algorithmic graph theory part v - approximation algorithms for
TRANSCRIPT
![Page 1: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/1.jpg)
Algorithmic Graph TheoryPart V - Approximation Algorithms
for Graph Problems
Martin [email protected]
University of Primorska, Koper, Slovenia
Dipartimento di InformaticaUniversita degli Studi di Verona, March 2013
1 / 26
![Page 2: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/2.jpg)
Coping with NP -hardness
There are several approaches on how to deal with (theintractability of) NP -hard problems:
polynomial algorithms for particular input instances
approximation algorithms
heuristics, local optimization
“efficient” exponential algorithms
randomized algorithms
parameterized complexity(fixed-parameter tractable (FPT) algorithms)
2 / 26
![Page 3: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/3.jpg)
What we’ll do
1 Basic Definitions.2 2-Approximation Algorithm for Vertex Cover.3 Approximation Algorithms for the Metric TSP.
2 / 26
![Page 4: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/4.jpg)
BASICS OF APPROXIMATION ALGORITHMS.
2 / 26
![Page 5: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/5.jpg)
Coping with NP -hardness
Heuristics:
intuitive algorithms;
3 / 26
![Page 6: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/6.jpg)
Coping with NP -hardness
Heuristics:
intuitive algorithms;
guaranteed to run in polynomial time;
3 / 26
![Page 7: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/7.jpg)
Coping with NP -hardness
Heuristics:
intuitive algorithms;
guaranteed to run in polynomial time;
no guarantee on quality of solution.
3 / 26
![Page 8: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/8.jpg)
Coping with NP -hardness
Heuristics:
intuitive algorithms;
guaranteed to run in polynomial time;
no guarantee on quality of solution.
Approximation algorithms:
guaranteed to run in polynomial time;
3 / 26
![Page 9: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/9.jpg)
Coping with NP -hardness
Heuristics:
intuitive algorithms;
guaranteed to run in polynomial time;
no guarantee on quality of solution.
Approximation algorithms:
guaranteed to run in polynomial time;
guaranteed to find “high quality” solution, say within 1% ofoptimum;
3 / 26
![Page 10: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/10.jpg)
Coping with NP -hardness
Heuristics:
intuitive algorithms;
guaranteed to run in polynomial time;
no guarantee on quality of solution.
Approximation algorithms:
guaranteed to run in polynomial time;
guaranteed to find “high quality” solution, say within 1% ofoptimum;
Obstacle:need to prove a solution’s value is close to optimum,without even knowing what the optimum value is!
3 / 26
![Page 11: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/11.jpg)
Approximation Algorithms and Schemes
ρ-approximation algorithm :
An algorithm A for an optimization problem Π that runs inpolynomial time.
4 / 26
![Page 12: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/12.jpg)
Approximation Algorithms and Schemes
ρ-approximation algorithm :
An algorithm A for an optimization problem Π that runs inpolynomial time.
For every instance of Π, A outputs a feasible solution withobjective function value within ratio ρ of true optimum forthat instance.
4 / 26
![Page 13: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/13.jpg)
Approximation Algorithms and Schemes
ρ-approximation algorithm :
An algorithm A for an optimization problem Π that runs inpolynomial time.
For every instance of Π, A outputs a feasible solution withobjective function value within ratio ρ of true optimum forthat instance.
More specifically:
for minimization problems:for every instance I, we have fA(I) ≤ ρ · OPT(I) , where
4 / 26
![Page 14: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/14.jpg)
Approximation Algorithms and Schemes
ρ-approximation algorithm :
An algorithm A for an optimization problem Π that runs inpolynomial time.
For every instance of Π, A outputs a feasible solution withobjective function value within ratio ρ of true optimum forthat instance.
More specifically:
for minimization problems:for every instance I, we have fA(I) ≤ ρ · OPT(I) , wherefA(I) is the value of the solution returned by the algorithm,and
4 / 26
![Page 15: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/15.jpg)
Approximation Algorithms and Schemes
ρ-approximation algorithm :
An algorithm A for an optimization problem Π that runs inpolynomial time.
For every instance of Π, A outputs a feasible solution withobjective function value within ratio ρ of true optimum forthat instance.
More specifically:
for minimization problems:for every instance I, we have fA(I) ≤ ρ · OPT(I) , wherefA(I) is the value of the solution returned by the algorithm,andOPT(I) is the optimal solution value.
for maximization problems: fA(I) ≥ OPT(I)/ρ .
4 / 26
![Page 16: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/16.jpg)
Approaches to the Design of ApproximationAlgorithms
There exist several approaches to the design of approximationalgorithms:
combinatorial algorithms,
5 / 26
![Page 17: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/17.jpg)
Approaches to the Design of ApproximationAlgorithms
There exist several approaches to the design of approximationalgorithms:
combinatorial algorithms,
algorithms based on linear programming,
5 / 26
![Page 18: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/18.jpg)
Approaches to the Design of ApproximationAlgorithms
There exist several approaches to the design of approximationalgorithms:
combinatorial algorithms,
algorithms based on linear programming,
randomized algorithms,
etc.
5 / 26
![Page 19: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/19.jpg)
2-APPROXIMATION ALGORITHMFOR THE VERTEX COVER PROBLEM
5 / 26
![Page 20: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/20.jpg)
The Vertex Cover Problem
Recall:vertex cover in a graph G = (V ,E):a subset C ⊆ V such that for all e ∈ E , e ∩C 6= ∅
tocka v pokritju
G
6 / 26
![Page 21: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/21.jpg)
The Vertex Cover Problem
Recall:vertex cover in a graph G = (V ,E):a subset C ⊆ V such that for all e ∈ E , e ∩C 6= ∅
a vertex in the cover
G
6 / 26
![Page 22: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/22.jpg)
The Vertex Cover Problem
Consider the optimization version of the VERTEX COVER
problem:
MINIMUM VERTEX COVER
Input: Graph G = (V ,E).Task: Find a minimum vertex cover in G.
7 / 26
![Page 23: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/23.jpg)
The Vertex Cover Problem
Consider the optimization version of the VERTEX COVER
problem:
MINIMUM VERTEX COVER
Input: Graph G = (V ,E).Task: Find a minimum vertex cover in G.
In bipartite graphs, the problem can be solved optimally inpolynomial time.
7 / 26
![Page 24: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/24.jpg)
The Vertex Cover Problem
Consider the optimization version of the VERTEX COVER
problem:
MINIMUM VERTEX COVER
Input: Graph G = (V ,E).Task: Find a minimum vertex cover in G.
In bipartite graphs, the problem can be solved optimally inpolynomial time.For general graphs, the problem is NP-hard.
7 / 26
![Page 25: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/25.jpg)
2-Approximation Algorithm for Vertex Cover
Approx-Cover :
8 / 26
![Page 26: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/26.jpg)
2-Approximation Algorithm for Vertex Cover
Approx-Cover :C := ∅;
8 / 26
![Page 27: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/27.jpg)
2-Approximation Algorithm for Vertex Cover
Approx-Cover :C := ∅;while (∃e = uv ∈ E)(u, v ∈ V \C) do
8 / 26
![Page 28: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/28.jpg)
2-Approximation Algorithm for Vertex Cover
Approx-Cover :C := ∅;while (∃e = uv ∈ E)(u, v ∈ V \C) do
C := C ∪ {u, v}
8 / 26
![Page 29: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/29.jpg)
2-Approximation Algorithm for Vertex Cover
Approx-Cover :C := ∅;while (∃e = uv ∈ E)(u, v ∈ V \C) do
C := C ∪ {u, v}end while
8 / 26
![Page 30: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/30.jpg)
2-Approximation Algorithm for Vertex Cover
Approx-Cover :C := ∅;while (∃e = uv ∈ E)(u, v ∈ V \C) do
C := C ∪ {u, v}end whilereturn C.
The algorithm computes an inclusion-wise maximal matchingM and returns the union of all edges in the matching.
8 / 26
![Page 31: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/31.jpg)
2-Approximation Algorithm for Vertex Cover
ClaimApprox-Cover is a 2-approximation algorithm for the MINIMUM
VERTEX COVER problem.
9 / 26
![Page 32: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/32.jpg)
2-Approximation Algorithm for Vertex Cover
ClaimApprox-Cover is a 2-approximation algorithm for the MINIMUM
VERTEX COVER problem.
Proof:
9 / 26
![Page 33: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/33.jpg)
2-Approximation Algorithm for Vertex Cover
ClaimApprox-Cover is a 2-approximation algorithm for the MINIMUM
VERTEX COVER problem.
Proof:The stopping criterion of the while loop guarantees that C is acover.
9 / 26
![Page 34: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/34.jpg)
2-Approximation Algorithm for Vertex Cover
ClaimApprox-Cover is a 2-approximation algorithm for the MINIMUM
VERTEX COVER problem.
Proof:The stopping criterion of the while loop guarantees that C is acover.Clearly, the algorithm can be implemented to run in polynomialtime.
9 / 26
![Page 35: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/35.jpg)
2-Approximation Algorithm for Vertex Cover
ClaimApprox-Cover is a 2-approximation algorithm for the MINIMUM
VERTEX COVER problem.
Proof:The stopping criterion of the while loop guarantees that C is acover.Clearly, the algorithm can be implemented to run in polynomialtime.Let M be the maximal matching consisting of all edges chosenby the algorithm.
9 / 26
![Page 36: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/36.jpg)
2-Approximation Algorithm for Vertex Cover
ClaimApprox-Cover is a 2-approximation algorithm for the MINIMUM
VERTEX COVER problem.
Proof:The stopping criterion of the while loop guarantees that C is acover.Clearly, the algorithm can be implemented to run in polynomialtime.Let M be the maximal matching consisting of all edges chosenby the algorithm.Every vertex cover must contain at least one vertex of eachedge of M,
9 / 26
![Page 37: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/37.jpg)
2-Approximation Algorithm for Vertex Cover
ClaimApprox-Cover is a 2-approximation algorithm for the MINIMUM
VERTEX COVER problem.
Proof:The stopping criterion of the while loop guarantees that C is acover.Clearly, the algorithm can be implemented to run in polynomialtime.Let M be the maximal matching consisting of all edges chosenby the algorithm.Every vertex cover must contain at least one vertex of eachedge of M,hence OPT ≥ |M|
9 / 26
![Page 38: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/38.jpg)
2-Approximation Algorithm for Vertex Cover
ClaimApprox-Cover is a 2-approximation algorithm for the MINIMUM
VERTEX COVER problem.
Proof:The stopping criterion of the while loop guarantees that C is acover.Clearly, the algorithm can be implemented to run in polynomialtime.Let M be the maximal matching consisting of all edges chosenby the algorithm.Every vertex cover must contain at least one vertex of eachedge of M,hence OPT ≥ |M|and consequently
|C| = 2|M| ≤ 2 · OPT .
9 / 26
![Page 39: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/39.jpg)
Other (In)approximability issues
The factor of 2 in the analysis cannot be improved:
10 / 26
![Page 40: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/40.jpg)
Other (In)approximability issues
The factor of 2 in the analysis cannot be improved:If G = Kn,n, then the algorithm returns C = V (Kn,n), whilean optimal solution is of size n (either part of thebipartition).
10 / 26
![Page 41: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/41.jpg)
Other (In)approximability issues
The factor of 2 in the analysis cannot be improved:If G = Kn,n, then the algorithm returns C = V (Kn,n), whilean optimal solution is of size n (either part of thebipartition).
Remark:It can be shown that a natural greedy algorithm is not aρ-approximation, for any ρ.
10 / 26
![Page 42: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/42.jpg)
Other (In)approximability issues
The factor of 2 in the analysis cannot be improved:If G = Kn,n, then the algorithm returns C = V (Kn,n), whilean optimal solution is of size n (either part of thebipartition).
Remark:It can be shown that a natural greedy algorithm is not aρ-approximation, for any ρ.[Greedy: start with C = ∅. Until C is not a vertex cover,peel off and add to C a vertex of maximum degree in thecurrent graph.]
10 / 26
![Page 43: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/43.jpg)
Other (In)approximability issues
The factor of 2 in the analysis cannot be improved:If G = Kn,n, then the algorithm returns C = V (Kn,n), whilean optimal solution is of size n (either part of thebipartition).
Remark:It can be shown that a natural greedy algorithm is not aρ-approximation, for any ρ.[Greedy: start with C = ∅. Until C is not a vertex cover,peel off and add to C a vertex of maximum degree in thecurrent graph.]
Inapproximability of vertex cover:If there exists a polynomial 1.36-approximation algorithmfor MINIMUM VERTEX COVER, then P = NP (Dinur-Safra 2005).
10 / 26
![Page 44: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/44.jpg)
Other (In)approximability issues
The factor of 2 in the analysis cannot be improved:If G = Kn,n, then the algorithm returns C = V (Kn,n), whilean optimal solution is of size n (either part of thebipartition).
Remark:It can be shown that a natural greedy algorithm is not aρ-approximation, for any ρ.[Greedy: start with C = ∅. Until C is not a vertex cover,peel off and add to C a vertex of maximum degree in thecurrent graph.]
Inapproximability of vertex cover:If there exists a polynomial 1.36-approximation algorithmfor MINIMUM VERTEX COVER, then P = NP (Dinur-Safra 2005).No ρ-approximation algorithm for MINIMUM VERTEX
COVER is known with ρ < 2.10 / 26
![Page 45: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/45.jpg)
APPROXIMATION ALGORITHMSFOR THE TRAVELING SALESMAN PROBLEM
10 / 26
![Page 46: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/46.jpg)
The Traveling Salesman Problem
TRAVELING SALESMAN (TSP)
Input: Graph G = (V ,E), a cost function c : E → R+.Task: Find a Hamiltonian cycle in G of smallest total cost.
11 / 26
![Page 47: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/47.jpg)
The Traveling Salesman Problem
Proposition
If there exists a ρ-approximation algorithm for TSP for someρ ≥ 1, then P = NP.
12 / 26
![Page 48: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/48.jpg)
The Traveling Salesman Problem
Proposition
If there exists a ρ-approximation algorithm for TSP for someρ ≥ 1, then P = NP.
Proof:Suppose that A is a ρ-approximation algorithm for TSP.
12 / 26
![Page 49: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/49.jpg)
The Traveling Salesman Problem
Proposition
If there exists a ρ-approximation algorithm for TSP for someρ ≥ 1, then P = NP.
Proof:Suppose that A is a ρ-approximation algorithm for TSP.We will show how to decide in polynomial time, using A, whether a givengraph G is Hamiltonian (which is an NP-complete problem).
12 / 26
![Page 50: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/50.jpg)
The Traveling Salesman Problem
Proposition
If there exists a ρ-approximation algorithm for TSP for someρ ≥ 1, then P = NP.
Proof:Suppose that A is a ρ-approximation algorithm for TSP.We will show how to decide in polynomial time, using A, whether a givengraph G is Hamiltonian (which is an NP-complete problem).HAMILTONIAN CYCLE: determine whether a given graph contains a cyclegoing through every vertex exactly once.
12 / 26
![Page 51: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/51.jpg)
The Traveling Salesman Problem
Proposition
If there exists a ρ-approximation algorithm for TSP for someρ ≥ 1, then P = NP.
Proof:Suppose that A is a ρ-approximation algorithm for TSP.We will show how to decide in polynomial time, using A, whether a givengraph G is Hamiltonian (which is an NP-complete problem).HAMILTONIAN CYCLE: determine whether a given graph contains a cyclegoing through every vertex exactly once.I = instance for HAMILTONIAN CYCLE: a graph G = (V ,E).
12 / 26
![Page 52: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/52.jpg)
The Traveling Salesman Problem
Proposition
If there exists a ρ-approximation algorithm for TSP for someρ ≥ 1, then P = NP.
Proof:Suppose that A is a ρ-approximation algorithm for TSP.We will show how to decide in polynomial time, using A, whether a givengraph G is Hamiltonian (which is an NP-complete problem).HAMILTONIAN CYCLE: determine whether a given graph contains a cyclegoing through every vertex exactly once.I = instance for HAMILTONIAN CYCLE: a graph G = (V ,E).Let G′ = KV (complete graph), c :
(V2
)
→ R+, where
c(e) ={
1, if e ∈ E ;ρ|V |+ 1, otherwise.
12 / 26
![Page 53: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/53.jpg)
The Traveling Salesman Problem
Proposition
If there exists a ρ-approximation algorithm for TSP for someρ ≥ 1, then P = NP.
Proof:Suppose that A is a ρ-approximation algorithm for TSP.We will show how to decide in polynomial time, using A, whether a givengraph G is Hamiltonian (which is an NP-complete problem).HAMILTONIAN CYCLE: determine whether a given graph contains a cyclegoing through every vertex exactly once.I = instance for HAMILTONIAN CYCLE: a graph G = (V ,E).Let G′ = KV (complete graph), c :
(V2
)
→ R+, where
c(e) ={
1, if e ∈ E ;ρ|V |+ 1, otherwise.
Notice: G is Hamiltonian if and only if G′ has a Hamiltonian cycle of total cost|V |.
12 / 26
![Page 54: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/54.jpg)
The Traveling Salesman Problem
Proposition
If there exists a ρ-approximation algorithm for TSP for someρ ≥ 1, then P = NP.
Proof:Suppose that A is a ρ-approximation algorithm for TSP.We will show how to decide in polynomial time, using A, whether a givengraph G is Hamiltonian (which is an NP-complete problem).HAMILTONIAN CYCLE: determine whether a given graph contains a cyclegoing through every vertex exactly once.I = instance for HAMILTONIAN CYCLE: a graph G = (V ,E).Let G′ = KV (complete graph), c :
(V2
)
→ R+, where
c(e) ={
1, if e ∈ E ;ρ|V |+ 1, otherwise.
Notice: G is Hamiltonian if and only if G′ has a Hamiltonian cycle of total cost|V |.Let Γ be the solution to the TSP given (KV , c), computed by our algorithm A.
12 / 26
![Page 55: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/55.jpg)
The Traveling Salesman Problem
Proposition
If there exists a ρ-approximation algorithm for TSP for someρ ≥ 1, then P = NP.
Proof:Suppose that A is a ρ-approximation algorithm for TSP.We will show how to decide in polynomial time, using A, whether a givengraph G is Hamiltonian (which is an NP-complete problem).HAMILTONIAN CYCLE: determine whether a given graph contains a cyclegoing through every vertex exactly once.I = instance for HAMILTONIAN CYCLE: a graph G = (V ,E).Let G′ = KV (complete graph), c :
(V2
)
→ R+, where
c(e) ={
1, if e ∈ E ;ρ|V |+ 1, otherwise.
Notice: G is Hamiltonian if and only if G′ has a Hamiltonian cycle of total cost|V |.Let Γ be the solution to the TSP given (KV , c), computed by our algorithm A.
(a) If c(Γ) ≤ ρ|V |, then Γ ⊆ E , hence G is Hamiltonian.
12 / 26
![Page 56: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/56.jpg)
The Traveling Salesman Problem
Proposition
If there exists a ρ-approximation algorithm for TSP for someρ ≥ 1, then P = NP.
Proof:Suppose that A is a ρ-approximation algorithm for TSP.We will show how to decide in polynomial time, using A, whether a givengraph G is Hamiltonian (which is an NP-complete problem).HAMILTONIAN CYCLE: determine whether a given graph contains a cyclegoing through every vertex exactly once.I = instance for HAMILTONIAN CYCLE: a graph G = (V ,E).Let G′ = KV (complete graph), c :
(V2
)
→ R+, where
c(e) ={
1, if e ∈ E ;ρ|V |+ 1, otherwise.
Notice: G is Hamiltonian if and only if G′ has a Hamiltonian cycle of total cost|V |.Let Γ be the solution to the TSP given (KV , c), computed by our algorithm A.
(a) If c(Γ) ≤ ρ|V |, then Γ ⊆ E , hence G is Hamiltonian.
(b) If c(Γ) > ρ|V |, then OPT > |V |, hence G is not Hamiltonian.12 / 26
![Page 57: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/57.jpg)
A Heuristic for TSP
Approx-TSP (G, c)
Find a minimum spanning tree T = (V ,ET ) for (G, c).
13 / 26
![Page 58: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/58.jpg)
A Heuristic for TSP
Approx-TSP (G, c)
Find a minimum spanning tree T = (V ,ET ) for (G, c).
1
2
3 4 5
6 7
8 9
1
2
3 4 5
6 7
8 9
input instance
(The cost of an egde connecting two vertices is equal to the Euclidean distance between them.)
minimum spanning tree T
13 / 26
![Page 59: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/59.jpg)
A Heuristic for TSP
Approx-TSP (G, c)
Find a minimum spanning tree T = (V ,ET ) for (G, c).
W ← walk in which each edge appears exactly twice(obtained for example with depth-first search)
14 / 26
![Page 60: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/60.jpg)
A Heuristic for TSP
Approx-TSP (G, c)
Find a minimum spanning tree T = (V ,ET ) for (G, c).
W ← walk in which each edge appears exactly twice(obtained for example with depth-first search)
H ← cycle that visits vertices in the order as they appear inW for the first time
14 / 26
![Page 61: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/61.jpg)
A Heuristic for TSP
Approx-TSP (G, c)
Find a minimum spanning tree T = (V ,ET ) for (G, c).
W ← walk in which each edge appears exactly twice(obtained for example with depth-first search)
H ← cycle that visits vertices in the order as they appear inW for the first time
1
2
3 4 5
6 7
8 9
1
2
3 4 5
6
7
8 9
walk W traveling salesman tour determined by W
(obtained with DFS)
14 / 26
![Page 62: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/62.jpg)
A Heuristic for TSP
Approx-TSP (G, c)
Find a minimum spanning tree T = (V ,ET ) for (G, c).
W ← walk in which each edge appears exactly twice(obtained for example with depth-first search)
H ← cycle that visits vertices in the order as they appear inW for the first time
1
2
3 4 5
6
7
8 9
1
2
3 4 5
6
7
8 9
optimal solution traveling salesman tour determined by W
cost ≈ 15, 72cost ≈ 13, 95
15 / 26
![Page 63: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/63.jpg)
The Metric TSP
METRIC TSP: TSP in which the cost function obeys the triangleinequality:For all u, v ,w ∈ V :
c(uw) ≤ c(uv) + c(vw) .
16 / 26
![Page 64: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/64.jpg)
The Metric TSP
METRIC TSP: TSP in which the cost function obeys the triangleinequality:For all u, v ,w ∈ V :
c(uw) ≤ c(uv) + c(vw) .
A reduction from the HAMILTONIAN CYCLE problem shows:
Proposition
METRIC TSP is NP-hard.
16 / 26
![Page 65: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/65.jpg)
The Metric TSP: a 2-approximation
Proposition
APPROX-TSP is a 2-approximation algorithm for the METRIC
TSP problem.
17 / 26
![Page 66: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/66.jpg)
The Metric TSP: a 2-approximation
Proposition
APPROX-TSP is a 2-approximation algorithm for the METRIC
TSP problem.
Proof:Let H∗ be an optimal tour. We need to show: c(H) ≤ 2 · c(H∗).
17 / 26
![Page 67: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/67.jpg)
The Metric TSP: a 2-approximation
Proposition
APPROX-TSP is a 2-approximation algorithm for the METRIC
TSP problem.
Proof:Let H∗ be an optimal tour. We need to show: c(H) ≤ 2 · c(H∗).
c(T ) ≤ c(H∗), since removing any edge of H∗ results in aspanning tree.
1
2
3 4 5
6 7
8 9
1
2
3 4 5
6
7
8 9
optimal solution H∗ minus one edge
minimum spanning tree T17 / 26
![Page 68: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/68.jpg)
The Metric TSP: a 2-approximation
Proposition
APPROX-TSP is a 2-approximation algorithm for the METRIC
TSP problem.
Proof:c(T ) ≤ c(H∗), since removing any edge of H∗ results in aspanning tree.c(W ) = 2 · c(T ), since every edge is visited exactly twice.
1
2
3 4 5
6 7
8 9
walk W
1
2
3 4 5
6 7
8 9
minimum spanning tree T
18 / 26
![Page 69: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/69.jpg)
The Metric TSP: a 2-approximation
Proposition
APPROX-TSP is a 2-approximation algorithm for the METRIC
TSP problem.
Proof:c(T ) ≤ c(H∗), since removing any edge of H∗ results in aspanning tree.c(W ) = 2 · c(T ), since every edge is visited exactly twice.c(H) ≤ ·c(W ), due to the triangle inequality.
1
2
3 4 5
6 7
8 9
1
2
3 4 5
6
7
8 9
walk W traveling salesman tour determined by W 19 / 26
![Page 70: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/70.jpg)
The Metric TSP: Christofides’ Algorithm
Theorem (Christofides, 1976)
There exists a 1.5-approximation algorithm for the METRIC
TSP problem.
20 / 26
![Page 71: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/71.jpg)
The Metric TSP: Christofides’ Algorithm
Theorem (Christofides, 1976)
There exists a 1.5-approximation algorithm for the METRIC
TSP problem.
Christofides-TSP (G, c)
Find a minimum spanning tree T = (V ,ET ) for (G, c).Find a minimum cost perfect matching M connectingvertices of odd degree in T .
1
2
3 4 5
6 7
8 9
1
67
9
matching Mminimum spanning tree T
vertices of odd degree are red
20 / 26
![Page 72: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/72.jpg)
The Metric TSP: Christofides’ Algorithm
Theorem (Christofides, 1976)
There exists a 1.5-approximation algorithm for the METRIC
TSP problem.
Christofides-TSP (G, c)
Find a minimum spanning tree T = (V ,ET ) for (G, c).Find a minimum cost perfect matching M connectingvertices of odd degree in T .G′ ← T ∪ME ← Euler tour in G′ (graph G′ is Eulerian, since it is connected
and has all vertices of even degree).1
2
3 4 5
67
8 9
G′
1
2
3 4 5
67
8 9
E = Euler tour in G′
21 / 26
![Page 73: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/73.jpg)
The Metric TSP: Christofides’ Algorithm
Proof:Let H∗ be an optimal tour. We need to show: c(H) ≤ 1.5 ·c(H∗).
22 / 26
![Page 74: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/74.jpg)
The Metric TSP: Christofides’ Algorithm
Proof:Let H∗ be an optimal tour. We need to show: c(H) ≤ 1.5 ·c(H∗).
c(T ) ≤ c(H∗), as before
22 / 26
![Page 75: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/75.jpg)
The Metric TSP: Christofides’ Algorithm
Proof:Let H∗ be an optimal tour. We need to show: c(H) ≤ 1.5 ·c(H∗).
c(T ) ≤ c(H∗), as beforec(M) ≤ (1/2) · c(Γ∗) ≤ (1/2) · c(H∗), where Γ∗ is anoptimal cycle on the odd vertices of T .
The first inequality follows since Γ∗ is the union of twoperfect matchings.The second inequality follows from the triangle inequality.
1
67
9
matching M
1
67
9
cycle Γ∗
22 / 26
![Page 76: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/76.jpg)
The Metric TSP: Christofides’ Algorithm
Proof:
c(T ) ≤ c(H∗), as before
c(M) ≤ (1/2) · c(Γ∗) ≤ (1/2) · c(H∗), where Γ∗ is anoptimal cycle on the odd vertices of T .
Due to the triangle inquality:c(H) ≤ c(M) + c(T ) ≤ (3/2) · c(H∗).
1
2
3 4 5
67
8 9
G′
1
2
3 4 5
6
7
8 9
H = traveling salesman tour
23 / 26
![Page 77: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/77.jpg)
What we did – Week 1
1 Tue March 5: Review of basic notions in graph theory,algorithms and complexity X
2 Wed March 6: Graph colorings X
3 Thu March 7: Perfect graphs and their subclasses, part 1 X
4 Fri March 8: Perfect graphs and their subclasses, part 2 X
24 / 26
![Page 78: Algorithmic Graph Theory Part V - Approximation Algorithms for](https://reader030.vdocuments.us/reader030/viewer/2022020703/61fb29332e268c58cd5adf3c/html5/thumbnails/78.jpg)
What we’ll do – Week 2
1 Tue March 19: Further examples of tractable problems,part 1 X
2 Wed March 20:Further examples of tractable problems, part 2 X
Approximation algorithms for graph problems X
3 Thu March 21: Lectio Magistralis lecture, “Graph classes:interrelations, structure, and algorithmic issues”
25 / 26