Minors & Algorithms - ICERM...Algorithms via Minors β’Theorem: Polynomial test for a fixed minor...
Transcript of Minors & Algorithms - ICERM...Algorithms via Minors β’Theorem: Polynomial test for a fixed minor...
Minors
β’ π»π» is a minor of πΊπΊ ifπΊπΊ can reach π»π» via edge deletions edge contractions
v w
v wdelete
vwcontract
πΎπΎ3,3 deletecontract
is a minor of
Wagnerβs Conjecture[Robertson & Seymour 2004]β’ Minor relation is a well-quasi-ordering:
Every infinite graph sequence πΊπΊ1,πΊπΊ2,πΊπΊ3, β¦has some πΊπΊππ a minor of πΊπΊππ (with ππ < ππ)
β¦or equivalentlyβ¦β’ Every minor-closed graph family is
characterized by a finite set of excluded minors Proof by contradiction: Infinite excluded minors
would have a minor relation among them In some sense necessarily nonconstructive
[Friedman, Robertson, Seymour 1987]
Wagnerβs Conjecture[Robertson & Seymour 2004]β’ Every minor-closed graph family is
characterized by a finite set of excluded minorsβ’ Examples: Forests: πΎπΎ3 Outerplanar: πΎπΎ4 & πΎπΎ2,3
Series-parallel: πΎπΎ4 Planar: πΎπΎ5 & πΎπΎ3,3
[Wagnerβs 1937] Linklessly embeddable:
Petersen family [Robertson & Seymour 1995]
Algorithms via Minors
β’ Theorem: Polynomial test for a fixed minor π»π» ππ π»π» ππ3 [GM XII, 1995] ππ π»π» ππ2 [Kawarabayashi, Kobayashi, Reed 2012]
β’ Corollary: Any minor-closed graph property has a polynomial-time decision algorithm But algorithm is existential:
in general, donβt know whichexcluded minors to test for
Warning:Graph Minor Constantsβ’ Example: Planar minor testing [GM5 1986],
analyzed by David Johnson [1987] Running time = ππ ππ3
Lead constant β 222π₯π₯
, π₯π₯ = 222β β β
222β β β
β¦
β
β
12ππ π»π»
Fixed-Parameter Algorithmsvia Minorsβ’ Graph parameter is minor-closed if it only
decreases when deleting or contracting Examples: vertex cover, feedback vertex set
β’ Corollary: Any minor-closed graph parameter has an ππ ππ ππ2 algorithm for deciding β€ ππ βFixed-parameter tractableβ But a different, unknown algorithm for each ππ Not really a (uniform) fixed-parameter algorithm[Fellows & Langston 1988]
Fixed-Parameter Algorithmsvia Minorsβ’ Theorem: Any minor-closed graph parameter nonzero on some planar graph, at least the sum over connected components in a
disconnected graph, and fixed-parameter tractable with respect to treewidth
can be solved in ππ ππ ππ by a known algorithm[Demaine & Hajiaghayi 2007]
What Are π―π―-minor-free Graphs Like?
β’ Bounded average degree (& degeneracy):ππ ππ π»π» log ππ π»π»
[Kostochka 1982; Thomason 1984]
β’ ππ ππ treewidth [Grohe 2003]β’ β ππ ππ balanced separators
(like Lipton-Tarjan planar separator theorem)[Alon, Seymour, Thomas 1990]
β’ Kinda like planar graphs?
π―π―-Minor-Free Graphs[Robertson & Seymour 2003]
Structure Theorem:β’ π»π»-minor-free graph looks like a tree of
βalmost embeddableβ graphs: Base graph drawn on
surface of genus ππ(π»π») ππ(π»π») vortex faces
filled with graphsof pathwidth ππ(π»π») ππ(π»π») apex vertices
connected to anything
π―π―-Minor-Free Graphs[Robertson & Seymour 2003]
Structure Theorem:β’ π»π»-minor-free graph looks like a tree of
βalmost embeddableβ graphs
Structure Algorithms:β’ ππππ π»π» [Demaine, Hajiaghayi,
Kawarabayashi β FOCS 2005]
β’ ππ ππ3 [Kawarabayashi &Wollan β STOC 2011]
β’ ππ ππ2 [Grohe, Kawarabayashi,Reed β SODA 2013]
Grid Minors vs. Treewidth
β’ Every π»π»-minor-free graph of treewidth β₯ππ π»π» ππ has an ππ Γ ππ grid minor[Demaine & Hajiaghayi 2005] Previous bounds exponential in ππ (and π»π») [GM5, β¦]
ππ π»π» = |ππ π»π» |ππ |πΈπΈ π»π» | [Kawarabayashi & Kobayashi 2012]
β’ Every graph of treewidth β₯ ππ1 ππππ2 has an ππ Γ ππgrid minor [Chekuri & Chuzhoy 2013] ππ2 > 2 necessary [Robertson, Seymour, Thomas 1994]
Previous bounds exponential in ππ [RST94]
β’ Almost-embeddable graphs: Remove apices by deletion Contract each vortex to a vertex Pull apart each handle
[Demaine, Fomin, Hajiaghayi, Thilikos 2004]
Planar graph has large grid minor [RST94]
β’ Tree takes max of treewidths, so onealmost-embeddable graph has full treewidth But may not be π»π»-minor-free from tree joins βApproximateβ graph to make minor of original
graph β π»π»-minor-free [Demaine & Hajiaghayi 2005]
Grid Minors vs. Treewidthin π―π―-Minor-Free Graphs
Apex-Minor-Free Graphs
β’ Bounded local treewidth =radius-ππ neighborhood around every vertex π£π£has treewidth β€ ππ ππ Examples: planar and bounded-genus graphs
β’ Minor-closed graph family hasbounded local treewidth βexcludes an apex graph =minor of planar graph + one vertex[Eppstein 2000] Call such graphs apex-minor-free
ππ = 1ππ = 2
π£π£
ππ = 1ππ = 2
π£π£
Apex-Minor-Free Graphs
β’ Structure Theorem:Apices attach only to vortices,allowing bounded-treewidth vortices[Demaine, Hajiaghayi, Kawarabayashi 2009] Additive +2 approximation for chromatic number
(extending from bounded genus [Thomassen 1997])β’ Bounded local treewidth ππ ππ = 22ππ ππ
[Eppstein 2000] ππ ππ = 2ππ ππ [Demaine & Hajiaghayi 2004a] ππ ππ = ππ ππ [Demaine & Hajiaghayi 2004b]
[EPTASs: 222ππ β1 ππ
ππππ 1 β 2ππ β1 ππ ππππ 1 ]
odd-π»π»-minor-free
Beyond Graph Minors
π»π»-minor-free
apex-minor-free
bounded genus
planar
Minors
β’ Equivalently, π»π» is a minor of πΊπΊ if Every vertex π£π£ in π»π» has a corresp. tree πππ£π£ in πΊπΊ Any edge π£π£,π€π€ in π»π» has a corresponding edge
connecting πππ£π£ to πππ€π€ in πΊπΊ
is a minor of
K3,3
v w
v wdelete
vwcontract
deletecontract
βmodelβ
Odd Minors
β’ π»π» is an odd minor of πΊπΊ if we canalso 2-color the vertices in πΊπΊ so that Tree edges are bichromatic πππ£π£-to-πππ€π€ edges are monochromatic
K3,3
v w
v wdelete
vwcontract
deletecontract
is an oddminor of
Odd Minors
β’ π»π» is an odd minor of πΊπΊ if we canalso 2-color the vertices in πΊπΊ so that Tree edges are bichromatic πππ£π£-to-πππ€π€ edges are monochromatic
v w
v wdelete
vwcontract
is an oddminor of
K3,3 deletecontract
Odd Minors
β’ π»π» is an odd minor of πΊπΊ if we canalso 2-color the vertices in πΊπΊ so that Tree edges are bichromatic πππ£π£-to-πππ€π€ edges are monochromatic
K3,3 deletecontract
v w
v wdelete
vwcontract
is an oddminor of
Odd-Minor-Free Graphs[Demaine, Hajiaghayi, Kawarabayashi 2010]
β’ Structure Theorem: Every odd-π»π»-minor-free graph can be written as a tree of graphs joined along ππ(π»π»)-size cliques, where each term is Bounded-genus graph
+ ππ(π»π») vortices+ ππ π»π» apices(joined at only 3 surface vertices) Bipartite graph
+ ππ(π»π») apices(joined at only 1 bipartite vertex)
β’ ππππ π»π» algorithm ππ π»π» ππππ 1 simpler decomp. [Tazari 2012]
any problem satisfying:
Bidimensionality Overview
closed under contraction [& deletion]
OPT costly on ππ Γ ππ grid
minor π»π»
subexponential fixed-parameter
algorithm: 2ππ OPT ππππ 1
efficient PTAS: ππ ππ ππππ 1 time
linear kernelization: ππ β ππ(OPT)
β¦ on π»π»-minor-free graphs
History of Subexponential FPT
β’ In the early days, most fixed-parameter algorithms ran in 2ππ ππππ ππππ 1 time, for ππ β₯ 1
β’ Natural question: Is 2ππ ππ ππππ 1 possible?
History of Subexponential FPT
β’ Vertex Cover: choose smallest set of verticesto cover every edge (on either endpoint)
β’ Vertex Cover can be kernelized to β€ 2ππ verticesβ’ Lipton & Tarjanβs separator approach [1980]
solves Planar Independent Set in 2ππ ππ time Min. Vertex Cover + Max. Independent Set = |ππ|
β’ β ππππ 1 + 2ππ ππ time
v w v w v w v wcover
History of Subexponential FPT
β’ What about problems like these? Dominating Set: Cover all vertices
with ππ vertex neighborhoods Feedback Vertex Set: Cover cycles with ππ vertices Long Path: Is there a simple path of length β₯ ππ?
β’ No linear (or even polynomial) kernel was known for these problems then
β’ We now know that some, e.g. Long Path,have no polynomial kernel (unless NP β
βcoNP poly), even for planar graphs[Bodlaender, Downey, Fellows, Hermelin 2009]
History of Subexponential FPT
β’ 2ππ ππ ππππ 1 for planar Dominating Set[Alber, Bodlaender, Fernau, Kloks, Niedermeier 2000]
β’ Same approach extended to Dominating Set variations, Feedback Vertex Set, etc.[Alber et al. 2002; Kanj & PerkoviΔ 2002; Fomin & Thilikos2003; Alber, Fernau, Niedermeier 2004; Chang, Kloks, Lee 2001; Kloks, Lee, Liu 2002; Gutin, Kloks, Lee 2001]
any problem satisfying:
Bidimensionality Overview
closed under contraction [& deletion]
OPT costly on ππ Γ ππ grid
minor π»π»
parameter-treewidth
bound
subexponential fixed-parameter
algorithm
efficient PTAS
linear kernel
β¦ on π»π»-minor-free graphs
Bidimensionality (version 1)[Demaine, Fomin, Hajiaghayi, Thilikos 2004]
β’ Parameter ππ = ππ(πΊπΊ) is bidimensional if Closed under minors:ππ only decreaseswhen deleting orcontracting edges
and
Large on grids:For the ππ Γ ππ grid, ππ = Ξ© ππ2 r
r
v w
v wdelete
vwcontract
Example 1: Vertex Cover
β’ ππ = minimum number of vertices required to cover every edge (on either endpoint)
β’ Closed under minors:β still a cover(only fewer edges)
β still a cover,possibly 1 smaller
v w v w v w v wcover
v w
v wdelete
vwcontract
Example 1: Vertex Cover
β’ ππ = minimum number of vertices required to cover every edge (on either endpoint)
β’ Large on grids: Matching of size Ξ© ππ2
Every edge must be coveredby a different vertex
v w v w v w v wcover
r
r
Example 2: Feedback Vertex Set
β’ ππ = minimum number of vertices required to cover every cycle (on some vertex)
β’ Closed under minors:β still a cover(only break cycles)
β still a cover,possibly 1 smaller
v w
u
v w
u
v w
u
v w
uβ¦
v w
v wdelete
vwcontract
cover
Example 2: Feedback Vertex Set
β’ ππ = minimum number of vertices required to cover every cycle (on some vertex)
β’ Large on grids: Ξ© ππ2 vertex-disjoint cycles Every cycle must be covered
by a different vertexr
r
v w
u
v w
u
v w
u
v w
uβ¦
cover
Bidimensional β RelateParameter & Treewidthβ’ Theorem 1: If a parameter ππ is
bidimensional, then it satisfiesparameter-treewidth bound
treewidth = ππ ππin π»π»-minor-free graphs[Demaine, Fomin, Hajiaghayi, Thilikos 2004;Demaine & Hajiaghayi 2005]
β’ Proof sketch:Treewidth π€π€ β Ξ© π€π€ Γ Ξ© π€π€ grid minor
β ππ = Ξ© π€π€2 [bidimensional]
&
Bidimensional βSubexponential FPTβ’ Theorem 2: If a parameter ππ is bidimensional, and fixed-parameter tractable on graphs of
bounded treewidth: β treewidth ππππ(1) timethen it has a subexponential fixed-parameter algorithm, running in β ππ ππππ 1 time,in π»π»-minor-free graphs
β’ Proof sketch: If (approx.) treewidth = ππ ππ , answer NO Else run bounded-treewidth algorithm
&
any problem satisfying:
Bidimensionality Overview
closed under contraction [& deletion]
OPT costly on ππ Γ ππ grid
minor π»π»
parameter-treewidth
bound
subexponential fixed-parameter
algorithm
efficient PTAS
linear kernel
β¦ on π»π»-minor-free graphs
any problem satisfying:
Bidimensionality Overview
closed under contraction [& deletion]
OPT costly on ππ Γ ππ grid
minor π»π»
parameter-treewidth
bound
subexponential fixed-parameter
algorithm
efficient PTAS
linear kernel
β¦ on π»π»-minor-free graphs
Bidimensional βLinear Kernelβ’ Theorem 4: If a parameter is bidimensional, satisfies the βseparation propertyβ, and has βfinite integer indexβ,
then it has a linear kernelin π»π»-minor-free graphs[Fomin, Lokshtanov, Saurabh, Thilikos 2009]
&
ππk k
ππ(ππ)poly
Bidimensional βLinear Kernelβ’ Protrusion:
β’ Idea: Kernelize protrusions
rest ofgraphtreewidth
ππ 1cut
rest ofgraphsize
ππ 1cut
Bidimensionality (version 2)[Demaine, Fomin, Hajiaghayi, Thilikos 2004]
β’ Parameter ππ is contraction-bidimensional if Closed under contractions:ππ only decreaseswhen contracting edges
and
Large on gridoids:For ππ Γ ππ βgrid-like graphsβ,ππ = Ξ© ππ2oTriangulated + few extra edges
r
r
v w
vwcontract
Bidimensionality (version 3)[Fomin, Golovach, Thilikos 2009]
β’ Parameter ππ is contraction-bidimensional if Closed under contractions:ππ only decreaseswhen contracting edges
and
Large on πͺπͺ graphs:For naturally triangulatedππ Γ ππ grid graphs, ππ = Ξ© ππ2
v w
vwcontract
Contraction-Bidimensional Problemsβ’ Minimum maximal matchingβ’ Face cover (planar graphs)β’ Dominating setβ’ Edge dominating setβ’ ππ-dominating setβ’ Connected β¦ dominating setβ’ Unweighted TSP tourβ’ Chordal completion (fill-in)
v w
vwcontract
Contraction-Bidimensional β Good
β’ Theorem 6: As before, obtain parameter-treewidth bound subexponential FPT efficient PTASs linear kernel
for contraction-bidimensional parametersin any graph family excluding an apex minor
[Demaine, Fomin, Hajiaghayi, Thilikos 2004;Demaine & Hajiaghayi 2005 & 2005;Fomin, Golovach, Thilikos 2009;Fomin, Lokshtanov, Saurabh, Thilikos 2009]
Separators via Bidimensionality
β’ Number of vertices is minor-bidimensionalβ’ Parameter-treewidth boundβ treewidth = ππ ππ = ππ ππ
β’ Corollary: For any fixed graph π»π», everyπ»π»-minor-free graph has treewidth ππ ππand hence separator of size ππ ππ[Alon, Seymour, Thomas 1990; Grohe 2003]
Local Treewidth viaBidimensionalityβ’ Diameter is a contraction-bidimensional
parameter except Ξ© ππ , not Ξ© ππ2 , in ππ Γ ππ gridβ’ β Treewidth = ππ diameter
in apex-minor-free graphsβ’ β Treewidth of radius-ππ neighborhood = ππ ππ
[Demaine & Hajiaghayi 2004]
ππ = 1ππ = 2
π£π£
General Graphs via Bidimensionality
β’ Theorem: Any minor-closed graph parameter nonzero on some planar graph ππ, at least the sum over connected components in a
disconnected graph, and fixed-parameter tractable with respect to treewidth
can be solved in ππ ππ ππ by a known algorithm[Demaine & Hajiaghayi 2007]
β’ Proof sketch: Treewidth > ππ1(π₯π₯ ππ)ππ2 β π₯π₯ ππ Γ (π₯π₯ ππ) grid minorβ ππ2 of π₯π₯ Γ π₯π₯ grid minors, with ππ minorβ OPT β₯ ππ2 β answer NO
Beyond Apex-Minor-Free
β’ Conjecture: Algorithms for contraction-bidimensional problems generalize toπ»π»-minor-free graphs, even for anon-apex graph π»π» True for e.g. dominating set subexponential FPT
(but parameter-treewidth bound false)[Demaine, Fomin, Hajiaghayi, Thilikos 2004] Stronger form of contraction-
bidimensionality works[Fomin, Golovach, Thilikos 2009]
New York Times
Beyond π―π―-Minor-Free Graphsβ’ Conjecture: Most of bidimensionality theory
generalizes to fixed powers ofπ»π»-minor-free graphs, e.g., map graphs Treewidth Ξ© ππ7β ππ Γ ππ grid minor[Demaine, Hajiaghayi,Kawarabayashi 2009] Subexponential FPT for
dominating set in map graphs[Demaine, Fomin, Hajiaghayi, Thilikos 2003] Subexponential FPT & EPTASs for map graphs
by βcleaningβ cliques [Fomin, Lokshtanov, 2012]
Beyond π―π―-Minor-Free Graphs
β’ Excluded topological minors Linear kernel for (connected) dominating set
[Fomin, Lokshtanov, Saurabh, Thilikos 2013] General framework for linear kernels
[Langer, Reidl, Rossmanith, Sikdar 2012]
β’ Directed graphs Subexponential fixed-parameter algorithms for e.g.
Directed Hamiltonian Path via bidimensionality[Dorn, Fomin, Lokshtanov, Raman, Saurabh 2010] PTASs?
Beyond: Subset Problems
β’ What if only some of the nodes are βcriticalβ?β’ Steiner tree, subset TSP, etc. have PTASs
up to bounded-genus graphs[Borradaile, Mathieu, Klein 2007;Borradaile, Demaine, Tazari 2009]
β’ Steiner forest has PTAS in planar graphs[Bateni, Hajiaghayi, Marx 2010]
β’ Wanted: General framework
Contractions not Well-Quasi-Ordered [Demaine, Hajiaghayi, Kawarabayashi 2009]
β’ No πΎπΎ2,ππ can be contracted into πΎπΎ2,ππ
β’ Well-quasi-ordered for triangulated planar graphs 2-connected outerplanar graphs trees
β’ Minor-closed β finite excluded contractions
β¦
Simplifying Graph Decomposition[Demaine, Hajiaghayi, Kawarabayashi 2005; DeVos et al. 2004]
β’ π»π»-minor-free graphs can have vertices or edges partitioned into ππ pieces such that deleting any one piece results in bounded treewidth For PTAS, set ππ β 1/ππ ππππ π»π» algorithm ππ π»π» ππππ 1 algorithm
[Tazari 2012]
β’ Previously known for planar graphs [Baker 1994]
apex-minor-free [Eppstein 2000]
Simplifying Graph Decomposition[Demaine, Hajiaghayi, Kawarabayashi 2010]
β’ Odd-π»π»-minor-free graphs can have their vertices or edges partitioned into 2 piecessuch that deleting any one piece results in bounded treewidth ππππ π»π» algorithm ππ π»π» ππππ 1 algorithm
[Tazari 2012]
Contraction Decomposition[Demaine, Hajiaghayi, Kawarabayashi 2011]
β’ Theorem: π»π»-minor-free graphs can have their edges partitioned into ππ pieces such that contracting any one piece results in bounded treewidth Polynomial-time algorithm Previously known for
o planar [Klein 2005, 2006]
o bounded-genus [Demaine, Hajiaghayi, Mohar 2007]
o apex-minor-free[Demaine, Hajiaghayi, Kawarabayashi 2009]
Applications ofContraction Decompositionβ’ PTAS for Traveling Salesman Problem
in weighted π»π»-minor-free graphs Decade-old problem by [Grohe]
β’ PTAS for minimum-sizeππ-edge-connected submultigraph
β’ Fixed-parameter algorithm forππ-cut & bisection
[Demaine, Hajiaghayi, Kawarabayashi 2011]
Fixed-Parameter Algorithmsvia Contraction Decomposition
β’ Bisection: Cut graph into equal halveswith β€ ππ edges between them
β’ FPT in π»π»-minor-free graphs: Closed under contraction
(but not minors) Contraction decomposition with ππ + 1 layers
avoids OPT in some contraction Vertices become weighted Can still solve weighted bounded-treewidth β Solve in 2 οΏ½ππ ππ ππ + ππππ 1 time
Fixed-Parameter Algorithmsvia Contraction Decomposition
β’ ππ-cut: Remove fewest edges to makeat least ππ connected components
β’ FPT in π»π»-minor-free graphs: Average degree πππ»π» = ππ π»π» lgπ»π» β OPT β€ πππ»π» ππ β Contraction decomposition with πππ»π» ππ + 1 layers
avoids OPT in some contraction β Solve in 2 οΏ½ππ ππ ππ + ππππ 1 time
β’ Generalization to arbitrary graphs [Kawarabayashi & Thorup 2011]
Radial Coloring for Bounded Genus[Demaine, Hajiaghayi, Kawarabayashi 2011]
β’ Start with tight precolored edges: Treewidth of radius-ππ neighborhood is ππ(ππ) ππ = ππ(1) induced connected components
β’ Color edges at radial distance ππwith color ππ mod ππ
Radial Coloring for Bounded Genus[Demaine, Hajiaghayi, Kawarabayashi 2011]
β’ Contract color ππ Each connected
component βarticulation point Split these apart β
blobs, connected in DAGβ’ Outdegree > 1 β splitβ’ Indegree > 1 β rejoin β Indegree β€ ππ + ππ
(initial component or handle)
Radial Coloring for Bounded Genus[Demaine, Hajiaghayi, Kawarabayashi 2011]
β’ Split nonroot blob intoβ€ ππ + ππ chunksby closest BFS seed
β’ Each chunk hasradial diameter ππ ππ
β’ Nonroot blob has radial diameter, sotreewidth ππ ππ ππ + ππ [Eppstein 2000]
β’ Blob DAG = tree + β€ ππ extra edges Tree of 1-sums doesnβt affect treewidth Add extra edges to all bags β +ππ ππ
Contraction Decomp.for π―π―-Minor-Free
β’ π»π»-minor-free graph = βtreeβof βalmost-embeddable graphsβ [GM16]
β’ Each almost-embeddable graph has contraction decomposition: Bounded genus done Apices easy:
increase treewidthof anything by ππ(1) Vortices similar[Demaine, Hajiaghayi, Mohar 2007]
Hard Part of the Problem[Demaine, Hajiaghayi, Mohar 2007]
β’ Tree βjoinsβ between two graphs areclique sums: Attach at clique of same size Delete any desired edges
πΊπΊ2πΊπΊ1 1
2
34
5
πΎπΎ5 πΎπΎ5πΊπΊ = πΊπΊ1 β πΊπΊ2
1
2
34
5
destroyscontraction
property
Solution Idea[Demaine, Hajiaghayi, Kawarabayashi 2011]
β’ Clique sums either: [Graph Minors] Vortices β apices
β no contraction necessary πΎπΎ3 in bounded-genus part
(hard part)
πΊπΊ2πΊπΊ1 1
2
34
5
πΎπΎ5 πΎπΎ5πΊπΊ = πΊπΊ1 β πΊπΊ2
1
2
34
5
Solution Idea[Demaine, Hajiaghayi, Kawarabayashi 2011]
β’ Find a path in child πΊπΊ2 to simulate the effectof each deleted edge in πΎπΎ3 Can make pieces πΊπΊππ 3-edge-connected
β’ Need these paths to all contract together,by putting them all in piece 1
a
b c
Solution Idea[Demaine, Hajiaghayi, Kawarabayashi 2011]
β’ Claim: Precoloring ππ(1) shortest pathsdoesnβt hurt contraction decomposition
β’ Shortcut paths to shortest paths(in radial graph β primal + dual) Two paths ππ β ππ; one path ππ β ππ-ππ path
a
b c
πΆπΆ ππ Shortest Paths are Tight[Demaine, Hajiaghayi, Kawarabayashi 2011]
β’ Theorem: Treewidth of radius-ππ neighborhood of ππ 1 shortest paths is ππ ππ
β’ If not, large gridwithin theneighborhood
β’ Intuition: Thenthere should bea shortcut:2ππ βͺ ππ(ππ) Contradiction
πΆπΆ ππ Shortest Paths are Tight[Demaine, Hajiaghayi, Kawarabayashi 2011]
β’ Theorem: Treewidth of radius-ππ neighborhood of ππ 1 shortest paths is ππ ππ Path is ππ-dive intoΞ ππ Γ Ξ ππ wall β ππ-rainbow By Mengerβs
Theorem, get Ξ ππβcross pathsβ Build Ξ© ππ Γ Ξ© ππ
wall β₯ 2 ππ awayfrom shortest path Repeat for each path