Parameterized

Reoptimization

F. Reidl

Reoptimization

Parameterized
Complexity

Parameterized Reoptimization Parameterized

Reoptimization

Felix Reidl

Lehr- und Forschungsgebiet Theoretische Informatik

RWTH Aachen University

August 17, 2010

 Parameterized
Contents Reoptimization

F. Reidl

Reoptimization

Parameterized
Complexity

Parameterized
Reoptimization
Reoptimization

Parameterized Complexity

Parameterized Reoptimization
 Parameterized
Reoptimization

F. Reidl

The premise for this talk

Reoptimization
I Theoretical scenario: solve problem, done. Parameterized
Complexity
I Real-life scenario: problem must be solved again and Parameterized
again Reoptimization

I Small updates: instance changes only slightly

I Do we have to throw previous solutions away?
 Parameterized
Reoptimization

F. Reidl

The premise for this talk

Reoptimization
I Theoretical scenario: solve problem, done. Parameterized
Complexity
I Real-life scenario: problem must be solved again and Parameterized
again Reoptimization

I Small updates: instance changes only slightly

I Do we have to throw previous solutions away?

Definition (Reoptimization Scenario)

Input: Instance I, optimal solution S for I, modified
instance I 0
Problem: Find an optimal solution S 0 for I 0

Note: Difference between I 0 and I is “small”

 Parameterized
Reoptimization

F. Reidl

Reoptimization

Parameterized
Complexity
Previous work in this direction
Parameterized
I So far only tackled with approximation algorithms Reoptimization

I No PTAS ⇒ solution quality deteriorates

 Parameterized
Reoptimization

F. Reidl

Reoptimization

Parameterized
Complexity
Previous work in this direction
Parameterized
I So far only tackled with approximation algorithms Reoptimization

I No PTAS ⇒ solution quality deteriorates

I PTAS known for Vertex Cover, Independent
Set, Dominating Set [Bilò,Widmayer,Zych] . . .
 Parameterized
Reoptimization

F. Reidl

Reoptimization

Parameterized
Complexity
Previous work in this direction
Parameterized
I So far only tackled with approximation algorithms Reoptimization

I No PTAS ⇒ solution quality deteriorates

I PTAS known for Vertex Cover, Independent
Set, Dominating Set [Bilò,Widmayer,Zych] . . .
I . . .but rely on optimal solution!
 Parameterized
Reoptimization

F. Reidl

Reoptimization

Parameterized
Complexity
Previous work in this direction
Parameterized
I So far only tackled with approximation algorithms Reoptimization

I No PTAS ⇒ solution quality deteriorates

I PTAS known for Vertex Cover, Independent
Set, Dominating Set [Bilò,Widmayer,Zych] . . .
I . . .but rely on optimal solution!
⇒ Solving exactly might be a better idea
 Parameterized
Reoptimization

F. Reidl
Parameterized Complexity
Reoptimization
I Motivation: more fine-grained look at NP-hard problems
Parameterized
I Isolate “difficult part” of the problem Complexity

Parameterized
Reoptimization
 Parameterized
Reoptimization

F. Reidl
Parameterized Complexity
Reoptimization
I Motivation: more fine-grained look at NP-hard problems
Parameterized
I Isolate “difficult part” of the problem Complexity

Parameterized
Reoptimization
Definition (Parameterized language)
A parameterized language L over an alphabet Σ is a subset
of Σ∗ × N. For any tuple (I, k) ∈ Σ∗ × N we call k the
parameter.
 Parameterized
Reoptimization

F. Reidl
Parameterized Complexity
Reoptimization
I Motivation: more fine-grained look at NP-hard problems
Parameterized
I Isolate “difficult part” of the problem Complexity

Parameterized
Reoptimization
Definition (Parameterized language)
A parameterized language L over an alphabet Σ is a subset
of Σ∗ × N. For any tuple (I, k) ∈ Σ∗ × N we call k the
parameter.

Definition (Fixed parameter tractable)

A parameterized language L is contained in the class F P T if
there exists an algorithm that decides L in time f (k)nO(1)
for an arbitrary function f .
 Parameterized
Reoptimization

F. Reidl

The famous positive example: Reoptimization

Parameterized
Definition (k-Vertex Cover) Complexity

Parameterized
Input: Graph G, integer k Reoptimization

Parameter: k
Problem: Has G a vertex cover of at most size k?
 Parameterized
Reoptimization

F. Reidl

The famous positive example: Reoptimization

Parameterized
Definition (k-Vertex Cover) Complexity

Parameterized
Input: Graph G, integer k Reoptimization

Parameter: k
Problem: Has G a vertex cover of at most size k?

I Bounded search tree: pick any uncovered edge (u, v),

include either u or v in vertex cover
I After at most k “decisions” we are done
I Search tree has at
most 2k leafs, therefore algorithm
runs in O 2k nO(1)
 Parameterized
Reoptimization

F. Reidl

Reoptimization

Parameterized
Complexity

Parameterized
Reoptimization
The famous negative examples:
I F P T ⊆ W [1] ⊆ W [2] . . ., most likely F P T 6= W [1]
I k-Independent Set and k-Clique W [1]-hard,
k-Dominating Set W [2]-hard
 Parameterized
Reoptimization
Definition (FPT-Reoptimizable) F. Reidl

A problem L is FPT-Reoptimizable (under a specified Reoptimization

modification) if the corresponding reoptimization problem Parameterized
Complexity
can be solved in FPT-time.
Parameterized
Reoptimization
 Parameterized
Reoptimization
Definition (FPT-Reoptimizable) F. Reidl

A problem L is FPT-Reoptimizable (under a specified Reoptimization

modification) if the corresponding reoptimization problem Parameterized
Complexity
can be solved in FPT-time.
Parameterized
Reoptimization
Let’s look at Vertex Cover again.
Definition (param.Vertex Cover reoptimization)
Input: Graph G with optimal vertex cover S
Modification: Delete edge e from G to obtain G0
Parameter: k := |S|
Problem: Find an optimal vertex cover S 0 for G0
 Parameterized
Reoptimization
Definition (FPT-Reoptimizable) F. Reidl

A problem L is FPT-Reoptimizable (under a specified Reoptimization

modification) if the corresponding reoptimization problem Parameterized
Complexity
can be solved in FPT-time.
Parameterized
Reoptimization
Let’s look at Vertex Cover again.
Definition (param.Vertex Cover reoptimization)
Input: Graph G with optimal vertex cover S
Modification: Delete edge e from G to obtain G0
Parameter: k := |S|
Problem: Find an optimal vertex cover S 0 for G0

Trivially reoptmizable: use parameterized algorithm for

k-Vertex Cover to solve.
 Parameterized
Reoptimization
Definition (FPT-Reoptimizable) F. Reidl

A problem L is FPT-Reoptimizable (under a specified Reoptimization

modification) if the corresponding reoptimization problem Parameterized
Complexity
can be solved in FPT-time.
Parameterized
Reoptimization
Let’s look at Vertex Cover again.
Definition (param.Vertex Cover reoptimization)
Input: Graph G with optimal vertex cover S
Modification: Delete edge e from G to obtain G0
Parameter: k := |S|
Problem: Find an optimal vertex cover S 0 for G0

Trivially reoptmizable: use parameterized algorithm for

k-Vertex Cover to solve.

Can we do better?
 Parameterized
Reoptimization

F. Reidl

Reoptimization

Parameterized
Complexity

Parameterized
Reoptimization
Can we do better?

No
 Parameterized
Reoptimization

F. Reidl

Reoptimization

Parameterized
Complexity

Parameterized
Reoptimization
Can we do better?

No
(Not by much.)
 Parameterized
An iterative algorithm Reoptimization

F. Reidl

Reoptimization

Parameterized
Complexity

Parameterized
Reoptimization
 Parameterized
The “compression routine” Reoptimization

F. Reidl

Reoptimization

Parameterized
Complexity

Parameterized
Reoptimization
 Parameterized
The “compression routine” Reoptimization

F. Reidl

Reoptimization

Parameterized
Complexity

Parameterized
Reoptimization
 Parameterized
General result Reoptimization

F. Reidl

Reoptimization

Parameterized
For all Vertex Cover-like problems the following holds: Complexity

Parameterized
Theorem Reoptimization

Given a reoptimization algorithm A with running time

f (k)nO(1) , one can solve the underlying parameterized
problem in time O f (k)nO(1) .


 Parameterized
General result Reoptimization

F. Reidl

Reoptimization

Parameterized
For all Vertex Cover-like problems the following holds: Complexity

Parameterized
Theorem Reoptimization

Given a reoptimization algorithm A with running time

f (k)nO(1) , one can solve the underlying parameterized
problem in time O f (k)nO(1) .

“Having an optimal solution for a similar instance does not

help”
 Parameterized
General result Reoptimization

F. Reidl

Reoptimization

Parameterized
For all Vertex Cover-like problems the following holds: Complexity

Parameterized
Theorem Reoptimization

Given a reoptimization algorithm A with running time

f (k)nO(1) , one can solve the underlying parameterized
problem in time O f (k)nO(1) .

“Having an optimal solution for a similar instance does not

help”

It does not matter what operation (edge/vertex

deletion/addition) the reoptimization algorithm supports.
 Parameterized
Affected problems Reoptimization

F. Reidl

Reoptimization
The theorem holds for the following problems: Parameterized
Complexity
I Vertex Cover Parameterized
Reoptimization
I Feedback Vertex Set
I Vertex-Deletion Bipartization
I Cluster Vertex Deletion
 Parameterized
Affected problems Reoptimization

F. Reidl

Reoptimization
The theorem holds for the following problems: Parameterized
Complexity
I Vertex Cover Parameterized
Reoptimization
I Feedback Vertex Set
I Vertex-Deletion Bipartization
I Cluster Vertex Deletion
...
I Connected [Insert from above]
 Parameterized
Affected problems Reoptimization

F. Reidl

Reoptimization
The theorem holds for the following problems: Parameterized
Complexity
I Vertex Cover Parameterized
Reoptimization
I Feedback Vertex Set
I Vertex-Deletion Bipartization
I Cluster Vertex Deletion
...
I Connected [Insert from above]

Other problems (same result with slightly different approach) :

I Steiner Tree
 Parameterized
Affected problems Reoptimization

F. Reidl

Reoptimization
The theorem holds for the following problems: Parameterized
Complexity
I Vertex Cover Parameterized
Reoptimization
I Feedback Vertex Set
I Vertex-Deletion Bipartization
I Cluster Vertex Deletion
...
I Connected [Insert from above]

Other problems (same result with slightly different approach) :

I Steiner Tree
I Dominating Set (Therefore not FPT-Reoptimizable)
 Parameterized
Reoptimization

F. Reidl

Reoptimization

Parameterized
Complexity

Parameterized
Reoptimization

Thanks!
Questions?