![Page 1: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/1.jpg)
Reinhard Pichler
Vienna University of Technology
SCCC 2013
Coping with High Complexity:
Structure Matters
![Page 2: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/2.jpg)
SCCC 2013 Reinhard Pichler 2
Roadmap
I. High complexity everywhere
II. Parameterized Complexity
III. What is a Parameter?
IV. How to Make Use of Parameters?
V. Conclusion and Research Opportunities
![Page 3: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/3.jpg)
SCCC 2013 Reinhard Pichler 3
Part I:High complexity everywhere
![Page 4: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/4.jpg)
SCCC 2013 Reinhard Pichler 4
A starting point
Theory of Computation: 1930s
– What can be computed?
Theory of Tractability: 1970s
– What can be computed efficiently?
Turing
KarpCook
![Page 5: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/5.jpg)
SCCC 2013 Reinhard Pichler 5
Computational hardness
NP-completeness:• small solutions• easy to verify• huge search space
thousands of NP-complete problems
![Page 6: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/6.jpg)
SCCC 2013 Reinhard Pichler 6
NP-complete problems
Combinatorial auctions(e.g. London Transport)
Scheduling(e.g. work schedules)
Graph layout problems(e.g. circuit design)
![Page 7: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/7.jpg)
SCCC 2013 Reinhard Pichler 7
Satisfiability
Is the following formula φ satisfiable?
φ = (a b) (b c) (¬a d) (¬b ¬c ¬d)∨ ∧ ∨ ∧ ∨ ∧ ∨ ∨
![Page 8: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/8.jpg)
SCCC 2013 Reinhard Pichler 8
Satisfiability
Is the following formula φ satisfiable?
φ = (a b) (b c) (¬a d) (¬b ¬c ¬d)∨ ∧ ∨ ∧ ∨ ∧ ∨ ∨
Work(8-9) Work(9-10) Work(10-11)→Break(11-12)∧ ∧
Work(8-9) Work(10-11)→(Break(9-10) Work(9-10))∧ ∨
very expressive NP-complete problem
![Page 9: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/9.jpg)
SCCC 2013 Reinhard Pichler 9
PSPACE
PH
The hierarchy of complexity
NP
P
Intr
acta
ble
![Page 10: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/10.jpg)
SCCC 2013 Reinhard Pichler 10
How to deal with intractability
Incomplete methods● Approximation algorithms● Heuristics● Randomized algorithms
not exact
![Page 11: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/11.jpg)
SCCC 2013 Reinhard Pichler 11
How to deal with intractability
Incomplete methods● Approximation algorithms● Heuristics● Randomized algorithms
Exact methods● Brute force● “Islands of tractability”● ...
not exact
![Page 12: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/12.jpg)
SCCC 2013 Reinhard Pichler 12
Islands of tractability
Easy to solve special cases
For the Satisfiability problem● Horn formulas:
Work(8-9) Work(9-10) Work(10-11)→Break(11-12)∧ ∧
● Krom formulas:
¬Work(8-9) ¬Break(8-9)∨
However: Such “islands” are not robust!
![Page 13: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/13.jpg)
SCCC 2013 Reinhard Pichler 13
Summary of Part I
● High complexity is ubiquitous● There are many methods to tackle intractable
problems, ...● … but can we have a method that is
– efficient,
– exact and
– robust?
![Page 14: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/14.jpg)
SCCC 2013 Reinhard Pichler 14
Part II:Parameterized Complexity
![Page 15: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/15.jpg)
SCCC 2013 Reinhard Pichler 15
Parameterized Complexity
✔ efficient,✔ exact and✔ robust.
Main idea:Exploit structural
properties of problem instances.
Main idea:Exploit structural
properties of problem instances.
![Page 16: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/16.jpg)
SCCC 2013 Reinhard Pichler 16
Example: subway emergencies
Problem:
emergency teams for every line segment
For example:
There has to be a team either at Santa Ana or at Los Héroes to cover the line segment in between.
![Page 17: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/17.jpg)
SCCC 2013 Reinhard Pichler 17
Vertex Cover
● Given a graph G and integer k, is there a set of vertices S of size at most k such that for every edge {a,b}, S contains a or b?
● NP-complete● the best known exact algorithms require
exponential time
![Page 18: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/18.jpg)
SCCC 2013 Reinhard Pichler 18
Real-world versus random108 stations,connected
108 nodes,connected
![Page 19: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/19.jpg)
SCCC 2013 Reinhard Pichler 19
The classical point of view
Classical Complexity Theory● measures complexity only in terms of size of an
instance● one-dimensional: O(f(n))● tractable means polynomial time
![Page 20: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/20.jpg)
SCCC 2013 Reinhard Pichler 20
The classical point of view
Classical Complexity Theory● measures complexity only in terms of size of an
instance● one-dimensional: O(f(n))● tractable means polynomial time
Criticism:We (almost) always know more about the input than
its number of bits.
Criticism:We (almost) always know more about the input than
its number of bits.
![Page 21: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/21.jpg)
SCCC 2013 Reinhard Pichler 21
The parameterized point of view
Parameterized Complexity Theory● takes structural properties of problem
instances into account (parameter k)● multi-dimensional: O(f(n,k))● tractability depends on parameter
![Page 22: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/22.jpg)
SCCC 2013 Reinhard Pichler 22
The parameterized point of view
Parameterized Complexity Theory● takes structural properties of problem
instances into account (parameter k)● multi-dimensional: O(f(n,k))● tractability depends on parameter
Benefit: Allows a more fine-grained
notion of tractability.
Benefit: Allows a more fine-grained
notion of tractability.
![Page 23: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/23.jpg)
SCCC 2013 Reinhard Pichler 23
Fixed-parameter tractability (FPT)
● ideal outcome of a parameterized complexity analysis
● O(f(k) * nd)● example: 2k * n2
● consequence: tractable for parameter of bounded size.
![Page 24: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/24.jpg)
SCCC 2013 Reinhard Pichler 24
Fixed-parameter tractability (FPT)
● ideal outcome of a parameterized complexity analysis
● O(f(k) * nd)● example: 2k * n2
● consequence: tractable for parameter of bounded size.
Example: Vertex Cover with bounded treewidth
![Page 25: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/25.jpg)
SCCC 2013 Reinhard Pichler 25
The class XP
● “second prize”● O(nf(k))● example: nk
● consequence: less favorable than FPT but still tractable for parameter of bounded size
![Page 26: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/26.jpg)
SCCC 2013 Reinhard Pichler 26
The class XP
● “second prize”● O(nf(k))● example: nk
● consequence: less favorable than FPT but still tractable for parameter of bounded size
Example: Conjunctive query evaluation with queries of bounded hypertree-width
![Page 27: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/27.jpg)
SCCC 2013 Reinhard Pichler 27
Comparison: FPT and XP
FPT XP
2k * 1002 100k
XP
FPT
Example:input size 100
classical point of view:2100 > 1030
![Page 28: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/28.jpg)
SCCC 2013 Reinhard Pichler 28
Summary of Part II
● Real-world problems often have structure● Classical complexity only considers the size of a
problem instance● Parameterized complexity takes problem
parameters into account➔ Exploit inherent structure of problem instances
![Page 29: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/29.jpg)
SCCC 2013 Reinhard Pichler 29
Part III:What is a Parameter?
![Page 30: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/30.jpg)
SCCC 2013 Reinhard Pichler 30
Boolean Satisfiability Revisited
φ = (a b) (b c) (¬a d) (¬b ¬c ¬d)∨ ∧ ∨ ∧ ∨ ∧ ∨ ∨
![Page 31: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/31.jpg)
SCCC 2013 Reinhard Pichler 31
Boolean Satisfiability Revisited
● Clause size = 3● Positive clause size = 2● Negative clause size = 3● Number of non-Horn clauses = 2● Number a variable occurs positively = 2● Number of variables that occur as positive literals = 4● … as negative literals … = 4 ● etc.
φ = (a b) (b c) (¬a d) (¬b ¬c ¬d)∨ ∧ ∨ ∧ ∨ ∧ ∨ ∨
![Page 32: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/32.jpg)
SCCC 2013 Reinhard Pichler 32
Parameterized Complexity Analysis
● Parameter “number of clauses”– 1.24k * poly(n)
● Parameter “number of variables”– 1.49k * poly(n)
● Parameter “clause size”– NP-complete for k=3
– O(2f(n))
![Page 33: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/33.jpg)
SCCC 2013 Reinhard Pichler 33
Generalization of Special Cases
● SAT– almost Horn / almost Krom ⇒ backdoors
● Graph Problems– almost trees ⇒ treewidth
● CSPs / CQs– almost acyclic ⇒ hypertree-width
![Page 34: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/34.jpg)
SCCC 2013 Reinhard Pichler 34
Minimal Model Satisfiability (MMSAT)
Instance: CNF formulas φ and π.
Question: Is there a subset minimal model of φ that also satisfies π ?
● Important subtask in non-monotonic reasoning.● MMSAT captures the complexity of computing a
minimal model.
![Page 35: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/35.jpg)
SCCC 2013 Reinhard Pichler 35
Minimal Model Satisfiability
k maximum weight of the minimal model
d clause size
d+,d- positive/negative clause size
h number of non-Horn clauses
p number a variable occurs positively
v+,v- number of variables that occur as positive/negative literals in φ or in π
dπ+positive clause size in π
||π|| length of π
b size of a strong Horn backdoor set
[Lackner, Pfandler 2012]
![Page 36: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/36.jpg)
SCCC 2013 Reinhard Pichler 36
Summary of Part III
● Any characteristics of problem instances may serve as parameters.
● Usefulness of a parameter depends on:– FPT (or at least XP) result
– Application context
![Page 37: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/37.jpg)
SCCC 2013 Reinhard Pichler 37
Part IV:How to Make Use of Parameters?
![Page 38: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/38.jpg)
SCCC 2013 Reinhard Pichler 38
Toolbox of Parameterized Complexity
W[i]-hardness
Kernel lower bounds
Exponential TimeHypothesis
Bounded search trees
Iterative compression
Logical meta-theorems
Color coding
Graph minors
Kernelization
Hardness ToolsHardness Tools
AlgorithmicTools
AlgorithmicTools
![Page 39: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/39.jpg)
SCCC 2013 Reinhard Pichler 39
Toolbox of Parameterized Complexity
W[i]-hardness
Kernel lower bounds
Exponential TimeHypothesis
Bounded search trees
Iterative compression
Logical meta-theorems
Color coding
Graph minors
Kernelization
Hardness ToolsHardness Tools
AlgorithmicTools
AlgorithmicTools
![Page 40: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/40.jpg)
SCCC 2013 Reinhard Pichler 40
Tree Decomposition
● Tree with a vertex set (= “bag”) associated with every node.
● For every edge (v,w): there is a bag containing both v and w.
● For every v: the nodes that contain v form a connected subtree.
![Page 41: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/41.jpg)
SCCC 2013 Reinhard Pichler 41
Tree Decomposition
● Tree with a vertex set (= “bag”) associated with every node.
● For every edge (v,w): there is a bag containing both v and w.
● For every v: the nodes that contain v form a connected subtree.
![Page 42: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/42.jpg)
SCCC 2013 Reinhard Pichler 42
Tree Decomposition
● Tree with a vertex set (= “bag”) associated with every node.
● For every edge (v,w): there is a bag containing both v and w.
● For every v: the nodes that contain v form a connected subtree.
![Page 43: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/43.jpg)
SCCC 2013 Reinhard Pichler 43
Tree Decomposition
● Tree with a vertex set (= “bag”) associated with every node.
● For every edge (v,w): there is a bag containing both v and w.
● For every v: the nodes that contain v form a connected subtree.
![Page 44: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/44.jpg)
SCCC 2013 Reinhard Pichler 44
Tree Decomposition
Graph
● Tree with a vertex set (= “bag”) associated with every node.
● For every edge (v,w): there is a bag containing both v and w.
● For every v: the nodes that contain v form a connected subtree.
Structure
● Tree with a set of domain elements (= “bag”) associated with every node.
● For every tuple (a1,...,ak) in any relation Ri: there is a bag containing {a1,...,ak}.
● For every a: the nodes that contain a form a connected subtree.
![Page 45: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/45.jpg)
SCCC 2013 Reinhard Pichler 45
Treewidth
● The width of a tree decomposition is the maximum bag size – 1.
● The treewidth of G is the minimum width over all tree decompositions of G.
● For fixed k, it is feasible in linear time to decide / compute tree decomposition of width ≤ k. [Bodlaender, 1996]
![Page 46: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/46.jpg)
SCCC 2013 Reinhard Pichler 46
Treewidth of a CNF Formula
● Represent CNF formula F as finite structure A(F) over signature τ = {cl, var, pos, neg}– cl(c), var(x): c is a clause (x is a variable) in F
– pos(x,c), neg(x,c): x occurs unnegated (negated) in clause c
● Treewidth of F is defined as the treewidth of A(F)
![Page 47: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/47.jpg)
SCCC 2013 Reinhard Pichler 47
Treewidth of a CNF Formula
Remark: this corresponds to the incidence graph of F.
● Represent CNF formula F as finite structure A(F) over signature τ = {cl, var, pos, neg}– cl(c), var(x): c is a clause (x is a variable) in F
– pos(x,c), neg(x,c): x occurs unnegated (negated) in clause c
● Treewidth of F is defined as the treewidth of A(F)
![Page 48: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/48.jpg)
SCCC 2013 Reinhard Pichler 48
Treewidth of CNF: Example
Let F = (x1 ¬x∨ 2 x∨ 3) (x∧ 2 ¬x∨ 3 x∨ 4)
● Then A(F) consists of:
– cl(c1), cl(c2)
– var(x1), var(x2), var(x3), var(x4)
– pos(x1,c1), pos(x3,c1), pos(x2,c2), pos(x4, c2)
– neg(x2,c1), neg(x3,c2)
![Page 49: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/49.jpg)
SCCC 2013 Reinhard Pichler 49
Treewidth of CNF: Example
● cl(c1), cl(c2)
● var(x1), var(x2), var(x3), var(x4)
● pos(x1,c1), pos(x3,c1), pos(x2,c2), pos(x4,c2)
● neg(x2,c1), neg(x3,c2)
F has treewidth = 2.F has treewidth = 2.
![Page 50: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/50.jpg)
SCCC 2013 Reinhard Pichler 50
Monadic Second-Order Logic (MSO)
● MSO extends first-order logic by the use of set variables.
● An MSO formula allows the following atoms:
– Relational atoms, e.g. R(x1,...,xn).
– Equational atoms, e.g. x=y.
– Atoms based on set variables, e.g. X(y).
![Page 51: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/51.jpg)
SCCC 2013 Reinhard Pichler 51
Monadic Second-Order Logic (MSO)
● MSO extends first-order logic by the use of set variables.
● An MSO formula allows the following atoms:
– Relational atoms, e.g. R(x1,...,xn).
– Equational atoms, e.g. x=y.
– Atoms based on set variables, e.g. X(y).
Courcelle's Theorem: Any property of finitestructures which is definable in MSO canbe decided in time O(f(k)*n), where n is thesize of the structure and k is its treewidth.
Courcelle's Theorem: Any property of finitestructures which is definable in MSO canbe decided in time O(f(k)*n), where n is thesize of the structure and k is its treewidth.
![Page 52: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/52.jpg)
SCCC 2013 Reinhard Pichler 52
From MSO to Algorithms
Various (automatic) constructions of FPT-algorithms from MSO-encodings exist:– Using correspondence between finite tree automata and MSO
on trees [Courcelle 1990; Arnborg et al. 1991; Flum et al. 2001]
– Finite model theory (k-types) [Grohe 1999]
– Datalog [Gottlob, P., Wei 2007]
– Games [Kneis, Langer, Rossmanith, 2011]
– New automata models (Courcelle, Durand 2011)
– Answer-set programming [Bliem, P., Woltran 2013]
![Page 53: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/53.jpg)
SCCC 2013 Reinhard Pichler 53
Applying Courcelle's Theorem
Theorem (folklore):SAT is FPT w.r.t. treewidth.Theorem (folklore):SAT is FPT w.r.t. treewidth.
![Page 54: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/54.jpg)
SCCC 2013 Reinhard Pichler 54
Applying Courcelle's Theorem
Proof: MSO-encoding.
Let F be a CNF formula and X a set of variables.
Encoding of X⊨F:
( c)cl(c)→( z)[(pos(z,c) X(z)) (neg(z,c) ¬X(z))]∀ ∃ ∧ ∨ ∧
Encoding of SAT:
( X) X F∃ ⊨
Theorem (folklore):SAT is FPT w.r.t. treewidth.Theorem (folklore):SAT is FPT w.r.t. treewidth.
![Page 55: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/55.jpg)
SCCC 2013 Reinhard Pichler 55
Minimal Model Satisfiability
Theorem [Gottlob, P., Wei, 2006]:Minimal Model SAT is FPT w.r.t. treewidth.Theorem [Gottlob, P., Wei, 2006]:Minimal Model SAT is FPT w.r.t. treewidth.
● Instance: CNF formulas F and π● Question: Is there a subset minimal model of φ that also
satisfies π?
![Page 56: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/56.jpg)
SCCC 2013 Reinhard Pichler 56
Minimal Model Satisfiability
● Instance: CNF formulas φ and π● Question: Is there a subset minimal model of φ that also
satisfies π?
Proof: MSO-encoding.
Encoding of Y X:⊊∀x (Y(x)→X(x)) y (X(y) ¬Y(y))∧ ∃ ∧
Encoding of Minimal Model SAT:
∃X, X φ X π Y (Y X → ¬Y φ)⊨ ∧ ⊨ ∧ ∀ ⊊ ⊨
Theorem [Gottlob, P., Wei, 2006]:Minimal Model SAT is FPT w.r.t. treewidth.Theorem [Gottlob, P., Wei, 2006]:Minimal Model SAT is FPT w.r.t. treewidth.
![Page 57: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/57.jpg)
SCCC 2013 Reinhard Pichler 57
Treewidth as a Key to Tractable Reasoning
● Minimal models● Various forms of closed-world reasoning● Disjunctive logic programming● Propositional abduction
[Gottlob, P., Wei, 2006]
![Page 58: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/58.jpg)
SCCC 2013 Reinhard Pichler 58
Part V:Conclusion and Research Opportunities
![Page 59: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/59.jpg)
SCCC 2013 Reinhard Pichler 59
Conclusion
Parameterized complexity is a viable way to tackle intractable problems:● Real-world problems often have structure● Classical complexity only considers the size of a
problem instance● Parameterized complexity takes problem
parameters into account
![Page 60: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/60.jpg)
SCCC 2013 Reinhard Pichler 60
The Gentle Revolution of Parameterized Complexity
● Bioinformatics, Operations Research, Optimization, Automated Reasoning, etc.
● STOC, FOCS, SODA, IJCAI, …
● 4 Monographs– [Downey&Fellows 1999]
– [Flum&Grohe 2006]
– [Niedermeier 2006]
– [Downey&Fellows 2013]
● The Computer Journal (BCS) two special issues in 2008
Papers containing “parameterized complexity” or “fixed-parameter tractable” published per year.
Source: Google Scholar
2000 2002 2004 2006 2008 2010 20120
100
200
300
400
500
600
700
800
900
1000
![Page 61: Coping with High Complexity: Structure MattersStructure Matters. SCCC 2013 Reinhard Pichler 2 Roadmap I. High complexity everywhere II. Parameterized Complexity III. What is a Parameter?](https://reader033.vdocuments.site/reader033/viewer/2022042103/5e80851f24a2a75c577f9109/html5/thumbnails/61.jpg)
SCCC 2013 Reinhard Pichler 61
Research Opportunities
● Parameterized complexity theory: – New algorithmic methods
– Kernelization (formal model of preprocessing)
– Relationship with other approaches (approximation, heuristics)
● Further applications