/18.jpg)
Talk Outline
1. Background and Previous Work 2. Our Results 3. Proof Ideas
a. (General) The edge-‐flip chains connects. b. (Dyadic) When λ < 1, the edge-‐flip chain is poly. c. (Both) When λ > 1, the edge-‐flip chain is exp. d. (General) When λ < 1, the edge-‐flip chain is exp.
![Page 19: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/19.jpg)
Proof Sketch: Connec+vity
Thm 1: The Edge-‐Flip Chain connects the set of dissec4ons of the n x n laDce region into n rectangles of area n.
It’s not immediately obvious that a single valid move even exists! Proof sketch: Double induc4on on “h-‐regions”:
• Simply-‐connected subset of rectangles from a dissec4on • All rectangles have height at most h • All ver4cal sec4ons on the boundary have height c.h (for some integer c)
For n =16, an 8-‐region and a 4-‐region.
![Page 20: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/20.jpg)
) ) )
Proof Sketch: Connec+vity
Thm 1: The Edge-‐Flip Chain connects the set of dissec4ons of the n x n laDce region into n rectangles of size n.
It’s not immediately obvious that a single valid move even exists!
Proof sketch: Double induc4on on “h-‐regions”: • Prove can 4le every h-‐region with all rectangles of height h
• Inside every h-‐region, find an h/2-‐region or an h-‐region with smaller area
I.H.
h/2
h
h/2
![Page 21: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/21.jpg)
Proof Sketch: Connec+vity
Thm 1: The Edge-‐Flip Chain connects the set of dissec4ons of the n x n laDce region into n rectangles of size n.
It’s not immediately obvious that a single valid move even exists! Proof sketch: Double induc4on on “h-‐regions”:
n-‐region
• Prove can 4le every h-‐region with all rectangles of height h
![Page 22: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/22.jpg)
1. Background and Previous Work 2. Our Results 3. Proof Ideas
a. (General) The edge-‐flip chains connects. b. (Dyadic) When λ < 1, the edge-‐flip chain is poly. c. (Both) When λ > 1, the edge-‐flip chain is exp. d. (General) When λ < 1, the edge-‐flip chain is exp.
Talk Outline
![Page 23: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/23.jpg)
Fast Mixing for Dyadic Tilings
Proof Technique: Path coupling with an exponen4al metric
[Kenyon, Mossel, Perez ‘01][Greenberg, Pascoe, Randall ‘09]
Thm: For any constant λ < 1 , the edge-‐flip chain on the set of dyadic 4lings converges in 4me O(n2 log n).
For two configura4ons differing by flipping edge f to edge e, let the distance between them be λ|f|-|e|.
�c = 1
Polynomial Mixing
�:Exponential Mixing
![Page 24: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/24.jpg)
1. Background and Previous Work 2. Our Results 3. Proof Ideas
a. (General) The edge-‐flip chains connects. b. (Dyadic) When λ < 1, the edge-‐flip chain is poly. c. (Both) When λ > 1, the edge-‐flip chain is exp. d. (General) When λ < 1, the edge-‐flip chain is exp.
Talk Outline
![Page 25: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/25.jpg)
Slow Mixing when λ>1
Proof idea: Show that a “boxleneck” exists.
Thm: For any constant λ > 1 , the edge-‐flip chain requires 4me exp(Ω(n2)).
S S
![Page 26: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/26.jpg)
The “Boxleneck”
At least one 1 x n rectangle
No 1 x n or n x 1 rectangles
At least one n x 1 rectangle
![Page 27: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/27.jpg)
1. Background and Previous Work 2. Our Results 3. Proof Ideas
a. (General) The edge-‐flip chains connects. b. (Dyadic) When λ < 1, the edge-‐flip chain is poly. c. (Both) When λ > 1, the edge-‐flip chain is exp. d. (General) When λ < 1, the edge-‐flip chain is exp.
Talk Outline
![Page 28: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/28.jpg)
Slow Mixing when λ<1
Proof idea: Show that a “boxleneck” exists.
Thm: For any constant λ < 1 , the edge-‐flip chain on rectangular dissec4ons requires 4me exp(Ω(n log n)).
Key Ideas: 1. In order to remove a “bar” you need
two bars next to each other. 2. If you have 2 bars you must also have
lots of other thin rectangles.
� = 1
Exponential Mixing
�:Exponential Mixing
![Page 29: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/29.jpg)
The “Boxleneck”
Separa4on ≥ n/2 + 2 Separa4on < n/2 + 2
At least 4 bars and separa4on = n/2 + 2
• Pair up the bars le| to right • The separa2on is the sum of the “gaps”
Separa4on = g1 + g2 + 4 g1 g2
![Page 30: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/30.jpg)
The “Boxleneck”
At least 4 bars and separa4on = n/2 + 2
g1 = d1 = n/2 -2
= 0111 . . . 1110 (in binary) g1 d1
![Page 31: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/31.jpg)
The “Boxleneck”
At least 4 bars and separa4on = n/2 + 2
g1 = d1 = n/2 -2
= 0111 . . . 1110 (in binary) . . . d1
![Page 32: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/32.jpg)
Summary and Open Problems
1. What happens when λ = 1 for dyadic and general 4lings?
2. When does bias speed up or slow down a chain?
3. Is there a MC that is polynomial when the EF chain is not?
Dyadic Tilings:
General Rectangle Dissec4ons:
?
?
�c = 1
Polynomial Mixing
�:Exponential Mixing
� = 1
Exponential Mixing
�:Exponential Mixing
![Page 33: Phase&Transi+ons&in&Random&Dyadic&Tilings& … · Phase&Transi+ons&in&Random&Dyadic&Tilings& and&Rectangular&Dissec+ons& Sarah%Cannon,%Sarah%Miracle%and%DanaRandall% GeorgiaIns4tute%of%Technology%](https://reader036.vdocument.in/reader036/viewer/2022062303/5f7b23531f1b052482026174/html5/thumbnails/33.jpg)
Thank you!