covering graphs using trees and starshyde.eng.tau.ac.il/talks/k_tree_cover.pdfcover the vertices of...
TRANSCRIPT
![Page 1: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/1.jpg)
Covering GraphsUsing
Trees and StarsGuy Even (Tel-Aviv), Naveen Garg (Delhi),
and
Jochen Konemann, R. Ravi and A. Sinha (Pittsburgh)
Third Haifa Workshop on Interdisciplinary Applications of Graph Theory, Combinatorics
and Computing
– p.1
![Page 2: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/2.jpg)
Motivation
Consider a TSP instance with a largeoptimal tour, e.g. w(tour∗) > 1000.
Suppose regulation dictates: agent maytravel at most 100 km.
⇒ must employ multiple agents.
⇒ k-traveling salespeople problem.Cover the vertices of a graph with k
tours.Balance the load of the agents:minimize the maximum tour.
MST is a constant ratio approx of a mintour ⇒ k-Tree Cover Problem.
– p.2
![Page 3: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/3.jpg)
Motivation
Consider a TSP instance with a largeoptimal tour, e.g. w(tour∗) > 1000.
Suppose regulation dictates: agent maytravel at most 100 km.
⇒ must employ multiple agents.
⇒ k-traveling salespeople problem.Cover the vertices of a graph with k
tours.Balance the load of the agents:minimize the maximum tour.
MST is a constant ratio approx of a mintour ⇒ k-Tree Cover Problem.
– p.2
![Page 4: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/4.jpg)
Motivation
Consider a TSP instance with a largeoptimal tour, e.g. w(tour∗) > 1000.
Suppose regulation dictates: agent maytravel at most 100 km.
⇒ must employ multiple agents.
⇒ k-traveling salespeople problem.Cover the vertices of a graph with k
tours.Balance the load of the agents:minimize the maximum tour.
MST is a constant ratio approx of a mintour ⇒ k-Tree Cover Problem.
– p.2
![Page 5: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/5.jpg)
Motivation
Consider a TSP instance with a largeoptimal tour, e.g. w(tour∗) > 1000.
Suppose regulation dictates: agent maytravel at most 100 km.
⇒ must employ multiple agents.
⇒ k-traveling salespeople problem.
Cover the vertices of a graph with k
tours.Balance the load of the agents:minimize the maximum tour.
MST is a constant ratio approx of a mintour ⇒ k-Tree Cover Problem.
– p.2
![Page 6: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/6.jpg)
Motivation
Consider a TSP instance with a largeoptimal tour, e.g. w(tour∗) > 1000.
Suppose regulation dictates: agent maytravel at most 100 km.
⇒ must employ multiple agents.
⇒ k-traveling salespeople problem.Cover the vertices of a graph with k
tours.
Balance the load of the agents:minimize the maximum tour.
MST is a constant ratio approx of a mintour ⇒ k-Tree Cover Problem.
– p.2
![Page 7: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/7.jpg)
Motivation
Consider a TSP instance with a largeoptimal tour, e.g. w(tour∗) > 1000.
Suppose regulation dictates: agent maytravel at most 100 km.
⇒ must employ multiple agents.
⇒ k-traveling salespeople problem.Cover the vertices of a graph with k
tours.Balance the load of the agents:minimize the maximum tour.
MST is a constant ratio approx of a mintour ⇒ k-Tree Cover Problem.
– p.2
![Page 8: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/8.jpg)
Motivation
Consider a TSP instance with a largeoptimal tour, e.g. w(tour∗) > 1000.
Suppose regulation dictates: agent maytravel at most 100 km.
⇒ must employ multiple agents.
⇒ k-traveling salespeople problem.Cover the vertices of a graph with k
tours.Balance the load of the agents:minimize the maximum tour.
MST is a constant ratio approx of a mintour ⇒ k-Tree Cover Problem.
– p.2
![Page 9: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/9.jpg)
k-Tree Cover Problem
Input: (i) integer k and (ii) G = (V,E) - anundirected graph with positive integraledge weights w : E → IN+.
k-tree cover: a set T of trees {Ti}i such thatV =
⋃ki=1 V (Ti).
cost : cost(T ) = maxTi∈T w(Ti).
goal : find a minimum cost k-tree cover.
remark : trees may share nodes & edges in atree cover.
– p.3
![Page 10: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/10.jpg)
k-Tree Cover Problem
Input: (i) integer k and (ii) G = (V,E) - anundirected graph with positive integraledge weights w : E → IN+.
k-tree cover: a set T of trees {Ti}i such thatV =
⋃ki=1 V (Ti).
cost : cost(T ) = maxTi∈T w(Ti).
goal : find a minimum cost k-tree cover.
remark : trees may share nodes & edges in atree cover.
– p.3
![Page 11: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/11.jpg)
k-Tree Cover Problem
Input: (i) integer k and (ii) G = (V,E) - anundirected graph with positive integraledge weights w : E → IN+.
k-tree cover: a set T of trees {Ti}i such thatV =
⋃ki=1 V (Ti).
cost : cost(T ) = maxTi∈T w(Ti).
goal : find a minimum cost k-tree cover.
remark : trees may share nodes & edges in atree cover.
– p.3
![Page 12: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/12.jpg)
k-Tree Cover Problem
Input: (i) integer k and (ii) G = (V,E) - anundirected graph with positive integraledge weights w : E → IN+.
k-tree cover: a set T of trees {Ti}i such thatV =
⋃ki=1 V (Ti).
cost : cost(T ) = maxTi∈T w(Ti).
goal : find a minimum cost k-tree cover.
remark : trees may share nodes & edges in atree cover.
cost=9
12
2
8
51 2
112
5
– p.3
![Page 13: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/13.jpg)
k-Tree Cover Problem
Input: (i) integer k and (ii) G = (V,E) - anundirected graph with positive integraledge weights w : E → IN+.
k-tree cover: a set T of trees {Ti}i such thatV =
⋃ki=1 V (Ti).
cost : cost(T ) = maxTi∈T w(Ti).
goal : find a minimum cost k-tree cover.
remark : trees may share nodes & edges in atree cover.
cost=9
12
2
8
51 2
112
5
– p.3
![Page 14: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/14.jpg)
k-Tree Cover Problem
Input: (i) integer k and (ii) G = (V,E) - anundirected graph with positive integraledge weights w : E → IN+.
k-tree cover: a set T of trees {Ti}i such thatV =
⋃ki=1 V (Ti).
cost : cost(T ) = maxTi∈T w(Ti).
goal : find a minimum cost k-tree cover.
remark : trees may share nodes & edges in atree cover.
– p.3
![Page 15: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/15.jpg)
k-Tree Cover Problem: the rooted case
Roots: Input contains also a set of roots:
R = {r1, r2, . . . , rk}.
k-rooted tree cover: a k-tree cover {Ti}i suchthat
∀i : ri ∈ Ti.
motivation : agents start their tour in differentlocations.
– p.4
![Page 16: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/16.jpg)
k-Tree Cover Problem: the rooted case
Roots: Input contains also a set of roots:
R = {r1, r2, . . . , rk}.
k-rooted tree cover: a k-tree cover {Ti}i suchthat
∀i : ri ∈ Ti.
motivation : agents start their tour in differentlocations.
roots
– p.4
![Page 17: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/17.jpg)
k-Tree Cover Problem: the rooted case
Roots: Input contains also a set of roots:
R = {r1, r2, . . . , rk}.
k-rooted tree cover: a k-tree cover {Ti}i suchthat
∀i : ri ∈ Ti.
motivation : agents start their tour in differentlocations.
roots
– p.4
![Page 18: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/18.jpg)
k-Tree Cover Problem: the rooted case
Roots: Input contains also a set of roots:
R = {r1, r2, . . . , rk}.
k-rooted tree cover: a k-tree cover {Ti}i suchthat
∀i : ri ∈ Ti.
motivation : agents start their tour in differentlocations.
– p.4
![Page 19: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/19.jpg)
k-Tree Cover Problem: the rooted case
Roots: Input contains also a set of roots:
R = {r1, r2, . . . , rk}.
k-rooted tree cover: a k-tree cover {Ti}i suchthat
∀i : ri ∈ Ti.
motivation : agents start their tour in differentlocations.
– p.4
![Page 20: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/20.jpg)
k-Tree Cover Problem: the rooted case
Roots: Input contains also a set of roots:
R = {r1, r2, . . . , rk}.
k-rooted tree cover: a k-tree cover {Ti}i suchthat
∀i : ri ∈ Ti.
motivation : agents start their tour in differentlocations.
– p.4
![Page 21: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/21.jpg)
k-Tree Cover Problem: the rooted case
Roots: Input contains also a set of roots:
R = {r1, r2, . . . , rk}.
k-rooted tree cover: a k-tree cover {Ti}i suchthat
∀i : ri ∈ Ti.
motivation : agents start their tour in differentlocations.
– p.4
![Page 22: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/22.jpg)
Star Covers
k-Star Cover: a k-tree cover {Ti}i in which
∀i : Ti is a star
rooted/unrooted versions: in rooted version,stars must be rooted at R. Namely, ri isthe root of Ti.
motivation : agents must return to base aftereach visit.
– p.5
![Page 23: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/23.jpg)
Star Covers
k-Star Cover: a k-tree cover {Ti}i in which
∀i : Ti is a star
rooted/unrooted versions: in rooted version,stars must be rooted at R. Namely, ri isthe root of Ti.
motivation : agents must return to base aftereach visit.
– p.5
![Page 24: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/24.jpg)
Star Covers
k-Star Cover: a k-tree cover {Ti}i in which
∀i : Ti is a star
rooted/unrooted versions: in rooted version,stars must be rooted at R. Namely, ri isthe root of Ti.
motivation : agents must return to base aftereach visit.
– p.5
![Page 25: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/25.jpg)
Star Covers
k-Star Cover: a k-tree cover {Ti}i in which
∀i : Ti is a star
rooted/unrooted versions: in rooted version,stars must be rooted at R. Namely, ri isthe root of Ti.
motivation : agents must return to base aftereach visit.
– p.5
![Page 26: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/26.jpg)
Star Covers
k-Star Cover: a k-tree cover {Ti}i in which
∀i : Ti is a star
rooted/unrooted versions: in rooted version,stars must be rooted at R. Namely, ri isthe root of Ti.
motivation : agents must return to base aftereach visit.
– p.5
![Page 27: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/27.jpg)
Star Covers
k-Star Cover: a k-tree cover {Ti}i in which
∀i : Ti is a star
rooted/unrooted versions: in rooted version,stars must be rooted at R. Namely, ri isthe root of Ti.
motivation : agents must return to base aftereach visit.
– p.5
![Page 28: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/28.jpg)
Star Covers
k-Star Cover: a k-tree cover {Ti}i in which
∀i : Ti is a star
rooted/unrooted versions: in rooted version,stars must be rooted at R. Namely, ri isthe root of Ti.
motivation : agents must return to base aftereach visit.
– p.5
![Page 29: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/29.jpg)
Star Covers
k-Star Cover: a k-tree cover {Ti}i in which
∀i : Ti is a star
rooted/unrooted versions: in rooted version,stars must be rooted at R. Namely, ri isthe root of Ti.
motivation : agents must return to base aftereach visit.
– p.5
![Page 30: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/30.jpg)
Star Covers
k-Star Cover: a k-tree cover {Ti}i in which
∀i : Ti is a star
rooted/unrooted versions: in rooted version,stars must be rooted at R. Namely, ri isthe root of Ti.
motivation : agents must return to base aftereach visit.
– p.5
![Page 31: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/31.jpg)
Related work
k-Traveling Repairman: Cover with tours, O(1)-approxminimize average latency. [Fakcharoenphol, Harrelson,Rao 2003]
k-Traveling Salesman: Cover with tours, O(1)-approxminimize total length. [Haimovich, Rinooy Kan, Stougie1988]
Vehicle Routing: Vast amount of work, e.g. Survey[Toth, Vigo, 2002]
Clustering is like covering with stars: Minimizemaximum edge - k center [Dyer, Frieze, 1985], Minimizesum of edge lengths k median [Arya, et al 2001],Minimize sum of star radii [Charikar, Panigrahy, 2001].
– p.6
![Page 32: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/32.jpg)
Related work
k-Traveling Repairman: Cover with tours, O(1)-approxminimize average latency. [Fakcharoenphol, Harrelson,Rao 2003]
k-Traveling Salesman: Cover with tours, O(1)-approxminimize total length. [Haimovich, Rinooy Kan, Stougie1988]
Vehicle Routing: Vast amount of work, e.g. Survey[Toth, Vigo, 2002]
Clustering is like covering with stars: Minimizemaximum edge - k center [Dyer, Frieze, 1985], Minimizesum of edge lengths k median [Arya, et al 2001],Minimize sum of star radii [Charikar, Panigrahy, 2001].
– p.6
![Page 33: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/33.jpg)
Related work
k-Traveling Repairman: Cover with tours, O(1)-approxminimize average latency. [Fakcharoenphol, Harrelson,Rao 2003]
k-Traveling Salesman: Cover with tours, O(1)-approxminimize total length. [Haimovich, Rinooy Kan, Stougie1988]
Vehicle Routing: Vast amount of work, e.g. Survey[Toth, Vigo, 2002]
Clustering is like covering with stars: Minimizemaximum edge - k center [Dyer, Frieze, 1985], Minimizesum of edge lengths k median [Arya, et al 2001],Minimize sum of star radii [Charikar, Panigrahy, 2001].
– p.6
![Page 34: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/34.jpg)
Related work - cont
Chandra Chekuri & Amit Kumar - similar results.
Arkin, Hassin, & Levin - approx algorithms for manysimilar problems:
O(1)-approx for unrooted k-path cover.O(1)-approx for unrooted B-star cover.many other problems...
– p.7
![Page 35: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/35.jpg)
Related work - cont
Chandra Chekuri & Amit Kumar - similar results.
Arkin, Hassin, & Levin - approx algorithms for manysimilar problems:
O(1)-approx for unrooted k-path cover.O(1)-approx for unrooted B-star cover.many other problems...
– p.7
![Page 36: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/36.jpg)
Related work - cont
Chandra Chekuri & Amit Kumar - similar results.
Arkin, Hassin, & Levin - approx algorithms for manysimilar problems:
O(1)-approx for unrooted k-path cover.
O(1)-approx for unrooted B-star cover.many other problems...
– p.7
![Page 37: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/37.jpg)
Related work - cont
Chandra Chekuri & Amit Kumar - similar results.
Arkin, Hassin, & Levin - approx algorithms for manysimilar problems:
O(1)-approx for unrooted k-path cover.O(1)-approx for unrooted B-star cover.
many other problems...
– p.7
![Page 38: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/38.jpg)
Related work - cont
Chandra Chekuri & Amit Kumar - similar results.
Arkin, Hassin, & Levin - approx algorithms for manysimilar problems:
O(1)-approx for unrooted k-path cover.O(1)-approx for unrooted B-star cover.many other problems...
– p.7
![Page 39: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/39.jpg)
Results
Hardness: All 4 problems are NP-complete. (reductionfrom Bin-Packing).
k-tree cover: 4-approximation algorithm. Stronglypolynomial versions are (4 + ε)-approx.
k-star cover:Unrooted version: (4, 4)-bicriteria approximationalgorithm (i.e. 4k stars of cost 4 · OPTk). Extendmethod of [Shmoys, Tardos, & Aardal, 1997] forcapacitated facility location.Rooted version: equivalent to min. makespan of k
machines and n jobs. 2-approximation of [Shmoys &Tardos, 1993].
– p.8
![Page 40: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/40.jpg)
Results
Hardness: All 4 problems are NP-complete. (reductionfrom Bin-Packing).
k-tree cover: 4-approximation algorithm. Stronglypolynomial versions are (4 + ε)-approx.
k-star cover:Unrooted version: (4, 4)-bicriteria approximationalgorithm (i.e. 4k stars of cost 4 · OPTk). Extendmethod of [Shmoys, Tardos, & Aardal, 1997] forcapacitated facility location.Rooted version: equivalent to min. makespan of k
machines and n jobs. 2-approximation of [Shmoys &Tardos, 1993].
– p.8
![Page 41: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/41.jpg)
Results
Hardness: All 4 problems are NP-complete. (reductionfrom Bin-Packing).
k-tree cover: 4-approximation algorithm. Stronglypolynomial versions are (4 + ε)-approx.
k-star cover:
Unrooted version: (4, 4)-bicriteria approximationalgorithm (i.e. 4k stars of cost 4 · OPTk). Extendmethod of [Shmoys, Tardos, & Aardal, 1997] forcapacitated facility location.Rooted version: equivalent to min. makespan of k
machines and n jobs. 2-approximation of [Shmoys &Tardos, 1993].
– p.8
![Page 42: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/42.jpg)
Results
Hardness: All 4 problems are NP-complete. (reductionfrom Bin-Packing).
k-tree cover: 4-approximation algorithm. Stronglypolynomial versions are (4 + ε)-approx.
k-star cover:Unrooted version: (4, 4)-bicriteria approximationalgorithm (i.e. 4k stars of cost 4 · OPTk). Extendmethod of [Shmoys, Tardos, & Aardal, 1997] forcapacitated facility location.
Rooted version: equivalent to min. makespan of k
machines and n jobs. 2-approximation of [Shmoys &Tardos, 1993].
– p.8
![Page 43: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/43.jpg)
Results
Hardness: All 4 problems are NP-complete. (reductionfrom Bin-Packing).
k-tree cover: 4-approximation algorithm. Stronglypolynomial versions are (4 + ε)-approx.
k-star cover:Unrooted version: (4, 4)-bicriteria approximationalgorithm (i.e. 4k stars of cost 4 · OPTk). Extendmethod of [Shmoys, Tardos, & Aardal, 1997] forcapacitated facility location.Rooted version: equivalent to min. makespan of k
machines and n jobs. 2-approximation of [Shmoys &Tardos, 1993].
– p.8
![Page 44: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/44.jpg)
approximation algorithm: k-rooted tree coverInput: graph, roots, and B - “guess” of opt. cost.
1. contract roots.
2. compute MST.
3. un-contract roots: forest of treesrooted at roots.
4. edge-decompose trees:w(subtrees) ∈ [B, 2B),w(leftovers) < B.
5. max match subtrees to roots (ifdist ≤ B).
6. if not all subtrees are matched⇒B < B∗.
7. else return ∀ri: leftover +matched subtree.
– p.9
![Page 45: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/45.jpg)
approximation algorithm: k-rooted tree coverInput: graph, roots, and B - “guess” of opt. cost.
1. contract roots.
2. compute MST.
3. un-contract roots: forest of treesrooted at roots.
4. edge-decompose trees:w(subtrees) ∈ [B, 2B),w(leftovers) < B.
5. max match subtrees to roots (ifdist ≤ B).
6. if not all subtrees are matched⇒B < B∗.
7. else return ∀ri: leftover +matched subtree.
– p.9
![Page 46: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/46.jpg)
approximation algorithm: k-rooted tree coverInput: graph, roots, and B - “guess” of opt. cost.
1. contract roots.
2. compute MST.
3. un-contract roots: forest of treesrooted at roots.
4. edge-decompose trees:w(subtrees) ∈ [B, 2B),w(leftovers) < B.
5. max match subtrees to roots (ifdist ≤ B).
6. if not all subtrees are matched⇒B < B∗.
7. else return ∀ri: leftover +matched subtree.
– p.9
![Page 47: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/47.jpg)
approximation algorithm: k-rooted tree coverInput: graph, roots, and B - “guess” of opt. cost.
1. contract roots.
2. compute MST.
3. un-contract roots: forest of treesrooted at roots.
4. edge-decompose trees:w(subtrees) ∈ [B, 2B),w(leftovers) < B.
5. max match subtrees to roots (ifdist ≤ B).
6. if not all subtrees are matched⇒B < B∗.
7. else return ∀ri: leftover +matched subtree.
– p.9
![Page 48: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/48.jpg)
approximation algorithm: k-rooted tree coverInput: graph, roots, and B - “guess” of opt. cost.
1. contract roots.
2. compute MST.
3. un-contract roots: forest of treesrooted at roots.
4. edge-decompose trees:w(subtrees) ∈ [B, 2B),w(leftovers) < B.
5. max match subtrees to roots (ifdist ≤ B).
6. if not all subtrees are matched⇒B < B∗.
7. else return ∀ri: leftover +matched subtree.
– p.9
![Page 49: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/49.jpg)
approximation algorithm: k-rooted tree coverInput: graph, roots, and B - “guess” of opt. cost.
1. contract roots.
2. compute MST.
3. un-contract roots: forest of treesrooted at roots.
4. edge-decompose trees:w(subtrees) ∈ [B, 2B),w(leftovers) < B.
5. max match subtrees to roots (ifdist ≤ B).
6. if not all subtrees are matched⇒B < B∗.
7. else return ∀ri: leftover +matched subtree.
– p.9
![Page 50: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/50.jpg)
approximation algorithm: k-rooted tree coverInput: graph, roots, and B - “guess” of opt. cost.
1. contract roots.
2. compute MST.
3. un-contract roots: forest of treesrooted at roots.
4. edge-decompose trees:w(subtrees) ∈ [B, 2B),w(leftovers) < B.
5. max match subtrees to roots (ifdist ≤ B).
6. if not all subtrees are matched⇒B < B∗.
7. else return ∀ri: leftover +matched subtree.
– p.9
![Page 51: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/51.jpg)
approximation algorithm: k-rooted tree coverInput: graph, roots, and B - “guess” of opt. cost.
1. contract roots.
2. compute MST.
3. un-contract roots: forest of treesrooted at roots.
4. edge-decompose trees:w(subtrees) ∈ [B, 2B),w(leftovers) < B.
5. max match subtrees to roots (ifdist ≤ B).
6. if not all subtrees are matched⇒B < B∗.
7. else return ∀ri: leftover +matched subtree.
– p.9
![Page 52: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/52.jpg)
approximation algorithm: k-rooted tree coverInput: graph, roots, and B - “guess” of opt. cost.
1. contract roots.
2. compute MST.
3. un-contract roots: forest of treesrooted at roots.
4. edge-decompose trees:w(subtrees) ∈ [B, 2B),w(leftovers) < B.
5. max match subtrees to roots (ifdist ≤ B).
6. if not all subtrees are matched⇒B < B∗.
7. else return ∀ri: leftover +matched subtree.
– p.9
![Page 53: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/53.jpg)
approximation algorithm: k-rooted tree coverInput: graph, roots, and B - “guess” of opt. cost.
1. contract roots.
2. compute MST.
3. un-contract roots: forest of treesrooted at roots.
4. edge-decompose trees:w(subtrees) ∈ [B, 2B),w(leftovers) < B.
5. max match subtrees to roots (ifdist ≤ B).
6. if not all subtrees are matched⇒B < B∗.
7. else return ∀ri: leftover +matched subtree.
– p.9
![Page 54: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/54.jpg)
approximation algorithm: k-rooted tree coverInput: graph, roots, and B - “guess” of opt. cost.
1. contract roots.
2. compute MST.
3. un-contract roots: forest of treesrooted at roots.
4. edge-decompose trees:w(subtrees) ∈ [B, 2B),w(leftovers) < B.
5. max match subtrees to roots (ifdist ≤ B).
6. if not all subtrees are matched⇒B < B∗.
7. else return ∀ri: leftover +matched subtree.
– p.9
![Page 55: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/55.jpg)
approximation algorithm: k-rooted tree coverInput: graph, roots, and B - “guess” of opt. cost.
1. contract roots.
2. compute MST.
3. un-contract roots: forest of treesrooted at roots.
4. edge-decompose trees:w(subtrees) ∈ [B, 2B),w(leftovers) < B.
5. max match subtrees to roots (ifdist ≤ B).
6. if not all subtrees are matched⇒B < B∗.
7. else return ∀ri: leftover +matched subtree.
– p.9
![Page 56: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/56.jpg)
approximation algorithm: k-rooted tree coverInput: graph, roots, and B - “guess” of opt. cost.
1. contract roots.
2. compute MST.
3. un-contract roots: forest of treesrooted at roots.
4. edge-decompose trees:w(subtrees) ∈ [B, 2B),w(leftovers) < B.
5. max match subtrees to roots (ifdist ≤ B).
6. if not all subtrees are matched⇒B < B∗.
7. else return ∀ri: leftover +matched subtree.
– p.9
![Page 57: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/57.jpg)
approximation algorithm: k-rooted tree coverInput: graph, roots, and B - “guess” of opt. cost.
1. contract roots.
2. compute MST.
3. un-contract roots: forest of treesrooted at roots.
4. edge-decompose trees:w(subtrees) ∈ [B, 2B),w(leftovers) < B.
5. max match subtrees to roots (ifdist ≤ B).
6. if not all subtrees are matched⇒B < B∗.
7. else return ∀ri: leftover +matched subtree.
– p.9
![Page 58: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/58.jpg)
4-approx algorithm : k-rooted tree cover
Claim: success ⇒ cost(cover) ≤ 4 · B.
Claim: fail ⇒ B < B∗.
Binary search on value of B ⇒ (weakly) polynomial4-approx algorithm.
Scaling ⇒ strongly polynomial (4 + ε)-approx algorithm.
– p.10
![Page 59: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/59.jpg)
4-approx algorithm : k-rooted tree cover
Claim: success ⇒ cost(cover) ≤ 4 · B.
Claim: fail ⇒ B < B∗.
Binary search on value of B ⇒ (weakly) polynomial4-approx algorithm.
Scaling ⇒ strongly polynomial (4 + ε)-approx algorithm.
– p.10
![Page 60: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/60.jpg)
4-approx algorithm : k-rooted tree cover
Claim: success ⇒ cost(cover) ≤ 4 · B.
Claim: fail ⇒ B < B∗.
Binary search on value of B ⇒ (weakly) polynomial4-approx algorithm.
Scaling ⇒ strongly polynomial (4 + ε)-approx algorithm.
– p.10
![Page 61: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/61.jpg)
4-approx algorithm : k-rooted tree cover
Claim: success ⇒ cost(cover) ≤ 4 · B.
Claim: fail ⇒ B < B∗.
Binary search on value of B ⇒ (weakly) polynomial4-approx algorithm.
Scaling ⇒ strongly polynomial (4 + ε)-approx algorithm.
– p.10
![Page 62: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/62.jpg)
Claim: success ⇒ cost(cover) ≤ 4 · B
Each tree in tree cover may consist of:
a rooted leftover subtree ⇒ cost(leftover) < B.
a matched subtree ⇒ cost(subtree) < 2 · B.
matching edge ⇒ cost(edge) ≤ B.
⇒ weight of every tree in solution is < 4 · B. QED
– p.11
![Page 63: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/63.jpg)
Claim: success ⇒ cost(cover) ≤ 4 · B
Each tree in tree cover may consist of:
a rooted leftover subtree ⇒ cost(leftover) < B.
a matched subtree ⇒ cost(subtree) < 2 · B.
matching edge ⇒ cost(edge) ≤ B.
⇒ weight of every tree in solution is < 4 · B. QED
– p.11
![Page 64: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/64.jpg)
Claim: success ⇒ cost(cover) ≤ 4 · B
Each tree in tree cover may consist of:
a rooted leftover subtree ⇒ cost(leftover) < B.
a matched subtree ⇒ cost(subtree) < 2 · B.
matching edge ⇒ cost(edge) ≤ B.
⇒ weight of every tree in solution is < 4 · B. QED
– p.11
![Page 65: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/65.jpg)
Claim: success ⇒ cost(cover) ≤ 4 · B
Each tree in tree cover may consist of:
a rooted leftover subtree ⇒ cost(leftover) < B.
a matched subtree ⇒ cost(subtree) < 2 · B.
matching edge ⇒ cost(edge) ≤ B.
⇒ weight of every tree in solution is < 4 · B. QED
– p.11
![Page 66: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/66.jpg)
Claim: success ⇒ cost(cover) ≤ 4 · B
Each tree in tree cover may consist of:
a rooted leftover subtree ⇒ cost(leftover) < B.
a matched subtree ⇒ cost(subtree) < 2 · B.
matching edge ⇒ cost(edge) ≤ B.
⇒ weight of every tree in solution is < 4 · B. QED
– p.11
![Page 67: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/67.jpg)
Claim: fail ⇒ B < B∗
Assume for sake of contradiction:
B ≥ B∗ and matching failed.
Hall’s Theorem: if matching failed, then
∃S ⊆ subtrees : |S| > |N(S)|.
Fix OPT:
T ∗ 4
= {T ∗1 , . . . , T ∗
k } where rj ∈ T ∗j .
T ∗(S)4
= {T ∗j | ∃Si ∈ S : T ∗
j ∩ Si 6= ∅}.
Note: T ∗j ∩ Si 6= ∅ ⇒ w(rj, Si) ≤ B∗ ≤ B.
⇒ |T ∗(S)| ≤ |N(S)|.
w(rj , Si) ≤ B
subtrees roots
Sirj
T∗
j
rj
Si
≤ B∗
– p.12
![Page 68: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/68.jpg)
Claim: fail ⇒ B < B∗
Assume for sake of contradiction:
B ≥ B∗ and matching failed.
Hall’s Theorem: if matching failed, then
∃S ⊆ subtrees : |S| > |N(S)|.
Fix OPT:
T ∗ 4
= {T ∗1 , . . . , T ∗
k } where rj ∈ T ∗j .
T ∗(S)4
= {T ∗j | ∃Si ∈ S : T ∗
j ∩ Si 6= ∅}.
Note: T ∗j ∩ Si 6= ∅ ⇒ w(rj, Si) ≤ B∗ ≤ B.
⇒ |T ∗(S)| ≤ |N(S)|.
w(rj , Si) ≤ B
subtrees roots
Sirj
T∗
j
rj
Si
≤ B∗
– p.12
![Page 69: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/69.jpg)
Claim: fail ⇒ B < B∗
Assume for sake of contradiction:
B ≥ B∗ and matching failed.
Hall’s Theorem: if matching failed, then
∃S ⊆ subtrees : |S| > |N(S)|.
Fix OPT:
T ∗ 4
= {T ∗1 , . . . , T ∗
k } where rj ∈ T ∗j .
T ∗(S)4
= {T ∗j | ∃Si ∈ S : T ∗
j ∩ Si 6= ∅}.
Note: T ∗j ∩ Si 6= ∅ ⇒ w(rj, Si) ≤ B∗ ≤ B.
⇒ |T ∗(S)| ≤ |N(S)|.
w(rj , Si) ≤ B
subtrees roots
Sirj
T∗
j
rj
Si
≤ B∗
– p.12
![Page 70: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/70.jpg)
Claim: fail ⇒ B < B∗
Assume for sake of contradiction:
B ≥ B∗ and matching failed.
Hall’s Theorem: if matching failed, then
∃S ⊆ subtrees : |S| > |N(S)|.
Fix OPT:
T ∗ 4
= {T ∗1 , . . . , T ∗
k } where rj ∈ T ∗j .
T ∗(S)4
= {T ∗j | ∃Si ∈ S : T ∗
j ∩ Si 6= ∅}.
Note: T ∗j ∩ Si 6= ∅ ⇒ w(rj, Si) ≤ B∗ ≤ B.
⇒ |T ∗(S)| ≤ |N(S)|.
w(rj , Si) ≤ B
subtrees roots
Sirj
T∗
j
rj
Si
≤ B∗
– p.12
![Page 71: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/71.jpg)
Claim: fail ⇒ B < B∗
Assume for sake of contradiction:
B ≥ B∗ and matching failed.
Hall’s Theorem: if matching failed, then
∃S ⊆ subtrees : |S| > |N(S)|.
Fix OPT:
T ∗ 4
= {T ∗1 , . . . , T ∗
k } where rj ∈ T ∗j .
T ∗(S)4
= {T ∗j | ∃Si ∈ S : T ∗
j ∩ Si 6= ∅}.
Note: T ∗j ∩ Si 6= ∅ ⇒ w(rj, Si) ≤ B∗ ≤ B.
⇒ |T ∗(S)| ≤ |N(S)|.
w(rj , Si) ≤ B
subtrees roots
Sirj
T∗
j
rj
Si
≤ B∗
– p.12
![Page 72: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/72.jpg)
Claim: fail ⇒ B < B∗
Assume for sake of contradiction:
B ≥ B∗ and matching failed.
Hall’s Theorem: if matching failed, then
∃S ⊆ subtrees : |S| > |N(S)|.
Fix OPT:
T ∗ 4
= {T ∗1 , . . . , T ∗
k } where rj ∈ T ∗j .
T ∗(S)4
= {T ∗j | ∃Si ∈ S : T ∗
j ∩ Si 6= ∅}.
Note: T ∗j ∩ Si 6= ∅ ⇒ w(rj, Si) ≤ B∗ ≤ B.
⇒ |T ∗(S)| ≤ |N(S)|.
w(rj , Si) ≤ B
subtrees roots
Sirj
T∗
j
rj
Si
≤ B∗
– p.12
![Page 73: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/73.jpg)
Claim: fail ⇒ B < B∗ - (cont.)
recall:B ≥ B∗ and |T ∗(S)| ≤ |N(S)| < |S|.
∀j : w(T ∗j ) ≤ B∗ ⇒ w(T ∗(S)) ≤ B∗ · |T ∗(S)|
∀i : w(Si) ∈ [B, 2B) ⇒ w(S) ≥ B · |S|.
⇒ w(S) > w(T ∗(S)).
ButT ′ 4
= MST + T ∗(S) − S
is a spanning tree and w(T ′) < w(MST ), contradiction. QED
– p.13
![Page 74: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/74.jpg)
Claim: fail ⇒ B < B∗ - (cont.)
recall:B ≥ B∗ and |T ∗(S)| ≤ |N(S)| < |S|.
∀j : w(T ∗j ) ≤ B∗ ⇒ w(T ∗(S)) ≤ B∗ · |T ∗(S)|
∀i : w(Si) ∈ [B, 2B) ⇒ w(S) ≥ B · |S|.
⇒ w(S) > w(T ∗(S)).
ButT ′ 4
= MST + T ∗(S) − S
is a spanning tree and w(T ′) < w(MST ), contradiction. QED
– p.13
![Page 75: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/75.jpg)
Claim: fail ⇒ B < B∗ - (cont.)
recall:B ≥ B∗ and |T ∗(S)| ≤ |N(S)| < |S|.
∀j : w(T ∗j ) ≤ B∗ ⇒ w(T ∗(S)) ≤ B∗ · |T ∗(S)|
∀i : w(Si) ∈ [B, 2B) ⇒ w(S) ≥ B · |S|.
⇒ w(S) > w(T ∗(S)).
ButT ′ 4
= MST + T ∗(S) − S
is a spanning tree and w(T ′) < w(MST ), contradiction. QED
– p.13
![Page 76: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/76.jpg)
Algorithm for Unrooted k-tree cover1. Prune edges we > B.
Let {Gi}i be components.
2. MSTi = MST of Gi.ki = bw(MSTi)
2Bc.
3. If∑
i(ki + 1) > k, return “fail”.
4. Decompose each MSTi into atmost ki+1 trees S1
i +. . .+Ski
i +Li
such that w(Sji ) ∈ [2B, 4B) and
w(Li) < 2B. Return “success”.
Long edges pruned
– p.14
![Page 77: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/77.jpg)
Algorithm for Unrooted k-tree cover1. Prune edges we > B.
Let {Gi}i be components.
2. MSTi = MST of Gi.ki = bw(MSTi)
2Bc.
3. If∑
i(ki + 1) > k, return “fail”.
4. Decompose each MSTi into atmost ki+1 trees S1
i +. . .+Ski
i +Li
such that w(Sji ) ∈ [2B, 4B) and
w(Li) < 2B. Return “success”.
MST
– p.14
![Page 78: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/78.jpg)
Algorithm for Unrooted k-tree cover1. Prune edges we > B.
Let {Gi}i be components.
2. MSTi = MST of Gi.ki = bw(MSTi)
2Bc.
3. If∑
i(ki + 1) > k, return “fail”.
4. Decompose each MSTi into atmost ki+1 trees S1
i +. . .+Ski
i +Li
such that w(Sji ) ∈ [2B, 4B) and
w(Li) < 2B. Return “success”.
MST
– p.14
![Page 79: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/79.jpg)
Algorithm for Unrooted k-tree cover1. Prune edges we > B.
Let {Gi}i be components.
2. MSTi = MST of Gi.ki = bw(MSTi)
2Bc.
3. If∑
i(ki + 1) > k, return “fail”.
4. Decompose each MSTi into atmost ki+1 trees S1
i +. . .+Ski
i +Li
such that w(Sji ) ∈ [2B, 4B) and
w(Li) < 2B. Return “success”. Final solution
– p.14
![Page 80: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/80.jpg)
Analysis
Claim: On success, each tree has weight no more than 4B.
Claim: On failure, B < B∗.
Alternatively, if B∗ ≤ B, then ki + 1 ≤ k∗i for all i.
– p.15
![Page 81: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/81.jpg)
Analysis
Claim: On success, each tree has weight no more than 4B.
Claim: On failure, B < B∗.
Alternatively, if B∗ ≤ B, then ki + 1 ≤ k∗i for all i.
– p.15
![Page 82: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/82.jpg)
Analysis
Claim: On success, each tree has weight no more than 4B.
Claim: On failure, B < B∗.
Alternatively, if B∗ ≤ B, then ki + 1 ≤ k∗i for all i.
– p.15
![Page 83: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/83.jpg)
Proof: B∗ ≤ B ⇒ ki + 1 ≤ k∗i
Let optimal solution cover Gi with {T ∗1 , . . . , T ∗
k∗
i
}.Augment it to span Gi by adding k∗i − 1 edges, so:
k∗
i∑
j=1
w(T ∗i ) + (k∗i − 1)B ≥ w(MSTi)
Since w(T ∗i ) ≤ B∗ ≤ B,
k∗i · B + (k∗i − 1)B ≥ w(MSTi).
⇒ k∗i ≥w(MSTi)
2B+
1
2> ki.
2
– p.16
![Page 84: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/84.jpg)
Proof: B∗ ≤ B ⇒ ki + 1 ≤ k∗i
Let optimal solution cover Gi with {T ∗1 , . . . , T ∗
k∗
i
}.
Augment it to span Gi by adding k∗i − 1 edges, so:
k∗
i∑
j=1
w(T ∗i ) + (k∗i − 1)B ≥ w(MSTi)
Since w(T ∗i ) ≤ B∗ ≤ B,
k∗i · B + (k∗i − 1)B ≥ w(MSTi).
⇒ k∗i ≥w(MSTi)
2B+
1
2> ki.
2
– p.16
![Page 85: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/85.jpg)
Proof: B∗ ≤ B ⇒ ki + 1 ≤ k∗i
Let optimal solution cover Gi with {T ∗1 , . . . , T ∗
k∗
i
}.Augment it to span Gi by adding k∗i − 1 edges, so:
k∗
i∑
j=1
w(T ∗i ) + (k∗i − 1)B ≥ w(MSTi)
Since w(T ∗i ) ≤ B∗ ≤ B,
k∗i · B + (k∗i − 1)B ≥ w(MSTi).
⇒ k∗i ≥w(MSTi)
2B+
1
2> ki.
2
– p.16
![Page 86: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/86.jpg)
Proof: B∗ ≤ B ⇒ ki + 1 ≤ k∗i
Let optimal solution cover Gi with {T ∗1 , . . . , T ∗
k∗
i
}.Augment it to span Gi by adding k∗i − 1 edges, so:
k∗
i∑
j=1
w(T ∗i ) + (k∗i − 1)B ≥ w(MSTi)
Since w(T ∗i ) ≤ B∗ ≤ B,
k∗i · B + (k∗i − 1)B ≥ w(MSTi).
⇒ k∗i ≥w(MSTi)
2B+
1
2> ki.
2
– p.16
![Page 87: Covering Graphs Using Trees and Starshyde.eng.tau.ac.il/Talks/k_tree_cover.pdfCover the vertices of a graph with k tours. Balance the load of the agents: minimize the maximum tour](https://reader033.vdocument.in/reader033/viewer/2022050306/5f6e7583bc3bce6f5f199040/html5/thumbnails/87.jpg)
Proof: B∗ ≤ B ⇒ ki + 1 ≤ k∗i
Let optimal solution cover Gi with {T ∗1 , . . . , T ∗
k∗
i
}.Augment it to span Gi by adding k∗i − 1 edges, so:
k∗
i∑
j=1
w(T ∗i ) + (k∗i − 1)B ≥ w(MSTi)
Since w(T ∗i ) ≤ B∗ ≤ B,
k∗i · B + (k∗i − 1)B ≥ w(MSTi).
⇒ k∗i ≥w(MSTi)
2B+
1
2> ki.
2
– p.16