approximability of combinatorial optimization problems on...
TRANSCRIPT
![Page 1: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/1.jpg)
Approximability of CombinatorialOptimization Problems on Power Law
Networks
Mikael Gast
Dept. of Computer Science andB-IT Research School,
University of Bonn
Ph.D. Thesis Defense
September 4, 2013
![Page 2: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/2.jpg)
First Observation
— Real world networks are not random, theyhave very small diameter and they possess apower law distribution of node degrees
Introduction First Observation 2 / 42
![Page 3: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/3.jpg)
Biological Networks
Example:Protein interactions ofArabidopsis Thaliana
Introduction Examples of Real World Networks 3 / 42
![Page 4: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/4.jpg)
Biological Networks
Example:Protein interactions ofArabidopsis Thaliana
100 101 102
100
101
102
103
Degree (i)
Frequency
(yi)
Introduction Examples of Real World Networks 3 / 42
![Page 5: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/5.jpg)
Technological Networks
Example:Network of InternetRouters
Introduction Examples of Real World Networks 4 / 42
![Page 6: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/6.jpg)
Technological Networks
Example:Network of InternetRouters
100 101 102 103
100
101
102
103
104
Degree (i)
Frequency
(yi)
Introduction Examples of Real World Networks 4 / 42
![Page 7: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/7.jpg)
Social Networks (Disease Spreading)
Example:Airport Network in theUnited States
Introduction Examples of Real World Networks 5 / 42
![Page 8: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/8.jpg)
Social Networks (Disease Spreading)
Example:Airport Network in theUnited States
101 102
100
101
Degree (i)
Frequency
(yi)
Introduction Examples of Real World Networks 5 / 42
![Page 9: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/9.jpg)
Social Networks (Info Spreading)
Example:Contact Network ofkarate club members
Introduction Examples of Real World Networks 6 / 42
![Page 10: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/10.jpg)
Social Networks (Info Spreading)
Example:Contact Network ofkarate club members
100 101
100
101
Degree (i)
Frequency
(yi)
Introduction Examples of Real World Networks 6 / 42
![Page 11: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/11.jpg)
Power Law Degree Distribution
Uniform random graph vs. power law random graph
Number of nodes yi having degree i: yi ∼ c · i−β
Erdos-Renyi Random Graph Power Law Graph (PLG)
Introduction Power Law Degree Distribution 7 / 42
![Page 12: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/12.jpg)
Power Law Degree Distribution
Uniform random graph vs. power law random graph
Number of nodes yi having degree i: yi ∼ c · i−β
100 101
100
101
Degree (i)
Frequency
(yi)
100 101
100
101
Degree (i)
Introduction Power Law Degree Distribution 7 / 42
![Page 13: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/13.jpg)
Power Law Degree Distribution
Uniform random graph vs. power law random graph
Number of nodes yi having degree i: yi ∼ c · i−β
100 101
100
101
Degree (i)
Frequency
(yi)
100 101
100
101
Degree (i)
Introduction Power Law Degree Distribution 7 / 42
![Page 14: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/14.jpg)
Motivation
— The study of combinatorial optimizationproblems on real world networks is motivatedby applications
Introduction Combinatorial Optimization on Real World Networks 8 / 42
![Page 15: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/15.jpg)
Example: Dominating Set Problem
Minimum dominatingset problem in realworld networks:
Optimal sensor orserver placementin wireless mobilenetworks
Search for keyplayers or nodesin social networks
Introduction Combinatorial Optimization on Real World Networks 9 / 42
![Page 16: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/16.jpg)
Example: Dominating Set Problem
Minimum dominatingset problem in realworld networks:
Optimal sensor orserver placementin wireless mobilenetworks
Search for keyplayers or nodesin social networks
Introduction Combinatorial Optimization on Real World Networks 9 / 42
![Page 17: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/17.jpg)
Example: Dominating Set Problem
Minimum dominatingset problem in realworld networks:
Optimal sensor orserver placementin wireless mobilenetworks
Search for keyplayers or nodesin social networks
Introduction Combinatorial Optimization on Real World Networks 9 / 42
![Page 18: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/18.jpg)
Other Properties
— Real world networks display a number ofother unique and characteristic topologicalproperties
Introduction Properties of Real World Networks 10 / 42
![Page 19: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/19.jpg)
Small Worlds
Real world networks behave like “Small Worlds”
Existence of bridging links across the network
Regular Ring Graph Small-World Graph Random Graph
p = 0 p = 1Increasing Rewiring Probability p
Introduction Properties of Real World Networks 11 / 42
![Page 20: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/20.jpg)
Small Worlds
Real world networks behave like “Small Worlds”
Existence of bridging links across the network
Regular Ring Graph Small-World Graph Random Graph
p = 0 p = 1Increasing Rewiring Probability p
Introduction Properties of Real World Networks 11 / 42
![Page 21: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/21.jpg)
Clustering Coefficients
Real world networks have large clusteringcoefficients
Clustering coefficient measures cliquishness
vvv
Cv = 0 Cv = 1Increasing Clustering Coefficient Cv
Introduction Properties of Real World Networks 12 / 42
![Page 22: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/22.jpg)
Clustering Coefficients
Real world networks have large clusteringcoefficients
Clustering coefficient measures cliquishness
vvv
Cv = 0 Cv = 1Increasing Clustering Coefficient Cv
Introduction Properties of Real World Networks 12 / 42
![Page 23: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/23.jpg)
Hyperbolicity
Real world networks have embedded hyperbolicgeometries
Relates to Gromov’s four-point condition forδ-hyperbolicity of a metric space
HyperbolicEuclideanSpherical
K > 0 K < 0Decreasing Curvature K
Introduction Properties of Real World Networks 13 / 42
![Page 24: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/24.jpg)
Hyperbolicity
Real world networks have embedded hyperbolicgeometries
Relates to Gromov’s four-point condition forδ-hyperbolicity of a metric space
HyperbolicEuclideanSpherical
K > 0 K < 0Decreasing Curvature K
Introduction Properties of Real World Networks 13 / 42
![Page 25: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/25.jpg)
Modeling
— There exists a large number of generatingmodels for power law graphs
Introduction Power Law Graph Models 14 / 42
![Page 26: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/26.jpg)
Preferential Attachment
Evolving random model for PLG’s:
The Preferential Attachment Model (Barabasi and
Albert, 1999)
v
u
After adding u, probability that u connects to somevertex v:
Pr(u, v) =
deg(v)/∑
i deg(vi) − 1 u 6= v
1/∑
i deg(vi) − 1 u = v
Introduction Power Law Graph Models 15 / 42
![Page 27: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/27.jpg)
Preferential Attachment
Evolving random model for PLG’s:
The Preferential Attachment Model (Barabasi and
Albert, 1999)
v
u
After adding u, probability that u connects to somevertex v:
Pr(u, v) =
deg(v)/∑
i deg(vi) − 1 u 6= v
1/∑
i deg(vi) − 1 u = v
Introduction Power Law Graph Models 15 / 42
![Page 28: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/28.jpg)
Static Degree Sequences
Static random model for PLG’s:
The Gα,β Model or ACL Model (Aiello, Chung, and Lu,
2001)
Ensures power-law degree distribution by fixing a degreesequence (y1, y2, . . . , y∆) via two parameters α,β andthen taking the space of random multigraphs with thisdegree sequence
Introduction Power Law Graph Models 16 / 42
![Page 29: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/29.jpg)
Static Degree Sequences
Static random model for PLG’s:
The Gα,β Model or ACL Model (Aiello, Chung, and Lu,
2001)
Ensures power-law degree distribution by fixing a degreesequence (y1, y2, . . . , y∆) via two parameters α,β andthen taking the space of random multigraphs with thisdegree sequence
Introduction Power Law Graph Models 16 / 42
![Page 30: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/30.jpg)
ACL Model for PLG’s
Definition of the ACL Model Gα,β:
For each 1 6 i 6 ∆ =⌊eα/β
⌋,
yi =
⌊eα
iβ⌋
if i > 1 or∑∆
i=1
⌊eαiβ⌋
is even
beαc+ 1 otherwise
α is the logarithm of the network size, β is thelog-log growth rate
Preliminaries ACL Model for Power Law Graphs 17 / 42
![Page 31: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/31.jpg)
ACL Model for PLG’s
Definition of the ACL Model Gα,β:
For each 1 6 i 6 ∆ =⌊eα/β
⌋,
yi =
⌊eα
iβ⌋
if i > 1 or∑∆
i=1
⌊eαiβ⌋
is even
beαc+ 1 otherwise
α is the logarithm of the network size, β is thelog-log growth rate
Preliminaries ACL Model for Power Law Graphs 17 / 42
![Page 32: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/32.jpg)
ACL Model for PLG’s
Definition of the ACL Model Gα,β:
For each 1 6 i 6 ∆ =⌊eα/β
⌋,
yi =
⌊eα
iβ⌋
if i > 1 or∑∆
i=1
⌊eαiβ⌋
is even
beαc+ 1 otherwise
α is the logarithm of the network size, β is thelog-log growth rate
Preliminaries ACL Model for Power Law Graphs 17 / 42
![Page 33: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/33.jpg)
ACL Model for PLG’s
Number of vertices:
n =
∆∑i=1
⌊eα
iβ
⌋≈
ζ(β) eα if β > 1
α eα if β = 1eα/β1−β if 0 < β < 1
Number of edges:
m =12
∆∑i=1
i⌊
eα
iβ
⌋≈
12ζ(β− 1) eα if β > 214α eα if β = 212
e2α/β
2−β if 0 < β < 2
Preliminaries ACL Model for Power Law Graphs 18 / 42
![Page 34: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/34.jpg)
ACL Model for PLG’s
Number of vertices:
n =
∆∑i=1
⌊eα
iβ
⌋≈
ζ(β) eα if β > 1
α eα if β = 1eα/β1−β if 0 < β < 1
Number of edges:
m =12
∆∑i=1
i⌊
eα
iβ
⌋≈
12ζ(β− 1) eα if β > 214α eα if β = 212
e2α/β
2−β if 0 < β < 2
Preliminaries ACL Model for Power Law Graphs 18 / 42
![Page 35: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/35.jpg)
ACL Random Model
The distribution of graphs G ∈ Gα,β over a sequence(y1, y2, . . . , y∆) or (deg(v1), deg(v2), . . . , deg(vn)) isgenerated as follows:
Preliminaries ACL Model for Power Law Graphs 19 / 42
![Page 36: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/36.jpg)
ACL Random Model
1 Generate set L of deg(v) distinct copies for eachvertex v ∈ V(G)
2 M := random matching on the elements of L3 For u, v ∈ V(G) number of edges u, v equals
number of edges m ∈M that join copies of u and v
. . . . . . . . .
deg(v) = 1 deg(v) = 2 deg(v) = 3
V(G)
L
Preliminaries ACL Model for Power Law Graphs 20 / 42
![Page 37: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/37.jpg)
ACL Random Model
1 Generate set L of deg(v) distinct copies for eachvertex v ∈ V(G)
2 M := random matching on the elements of L3 For u, v ∈ V(G) number of edges u, v equals
number of edges m ∈M that join copies of u and v
. . . . . . . . .
deg(v) = 1 deg(v) = 2 deg(v) = 3
V(G)
L
Preliminaries ACL Model for Power Law Graphs 20 / 42
![Page 38: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/38.jpg)
ACL Random Model
1 Generate set L of deg(v) distinct copies for eachvertex v ∈ V(G)
2 M := random matching on the elements of L3 For u, v ∈ V(G) number of edges u, v equals
number of edges m ∈M that join copies of u and v
. . . . . . . . .
deg(v) = 1 deg(v) = 2 deg(v) = 3
V(G)
L
edge multi-edgeself-loop
Preliminaries ACL Model for Power Law Graphs 20 / 42
![Page 39: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/39.jpg)
Overview of Main Results
![Page 40: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/40.jpg)
Overview of Results
Presented here:
Approximation lower bounds for Minimum
Dominating Set (Min-DS) in connected PLG’s
Approximation upper bounds for Minimum Vertex
Cover (Min-VC) in random PLG’s
Techniques:
Connected Embedding Approximation-Preserving(CEAP) reductions
Transforming hardness results for boundedoccurrence CSP’s and Set Cover
Overview Main Results 22 / 42
![Page 41: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/41.jpg)
Overview of Results
Presented here:
Approximation lower bounds for Minimum
Dominating Set (Min-DS) in connected PLG’s
Approximation upper bounds for Minimum Vertex
Cover (Min-VC) in random PLG’s
Techniques:
LP-relaxation and deterministic rounding algorithm
Upper and lower bounds on the size of half-integralsolutions in random PLG’s
Overview Main Results 22 / 42
![Page 42: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/42.jpg)
Overview of Results
Presented here:
Approximation lower bounds for Minimum
Dominating Set (Min-DS) in connected PLG’s
Approximation upper bounds for Minimum Vertex
Cover (Min-VC) in random PLG’s
Further results:
Approximation lower bounds for Min-VC inconnected PLG’s
Approximation upper bounds for Min-DS for β > 2
Overview Main Results 22 / 42
![Page 43: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/43.jpg)
Overview of Results
Presented here:
Approximation lower bounds for Minimum
Dominating Set (Min-DS) in connected PLG’s
Approximation upper bounds for Minimum Vertex
Cover (Min-VC) in random PLG’s
Further results:
Approximation lower bounds for Min-VC inconnected PLG’s
Approximation upper bounds for Min-DS for β > 2
Overview Main Results 22 / 42
![Page 44: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/44.jpg)
Techniques and Paradigms Used
![Page 45: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/45.jpg)
Techniques and Paradigms Used
Lower bound technique:
CEAP reductions (high level view)Embed bounded occurrence CSP and Set Cover
reduction instances G ′ into PLG Gα,β ∈ Gα,β
Achieve connectivity with reasonable cut sizesbetween G ′ and Gα,β \ G ′
Preserve hardness of approximation in theembedding construction
Overview Techniques and Paradigms 24 / 42
![Page 46: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/46.jpg)
Techniques and Paradigms Used
Lower bound technique:
CEAP reductions (high level view)Embed bounded occurrence CSP and Set Cover
reduction instances G ′ into PLG Gα,β ∈ Gα,β
Achieve connectivity with reasonable cut sizesbetween G ′ and Gα,β \ G ′
Preserve hardness of approximation in theembedding construction
Overview Techniques and Paradigms 24 / 42
![Page 47: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/47.jpg)
Techniques and Paradigms Used
Lower bound technique:
CEAP reductions (high level view)Embed bounded occurrence CSP and Set Cover
reduction instances G ′ into PLG Gα,β ∈ Gα,β
Achieve connectivity with reasonable cut sizesbetween G ′ and Gα,β \ G ′
Preserve hardness of approximation in theembedding construction
Overview Techniques and Paradigms 24 / 42
![Page 48: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/48.jpg)
Bounded Occurrence CSP Paradigm
Method: Bounded degree amplifier graphs(Berman and Karpinski, 1999)
v
Basic Idea:
Replace nodes corre-sponding to variablesby 3-regular amplifier
Overview Techniques and Paradigms 25 / 42
![Page 49: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/49.jpg)
Bounded Occurrence CSP Paradigm
Method: Bounded degree amplifier graphs(Berman and Karpinski, 1999)
v
Basic Idea:
Replace nodes corre-sponding to variablesby 3-regular amplifier
Overview Techniques and Paradigms 25 / 42
![Page 50: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/50.jpg)
Bounded Occurrence CSP Paradigm
From bounded occurrence CSP’s to vertex covers:Reduce bounded occurrence Hybrid (equationswith 2 and 3 variables) to Min-VC on degree dbounded graphs (d-Min-VC)
I Yields explicit lower bounds of 103102 for d = 3 and 55
54 ford = 4, 5 (Berman and Karpinski, 2003)
I For larger d assuming UGC: 2 − (2 + o(1)) log log dlog d
(Austrin, Khot, and M. Safra, 2009)
d-Min-VC serves as starting point for our CEAPreduction to Min-VC on PLG’s
Overview Techniques and Paradigms 26 / 42
![Page 51: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/51.jpg)
Bounded Occurrence CSP Paradigm
From bounded occurrence CSP’s to vertex covers:Reduce bounded occurrence Hybrid (equationswith 2 and 3 variables) to Min-VC on degree dbounded graphs (d-Min-VC)
I Yields explicit lower bounds of 103102 for d = 3 and 55
54 ford = 4, 5 (Berman and Karpinski, 2003)
I For larger d assuming UGC: 2 − (2 + o(1)) log log dlog d
(Austrin, Khot, and M. Safra, 2009)
d-Min-VC serves as starting point for our CEAPreduction to Min-VC on PLG’s
Overview Techniques and Paradigms 26 / 42
![Page 52: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/52.jpg)
Bounded Occurrence CSP Paradigm
From bounded occurrence CSP’s to vertex covers:Reduce bounded occurrence Hybrid (equationswith 2 and 3 variables) to Min-VC on degree dbounded graphs (d-Min-VC)
I Yields explicit lower bounds of 103102 for d = 3 and 55
54 ford = 4, 5 (Berman and Karpinski, 2003)
I For larger d assuming UGC: 2 − (2 + o(1)) log log dlog d
(Austrin, Khot, and M. Safra, 2009)
d-Min-VC serves as starting point for our CEAPreduction to Min-VC on PLG’s
Overview Techniques and Paradigms 26 / 42
![Page 53: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/53.jpg)
Bounded Occurrence CSP Paradigm
From bounded occurrence CSP’s to vertex covers:Reduce bounded occurrence Hybrid (equationswith 2 and 3 variables) to Min-VC on degree dbounded graphs (d-Min-VC)
I Yields explicit lower bounds of 103102 for d = 3 and 55
54 ford = 4, 5 (Berman and Karpinski, 2003)
I For larger d assuming UGC: 2 − (2 + o(1)) log log dlog d
(Austrin, Khot, and M. Safra, 2009)
d-Min-VC serves as starting point for our CEAPreduction to Min-VC on PLG’s
Overview Techniques and Paradigms 26 / 42
![Page 54: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/54.jpg)
Set Cover Paradigm
From set covering to dominating sets:
A Set Coverinstance (U, S)
A Min-DSinstance GU,S
GU,S instances will serve as starting point for ourCEAP reduction to Min-DS on PLG’s
Overview Techniques and Paradigms 27 / 42
![Page 55: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/55.jpg)
Set Cover Paradigm
From set covering to dominating sets:
A Set Coverinstance (U, S)
A Min-DSinstance GU,S
GU,S instances will serve as starting point for ourCEAP reduction to Min-DS on PLG’s
Overview Techniques and Paradigms 27 / 42
![Page 56: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/56.jpg)
Set Cover Paradigm
From set covering to dominating sets:
A Set Coverinstance (U, S)
A Min-DSinstance GU,S
GU,S instances will serve as starting point for ourCEAP reduction to Min-DS on PLG’s
Overview Techniques and Paradigms 27 / 42
![Page 57: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/57.jpg)
Set Cover Paradigm
From set covering to dominating sets:
A Set Coverinstance (U, S)
A Min-DSinstance GU,S
GU,S instances will serve as starting point for ourCEAP reduction to Min-DS on PLG’s
Overview Techniques and Paradigms 27 / 42
![Page 58: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/58.jpg)
Approximation Lower Bounds for
Minimum Dominating Set on
Connected Power Law Graphs
![Page 59: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/59.jpg)
Minimum Dominating Set
Definition (Min-DS)Input: A graph G = (V, E)
Output: A subset D ⊆ V such that for each vertexv ∈ V either v ∈ D or D∪N(v) 6= ∅
Objective: Minimize |D|
Main Results Approximation Lower Bounds for Min-DS on PLG’s 29 / 42
![Page 60: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/60.jpg)
Minimum Dominating Set
Definition (Min-DS)Input: A graph G = (V, E)
Output: A subset D ⊆ V such that for each vertexv ∈ V either v ∈ D or D∪N(v) 6= ∅
Objective: Minimize |D|
7→
Dominating Set
7→
Minimum Dominating Set
Main Results Approximation Lower Bounds for Min-DS on PLG’s 29 / 42
![Page 61: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/61.jpg)
Minimum Dominating Set
Definition (Min-DS)Input: A graph G = (V, E)
Output: A subset D ⊆ V such that for each vertexv ∈ V either v ∈ D or D∪N(v) 6= ∅
Objective: Minimize |D|
7→
Dominating Set
7→
Minimum Dominating Set
Main Results Approximation Lower Bounds for Min-DS on PLG’s 29 / 42
![Page 62: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/62.jpg)
Minimum Dominating Set
Definition (Min-DS)Input: A graph G = (V, E)
Output: A subset D ⊆ V such that for each vertexv ∈ V either v ∈ D or D∪N(v) 6= ∅
Objective: Minimize |D|
Approximability on general graphs:
Upper bound: ln n (Johnson, 1974; Lovasz, 1975)
Lower bound: (1 − o(1)) ln n (Feige, 1998)
Main Results Approximation Lower Bounds for Min-DS on PLG’s 29 / 42
![Page 63: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/63.jpg)
Minimum Dominating Set
Definition (Min-DS)Input: A graph G = (V, E)
Output: A subset D ⊆ V such that for each vertexv ∈ V either v ∈ D or D∪N(v) 6= ∅
Objective: Minimize |D|
Approximability on general graphs:
Upper bound: ln n (Johnson, 1974; Lovasz, 1975)
Lower bound: (1 − o(1)) ln n (Feige, 1998)
Main Results Approximation Lower Bounds for Min-DS on PLG’s 29 / 42
![Page 64: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/64.jpg)
Minimum Dominating Set
Definition (Min-DS)Input: A graph G = (V, E)
Output: A subset D ⊆ V such that for each vertexv ∈ V either v ∈ D or D∪N(v) 6= ∅
Objective: Minimize |D|
Approximability on general graphs:
Upper bound: ln n (Johnson, 1974; Lovasz, 1975)
Lower bound: (1 − o(1)) ln n (Feige, 1998)
Main Results Approximation Lower Bounds for Min-DS on PLG’s 29 / 42
![Page 65: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/65.jpg)
Previous Results
Approximability on PLG’s:
For all β > 0, NP-hard on simple disconnectedPLG’s (Ferrante, Pandurangan, and Park, 2008)
For all β > 1, APX-hard on disconnected power lawmultigraphs (Shen et al., 2012)
I Explicit inapproximability factors for 1 < β 6 2:
Simple PLG’s General PLG’s
1 +1
390(2ζ(β)3β − 1)1 +
13120ζ(β)3β
Main Results Approximation Lower Bounds for Min-DS on PLG’s 30 / 42
![Page 66: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/66.jpg)
Previous Results
Approximability on PLG’s:
For all β > 0, NP-hard on simple disconnectedPLG’s (Ferrante, Pandurangan, and Park, 2008)
For all β > 1, APX-hard on disconnected power lawmultigraphs (Shen et al., 2012)
I Explicit inapproximability factors for 1 < β 6 2:
Simple PLG’s General PLG’s
1 +1
390(2ζ(β)3β − 1)1 +
13120ζ(β)3β
Main Results Approximation Lower Bounds for Min-DS on PLG’s 30 / 42
![Page 67: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/67.jpg)
Previous Results
Approximability on PLG’s:
For all β > 0, NP-hard on simple disconnectedPLG’s (Ferrante, Pandurangan, and Park, 2008)
For all β > 1, APX-hard on disconnected power lawmultigraphs (Shen et al., 2012)
I Explicit inapproximability factors for 1 < β 6 2:
Simple PLG’s General PLG’s
1 +1
390(2ζ(β)3β − 1)1 +
13120ζ(β)3β
Main Results Approximation Lower Bounds for Min-DS on PLG’s 30 / 42
![Page 68: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/68.jpg)
Previous Results
Approximability on PLG’s:
For all β > 0, NP-hard on simple disconnectedPLG’s (Ferrante, Pandurangan, and Park, 2008)
For all β > 1, APX-hard on disconnected power lawmultigraphs (Shen et al., 2012)
I Explicit inapproximability factors for 1 < β 6 2:
Simple PLG’s General PLG’s
1 +1
390(2ζ(β)3β − 1)1 +
13120ζ(β)3β
Main Results Approximation Lower Bounds for Min-DS on PLG’s 30 / 42
![Page 69: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/69.jpg)
Hardness of Min-DS on PLG’s
Open QuestionsIs Minimum Dominating Set NP-hard andAPX-hard on connected PLG’s?
Can we close the gap between the constant lowerbounds on PLG’s and the general logarithmic lowerbound?
Can we extend the results to the range β ∈ [0, 1]?
Theorem (Gast, Hauptmann, and Karpinski, 2012)
For all β ∈ [0 + ε, 2] and ε > 0, Min-DS is hard toapproximate within Ω(ln(n) − cβ) on connected PLG’s
Main Results Approximation Lower Bounds for Min-DS on PLG’s 31 / 42
![Page 70: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/70.jpg)
Hardness of Min-DS on PLG’s
Open QuestionsIs Minimum Dominating Set NP-hard andAPX-hard on connected PLG’s?
Can we close the gap between the constant lowerbounds on PLG’s and the general logarithmic lowerbound?
Can we extend the results to the range β ∈ [0, 1]?
Theorem (Gast, Hauptmann, and Karpinski, 2012)
For all β ∈ [0 + ε, 2] and ε > 0, Min-DS is hard toapproximate within Ω(ln(n) − cβ) on connected PLG’s
Main Results Approximation Lower Bounds for Min-DS on PLG’s 31 / 42
![Page 71: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/71.jpg)
Hardness of Min-DS on PLG’s
Open QuestionsIs Minimum Dominating Set NP-hard andAPX-hard on connected PLG’s?
Can we close the gap between the constant lowerbounds on PLG’s and the general logarithmic lowerbound?
Can we extend the results to the range β ∈ [0, 1]?
Theorem (Gast, Hauptmann, and Karpinski, 2012)
For all β ∈ [0 + ε, 2] and ε > 0, Min-DS is hard toapproximate within Ω(ln(n) − cβ) on connected PLG’s
Main Results Approximation Lower Bounds for Min-DS on PLG’s 31 / 42
![Page 72: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/72.jpg)
Hardness of Min-DS on PLG’s
Open QuestionsIs Minimum Dominating Set NP-hard andAPX-hard on connected PLG’s?
Can we close the gap between the constant lowerbounds on PLG’s and the general logarithmic lowerbound?
Can we extend the results to the range β ∈ [0, 1]?
Theorem (Gast, Hauptmann, and Karpinski, 2012)
For all β ∈ [0 + ε, 2] and ε > 0, Min-DS is hard toapproximate within Ω(ln(n) − cβ) on connected PLG’s
Main Results Approximation Lower Bounds for Min-DS on PLG’s 31 / 42
![Page 73: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/73.jpg)
Reduction
Embedding technique (CEAP reduction):Map GU,S to Gα,β via scaling constructionconnecting to a multigraph wheel W
I Number of edges between GU,S and W isO(min|GU,S|, |W|)
Vertex set Γ separates GU,S from Gα,β \ GU,SI Hardness on maximal component GU,S is preserved
Maintain small set X to dominate all vertices in WI Min-DS is polynomially solvable on W
Main Results Approximation Lower Bounds for Min-DS on PLG’s 32 / 42
![Page 74: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/74.jpg)
Reduction
Embedding technique (CEAP reduction):Map GU,S to Gα,β via scaling constructionconnecting to a multigraph wheel W
I Number of edges between GU,S and W isO(min|GU,S|, |W|)
Vertex set Γ separates GU,S from Gα,β \ GU,SI Hardness on maximal component GU,S is preserved
Maintain small set X to dominate all vertices in WI Min-DS is polynomially solvable on W
Main Results Approximation Lower Bounds for Min-DS on PLG’s 32 / 42
![Page 75: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/75.jpg)
Reduction
Embedding technique (CEAP reduction):Map GU,S to Gα,β via scaling constructionconnecting to a multigraph wheel W
I Number of edges between GU,S and W isO(min|GU,S|, |W|)
Vertex set Γ separates GU,S from Gα,β \ GU,SI Hardness on maximal component GU,S is preserved
Maintain small set X to dominate all vertices in WI Min-DS is polynomially solvable on W
Main Results Approximation Lower Bounds for Min-DS on PLG’s 32 / 42
![Page 76: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/76.jpg)
Reduction
Embedding technique (CEAP reduction):Map GU,S to Gα,β via scaling constructionconnecting to a multigraph wheel W
I Number of edges between GU,S and W isO(min|GU,S|, |W|)
Vertex set Γ separates GU,S from Gα,β \ GU,SI Hardness on maximal component GU,S is preserved
Maintain small set X to dominate all vertices in WI Min-DS is polynomially solvable on W
Main Results Approximation Lower Bounds for Min-DS on PLG’s 32 / 42
![Page 77: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/77.jpg)
Reduction
Embedding technique (CEAP reduction):Map GU,S to Gα,β via scaling constructionconnecting to a multigraph wheel W
I Number of edges between GU,S and W isO(min|GU,S|, |W|)
Vertex set Γ separates GU,S from Gα,β \ GU,SI Hardness on maximal component GU,S is preserved
Maintain small set X to dominate all vertices in WI Min-DS is polynomially solvable on W
Main Results Approximation Lower Bounds for Min-DS on PLG’s 32 / 42
![Page 78: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/78.jpg)
Reduction
Embedding technique (CEAP reduction):Map GU,S to Gα,β via scaling constructionconnecting to a multigraph wheel W
I Number of edges between GU,S and W isO(min|GU,S|, |W|)
Vertex set Γ separates GU,S from Gα,β \ GU,SI Hardness on maximal component GU,S is preserved
Maintain small set X to dominate all vertices in WI Min-DS is polynomially solvable on W
Main Results Approximation Lower Bounds for Min-DS on PLG’s 32 / 42
![Page 79: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/79.jpg)
The Reduction
Γ ⊆ V2
GU,S
W
X
Main Results Approximation Lower Bounds for Min-DS on PLG’s 33 / 42
![Page 80: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/80.jpg)
Phase Transitions
ObservationFor β > 2, Min-DS on Gα,β PLG’s is in APX
Analysis of the phase transition:
Study of functional case βf = 2 + 1f (n)
I Hard to approximate within Ω(ln(n) − cβ) forf (n) = ω(log(n))
I In APX for f (n) = o(log(n)) (!)
Main Results Approximation Lower Bounds for Min-DS on PLG’s 34 / 42
![Page 81: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/81.jpg)
Phase Transitions
ObservationFor β > 2, Min-DS on Gα,β PLG’s is in APX
Analysis of the phase transition:
Study of functional case βf = 2 + 1f (n)
I Hard to approximate within Ω(ln(n) − cβ) forf (n) = ω(log(n))
I In APX for f (n) = o(log(n)) (!)
Main Results Approximation Lower Bounds for Min-DS on PLG’s 34 / 42
![Page 82: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/82.jpg)
Phase Transitions
ObservationFor β > 2, Min-DS on Gα,β PLG’s is in APX
Analysis of the phase transition:
Study of functional case βf = 2 + 1f (n)
I Hard to approximate within Ω(ln(n) − cβ) forf (n) = ω(log(n))
I In APX for f (n) = o(log(n)) (!)
Main Results Approximation Lower Bounds for Min-DS on PLG’s 34 / 42
![Page 83: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/83.jpg)
Phase Transitions
ObservationFor β > 2, Min-DS on Gα,β PLG’s is in APX
Analysis of the phase transition:
Study of functional case βf = 2 + 1f (n)
I Hard to approximate within Ω(ln(n) − cβ) forf (n) = ω(log(n))
I In APX for f (n) = o(log(n)) (!)
Main Results Approximation Lower Bounds for Min-DS on PLG’s 34 / 42
![Page 84: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/84.jpg)
Approximation Upper Bounds for
Minimum Vertex Cover on
Random Power Law Graphs
![Page 85: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/85.jpg)
Minimum Vertex Cover Problem
Definition (Min-VC)Input: A graph G = (V, E)
Output: A subset C ⊆ V such that each edgeu, v ∈ E has at least one endpoint in C
Objective: Minimize |C|
Main Results Approximation Upper Bounds for Min-VC on PLG’s 36 / 42
![Page 86: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/86.jpg)
Minimum Vertex Cover Problem
Definition (Min-VC)Input: A graph G = (V, E)
Output: A subset C ⊆ V such that each edgeu, v ∈ E has at least one endpoint in C
Objective: Minimize |C|
7→
Vertex Cover
7→
Minimum Vertex Cover
Main Results Approximation Upper Bounds for Min-VC on PLG’s 36 / 42
![Page 87: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/87.jpg)
Minimum Vertex Cover Problem
Definition (Min-VC)Input: A graph G = (V, E)
Output: A subset C ⊆ V such that each edgeu, v ∈ E has at least one endpoint in C
Objective: Minimize |C|
7→
Vertex Cover
7→
Minimum Vertex Cover
Main Results Approximation Upper Bounds for Min-VC on PLG’s 36 / 42
![Page 88: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/88.jpg)
Minimum Vertex Cover Problem
Definition (Min-VC)Input: A graph G = (V, E)
Output: A subset C ⊆ V such that each edgeu, v ∈ E has at least one endpoint in C
Objective: Minimize |C|
Approximability on general graphs:
Upper bound: 2 −Θ(1/√
log n) (Karakostas, 2009)
Lower bounds:I 2 − ε assuming UGC (Khot and Regev, 2008)I 1.3606 assuming P 6= NP (Dinur and S. Safra, 2005)
Main Results Approximation Upper Bounds for Min-VC on PLG’s 36 / 42
![Page 89: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/89.jpg)
Minimum Vertex Cover Problem
Definition (Min-VC)Input: A graph G = (V, E)
Output: A subset C ⊆ V such that each edgeu, v ∈ E has at least one endpoint in C
Objective: Minimize |C|
Approximability on general graphs:
Upper bound: 2 −Θ(1/√
log n) (Karakostas, 2009)
Lower bounds:I 2 − ε assuming UGC (Khot and Regev, 2008)I 1.3606 assuming P 6= NP (Dinur and S. Safra, 2005)
Main Results Approximation Upper Bounds for Min-VC on PLG’s 36 / 42
![Page 90: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/90.jpg)
Minimum Vertex Cover Problem
Definition (Min-VC)Input: A graph G = (V, E)
Output: A subset C ⊆ V such that each edgeu, v ∈ E has at least one endpoint in C
Objective: Minimize |C|
Approximability on general graphs:
Upper bound: 2 −Θ(1/√
log n) (Karakostas, 2009)
Lower bounds:I 2 − ε assuming UGC (Khot and Regev, 2008)I 1.3606 assuming P 6= NP (Dinur and S. Safra, 2005)
Main Results Approximation Upper Bounds for Min-VC on PLG’s 36 / 42
![Page 91: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/91.jpg)
Minimum Vertex Cover Problem
Definition (Min-VC)Input: A graph G = (V, E)
Output: A subset C ⊆ V such that each edgeu, v ∈ E has at least one endpoint in C
Objective: Minimize |C|
Approximability on general graphs:
Upper bound: 2 −Θ(1/√
log n) (Karakostas, 2009)
Lower bounds:I 2 − ε assuming UGC (Khot and Regev, 2008)I 1.3606 assuming P 6= NP (Dinur and S. Safra, 2005)
Main Results Approximation Upper Bounds for Min-VC on PLG’s 36 / 42
![Page 92: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/92.jpg)
Previous Results
Approximability on PLG’s:
ObservationThere exists practical evidence that Min-VC is easier toapproximate on PLG’s
The greedy algorithm often outperforms the2-approximation algorithm (Park and Lee, 2001)
Achieves average ratios of ∼ 1.02 on real worldnetwork topologies (M. O. Da Silva, Gimenez-Lugo, and
M. V. G. Da Silva, 2013)
Main Results Approximation Upper Bounds for Min-VC on PLG’s 37 / 42
![Page 93: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/93.jpg)
Previous Results
Approximability on PLG’s:
ObservationThere exists practical evidence that Min-VC is easier toapproximate on PLG’s
The greedy algorithm often outperforms the2-approximation algorithm (Park and Lee, 2001)
Achieves average ratios of ∼ 1.02 on real worldnetwork topologies (M. O. Da Silva, Gimenez-Lugo, and
M. V. G. Da Silva, 2013)
Main Results Approximation Upper Bounds for Min-VC on PLG’s 37 / 42
![Page 94: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/94.jpg)
Previous Results
Approximability on PLG’s:
ObservationThere exists practical evidence that Min-VC is easier toapproximate on PLG’s
The greedy algorithm often outperforms the2-approximation algorithm (Park and Lee, 2001)
Achieves average ratios of ∼ 1.02 on real worldnetwork topologies (M. O. Da Silva, Gimenez-Lugo, and
M. V. G. Da Silva, 2013)
Main Results Approximation Upper Bounds for Min-VC on PLG’s 37 / 42
![Page 95: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/95.jpg)
Previous Results
Approximability on PLG’s:
ObservationThere exists practical evidence that Min-VC is easier toapproximate on PLG’s
The greedy algorithm often outperforms the2-approximation algorithm (Park and Lee, 2001)
Achieves average ratios of ∼ 1.02 on real worldnetwork topologies (M. O. Da Silva, Gimenez-Lugo, and
M. V. G. Da Silva, 2013)
Main Results Approximation Upper Bounds for Min-VC on PLG’s 37 / 42
![Page 96: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/96.jpg)
Approximation of Min-VC on PLG’s
Open QuestionCan we give provable guarantees that Min-VC is easierto approximate on PLG’s?
Theorem (Gast and Hauptmann, 2012)
There exists an approximation algorithm for Min-VC onrandom Gα,β PLG’s with expected approximation ratio
ρ 6 2 −ζ(β) − 1 − 1
2β
2βζ(β− 1)ζ(β)
Main Results Approximation Upper Bounds for Min-VC on PLG’s 38 / 42
![Page 97: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/97.jpg)
Approximation of Min-VC on PLG’s
Open QuestionCan we give provable guarantees that Min-VC is easierto approximate on PLG’s?
Theorem (Gast and Hauptmann, 2012)
There exists an approximation algorithm for Min-VC onrandom Gα,β PLG’s with expected approximation ratio
ρ 6 2 −ζ(β) − 1 − 1
2β
2βζ(β− 1)ζ(β)
Main Results Approximation Upper Bounds for Min-VC on PLG’s 38 / 42
![Page 98: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/98.jpg)
Half Integral Solutions
Consider the following LP-Relaxation for Min-VC:
Main Results Approximation Upper Bounds for Min-VC on PLG’s 39 / 42
![Page 99: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/99.jpg)
Half Integral Solutions
minimizen∑
i=1
wixi,
subject to xi + xj > 1, for all edges e = vi, vj,
xi > 0, for all vertices vi ∈ V
There always exists optimal solution which ishalf-integral, i.e. ∀i : xi ∈ 0, 1/2, 1 andvi ∈ V0, V1/2, V1, respectively
A half-integral solution can be computed inpolynomial time (using algorithm for Min-VC orPerfect Matching in bipartite graphs)
Main Results Approximation Upper Bounds for Min-VC on PLG’s 39 / 42
![Page 100: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/100.jpg)
Half Integral Solutions
minimizen∑
i=1
wixi,
subject to xi + xj > 1, for all edges e = vi, vj,
xi > 0, for all vertices vi ∈ V
There always exists optimal solution which ishalf-integral, i.e. ∀i : xi ∈ 0, 1/2, 1 andvi ∈ V0, V1/2, V1, respectively
A half-integral solution can be computed inpolynomial time (using algorithm for Min-VC orPerfect Matching in bipartite graphs)
Main Results Approximation Upper Bounds for Min-VC on PLG’s 39 / 42
![Page 101: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/101.jpg)
Half Integral Solutions
minimizen∑
i=1
wixi,
subject to xi + xj > 1, for all edges e = vi, vj,
xi > 0, for all vertices vi ∈ V
There always exists optimal solution which ishalf-integral, i.e. ∀i : xi ∈ 0, 1/2, 1 andvi ∈ V0, V1/2, V1, respectively
A half-integral solution can be computed inpolynomial time (using algorithm for Min-VC orPerfect Matching in bipartite graphs)
Main Results Approximation Upper Bounds for Min-VC on PLG’s 39 / 42
![Page 102: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/102.jpg)
Approximation Algorithm
Start with half-integralsolution x : V → 0, 1/2, 1
Apply new deterministicrounding algorithm to x
Prove that algorithm achievesratio of 3/2 on subset V ′ ⊆
V of low-degree verticesand their neighborhood
Overall approximation ratio asconvex combination of ratio 3/2
on V ′ and ratio 2 on V \ V ′
Prove lower bounds on x(V ′)and upper bounds on x(V)
to determine the effect of therounding on global solution
Main Results Approximation Upper Bounds for Min-VC on PLG’s 40 / 42
![Page 103: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/103.jpg)
Approximation Algorithm
Start with half-integralsolution x : V → 0, 1/2, 1
Apply new deterministicrounding algorithm to x
Prove that algorithm achievesratio of 3/2 on subset V ′ ⊆
V of low-degree verticesand their neighborhood
Overall approximation ratio asconvex combination of ratio 3/2
on V ′ and ratio 2 on V \ V ′
Prove lower bounds on x(V ′)and upper bounds on x(V)
to determine the effect of therounding on global solution
Main Results Approximation Upper Bounds for Min-VC on PLG’s 40 / 42
![Page 104: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/104.jpg)
Approximation Algorithm
Start with half-integralsolution x : V → 0, 1/2, 1
Apply new deterministicrounding algorithm to x
Prove that algorithm achievesratio of 3/2 on subset V ′ ⊆
V of low-degree verticesand their neighborhood
Overall approximation ratio asconvex combination of ratio 3/2
on V ′ and ratio 2 on V \ V ′
Prove lower bounds on x(V ′)and upper bounds on x(V)
to determine the effect of therounding on global solution
Main Results Approximation Upper Bounds for Min-VC on PLG’s 40 / 42
![Page 105: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/105.jpg)
Approximation Algorithm
Start with half-integralsolution x : V → 0, 1/2, 1
Apply new deterministicrounding algorithm to x
Prove that algorithm achievesratio of 3/2 on subset V ′ ⊆
V of low-degree verticesand their neighborhood
Overall approximation ratio asconvex combination of ratio 3/2
on V ′ and ratio 2 on V \ V ′
Prove lower bounds on x(V ′)and upper bounds on x(V)
to determine the effect of therounding on global solution
Main Results Approximation Upper Bounds for Min-VC on PLG’s 40 / 42
![Page 106: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/106.jpg)
Approximation Algorithm
Start with half-integralsolution x : V → 0, 1/2, 1
Apply new deterministicrounding algorithm to x
Prove that algorithm achievesratio of 3/2 on subset V ′ ⊆
V of low-degree verticesand their neighborhood
Overall approximation ratio asconvex combination of ratio 3/2
on V ′ and ratio 2 on V \ V ′
Prove lower bounds on x(V ′)and upper bounds on x(V)
to determine the effect of therounding on global solution
Main Results Approximation Upper Bounds for Min-VC on PLG’s 40 / 42
![Page 107: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/107.jpg)
Open Problems and Further Research
Still improving on the presented resultsI Investigating the gap between upper and lower
approximation bound for Min-VC on PLG’sI Improving upper bounds for Min-DS on PLG’s whenβ 6 2 (in random or quasi-random models)
Exploit network hyperbolicity in biological andInternet based network design problems
Computational complexity of node and edgedeletion problems and information spreading indynamic networks (especially in biological settings)
Applicability of graph limit theory in order to gathertopological information of PLG generating processes
Conclusion Open Problems and Further Research 41 / 42
![Page 108: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/108.jpg)
Open Problems and Further Research
Still improving on the presented resultsI Investigating the gap between upper and lower
approximation bound for Min-VC on PLG’sI Improving upper bounds for Min-DS on PLG’s whenβ 6 2 (in random or quasi-random models)
Exploit network hyperbolicity in biological andInternet based network design problems
Computational complexity of node and edgedeletion problems and information spreading indynamic networks (especially in biological settings)
Applicability of graph limit theory in order to gathertopological information of PLG generating processes
Conclusion Open Problems and Further Research 41 / 42
![Page 109: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/109.jpg)
Open Problems and Further Research
Still improving on the presented resultsI Investigating the gap between upper and lower
approximation bound for Min-VC on PLG’sI Improving upper bounds for Min-DS on PLG’s whenβ 6 2 (in random or quasi-random models)
Exploit network hyperbolicity in biological andInternet based network design problems
Computational complexity of node and edgedeletion problems and information spreading indynamic networks (especially in biological settings)
Applicability of graph limit theory in order to gathertopological information of PLG generating processes
Conclusion Open Problems and Further Research 41 / 42
![Page 110: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/110.jpg)
Open Problems and Further Research
Still improving on the presented resultsI Investigating the gap between upper and lower
approximation bound for Min-VC on PLG’sI Improving upper bounds for Min-DS on PLG’s whenβ 6 2 (in random or quasi-random models)
Exploit network hyperbolicity in biological andInternet based network design problems
Computational complexity of node and edgedeletion problems and information spreading indynamic networks (especially in biological settings)
Applicability of graph limit theory in order to gathertopological information of PLG generating processes
Conclusion Open Problems and Further Research 41 / 42
![Page 111: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/111.jpg)
Open Problems and Further Research
Still improving on the presented resultsI Investigating the gap between upper and lower
approximation bound for Min-VC on PLG’sI Improving upper bounds for Min-DS on PLG’s whenβ 6 2 (in random or quasi-random models)
Exploit network hyperbolicity in biological andInternet based network design problems
Computational complexity of node and edgedeletion problems and information spreading indynamic networks (especially in biological settings)
Applicability of graph limit theory in order to gathertopological information of PLG generating processes
Conclusion Open Problems and Further Research 41 / 42
![Page 112: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/112.jpg)
Open Problems and Further Research
Still improving on the presented resultsI Investigating the gap between upper and lower
approximation bound for Min-VC on PLG’sI Improving upper bounds for Min-DS on PLG’s whenβ 6 2 (in random or quasi-random models)
Exploit network hyperbolicity in biological andInternet based network design problems
Computational complexity of node and edgedeletion problems and information spreading indynamic networks (especially in biological settings)
Applicability of graph limit theory in order to gathertopological information of PLG generating processes
Conclusion Open Problems and Further Research 41 / 42
![Page 113: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/113.jpg)
Open Problems and Further Research
Still improving on the presented resultsI Investigating the gap between upper and lower
approximation bound for Min-VC on PLG’sI Improving upper bounds for Min-DS on PLG’s whenβ 6 2 (in random or quasi-random models)
Exploit network hyperbolicity in biological andInternet based network design problems
Computational complexity of node and edgedeletion problems and information spreading indynamic networks (especially in biological settings)
Applicability of graph limit theory in order to gathertopological information of PLG generating processes
Conclusion Open Problems and Further Research 41 / 42
![Page 114: Approximability of Combinatorial Optimization Problems on ...theory.cs.uni-bonn.de/~gast/presentations/Talk_Thesis.pdf · Uniform random graph vs. power law random graph Number of](https://reader034.vdocuments.mx/reader034/viewer/2022042401/5f1068c07e708231d448f7c3/html5/thumbnails/114.jpg)
Thank you!