On independent set on B
1-EPG graphs
Marin Bougeret (Lirmm, Montpellier, France) Joint Work with S. Bessy, D.Goncalves, C. Paul
WAOA 2015
Onlindependent set on B
1-EPG graphs
Marin Bougeret (Lirmm, Montpellier, France) Joint Work with S. Bessy, D.Goncalves, C. Paul
WAOA 2015
On independent set on B
1-EPG graphs
Marin Bougeret (Lirmm, Montpellier, France) Joint Work with S. Bessy, D.Goncalves, C. Paul
WAOA 2015
Outline
1 Introduction on EPG-Graphs
2 Approximability
3 Fixed parameter tractability
4 / 29
Definition of EPG-Graphs
EPG (for Edge intersection graphs of Paths on a Grid) graphs introduced in [GLS09]
In EPG-graphG = (V,E):
each vertexv corresponds to a pathPv
{u,v} ∈E iff Pu,Pv share a grid edge
3 4 1
2
an EPG-representation Gis a{xyq}-EPG graph
P1
P2
P3
P4
G
Bk-EPG: graphs having a representation where every path has at most k bends
X-EPG ⊆B1-EPG (with X ⊆ {p,q,x,y}): paths can only have shapes inX
5 / 29
Definition of EPG-Graphs
EPG (for Edge intersection graphs of Paths on a Grid) graphs introduced in [GLS09]
In EPG-graphG = (V,E):
each vertexv corresponds to a pathPv
{u,v} ∈E iff Pu,Pv share a grid edge
3 4 1
2
an EPG-representation Gis a{xyq}-EPG graph
P1
P2
P3
P4
G
Bk-EPG: graphs having a representation where every path has at most k bends
X-EPG ⊆B1-EPG (with X ⊆ {p,q,x,y}): paths can only have shapes inX
5 / 29
Properties of EPG-Graphs
Class inclusions
max deg ∆⊆B∆-EPG [HKU14]
B0-EPG = interval graphs
B1-EPG ⊂2-Track graphs⊂B3-EPG
B1-EPG K3,3 induced free, K3,3−e induced free, Sn≥4 induced free [GLS09]
2−Track representation Track A
Track B I4B I2B I1B
I1A I3A I4A
row 2 row 3 horizontal track
vertical track
horizontal track
vertical track
⇒
column 1
row 1
a
6 / 29
Properties of EPG-Graphs
Class inclusions
max deg ∆⊆B∆-EPG [HKU14]
B0-EPG = interval graphs
B1-EPG ⊂2-Track graphs⊂B3-EPG
B1-EPG K3,3 induced free, K3,3−e induced free, Sn≥4 induced free [GLS09]
2−Track representation Track A
Track B I4B I2B I1B
I1A I3A I4A
row 2 row 3 horizontal track
vertical track
horizontal track
vertical track
⇒
column 1
row 1
a
6 / 29
Properties of EPG-Graphs
Class inclusions
max deg ∆⊆B∆-EPG [HKU14]
B0-EPG = interval graphs
B1-EPG ⊂2-Track graphs⊂B3-EPG
B1-EPG K3,3 induced free, K3,3−e induced free, Sn≥4 induced free [GLS09]
2−Track representation Track A
Track B I4B I2B I1B
I1A I3A I4A
row 2 row 3 horizontal track
vertical track
horizontal track
vertical track
⇒
column 1
row 1
a
6 / 29
Properties of EPG-Graphs
Class inclusions
max deg ∆⊆B∆-EPG [HKU14]
B0-EPG = interval graphs
B1-EPG ⊂2-Track graphs⊂B3-EPG
B1-EPG K3,3 induced free, K3,3−e induced free, Sn≥4 induced free [GLS09]
2−Track representation Track A
Track B I4B I2B I1B
I1A I3A I4A
row 2 row 3 horizontal track
vertical track
horizontal track
vertical track
⇒
column 1
row 1
a
6 / 29
Properties of EPG-Graphs
Bad news aboutB1-EPG
B1-EPG are not planar graphs (Kn∈ B1-EPG)
B1-EPG are not perfect graphs (Cn∈B1-EPG) (but G[N(v)]
weakly chordal)
B1-EPG do not benefit from the many results on "intersection graph class" as two paths can cross without creating an edge
N(v ) is a C
4v
7 / 29
Properties of EPG-Graphs
Bad news aboutB1-EPG
B1-EPG are not planar graphs (Kn∈ B1-EPG)
B1-EPG are not perfect graphs (Cn∈B1-EPG) (but G[N(v)]
weakly chordal)
B1-EPG do not benefit from the many results on "intersection graph class" as two paths can cross without creating an edge
N(v ) is a C
4v
7 / 29
Properties of EPG-Graphs
Bad news aboutB1-EPG
B1-EPG are not planar graphs (Kn∈ B1-EPG)
B1-EPG are not perfect graphs (Cn∈B1-EPG) (but G[N(v)]
weakly chordal)
B1-EPG do not benefit from the many results on "intersection graph class" as two paths can cross without creating an edge
N(v ) is a C
4v
7 / 29
Motivation
Considered problem: MIS (Maximum Independent Set) on B1-EPG graphs (supposing a representation is given)
In 2013, [EGM13] proved that MIS (and coloring) are NP-hard on B1-EPG and admits a 4-approximation algorithm
Can we say more ? (approximability and fixed parameter tractability with standard parameterization)
8 / 29
Motivation
Considered problem: MIS (Maximum Independent Set) on B1-EPG graphs (supposing a representation is given)
In 2013, [EGM13] proved that MIS (and coloring) are NP-hard on B1-EPG and admits a 4-approximation algorithm
Can we say more ? (approximability and fixed parameter tractability with standard parameterization)
8 / 29
Motivation
Considered problem: MIS (Maximum Independent Set) on B1-EPG graphs (supposing a representation is given)
In 2013, [EGM13] proved that MIS (and coloring) are NP-hard on B1-EPG and admits a 4-approximation algorithm
Can we say more ? (approximability and fixed parameter tractability with standard parameterization)
8 / 29
Outline
1 Introduction on EPG-Graphs
2 Approximability
3 Fixed parameter tractability
9 / 29
Outline approximability
Related work
simple 4 approximation onB1-EPG ([EGM13] or [BYHN+06]) no PTAS for MIS on 2-Track as max deg∆ =3⊆2-Track (recallB1-EPG ⊂2-Track)
Our contributions
no PTAS for MIS on {p}-EPG, even if each path has its vertical part or its horizontal part of length at most 3 PTAS for MIS on B1-EPG when each path has its horizontal part at mostc
10 / 29
Outline approximability
Related work
simple 4 approximation onB1-EPG ([EGM13] or [BYHN+06]) no PTAS for MIS on 2-Track as max deg∆ =3⊆2-Track (recallB1-EPG ⊂2-Track)
Our contributions
no PTAS for MIS on {p}-EPG, even if each path has its vertical part or its horizontal part of length at most 3 PTAS for MIS on B1-EPG when each path has its horizontal part at mostc
10 / 29
Outline approximability
Related work
simple 4 approximation onB1-EPG ([EGM13] or [BYHN+06]) no PTAS for MIS on 2-Track as max deg∆ =3⊆2-Track (recallB1-EPG ⊂2-Track)
Our contributions
no PTAS for MIS on {p}-EPG, even if each path has its vertical part or its horizontal part of length at most 3 PTAS for MIS on B1-EPG when each path has its horizontal part at mostc
10 / 29
Negative result
Lemma: (no PTAS for{p,q}-EPG)
There is a strict reduction from MAX-3-SAT(3) to{p,q}-EPG
Proof
Consider first the textbook reduction from MAX-3-SAT to MIS as : create one triangle for each clause
add an edge between any occurrence of literal and its negation assignment satisfying t clauses⇔ IS size t
As we reduce from MAX-3-SAT(3), the graph is{p,q}-EPG
C1 C2 C3
C1=not(x1)∨x2∨x3
C2=x1∨not(x2)∨x3
C3=x1∨x2∨not(x3)
x1 x1
not(x1) x3
x2 not(x2) x3 x2 not(x3)
11 / 29
Negative result
Lemma: (no PTAS for{p,q}-EPG)
There is a strict reduction from MAX-3-SAT(3) to{p,q}-EPG
Proof
Consider first the textbook reduction from MAX-3-SAT to MIS as : create one triangle for each clause
add an edge between any occurrence of literal and its negation assignment satisfying t clauses⇔ IS size t
As we reduce from MAX-3-SAT(3), the graph is{p,q}-EPG
C1 C2 C3
C1=not(x1)∨x2∨x3
C2=x1∨not(x2)∨x3
C3=x1∨x2∨not(x3)
x1 x1
not(x1) x3
x2 not(x2) x3 x2 not(x3)
11 / 29
Negative result
Lemma: (no PTAS for{p,q}-EPG)
There is a strict reduction from MAX-3-SAT(3) to{p,q}-EPG
Proof
Consider first the textbook reduction from MAX-3-SAT to MIS as : create one triangle for each clause
add an edge between any occurrence of literal and its negation assignment satisfying t clauses⇔ IS size t
As we reduce from MAX-3-SAT(3), the graph is{p,q}-EPG
C1 C2 C3
C1=not(x1)∨x2∨x3
C2=x1∨not(x2)∨x3
C3=x1∨x2∨not(x3)
x1 x1
not(x1) x3
x2 not(x2) x3 x2 not(x3)
11 / 29
Negative result
Lemma: (no PTAS for{p,q}-EPG)
There is a strict reduction from MAX-3-SAT(3) to{p,q}-EPG
Proof
Consider first the textbook reduction from MAX-3-SAT to MIS as : create one triangle for each clause
add an edge between any occurrence of literal and its negation assignment satisfying t clauses⇔ IS size t
As we reduce from MAX-3-SAT(3), the graph is {p,q}-EPG
C1 C2 C3
C1=not(x1)∨x2∨x3
C2=x1∨not(x2)∨x3
C3=x1∨x2∨not(x3)
x1 x1
not(x1) x3
x2 not(x2) x3 x2 not(x3)
C1 C2 C3
Line used forx1andnot(x1) Line used forx2andnot(x2) Line used forx3andnot(x3)
11 / 29
Negative result
How removing one type of shape ?
Given the previousG, subdivide 4 times each "literral" edge to get a new graph G0
G0 can now be drawn as{p}-EPG
∃ indep. set. S inG ⇔ ∃indep. set. S0 in G0 with
|S0|=|S|+2m
As ∃c such that m≤cOpt(G), we get an AP-reduction for MIS from{p,q}-EPG to{p}-EPG
C1 C2 C3
Line used forx3andnot(x3)
C1 C2 C3
x1 x1
not(x1) x3
x2 not(x2) x3 x2 not(x3)
12 / 29
Negative result
How removing one type of shape ?
Given the previousG, subdivide 4 times each "literral" edge to get a new graph G0
G0 can now be drawn as{p}-EPG
∃ indep. set. S inG ⇔ ∃indep. set. S0 in G0 with
|S0|=|S|+2m
As ∃c such that m≤cOpt(G), we get an AP-reduction for MIS from{p,q}-EPG to{p}-EPG
C1 C2 C3
Line used forx3andnot(x3)
C1 C2 C3
x1 x1
not(x1) x3
x2 not(x2) x3 x2 not(x3)
12 / 29
Negative result
How removing one type of shape ?
Given the previousG, subdivide 4 times each "literral" edge to get a new graph G0
G0 can now be drawn as{p}-EPG
∃ indep. set. S inG ⇔ ∃indep. set. S0 in G0 with
|S0|=|S|+2m
As ∃c such that m≤cOpt(G), we get an AP-reduction for MIS from{p,q}-EPG to{p}-EPG
C1 C2 C3
Line used forx3andnot(x3)
C1 C2 C3
x1 x1
not(x1) x3
x2 not(x2) x3 x2 not(x3)
12 / 29
Negative result
How removing one type of shape ?
Given the previousG, subdivide 4 times each "literral" edge to get a new graph G0
G0 can now be drawn as{p}-EPG
∃ indep. set. S inG ⇔ ∃indep. set. S0 in G0 with
|S0|=|S|+2m
As ∃c such that m≤cOpt(G), we get an AP-reduction for MIS from{p,q}-EPG to{p}-EPG
C1 C2 C3
x1 x1
not(x1) x3
x2 not(x2) x3 x2 not(x3)
Lines used forx3andnot(x3)
12 / 29
Negative result
How removing one type of shape ?
Given the previousG, subdivide 4 times each "literral" edge to get a new graph G0
G0 can now be drawn as{p}-EPG
∃ indep. set. S inG ⇔ ∃indep. set. S0 in G0 with
|S0|=|S|+2m
As ∃c such that m≤cOpt(G), we get an AP-reduction for MIS from{p,q}-EPG to{p}-EPG
C1 C2 C3
x1 x1
not(x1) x3
x2 not(x2) x3 x2 not(x3)
Lines used forx3andnot(x3)
12 / 29
Negative result
How removing one type of shape ?
Given the previousG, subdivide 4 times each "literral" edge to get a new graph G0
G0 can now be drawn as{p}-EPG
∃ indep. set. S inG ⇔ ∃indep. set. S0 in G0 with
|S0|=|S|+2m
As ∃c such that m≤cOpt(G), we get an AP-reduction for MIS from{p,q}-EPG to{p}-EPG
C1 C2 C3
x1 x1
not(x1) x3
x2 not(x2) x3 x2 not(x3)
Lines used forx3andnot(x3)
Theorem
There is no PTAS for MIS on{p}-EPG, even if each path has its vertical part or its horizontal part of length at most 3
12 / 29
Negative result
Can we improve this when both parts have constant size ? No.
Theorem
There is PTAS for MIS onB1-EPG when each path has its horizontal part at mostc (which remains NP-hard)
Proof
We will prove this using classical Baker shifting technique
13 / 29
Positive result
Proof
Letk be a large integer (goal: sol of size|S| ≥ |OPT|(1−1k)) Givend ∈[kc−1], letRd be the set of paths whose horizontal part crosses the vertical stripd,d +kc,d +2kc, ..
Let OPTd =OPT ∩Rd
∈/Rd
kc kc
∈/Rd
kc
∈Rd
∈Rd
d+kc d+2kc
0 c d
14 / 29
Positive result
Proof
Letk be a large integer (goal: sol of size|S| ≥ |OPT|(1−1k)) Givend ∈[kc−1], letRd be the set of paths whose horizontal part crosses the vertical stripd,d +kc,d +2kc, ..
Let OPTd =OPT ∩Rd
∈/Rd
kc kc
∈/Rd
kc
∈Rd
∈Rd
d +kc d+2kc
0 c d
14 / 29
Positive result
Proof
Letk be a large integer (goal: sol of size|S| ≥ |OPT|(1−1k)) Givend ∈[kc−1], letRd be the set of paths whose horizontal part crosses the vertical stripd,d +kc,d +2kc, ..
Let OPTd =OPT ∩Rd
∈/Rd
kc kc
∈/Rd
kc
∈Rd
∈Rd
d +kc d+2kc
0 c d
14 / 29
Positive result
Lemma
∃d0 such that |OPTd0| ≤ 1k|OPT|
Proof Pkc−1
d=0 |OPTd| ≤c|OPT|(as each vertex belongs to at most c different Rd)
as |OPTd0|kc≤Pkc−1
d=0 |OPTd|, we get the result
15 / 29
Positive result
Proof
Back to the PTAS proof:
thus, for each d we solve the problem optimally onG \Rd, and we output A: the best of these solutions
Previous Lemma ⇒ |A| ≥(1− 1k)|OPT|
16 / 29
Positive result
Proof
Back to the PTAS proof:
thus, for each d we solve the problem optimally onG \Rd, and we output A: the best of these solutions
Previous Lemma ⇒ |A| ≥(1− 1k)|OPT|
16 / 29
Positive result
Proof
It remains to solve the problem optimally onG\Rd:
G \Rd has several connected componentCCl, where each component is drawn on a grid of constant widthkc
we solve optimally on each CCl using a dyn. prog. algorithm
kc kc
kc
∈Rd
d+kc d+2kc
0 d
CC1 CC2 CC3
c
17 / 29
Positive result
Proof
It remains to solve the problem optimally onG\Rd:
G \Rd has several connected componentCCl, where each component is drawn on a grid of constant widthkc
we solve optimally on each CCl using a dyn. prog. algorithm
kc kc
kc
∈Rd
d +kc d+2kc
0 d
CC1 CC2 CC3
c
17 / 29
Positive result
Proof
It remains to solve the problem optimally onG\Rd:
G \Rd has several connected componentCCl, where each component is drawn on a grid of constant widthkc
we solve optimally on each CCl using a dyn. prog. algorithm
kc kc
kc
∈Rd
d +kc d+2kc
0 d
CC1 CC2 CC3
c
17 / 29
Outline
1 Introduction on EPG-Graphs
2 Approximability
3 Fixed parameter tractability
18 / 29
Outline fixed parameter tractability
We consider the problemOPT ≤k?parameterized by k.
Question: is this problem FPT ? (i.e. can be solved inf(k)poly(n))
Related work
B1-EPG ⊂2-Track ⊂B3-EPG+MIS is W1-hard on unit 2-Track [Jia10]⇒ MIS isW1-hard onB3-EPG
Our contributions
MIS is FPT on {x,y,p}-EPG MIS is W1-hard on B2-EPG
We will prove that MIS is FPT on{x}-EPG using a branching algorithm.
Definition
Given a pathP, let cor(P) = (x,y) be the coordinates of the corner ofP in the grid (orienting as usual ↑y →x)
19 / 29
Outline fixed parameter tractability
We consider the problemOPT ≤k?parameterized by k.
Question: is this problem FPT ? (i.e. can be solved inf(k)poly(n))
Related work
B1-EPG ⊂2-Track⊂ B3-EPG+MIS is W1-hard on unit 2-Track [Jia10]⇒ MIS isW1-hard onB3-EPG
Our contributions
MIS is FPT on {x,y,p}-EPG MIS is W1-hard on B2-EPG
We will prove that MIS is FPT on{x}-EPG using a branching algorithm.
Definition
Given a pathP, let cor(P) = (x,y) be the coordinates of the corner ofP in the grid (orienting as usual ↑y →x)
19 / 29
Outline fixed parameter tractability
We consider the problemOPT ≤k?parameterized by k.
Question: is this problem FPT ? (i.e. can be solved inf(k)poly(n))
Related work
B1-EPG ⊂2-Track⊂ B3-EPG+MIS is W1-hard on unit 2-Track [Jia10]⇒ MIS isW1-hard onB3-EPG
Our contributions
MIS is FPT on {x,y,p}-EPG MIS is W1-hard onB2-EPG
We will prove that MIS is FPT on{x}-EPG using a branching algorithm.
Definition
Given a pathP, let cor(P) = (x,y) be the coordinates of the corner ofP in the grid (orienting as usual ↑y →x)
19 / 29
Outline fixed parameter tractability
We consider the problemOPT ≤k?parameterized by k.
Question: is this problem FPT ? (i.e. can be solved inf(k)poly(n))
Related work
B1-EPG ⊂2-Track⊂ B3-EPG+MIS is W1-hard on unit 2-Track [Jia10]⇒ MIS isW1-hard onB3-EPG
Our contributions
MIS is FPT on {x,y,p}-EPG MIS is W1-hard onB2-EPG
We will prove that MIS is FPT on{x}-EPG using a branching algorithm.
Definition
Given a pathP, let cor(P) = (x,y) be the coordinates of the corner ofP in the grid (orienting as usual ↑y →x)
19 / 29
MIS is FPT on { x }-EPG
Proof
Let Pv0 (cor(Pv0) = (x0,y0)): a path whose corner is
"top-right most" (highest line of the grid, and then the right most)
Suppose |OPT|=k
We know that∃Pv∗ ∈N[Pv0]∩OPT (cor(Pv∗) = (x∗,y∗)) Let us find this Pv∗ (or aP0 ∈OPT0) using at most f(k) branches
y0
x0 Pv0
20 / 29
MIS is FPT on { x }-EPG
Proof
Let Pv0 (cor(Pv0) = (x0,y0)): a path whose corner is
"top-right most" (highest line of the grid, and then the right most)
Suppose |OPT|=k
We know that∃Pv∗ ∈N[Pv0]∩OPT (cor(Pv∗) = (x∗,y∗)) Let us find this Pv∗ (or aP0 ∈OPT0) using at most f(k) branches
y0
x0 Pv0
20 / 29
MIS is FPT on { x }-EPG
Proof
Let Pv0 (cor(Pv0) = (x0,y0)): a path whose corner is
"top-right most" (highest line of the grid, and then the right most)
Suppose |OPT|=k
We know that∃Pv∗ ∈N[Pv0]∩OPT (cor(Pv∗) = (x∗,y∗)) Let us find this Pv∗ (or aP0 ∈OPT0) using at most f(k) branches
y0
x0 Pv0
20 / 29
MIS is FPT on { x }-EPG
Proof
Let Pv0 (cor(Pv0) = (x0,y0)): a path whose corner is
"top-right most" (highest line of the grid, and then the right most)
Suppose |OPT|=k
We know that∃Pv∗ ∈N[Pv0]∩OPT (cor(Pv∗) = (x∗,y∗)) Let us find this Pv∗ (or aP0 ∈OPT0) using at most f(k) branches
y0
x0 Pv0
20 / 29
MIS is FPT on { x }-EPG
Proof
Let Pv0 (cor(Pv0) = (x0,y0)): a path whose corner is
"top-right most" (highest line of the grid, and then the right most)
Suppose |OPT|=k
We know that∃Pv∗ ∈N[Pv0]∩OPT (cor(Pv∗) = (x∗,y∗)) Let us find this Pv∗ (or aP0 ∈OPT0) using at most f(k) branches
y0
x0 Pv0
20 / 29
MIS is FPT on { x }-EPG
Proof
We branch on the following cases:
1 x∗ =x0 andy∗ =y0: easy, takePv0
2 x∗ =x0 andy∗ <y0
3 x∗ <x0 andy∗ =y0
Let us only treat case 2 here (case 3 works similarly)
y0
x0 Pv0
21 / 29
MIS is FPT on { x }-EPG
Proof
We branch on the following cases:
1 x∗ =x0 andy∗ =y0: easy, takePv0
2 x∗ =x0 andy∗ <y0
3 x∗ <x0 andy∗ =y0
Let us only treat case 2 here (case 3 works similarly)
y0
x0 Pv0
Pv∗
21 / 29
MIS is FPT on { x }-EPG
Proof
We branch on the following cases:
1 x∗ =x0 andy∗ =y0: easy, takePv0
2 x∗ =x0 andy∗ <y0
3 x∗ <x0 andy∗ =y0
Let us only treat case 2 here (case 3 works similarly)
y0
x0 Pv0
Pv∗
21 / 29
MIS is FPT on { x }-EPG
Proof
We branch on the following cases:
1 x∗ =x0 andy∗ =y0: easy, takePv0
2 x∗ =x0 andy∗ <y0
3 x∗ <x0 andy∗ =y0
Let us only treat case 2 here (case 3 works similarly)
y0
x0 Pv0 Pv∗
21 / 29
MIS is FPT on { x }-EPG
Proof
We branch on the following cases:
1 x∗ =x0 andy∗ =y0: easy, takePv0
2 x∗ =x0 andy∗ <y0
3 x∗ <x0 andy∗ =y0
Let us only treat case 2 here (case 3 works similarly)
y0
x0 Pv0
Pv∗
21 / 29
MIS is FPT on { x }-EPG
Proof
Let X be the "vertical" neighbors of Pv0 (Pv∗ ∈X).
Let partition X =G1∪G2..∪Gt
Sufficient to keep the left most pathPi in each Gi
X y0
x0 Pv0
G1 G2
Gt
22 / 29
MIS is FPT on { x }-EPG
Proof
Let X be the "vertical" neighbors of Pv0 (Pv∗ ∈X).
Let partition X =G1∪G2..∪Gt
Sufficient to keep the left most pathPi in each Gi
y0
x0 Pv0
P1 P2
Pt
22 / 29
MIS is FPT on { x }-EPG
Proof
Now, two cases are possible
1 Pv∗ ∈ {P1, . . . ,Pk}: easy, guess which Pi to take
2 Pv∗ ∈ {Pk+1, . . . ,Pt}:
as|OPT|=k, there existsPi∈ {P1, . . . ,Pk}such that there is noP0 ∈OPT in the "horizontal line" ofPi
guess thisPi and take it
y0
x0 Pv0
P1 P2
Pt
23 / 29
MIS is FPT on { x }-EPG
Proof
Now, two cases are possible
1 Pv∗ ∈ {P1, . . . ,Pk}: easy, guess which Pi to take
2 Pv∗ ∈ {Pk+1, . . . ,Pt}:
as|OPT|=k, there existsPi∈ {P1, . . . ,Pk}such that there is noP0 ∈OPT in the "horizontal line" ofPi
guess thisPi and take it
y0
x0 Pv0
P1 P2
Pt Pk
Pv∗
23 / 29
MIS is FPT on { x }-EPG
Proof
Now, two cases are possible
1 Pv∗ ∈ {P1, . . . ,Pk}: easy, guess which Pi to take
2 Pv∗ ∈ {Pk+1, . . . ,Pt}:
as|OPT|=k, there existsPi∈ {P1, . . . ,Pk}such that there is noP0 ∈OPT in the "horizontal line" ofPi
guess thisPi and take it
y0
x0 Pv0
P1 P2
Pt Pk
Pv∗
∈OPT ∈OPT
23 / 29
MIS is FPT on { x }-EPG
Proof
Now, two cases are possible
1 Pv∗ ∈ {P1, . . . ,Pk}: easy, guess which Pi to take
2 Pv∗ ∈ {Pk+1, . . . ,Pt}:
as|OPT|=k, there existsPi∈ {P1, . . . ,Pk}such that there is noP0 ∈OPT in the "horizontal line" ofPi
guess thisPi and take it
y0
x0 Pv0
P1 P2
Pt Pk
Pv∗
∈OPT ∈OPT
Pi
23 / 29
Conclusion on FPT
Positive results
MIS is FPT on{x}-EPG as we can find aPv∗∈OPT using at most f(k) branches
The previous appproach doesn’t work (even for 2 shapes)..
but we use the same kind of arguments to prove that MIS is FPT with 3 shapes
Negative results
MIS is W1-hard on B2-EPG
24 / 29
Conclusion on FPT
Positive results
MIS is FPT on{x}-EPG as we can find aPv∗∈OPT using at most f(k) branches
The previous appproach doesn’t work (even for 2 shapes)..
but we use the same kind of arguments to prove that MIS is FPT with 3 shapes
Negative results
MIS is W1-hard on B2-EPG
24 / 29
Conclusion on FPT
Positive results
MIS is FPT on{x}-EPG as we can find aPv∗∈OPT using at most f(k) branches
The previous appproach doesn’t work (even for 2 shapes)..
but we use the same kind of arguments to prove that MIS is FPT with 3 shapes
Negative results
MIS is W1-hard on B2-EPG
24 / 29
Conclusion on FPT
Positive results
MIS is FPT on{x}-EPG as we can find aPv∗∈OPT using at most f(k) branches
The previous appproach doesn’t work (even for 2 shapes)..
but we use the same kind of arguments to prove that MIS is FPT with 3 shapes
Negative results
MIS is W1-hard onB2-EPG
24 / 29
Conclusion
Approximability
No PTAS unlessP =NP
PTAS when one side is always small Simple 4 approximation
Open: c <4 approximation ?
Fixed parameter tractability FPT with 3 shapes W1-hard onB2-EPG Open: FPT onB1-EPG ?
25 / 29
Thank you for your attention ! ...
26 / 29
MIS isW1-hard on B2-EPG as:
0000 0000 0000 0000 0000 0000 0000 0000 00
1111 1111 1111 1111 1111 1111 1111 1111 11
0000 0000 0000 0000 0000 0000 00
1111 1111 1111 1111 1111 1111 11
0000 0000 0000 0000 0000 0000 00
1111 1111 1111 1111 1111 1111 11
0000 0000 0000 0000 0000 0000 00
1111 1111 1111 1111 1111 1111 11
0000 0000 0000 0000 00
1111 1111 1111 1111 11
000000 000000 000000 000000 000
111111 111111 111111 111111 111
0000 0000 0000 0000 0000 0000 0000 0000 00
1111 1111 1111 1111 1111 1111 1111 1111 11
000000 000000 000000 000000 000
111111 111111 111111 111111 111
0000 0000 0000 0000 0000 0000 0000 0000 00
1111 1111 1111 1111 1111 1111 1111 1111 11
0000 0000 0000 0000 0000 0000 0000 0000 00
1111 1111 1111 1111 1111 1111 1111 1111 11
0000 0000 0000 0000 0000 0000 00
1111 1111 1111 1111 1111 1111 11
0000 0000 0000 0000 00
1111 1111 1111 1111 11 000000
000000 000000 000000 000000 000000
111111 111111 111111 111111 111111 111111
000000 000000 000000 111111 111111 111111
000000 000000 000000 000000 000000 000000
111111 111111 111111 111111 111111 111111
000000 000000 000000 111111 111111 111111
linesL13,x linesL23,x
linesL32,x T3
T2 T1
T2,3
T1,3 T1,2
pathsa0`anda`fromT1,2
pathsb`0andb`fromT1,2
27 / 29
Bibliography
[BYHN+06] Reuven Bar-Yehuda, Magnús M Halldórsson, Joseph Seffi Naor, Hadas Shachnai, and Irina Shapira.
Scheduling split intervals.
SIAM J. Comput., 36(1):1–15, 2006.
[EGM13] Dror Epstein, MartinCharles Golumbic, and Gila Morgenstern.
Approximation algorithms forb1-epg graphs.
In WADS 2013: Algorithms and Data Structures, volume 8037 of Lecture Notes in Computer Science, pages 328–340. Springer Berlin Heidelberg, 2013.
[GLS09] Martin Charles Golumbic, Marina Lipshteyn, and Michal Stern.
Edge intersection graphs of single bend paths on a grid.
Networks, 54(3):130–138, 2009.
[HKU14] Daniel Heldt, Kolja Knauer, and Torsten Ueckerdt.
Edge-intersection graphs of grid paths: The bend-number.
Discrete Applied Mathematics, 167(0):144 – 162, 2014.
[Jia10] Minghui Jiang.
On the parameterized complexity of some optim ization problems related to multiple-interval graphs.
Theoretical Computer Science, 411:4253–4262, 2010.
28 / 29
Why does it fail for 2 shapes ?
Pv∗∈ {Pk+1, . . . ,Pt}
but we cannot restructure OPT to chose one of the{P1, . . . ,Pk} P4
P3
Pv0
P1
Pk=P2
Pv∗
29 / 29