randomized coloring with symmetry breaking › ~sriram › talks › randomcoloring.pdf · each...
TRANSCRIPT
![Page 1: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/1.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Randomized Coloring with Symmetry Breaking
Sriram V. Pemmaraju1
1Department of Computer ScienceThe University of Iowa
[email protected]/∼sriram/
MPII, Saarbrücken
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 2: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/2.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Randomized Coloring Procedure
COLORING STEP. Each vertex v picks a tentative coloruniformly at random from its palette.
REPAIR STEP. The tentative coloring may be improper; itis repaired by uncoloring each vertex v that receives thesame color as a neighbor.
The coloring may be completed deterministically (e.g.,greedily).
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 3: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/3.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Randomized Coloring Procedure
COLORING STEP. Each vertex v picks a tentative coloruniformly at random from its palette.
REPAIR STEP. The tentative coloring may be improper; itis repaired by uncoloring each vertex v that receives thesame color as a neighbor.
The coloring may be completed deterministically (e.g.,greedily).
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 4: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/4.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Randomized Coloring Procedure
COLORING STEP. Each vertex v picks a tentative coloruniformly at random from its palette.
REPAIR STEP. The tentative coloring may be improper; itis repaired by uncoloring each vertex v that receives thesame color as a neighbor.
The coloring may be completed deterministically (e.g.,greedily).
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 5: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/5.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Iterative Randomized Coloring Procedure
One application of RCP yields a “good,” partial coloringwith positive (though, usually very small) probability.
A “good” coloring is fixed and then extended by performingthe next iteration of the RCP on the uncolored graph.
This is an example of the semi-random method or Rödl Nibble.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 6: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/6.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Iterative Randomized Coloring Procedure
One application of RCP yields a “good,” partial coloringwith positive (though, usually very small) probability.
A “good” coloring is fixed and then extended by performingthe next iteration of the RCP on the uncolored graph.
This is an example of the semi-random method or Rödl Nibble.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 7: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/7.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Iterative Randomized Coloring Procedure
One application of RCP yields a “good,” partial coloringwith positive (though, usually very small) probability.
A “good” coloring is fixed and then extended by performingthe next iteration of the RCP on the uncolored graph.
This is an example of the semi-random method or Rödl Nibble.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 8: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/8.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Success Stories
Kim (CPC, 1995)
If girth(G) ≥ 5 then
χ(G) ≤ ∆
ln ∆(1 + o(1)).
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 9: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/9.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Success Stories
Johansson (1996)
There exists a constant ∆0: if girth(G) ≥ 4 and ∆ ≥ ∆0 then
χ(G) ≤ 9∆
ln ∆.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 10: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/10.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Success Stories
See “Graph Coloring and the Probabilistic Method” by Molloyand Reed for more examples.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 11: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/11.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Symmetry-Breaking Augmentation
Pick π, a permutation of the vertices.
Depending on the application, π can be chosen arbitrarily,randomly, or depending on the structure of the graph.
In the “repair” step, a vertex v is uncolored only if there is aneighbor of lower rank in π that has received the sametentative color.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 12: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/12.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Symmetry-Breaking Augmentation
Pick π, a permutation of the vertices.
Depending on the application, π can be chosen arbitrarily,randomly, or depending on the structure of the graph.
In the “repair” step, a vertex v is uncolored only if there is aneighbor of lower rank in π that has received the sametentative color.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 13: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/13.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Symmetry-Breaking Augmentation
Pick π, a permutation of the vertices.
Depending on the application, π can be chosen arbitrarily,randomly, or depending on the structure of the graph.
In the “repair” step, a vertex v is uncolored only if there is aneighbor of lower rank in π that has received the sametentative color.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 14: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/14.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
How does this help?
Key step in analyzing RCP is showing the concentration ofcertain random variables.
Concentration inequalities such as Azuma’s MartingaleInequalityare often used.
Symmetry-breaking makes it easier to apply Azuma’sInequality.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 15: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/15.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
How does this help?
Key step in analyzing RCP is showing the concentration ofcertain random variables.
Concentration inequalities such as Azuma’s MartingaleInequalityare often used.
Symmetry-breaking makes it easier to apply Azuma’sInequality.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 16: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/16.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
How does this help?
Key step in analyzing RCP is showing the concentration ofcertain random variables.
Concentration inequalities such as Azuma’s MartingaleInequalityare often used.
Symmetry-breaking makes it easier to apply Azuma’sInequality.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 17: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/17.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Azuma’s Martingale Inequality
LemmaLet X be a random variable determined by n trialsT1, T2, . . . , Tn, such that for each i, and any two possiblesequences of outcomes t1, t2, . . . , ti−1, ti and t1, t2, . . . , ti−1, ti ′:∣∣∣E [X |T1 = t1, . . . , Ti = ti ]− E [X |T1 = t1, . . . , Ti = ti ′]
∣∣∣ ≤ ci (1)
thenPr
[|X − E [X ]| > t
]≤ 2e−t2/(2
Pc2
i ). (2)
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 18: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/18.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Difficulty Applying Azuma’s Inequality
Pr[|X − E [X ]| > t
]≤ 2e−t2/(2
Pc2
i )
Showing ci = O(1) only leads to the bound e−ct2/n forsome positive constant c.
What is desired is eΩ(t2/E [X ]).
Then setting t =√
E [X ] · log E [X ] yields a sharpconcentration.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 19: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/19.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Difficulty Applying Azuma’s Inequality
Pr[|X − E [X ]| > t
]≤ 2e−t2/(2
Pc2
i )
Showing ci = O(1) only leads to the bound e−ct2/n forsome positive constant c.
What is desired is eΩ(t2/E [X ]).
Then setting t =√
E [X ] · log E [X ] yields a sharpconcentration.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 20: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/20.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Difficulty Applying Azuma’s Inequality
Pr[|X − E [X ]| > t
]≤ 2e−t2/(2
Pc2
i )
Showing ci = O(1) only leads to the bound e−ct2/n forsome positive constant c.
What is desired is eΩ(t2/E [X ]).
Then setting t =√
E [X ] · log E [X ] yields a sharpconcentration.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 21: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/21.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Difficulty Applying Azuma’s Inequality
Pr[|X − E [X ]| > t
]≤ 2e−t2/(2
Pc2
i )
Showing ci = O(1) only leads to the bound e−ct2/n forsome positive constant c.
What is desired is eΩ(t2/E [X ]).
Then setting t =√
E [X ] · log E [X ] yields a sharpconcentration.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 22: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/22.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Applications of Symmetry-Breaking
In proving existence of and deriving algorithms for frugalcolorings.
In proving existence of and deriving algorithms forweighted equitable colorings.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 23: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/23.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Applications of Symmetry-Breaking
In proving existence of and deriving algorithms for frugalcolorings.
In proving existence of and deriving algorithms forweighted equitable colorings.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 24: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/24.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Frugal Coloring
DefinitionA β-frugal coloring of a graph G is a proper vertex coloring inwhich no color appears more than β times in a neighborhood.
Lower Bound (Hind, Molloy and Reed, Combinatorica 1997):There are graphs for which every (∆ + 1)-coloring has frugality
Ω
(log ∆
log log ∆
).
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 25: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/25.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Frugal Coloring
DefinitionA β-frugal coloring of a graph G is a proper vertex coloring inwhich no color appears more than β times in a neighborhood.
Lower Bound (Hind, Molloy and Reed, Combinatorica 1997):There are graphs for which every (∆ + 1)-coloring has frugality
Ω
(log ∆
log log ∆
).
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 26: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/26.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Upper Bounds
Hind, Molloy, and Reed (1997): Every graph has a(∆ + 1)-coloring that is O(log5 ∆)-frugal.
This proof can be tightened to yield a (∆ + 1)-coloring thatis O(log3 ∆)-frugal.
This talk
Every graph has a (∆ + 1)-coloring that is O(
log2 ∆log log ∆
)-frugal.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 27: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/27.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Upper Bounds
Hind, Molloy, and Reed (1997): Every graph has a(∆ + 1)-coloring that is O(log5 ∆)-frugal.
This proof can be tightened to yield a (∆ + 1)-coloring thatis O(log3 ∆)-frugal.
This talk
Every graph has a (∆ + 1)-coloring that is O(
log2 ∆log log ∆
)-frugal.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 28: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/28.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Upper Bounds
Hind, Molloy, and Reed (1997): Every graph has a(∆ + 1)-coloring that is O(log5 ∆)-frugal.
This proof can be tightened to yield a (∆ + 1)-coloring thatis O(log3 ∆)-frugal.
This talk
Every graph has a (∆ + 1)-coloring that is O(
log2 ∆log log ∆
)-frugal.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 29: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/29.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Key Elements
Iterative RCP is used, with O(log ∆) iterations.
The symmetry-breaker π is chosen arbitrarily.
Vertices have associated activation probabilities: a vertexv has activation probability av ; it goes to sleep in aniteration with probability 1 − av .
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 30: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/30.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Key Elements
Iterative RCP is used, with O(log ∆) iterations.
The symmetry-breaker π is chosen arbitrarily.
Vertices have associated activation probabilities: a vertexv has activation probability av ; it goes to sleep in aniteration with probability 1 − av .
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 31: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/31.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Key Elements
Iterative RCP is used, with O(log ∆) iterations.
The symmetry-breaker π is chosen arbitrarily.
Vertices have associated activation probabilities: a vertexv has activation probability av ; it goes to sleep in aniteration with probability 1 − av .
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 32: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/32.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
More Details: An Iteration
Initially, each vertex v has palette P(v) = 1, 2, . . . ,∆ + 1.1 Vertex v picks a tentative color uniformly at random from
P(v).2 Vertex v computes its activation probability av and dozes
off with probability 1 − av (independently).3 A vertex that is awake examines the tentative colors picked
by lower ranked neighbors.4 If there is a “color conflict,” the vertex uncolors itself;
otherwise the vertex makes it color permanent.5 Vertex v deletes from P(v) all colors that became
permanent at neighbors during this iteration
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 33: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/33.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
More Details: An Iteration
Initially, each vertex v has palette P(v) = 1, 2, . . . ,∆ + 1.1 Vertex v picks a tentative color uniformly at random from
P(v).2 Vertex v computes its activation probability av and dozes
off with probability 1 − av (independently).3 A vertex that is awake examines the tentative colors picked
by lower ranked neighbors.4 If there is a “color conflict,” the vertex uncolors itself;
otherwise the vertex makes it color permanent.5 Vertex v deletes from P(v) all colors that became
permanent at neighbors during this iteration
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 34: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/34.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
More Details: An Iteration
Initially, each vertex v has palette P(v) = 1, 2, . . . ,∆ + 1.1 Vertex v picks a tentative color uniformly at random from
P(v).2 Vertex v computes its activation probability av and dozes
off with probability 1 − av (independently).3 A vertex that is awake examines the tentative colors picked
by lower ranked neighbors.4 If there is a “color conflict,” the vertex uncolors itself;
otherwise the vertex makes it color permanent.5 Vertex v deletes from P(v) all colors that became
permanent at neighbors during this iteration
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 35: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/35.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
More Details: An Iteration
Initially, each vertex v has palette P(v) = 1, 2, . . . ,∆ + 1.1 Vertex v picks a tentative color uniformly at random from
P(v).2 Vertex v computes its activation probability av and dozes
off with probability 1 − av (independently).3 A vertex that is awake examines the tentative colors picked
by lower ranked neighbors.4 If there is a “color conflict,” the vertex uncolors itself;
otherwise the vertex makes it color permanent.5 Vertex v deletes from P(v) all colors that became
permanent at neighbors during this iteration
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 36: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/36.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
More Details: An Iteration
Initially, each vertex v has palette P(v) = 1, 2, . . . ,∆ + 1.1 Vertex v picks a tentative color uniformly at random from
P(v).2 Vertex v computes its activation probability av and dozes
off with probability 1 − av (independently).3 A vertex that is awake examines the tentative colors picked
by lower ranked neighbors.4 If there is a “color conflict,” the vertex uncolors itself;
otherwise the vertex makes it color permanent.5 Vertex v deletes from P(v) all colors that became
permanent at neighbors during this iteration
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 37: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/37.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Overview of the Proof
Things to prove for each iteration:The iteration is frugal , i.e., each color is used at mostO( log ∆
log log ∆) times.
Sufficient progress is made in the iteration, i.e., maximumdegree shrinks by a constant factor.
Note that |P(v)| ≥ degree(v) + 1.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 38: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/38.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Overview of the Proof
Things to prove for each iteration:The iteration is frugal , i.e., each color is used at mostO( log ∆
log log ∆) times.
Sufficient progress is made in the iteration, i.e., maximumdegree shrinks by a constant factor.
Note that |P(v)| ≥ degree(v) + 1.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 39: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/39.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Overview of the Proof
Things to prove for each iteration:The iteration is frugal , i.e., each color is used at mostO( log ∆
log log ∆) times.
Sufficient progress is made in the iteration, i.e., maximumdegree shrinks by a constant factor.
Note that |P(v)| ≥ degree(v) + 1.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 40: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/40.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Palette Size Needs Controlling
With high probability, abouthalf of v ’s neighbors will becolored 1.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 41: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/41.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Overview of the Proof
Things to prove for each iteration:The iteration is frugal , i.e., each color is used at mostO( log ∆
log log ∆) times.
Sufficient progress is made in the iteration, i.e., maximumdegree shrinks by a constant factor.
Palette size does not shrink too much relative to ∆.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 42: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/42.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Overview of the Proof
Things to prove for each iteration:The iteration is frugal , i.e., each color is used at mostO( log ∆
log log ∆) times.
Sufficient progress is made in the iteration, i.e., maximumdegree shrinks by a constant factor.
Palette size does not shrink too much relative to ∆.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 43: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/43.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Overview of the Proof
Things to prove for each iteration:The iteration is frugal , i.e., each color is used at mostO( log ∆
log log ∆) times.
Sufficient progress is made in the iteration, i.e., maximumdegree shrinks by a constant factor.
Palette size does not shrink too much relative to ∆.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 44: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/44.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Main Theorem
TheoremSuppose at the beginning of an iteration, we have for all v
|P(v)| ≥ degree(v) + 1 |P(v)| ≥ ∆/4.
Then, for some C ⊆ V, there is a proper list coloring of G[C]such that
1 At most 9 log ∆log log ∆ vertices share a color in any neighborhood.
2 degree C(v) ≤ ∆ ·(1 − 1
e5
)+ 27 ·
√∆ log ∆ for all v
3 degreeC(v) ≤ ∆ · 1e5 + 27 ·
√∆ log ∆ for all v .
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 45: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/45.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Proof of Frugality
Pr[u is tentatively colored x ] =1
|P(u)|.
E[|neighbors of v tentatively colored x |
]=
∑u∈N(v)
1|P(u)|
≤ ∆
∆/4≤ 4.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 46: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/46.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Proof of Frugality
Since choice of tentative colors is independent, via ChernoffBounds we get
Pr[|neighbors of v tentatively colored x | > 9 log ∆
log log ∆
]<
1∆6 .
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 47: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/47.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Proof of Progress
L(v , x) = set of lower ranked neighbors of v containing x intheir palette. Pr[v is permanently colored] is
av ·∑
x∈P(v)
1|P(v)|
·
∏u∈L(v ,x)
(1 − 1
|P(u)|
)︸ ︷︷ ︸
qv
.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 48: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/48.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Understanding qv
qv =∑
x∈P(v)
1|P(v)|
·
∏u∈L(v ,x)
(1 − 1
|P(u)|
) .
This achieves minimum at
∑x∈P(v)
1|P(v)|
·(
1 − 1∆/4
)∆
>∑
x∈P(v)
1|P(v)|
· e−5 =1e5 .
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 49: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/49.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Understanding qv
qv =∑
x∈P(v)
1|P(v)|
·
∏u∈L(v ,x)
(1 − 1
|P(u)|
) .
This can be as high as
∑x∈P(v)
1|P(v)|
· 1 = 1.
For example when L(v , x) = ∅.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 50: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/50.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
The Choice of av
RecallPr[v is permanently colored] = av · qv .
Now set av = 1qv ·e5 . Then,
av ≤ 1 (since qv ≥ 1e5 ) and
Pr[v is permanently colored] = 1e5 .
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 51: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/51.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
The Choice of av
RecallPr[v is permanently colored] = av · qv .
Now set av = 1qv ·e5 . Then,
av ≤ 1 (since qv ≥ 1e5 ) and
Pr[v is permanently colored] = 1e5 .
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 52: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/52.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
The Choice of av
RecallPr[v is permanently colored] = av · qv .
Now set av = 1qv ·e5 . Then,
av ≤ 1 (since qv ≥ 1e5 ) and
Pr[v is permanently colored] = 1e5 .
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 53: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/53.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
The Choice of av
RecallPr[v is permanently colored] = av · qv .
Now set av = 1qv ·e5 . Then,
av ≤ 1 (since qv ≥ 1e5 ) and
Pr[v is permanently colored] = 1e5 .
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 54: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/54.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Results in Expectation
E [ |neighbors of v permanently colored| ] =degree(v)
e5 .
We expect v to have
degree(v)
e5 ≤ ∆
e5
colored neighbors and(1 − 1
e5
)· degree(v) ≤
(1 − 1
e5
)·∆
uncolored neighbors.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 55: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/55.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Azuma’s Martingale Inequality
LemmaLet X be a random variable determined by n trialsT1, T2, . . . , Tn, such that for each i, and any two possiblesequences of outcomes t1, t2, . . . , ti−1, ti and t1, t2, . . . , ti−1, ti ′:∣∣∣E [X |T1 = t1, . . . , Ti = ti ]− E [X |T1 = t1, . . . , Ti = ti ′]
∣∣∣ ≤ ci (3)
thenPr
[|X − E [X ]| > t
]≤ 2e−t2/(2
Pc2
i ). (4)
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 56: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/56.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Applying Azuma’s Inequality
X = number of neighbors of v that are permanently colored.Each trial Ti is the choice of
tentative color andwhether to stay awake
at a vertex u in the 2-neighborhood of v .
Number of trials can be as large as ∆2.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 57: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/57.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Applying Azuma’s Inequality
Temporarily ignore symmetry-breaking.Arbitrarily, order and label vertices in the 2-neighborhoodof v : u1, u2, . . . , uk .Ti is the choices made at ui .Target for t ∼ c ·
√∆ log ∆. So
e−t2/(2P
c2i ) = e−c2∆ log ∆/(2
Pc2
i ) = ∆−c2∆/(2P
c2i ).
If we can show that∑
c2i = O(∆), we would get a 1
∆5
upper bound, by choosing constant c large enough.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 58: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/58.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Applying Azuma’s Inequality
Temporarily ignore symmetry-breaking.Arbitrarily, order and label vertices in the 2-neighborhoodof v : u1, u2, . . . , uk .Ti is the choices made at ui .Target for t ∼ c ·
√∆ log ∆. So
e−t2/(2P
c2i ) = e−c2∆ log ∆/(2
Pc2
i ) = ∆−c2∆/(2P
c2i ).
If we can show that∑
c2i = O(∆), we would get a 1
∆5
upper bound, by choosing constant c large enough.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 59: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/59.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Applying Azuma’s Inequality
Temporarily ignore symmetry-breaking.Arbitrarily, order and label vertices in the 2-neighborhoodof v : u1, u2, . . . , uk .Ti is the choices made at ui .Target for t ∼ c ·
√∆ log ∆. So
e−t2/(2P
c2i ) = e−c2∆ log ∆/(2
Pc2
i ) = ∆−c2∆/(2P
c2i ).
If we can show that∑
c2i = O(∆), we would get a 1
∆5
upper bound, by choosing constant c large enough.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 60: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/60.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Applying Azuma’s Inequality
Temporarily ignore symmetry-breaking.Arbitrarily, order and label vertices in the 2-neighborhoodof v : u1, u2, . . . , uk .Ti is the choices made at ui .Target for t ∼ c ·
√∆ log ∆. So
e−t2/(2P
c2i ) = e−c2∆ log ∆/(2
Pc2
i ) = ∆−c2∆/(2P
c2i ).
If we can show that∑
c2i = O(∆), we would get a 1
∆5
upper bound, by choosing constant c large enough.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 61: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/61.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Applying Azuma’s Inequality
Recall,∣∣∣E [X |T1 = t1, . . . , Ti = ti ]− E [X |T1 = t1, . . . , Ti = ti ′]∣∣∣ ≤ ci
Note that∑
c2i may have ∆2 terms. To show
∑c2
i = O(∆), it is
not enough to show to show ci = O(1).
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 62: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/62.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Upper Bound on Expected Difference
We want to upper bound:∣∣∣E [X |T1 = t1, . . . , Ti = ti ]− E [X |T1 = t1, . . . , Ti = ti ′]∣∣∣
Bad example: expecteddifference can be as much asdegree(v).
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 63: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/63.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Symmetry-Breaking in Action
Order 2-neighborhood vertices by increasing rank in π.Reconsider∣∣∣E [X |T1 = t1, . . . , Ti = ti ]− E [X |T1 = t1, . . . , Ti = ti ′]
∣∣∣Changing T = ti to Ti = t ′i cannot affect verticesu1, u2, . . . , ui−1.It can only affect vertices ui+1, ui+2, . . . , un, but thesevertices affect X only in expectation.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 64: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/64.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Symmetry-Breaking in Action
Order 2-neighborhood vertices by increasing rank in π.Reconsider∣∣∣E [X |T1 = t1, . . . , Ti = ti ]− E [X |T1 = t1, . . . , Ti = ti ′]
∣∣∣Changing T = ti to Ti = t ′i cannot affect verticesu1, u2, . . . , ui−1.It can only affect vertices ui+1, ui+2, . . . , un, but thesevertices affect X only in expectation.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 65: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/65.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Symmetry-Breaking in Action
Order 2-neighborhood vertices by increasing rank in π.Reconsider∣∣∣E [X |T1 = t1, . . . , Ti = ti ]− E [X |T1 = t1, . . . , Ti = ti ′]
∣∣∣Changing T = ti to Ti = t ′i cannot affect verticesu1, u2, . . . , ui−1.It can only affect vertices ui+1, ui+2, . . . , un, but thesevertices affect X only in expectation.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 66: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/66.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Concentration Result
Using these observations we show that∑
c2i ≤ 81 ·∆.
LemmaLet Pv denote the number of neighbors of v that arepermanently colored in this iteration.
Pr[∣∣∣Pv −
degree(v)
e5
∣∣∣ > 27 ·√
∆ log ∆
]<
2∆4.5 .
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 67: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/67.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Proof Sketch
Si = uj | j > i and uj is adjacent to both v and ui.
These are the only vertices that are affected by the changein the tentative color at ui from ti to t ′i and can in turn affectX .
But only for those uj ∈ Si that tentatively choose color ti ort ′i .
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 68: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/68.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Proof Sketch
Si = uj | j > i and uj is adjacent to both v and ui.
These are the only vertices that are affected by the changein the tentative color at ui from ti to t ′i and can in turn affectX .
But only for those uj ∈ Si that tentatively choose color ti ort ′i .
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 69: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/69.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Proof Sketch
Si = uj | j > i and uj is adjacent to both v and ui.
These are the only vertices that are affected by the changein the tentative color at ui from ti to t ′i and can in turn affectX .
But only for those uj ∈ Si that tentatively choose color ti ort ′i .
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 70: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/70.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Proof Sketch
Pr[uj chooses tj or t ′j ] =2
|P(uj)|≤ 8
∆.
Therefore, by linearity of expectation, the expected numberof vertices in Si that choose ti or t ′i is at most |Si |/8∆.
Expected difference is bounded above by ci = |Si |/8∆.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 71: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/71.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Proof Sketch
Pr[uj chooses tj or t ′j ] =2
|P(uj)|≤ 8
∆.
Therefore, by linearity of expectation, the expected numberof vertices in Si that choose ti or t ′i is at most |Si |/8∆.
Expected difference is bounded above by ci = |Si |/8∆.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 72: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/72.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Proof Sketch
Pr[uj chooses tj or t ′j ] =2
|P(uj)|≤ 8
∆.
Therefore, by linearity of expectation, the expected numberof vertices in Si that choose ti or t ′i is at most |Si |/8∆.
Expected difference is bounded above by ci = |Si |/8∆.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 73: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/73.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Proof Sketch
We use inequalities:1 |Si | ≤ ∆
2∑
i |Si | ≤ ∆2
to get ∑i
|Si |2 ≤ ∆ · (∑
i
|Si |) ≤ ∆3.
Hence, ∑i
c2i ≤
∑i
(8|Si |∆
)2
≤ 64∆.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 74: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/74.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Final Step: Apply LLL
Let Bv denote the “bad event” thatfor some color x , vertex v has more than 9 log ∆
log log ∆ neighborscolored x or|Pv − degree(v)
e4 | > 27 ·√
∆ log ∆.
From the concentration results we get
Pr[Bv ] <(∆ + 1)
∆6 +2
∆4.5 ≤ 4∆4.5 .
Each Bv is mutually independent of all except ∆4 other bad
events.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 75: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/75.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Final Step: Apply LLL
Let Bv denote the “bad event” thatfor some color x , vertex v has more than 9 log ∆
log log ∆ neighborscolored x or|Pv − degree(v)
e4 | > 27 ·√
∆ log ∆.
From the concentration results we get
Pr[Bv ] <(∆ + 1)
∆6 +2
∆4.5 ≤ 4∆4.5 .
Each Bv is mutually independent of all except ∆4 other bad
events.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 76: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/76.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Final Step: Apply LLL
Let Bv denote the “bad event” thatfor some color x , vertex v has more than 9 log ∆
log log ∆ neighborscolored x or|Pv − degree(v)
e4 | > 27 ·√
∆ log ∆.
From the concentration results we get
Pr[Bv ] <(∆ + 1)
∆6 +2
∆4.5 ≤ 4∆4.5 .
Each Bv is mutually independent of all except ∆4 other bad
events.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 77: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/77.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Final Step: Apply LLL
Let Bv denote the “bad event” thatfor some color x , vertex v has more than 9 log ∆
log log ∆ neighborscolored x or|Pv − degree(v)
e4 | > 27 ·√
∆ log ∆.
From the concentration results we get
Pr[Bv ] <(∆ + 1)
∆6 +2
∆4.5 ≤ 4∆4.5 .
Each Bv is mutually independent of all except ∆4 other bad
events.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 78: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/78.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Other Applications of RCP with Symmetry-Breaking
Frugal colorings of special classes of graphs: d-inductivegraphs, sparse graphs.
Weighted equitable colorings.
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 79: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/79.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Last Slide
This is joint work with Aravind Srinivasan at University ofMaryland.
Thanks to Rajiv Raman for arranging my visit and talk.
Any questions?
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 80: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/80.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Last Slide
This is joint work with Aravind Srinivasan at University ofMaryland.
Thanks to Rajiv Raman for arranging my visit and talk.
Any questions?
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking
![Page 81: Randomized Coloring with Symmetry Breaking › ~sriram › talks › randomColoring.pdf · Each vertex v picks a tentative color uniformly at random from its palette. REPAIR STEP](https://reader034.vdocuments.site/reader034/viewer/2022042402/5f1120392b83bf29922cb9cc/html5/thumbnails/81.jpg)
IntroductionFrugal ColoringMain Theorem
Showing Concentration
Last Slide
This is joint work with Aravind Srinivasan at University ofMaryland.
Thanks to Rajiv Raman for arranging my visit and talk.
Any questions?
Sriram V. Pemmaraju Randomized Coloring with Symmetry Breaking