a rendezvous of complexity, logic, algebra, and ......graph-theoretic problems spatial and temporal...

38
A Rendezvous of Complexity, Logic, Algebra, and Combinatorics Hubie Chen Universitat Pompeu Fabra Barcelona, Spain BCAM - September 2011 Hubie Chen A Rendezvous

Upload: others

Post on 26-Jul-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

A Rendezvous of Complexity, Logic, Algebra, andCombinatorics

Hubie Chen

Universitat Pompeu FabraBarcelona, Spain

BCAM - September 2011

Hubie Chen A Rendezvous

Page 2: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

3-Colorability

Given an undirected graph...

...is it 3-colorable?

Idea: divide a set of objects into three groups so that no twoincompatible objects are placed into same group

Hubie Chen A Rendezvous

Page 3: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Boolean satisfiability

Given a conjunction of propositional clauses...

(¬a ∨ ¬b ∨ c) ∧ (¬c ∨ d) ∧ (a ∨ ¬d)

...is there a satisfying assignment to the variables?

a → 1b → 0c → 0d → 1

Hubie Chen A Rendezvous

Page 4: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Digraph acyclicity

Given a directed graph, does it contain a cycle?

Hubie Chen A Rendezvous

Page 5: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Efficient algorithms

A broad goal of computer science:

Develop fast, usable algorithms for computational problems

A broad goal of theoretical computer science:

Classify problems according to whether or notthey have “efficient” algorithms

That is, classify problems according to their inherentcomplexity...

Hubie Chen A Rendezvous

Page 6: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Classifying problems

Theoretical computer science: place problems into complexityclasses...

P: polynomial time

Theoretical notion of efficiency/tractabilityProblems for which there’s an algorithm running in time O(nk)Ex: digraph acyclicity

NP: non-deterministic polynomial time

P ⊆ NP, believed that NP is “much bigger”Idea: problems for which a solution can be efficiently verified

Exs: all three problems seen

NP-hard: problems “as hard as”/that “can simulate”all problems in NP

Exs: 3-colorability, boolean sat.

Generally believed that

P = NP ⇔ no NP-hard problem is in P

Hubie Chen A Rendezvous

Page 7: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

A dream

(Garey & Johnson ’79): Compendium of 100s of NP-hardproblems

Dream: to have an encyclopedia – a general classificationtheorem that, given a problem, would tell us if tractable ornot (if in P or if NP-hard)

This talk: discuss (my) research towards this dream...

Hubie Chen A Rendezvous

Page 8: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Classification

Classification projects

Identify a broad family of problemsTry to classify each problem as tractable/intractableOften, obtain generic conditions for tractability/intractability

Hubie Chen A Rendezvous

Page 9: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Exhibit #1: Constraint Satisfaction

Hubie Chen A Rendezvous

Page 10: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Examples

Given an undirected graph, is it 3-colorable?

Given a conjunction of propositional clauses, is there asatisfying assignment?

(¬a ∨ ¬b ∨ c) ∧ (¬c) ∧ (a)

Given a directed graph, does it contain a cycle?

Hubie Chen A Rendezvous

Page 11: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Animals

Are these problems inherently different animals?

...or can we place them into an organizational scheme?

Hubie Chen A Rendezvous

Page 12: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Constraint Satisfaction

All of these problems are constraint satisfaction problems!

The constraint satisfaction problem (CSP)

Given a primitive positive (pp) sentence

φ = ∃v1 . . . ∃vn (R(vi1 , . . . , vik ) ∧ . . .)

and a relational structure B, decide if B |= φ.

Idea: decide if there is an assignment to variables satisfying acollection of constraints

Examples:

graph coloring (vi = vertices, B = colors,one constraint E (vi , vj) for each edge)boolean satisfiability (vi = variables, B = 0, 1,one constraint for each clause)

Hubie Chen A Rendezvous

Page 13: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Constraint Satisfaction

For each structure B, we can define:

CSP(B)

Given a primitive positive (pp) sentence φ, decide if B |= φ.

Can capture many studied problems:

Graph 3-Coloring (B = K3), generalizationsBoolean sat. problems: 3-SAT, 2-SAT, Horn-SATDigraph acyclicity (B = (Q;<))Temporal Reasoning (e.g. B = (Q;≤, <, =))Systems of equations over various algebraic structures

Hubie Chen A Rendezvous

Page 14: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Constraint Satisfaction

Unified framework, captures many problems

Numerous application areas:

Assignment/scheduling problemsBoolean sat. problems & generalizationsGraph-theoretic problemsSpatial and temporal reasoningInterval reasoningAlgebraic equations problems

Research question:

Which problems CSP(B) are tractable (in poly-time), and why?

Known:

(Schaefer ’78): Classification for 2-element structures(Bulatov ’02): Classification for 3-element structures

Hubie Chen A Rendezvous

Page 15: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Constraint Satisfaction and Consistency

Consistency methods: simple, efficient, well-studied,used-in-practice inference methods that (in general) candetect unsatisfiability

What are the scope of various consistency methods?Which problems CSP(B) can they solve?

My work: study ofarc consistency and extensions(Bodirsky & Chen ’10), (Chen, Dalmau & Grussien ’11)local-to-global consistency(Bodirsky & Chen ’09), (Bodirsky, Chen & Wrona, ongoing)

Hubie Chen A Rendezvous

Page 16: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

An Algebraic Approach

Algebraic approach: associate to each struct. B an algebra AB

An operation f : Bm → B is a polymorphism of a relationU ⊆ Bk if for any choice of m tuples from U, applying fcoordinatewise yields a tuple (s1, . . . , sk) also in U.

(t11, t12, . . . t1k) ∈ U...

.... . .

...(tm1, tm2, . . . tmk) ∈ U

f ↓ f ↓ . . . f ↓

(s1, s2, . . . sk) ∈ U

An operation f is a polymorphism of a structure if it is apolymorphism of all relations (of the structure)

Hubie Chen A Rendezvous

Page 17: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

An Algebraic Approach

For a struct. B, define the algebra of polymorphisms AB to bethe algebra

with universe B , andwhose operations are all polymorphisms of B

Key Theorem (Geiger/Bodcharnuk et al. ’60s, Jeavons et al. ’90s):

If finite structs. B,B have AB = AB , thenB,B can pp-define (∃,∧) the same relations, andCSP(B), CSP(B) have same complexity(up to polytime reduction)

Use properties of algebra AB to understand struct. B

Hubie Chen A Rendezvous

Page 18: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Exhibit #2: Quantified Constraint Satisfaction

Hubie Chen A Rendezvous

Page 19: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Quantified Constraint Satisfaction

The quantified constraint satisfaction problem (QCSP)

Given a conjunctive positive (cp) sentence

φ = Q1v1 . . .Qnvn (R(vi1 , . . . , vik ) ∧ . . .)

(with Qi ∈ ∀, ∃), and a relational structure B, decide if B |= φ.

Can model situations with uncertainty, two-party interaction,... that cannot be modelled with CSP(complexity is higher: generally PSPACE-complete)

Wide range of applications: verification, game playing,adversarial reasoning, planning, ...

Hubie Chen A Rendezvous

Page 20: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Quantified Constraint Satisfaction

QCSP(B)

Given a conjunctive positive sentence φ, decide if B |= φ.

Framework goes back to:

(Aspvall, Plass, Tarjan ’79) - Quantified 2-SAT

(Karpinski, Kleine Buning, Schmitt ’87) - Quantified HornSAT

Research issues:

Search for generic sources of tractability / hardness

Develop algebraic approach

Attempt to establish systematic classifications

Hubie Chen A Rendezvous

Page 21: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Polynomially Generated Powers (PGP)

(Chen ’08, ’11): study of polynomially generated powers(PGP) property – algebraic property

A combinatorial property of algebras concerning the size ofgenerating sets for An

(Wiegold ’70s-’80s), (Quick & Ruskuc ’10), (Awang, Garrido,McLeman, Quick & Ruskuc), ...

Hubie Chen A Rendezvous

Page 22: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Polynomially Generated Powers (PGP)

Def: An algebra A has polynomially generated powers if thereis a polynomial p ∈ N[X ] such that, for each n ≥ 1,the algebra An has a generating set of size ≤ p(n)

Example: let G be a finite group with identity elt. eConsider tuples where all entries [with one possible exception]are = e. In G 4:

(g1, e, e, e)(e, g2, e, e)(e, e, g3, e)(e, e, e, g4)↓ ↓ ↓ ↓

(g1, g2, g3, g4)

Have generating set for Gn of size ≤ |G |n: linear growth

Hubie Chen A Rendezvous

Page 23: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Polynomially Generated Powers (PGP)

(Chen ’11): Gave broad conditions for reducing QCSP to CSPbased on PGP

Observation: Suppose that algebra AB has the PGP, and agenerating set for An

B can be computed in poly time (in n)

Then, Π2-QCSPc(B) reduces to CSPc(B):

Consider an instance Φ = ∀y1 . . . ∀yn∃x1 . . . ∃xmψCompute a generating set t1, . . . , tk for An

BLet Φ[ti ] be the CSP instance where the vars. yj arereplaced with the values in tiΦ is equivalent to the CSP instance Φ[t1] ∧ · · · ∧ Φ[tk ]

Hubie Chen A Rendezvous

Page 24: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Polynomially Generated Powers (PGP)

Thm (Chen ’11): Let AB be a 3-elt algebra[that’s idempotent and where CSP(B) tractable].

Either:

AB is switchable, has the PGP, and QCSP(B) tractable; orAB has exponentially generated powers (EGP):generating sets require size Ω(cn), with c > 1

Dichotomy theorem: all studied algebras have PGP or EGP

Reduces the classification of tractable 3-elt QCSP(B) to aparticular, robust class of algebras

Open questions:

Full QCSP classification for 3-elt structuresClassification of PGP/EGP on all finite algebras

Hubie Chen A Rendezvous

Page 25: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Exhibit #3: Constraint Satisfaction,from the other side

Hubie Chen A Rendezvous

Page 26: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Constraint Satisfaction, the other side

Recall...

The Constraint Satisfaction Problem (CSP)

Given a primitive positive sentence φ = ∃v1 . . . ∃vn ∧ αj anda structure B, decide if B |= φ.

Before, we restricted the structure. Now we restrict the formulas.

Remark: for a fixed sentence φ, the CSP on φ is in poly time

Question: on which sets of formulas is CSP tractable?What structural restrictions can we place on formulas, toensure tractability?

Hubie Chen A Rendezvous

Page 27: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Graph-based approach

Graph-based approach: given a CSP instance, look at graph wherevertices are variables, two variables linked if in a commonconstraint

Example:∃a∃b∃c∃d∃e∃f ∃g (R(a, b, c) ∧ S(b, c , d) ∧ T (d , e, f )

∧U(a, g) ∧W (f , g))

a

b

c

d

e

f

g

Hubie Chen A Rendezvous

Page 28: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Treewidth

Thm (Freuder ’90):When the graphs have bounded treewidth,the CSP is tractable.

Tree, tw=1 Cycle, tw=2 k-Clique,tw=(k-1)

Thm (Grohe ’07): This theorem is optimal–unbounded treewidth implies CSP intractability.

In fact, bounded treewidth essentially determines tractabilityon formulas of bounded arity.

Hubie Chen A Rendezvous

Page 29: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Hypergraph-based approach

Hypergraph-based approach: given a CSP instance, look athypergraph where vertices are variables, have a hyperedge for eachconstraint

Example:

∃a∃b∃c∃d∃e∃f ∃g (R(a, b, c) ∧ S(b, c , d) ∧ T (d , e, f )∧U(a, g) ∧W (f , g))

a

b

c

d

e

f

g

Hubie Chen A Rendezvous

Page 30: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Hypergraph-based approach

Basic question: classify sets of hypergraphs as tractable /intractable

(Gottlob et al., ’90s): Introduction of hypergraph complexitymeasures – hypertree width, generalized hypertree width

Showed that CSP is tractable when hypertree width is bounded

Thm (Chen & Dalmau ’05): CSPs are tractable when thegeneralized hypertree width is bounded.(Resolved primary open question of Gottlob et al. work)

Still open: which hypergraphs can be solvable in poly time?

Hubie Chen A Rendezvous

Page 31: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Exhibit #4: Quantified Constraint Satisfaction,from the other side

Hubie Chen A Rendezvous

Page 32: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Quantified Constraint Satisfaction, the other side

Recall...

The Quantified Constraint Satisfaction Problem (QCSP)

Given a conjunctive positive sentence φ = Q1v1 . . .Qnvn ∧ αj anda structure B, decide if B |= φ.

Before, we restricted the structure. Now we restrict the formulas.

Question: on which sets of formulas is QCSP tractable?What structural restrictions can we place on formulas, toensure tractability?

Hubie Chen A Rendezvous

Page 33: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Graph-based approach

Graph-based approach: given a QCSP instance, look atprefixed graph containing quantifier prefix & graph

Example:∀a∃b∃c∀d∃e∀f ∃g (R(a, b, c) ∧ S(b, c , d) ∧ T (d , e, f )

∧U(a, g) ∧W (f , g))

(∀a∃b∃c∀d∃e∀f ∃g ,

a

b

c

d

e

f

g)

Hubie Chen A Rendezvous

Page 34: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Prefixed Graphs

Prefixed graph perspective:now have graph PLUS quantifier prefix

Examples:

X1

X2

X3

Y

∃X1∃X2∃X3 . . . ∃Xn∀YTractable!

Y1

Y2

Y3

X

∀Y 1∀Y 2∀Y 3 . . . ∃Yn∃XIntractable!

Hubie Chen A Rendezvous

Page 35: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Prefixed Graphs

(Chen & Dalmau ’05): Notion of treewidth for prefixedgraphs, proof that bounded treewidth ⇒ tractability

Pulina and Tacchella (’10) study this treewidth measure,showing that an approximation

“is a marker of empirical hardness”, and “it is theonly parameter that succeeds consistently in beingso among several other syntactic parameters whichare plausible candidates”

Close connection between

treewidth notions ↔ variable elimination algorithms(used in practice)

Hubie Chen A Rendezvous

Page 36: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Prefixed Graphs

(Chen & Dalmau, recent):

New & refined notion of width for prefixed graphs(assigns a natural number w ∈ N to each prefixed graph)

Generalization of treewidth

Complete classification of all prefixed graphs:A set of prefixed graphs is tractable iff it has bounded width

Open: Classification for formulas of bounded arity.

Perhaps the following will help:

Theorem (Chen, Madelaine & Martin ’08):Entailment and equivalence are decidable (computable) on QCSP(∀, ∃,∧) sentences.

(New decidability result for entailment!Optimal in that entailment undecidable in positive (∀, ∃,∧,∨).)

Hubie Chen A Rendezvous

Page 37: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Closing

Hubie Chen A Rendezvous

Page 38: A Rendezvous of Complexity, Logic, Algebra, and ......Graph-theoretic problems Spatial and temporal reasoning Interval reasoning Algebraic equations problems Research question: Which

Themes

Study the computational complexity of fundamentalcomputational problems

Dream: to have an encyclopedic “master” classificationtheoremLook at broad families of problems,attempt to obtain comprehensive classification theoremsOften obtain general conditions for tractability/intractability

Use ideas from / interface with many areas: algorithms,constraint satisfaction, universal algebra, graph theory,combinatorics, logic ...

Hubie Chen A Rendezvous