Ranking based Techniques for Disambiguating B uchi Automata - - PowerPoint PPT Presentation

ranking based techniques for disambiguating b uchi
SMART_READER_LITE
LIVE PREVIEW

Ranking based Techniques for Disambiguating B uchi Automata - - PowerPoint PPT Presentation

Ranking based Techniques for Disambiguating B uchi Automata Hrishikesh Karmarkar Supratik Chakraborty January 29, 2011 Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B uchi Automata


slide-1
SLIDE 1

Ranking based Techniques for Disambiguating B¨ uchi Automata

Hrishikesh Karmarkar Supratik Chakraborty January 29, 2011

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-2
SLIDE 2

Non-deterministic B¨ uchi automata over words (NBW)

A 5−tuple (Σ, Q, Q0, δ, F), where Σ : Input alphabet Q : Finite set of states Q0 ⊆ Q: Initial states δ ⊆ Q × Σ × Q: State transition relation F : Set of final/accepting states

q1 q2 q3 q4 q5 qs a b a, b Final state Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-3
SLIDE 3

Runs and acceptance

A run of A on α ∈ Σω is a sequence ρ : N → Q such that

ρ(0) ∈ Q0 ρ(i + 1) ∈ δ(ρ(i), α(i))

q1 q2 q3 q4 q5 qs a b a, b Final state

α = abbbbb · · · , ρ1 = q1q2q2q2q2q2q2 · · ·

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-4
SLIDE 4

Runs and acceptance

A run of A on α ∈ Σω is a sequence ρ : N → Q such that

ρ(0) ∈ Q0 ρ(i + 1) ∈ δ(ρ(i), α(i))

An automaton may have several runs on α.

q1 q2 q3 q4 q5 qs a b a, b Final state

α = abbbbb · · · , ρ1 = q1q2q2q2q2q2q2 · · · ρ2 = q1q1q2q2q2q2q2 · · ·

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-5
SLIDE 5

Runs and acceptance

A run of A on α ∈ Σω is a sequence ρ : N → Q such that

ρ(0) ∈ Q0 ρ(i + 1) ∈ δ(ρ(i), α(i))

An automaton may have several runs on α. ρ is accepting iff inf(ρ) ∩ F = ∅ α is accepted by A (α ∈ L(A)) iff there is an accepting run of A on α.

q1 q2 q3 q4 q5 qs a b a, b Final state

α = abbbbb · · · , ρ1 = q1q2q2q2q2q2q2 · · · ρ2 = q1q1q2q2q2q2q2 · · ·

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-6
SLIDE 6

Ambiguous automata

A is ambiguous if there exists α ∈ L(A) such that there are ≥ 2 accepting runs of A on α Otherwise, A is unambiguous.

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-7
SLIDE 7

Ambiguous automata

A is ambiguous if there exists α ∈ L(A) such that there are ≥ 2 accepting runs of A on α Otherwise, A is unambiguous. An ambiguous NBW

q1 q2 q3 q4 q5 qs a b a, b Final state

α = abω ρ1 = q1qω

2

ρ2 = q1q1qω

2

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-8
SLIDE 8

Strongly Unambiguous automata

Final run of A on α: A run ρ starting from any state in Q such that inf(ρ) ∩ F = ∅.

A word ∈ L(A) may have 0 or more final runs A word ∈ L(A) has ≥ 1 final runs

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-9
SLIDE 9

Strongly Unambiguous automata

Final run of A on α: A run ρ starting from any state in Q such that inf(ρ) ∩ F = ∅.

A word ∈ L(A) may have 0 or more final runs A word ∈ L(A) has ≥ 1 final runs

NBW A is strongly unambiguous if for every α ∈ Σω, there is exactly one final run.

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-10
SLIDE 10

Strongly Unambiguous automata

Final run of A on α: A run ρ starting from any state in Q such that inf(ρ) ∩ F = ∅.

A word ∈ L(A) may have 0 or more final runs A word ∈ L(A) has ≥ 1 final runs

NBW A is strongly unambiguous if for every α ∈ Σω, there is exactly one final run. Not all unambiguous automata are strongly unambiguous.

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-11
SLIDE 11

Strongly Unambiguous automata

Final run of A on α: A run ρ starting from any state in Q such that inf(ρ) ∩ F = ∅.

A word ∈ L(A) may have 0 or more final runs A word ∈ L(A) has ≥ 1 final runs

NBW A is strongly unambiguous if for every α ∈ Σω, there is exactly one final run. Not all unambiguous automata are strongly unambiguous.

a b a b Deterministic (hence unambiguous) but not strongly unambiguous a b a b Strongly unambiguous

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-12
SLIDE 12

Containment relations

UNBW: Unambiguous NBW, SUNBW: Strongly unambiguous NBW, DBW: Deterministic B¨ uchi automata over words Expressive power-wise DBW NBW ≡ UNBW ≡ SUNBW Automata structure-wise

DBW NBW UNBW SUNBW

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-13
SLIDE 13

What this talk is about

Given an NBW, construct UNBW accepting the same language and using as few states as possible.

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-14
SLIDE 14

What this talk is about

Given an NBW, construct UNBW accepting the same language and using as few states as possible. Relevant earlier work: Arnold 1983: UNBW expressively equivalent to NBW Carton & Michel 2003: Effective construction of SUNBW, size bound O((12n)n) K¨ ahler and Wilke 2008: Effective construction of UNBW, size bound O((3n)n). Bousquet and L¨

  • ding 2010: Equivalence and inclusion

problems for SUNBW are poly-time

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-15
SLIDE 15

What this talk is about

Given an NBW, construct UNBW accepting the same language and using as few states as possible. Relevant earlier work: Arnold 1983: UNBW expressively equivalent to NBW Carton & Michel 2003: Effective construction of SUNBW, size bound O((12n)n) K¨ ahler and Wilke 2008: Effective construction of UNBW, size bound O((3n)n). Bousquet and L¨

  • ding 2010: Equivalence and inclusion

problems for SUNBW are poly-time Our contribution Effective construction of UNBW, size bound O(n2.(0.76n)n)

Same as best known bound for NBW complementation!

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-16
SLIDE 16

Why care about disambiguation?

Of course, a theoretically interesting problem Can it lead to a better understanding of what kinds of NBW admit easy determinization? Practical application? Seek inputs from the audience.

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-17
SLIDE 17

Run DAGs

q1 q2 q3 q4 q5 qs a b a, b Final state

Run DAG for aω

q1 q1 q1 q1 q1 q3 q3 q3 q3 q3 q5 q5 q5 q2 q2 q2 q2 q2 q4 q4 q4 q4 Vertices with qs not shown for clarity Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-18
SLIDE 18

Ranking run DAGs

Intuitively, assign a metric to each vertex in run DAG such that the metric changes in a desirable way only along “good” runs.

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-19
SLIDE 19

Ranking run DAGs

Intuitively, assign a metric to each vertex in run DAG such that the metric changes in a desirable way only along “good” runs. Early work by Michel (1984?), Klarlund (1991): Ranking functions/progress measures for B¨ uchi complementation

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-20
SLIDE 20

Ranking run DAGs

Intuitively, assign a metric to each vertex in run DAG such that the metric changes in a desirable way only along “good” runs. Early work by Michel (1984?), Klarlund (1991): Ranking functions/progress measures for B¨ uchi complementation Recent spurt of work triggered by similar metrics defined by Kupferman & Vardi (2001 onwards)

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-21
SLIDE 21

Ranking run DAGs

Intuitively, assign a metric to each vertex in run DAG such that the metric changes in a desirable way only along “good” runs. Early work by Michel (1984?), Klarlund (1991): Ranking functions/progress measures for B¨ uchi complementation Recent spurt of work triggered by similar metrics defined by Kupferman & Vardi (2001 onwards)

Schewe (2009) used this approach to match upper bound of NBW complementation within O(n2) of lower bound

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-22
SLIDE 22

Ranking run DAGs

Intuitively, assign a metric to each vertex in run DAG such that the metric changes in a desirable way only along “good” runs. Early work by Michel (1984?), Klarlund (1991): Ranking functions/progress measures for B¨ uchi complementation Recent spurt of work triggered by similar metrics defined by Kupferman & Vardi (2001 onwards)

Schewe (2009) used this approach to match upper bound of NBW complementation within O(n2) of lower bound We use Kupferman-Vardi style rankings

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-23
SLIDE 23

Kupferman-Vardi style ranking

n: Number of states in NBW V : Set of run DAG vertices r : V → {1, 2, . . . 2n + 1}: Ranking function

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-24
SLIDE 24

Kupferman-Vardi style ranking

n: Number of states in NBW V : Set of run DAG vertices r : V → {1, 2, . . . 2n + 1}: Ranking function Constraints on ranks Vertices corresponding to final states must not get odd ranks

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-25
SLIDE 25

Kupferman-Vardi style ranking

n: Number of states in NBW V : Set of run DAG vertices r : V → {1, 2, . . . 2n + 1}: Ranking function Constraints on ranks Vertices corresponding to final states must not get odd ranks Ranking cannot increase along any path in run DAG

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-26
SLIDE 26

Kupferman-Vardi style ranking

n: Number of states in NBW V : Set of run DAG vertices r : V → {1, 2, . . . 2n + 1}: Ranking function Constraints on ranks Vertices corresponding to final states must not get odd ranks Ranking cannot increase along any path in run DAG Odd ranking: Every path eventually trapped in an odd rank Even ranking otherwise

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-27
SLIDE 27

Example of KV-ranking

q1 q2 q3 q4 q5 qs a b a, b Final state

Example KV-ranking of run DAG for aω

q1 q1 q1 q1 q1 q3 q3 q3 q3 q3 q5 q5 q5 q2 q2 q2 q2 q2 q4 q4 q4 q4 Vertices with qs not shown for clarity 3 2 1 4 4 2 2 3 5 6 3 3 6 6 4 4 4 6 6 6 4 4 Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-28
SLIDE 28

Ranking based complementation

Theorem (Kupferman-Vardi 2001) An ω-word α ∈ L(A) iff there is an odd ranking of the run DAG of A on α

q1 q2 q3 q4 q5 qs a b a, b Final state

Example ranking for baω

q1 q3 q3 q3 q3 q3 q5 q5 q5 q2 q4 q4 q4 q4 Vertices with qs not shown for clarity 3 1 1 4 4 2 2 3 5 6 3 3 6 6

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-29
SLIDE 29

Applications of Kupferman-Vardi’s theorem

Series of followup work on NBW complementation using KV-ranking

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-30
SLIDE 30

Applications of Kupferman-Vardi’s theorem

Series of followup work on NBW complementation using KV-ranking Schewe (2009) finally gave a construction yielding a complement NBW of size O(n2.(0.76n)n)

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-31
SLIDE 31

Applications of Kupferman-Vardi’s theorem

Series of followup work on NBW complementation using KV-ranking Schewe (2009) finally gave a construction yielding a complement NBW of size O(n2.(0.76n)n)

Lower bound Ω((0.76n)n).

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-32
SLIDE 32

Applications of Kupferman-Vardi’s theorem

Series of followup work on NBW complementation using KV-ranking Schewe (2009) finally gave a construction yielding a complement NBW of size O(n2.(0.76n)n)

Lower bound Ω((0.76n)n).

Several optimizations possible on basic construction

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-33
SLIDE 33

Applications of Kupferman-Vardi’s theorem

Series of followup work on NBW complementation using KV-ranking Schewe (2009) finally gave a construction yielding a complement NBW of size O(n2.(0.76n)n)

Lower bound Ω((0.76n)n).

Several optimizations possible on basic construction One such set of optimizations leads to an unambiguous complementation construction, and a disambiguation construction too!

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-34
SLIDE 34

Applications of Kupferman-Vardi’s theorem

Series of followup work on NBW complementation using KV-ranking Schewe (2009) finally gave a construction yielding a complement NBW of size O(n2.(0.76n)n)

Lower bound Ω((0.76n)n).

Several optimizations possible on basic construction One such set of optimizations leads to an unambiguous complementation construction, and a disambiguation construction too!

Achieves same bound of O(n2, (0.76n)n).

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-35
SLIDE 35

Extending KV-ranks

Recall KV-ranking

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-36
SLIDE 36

Extending KV-ranks

Recall KV-ranking n: Number of states in NBW V : Set of run DAG vertices r : V → {1, 2, . . . 2n + 1} ∪{∞} : Ranking function

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-37
SLIDE 37

Extending KV-ranks

Recall KV-ranking n: Number of states in NBW V : Set of run DAG vertices r : V → {1, 2, . . . 2n + 1} ∪{∞} : Ranking function Constraints on ranks Vertices corresponding to final states must not get odd ranks Ranking cannot increase along any path in run DAG

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-38
SLIDE 38

Extending KV-ranks

Recall KV-ranking n: Number of states in NBW V : Set of run DAG vertices r : V → {1, 2, . . . 2n + 1} ∪{∞} : Ranking function Constraints on ranks Vertices corresponding to final states must not get odd ranks Ranking cannot increase along any path in run DAG Every path eventually trapped in odd rank or in ∞

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-39
SLIDE 39

Extending KV-ranks

Recall KV-ranking n: Number of states in NBW V : Set of run DAG vertices r : V → {1, 2, . . . 2n + 1} ∪{∞} : Ranking function Constraints on ranks Vertices corresponding to final states must not get odd ranks Ranking cannot increase along any path in run DAG Every path eventually trapped in odd rank or in ∞ We call this a full ranking of the run DAG.

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-40
SLIDE 40

Example of full ranking

q1 q2 q3 q4 q5 qs a b a, b Final state

Example full ranking for aω

q1 q1 q1 q1 q1 q3 q3 q3 q3 q3 q5 q5 q5 q2 q2 q2 q2 q2 q4 q4 q4 q4 Vertices with qs not shown for clarity 3 3 1 4 4 2 2 3 5 6 3 3 6 4 4 ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-41
SLIDE 41

Minimal full rankings

Given run DAG G, full ranking r∗ of G is minimal iff for all full rankings r of G, r∗(v) ≤ r(v) for all vertices v in G. Non-minimal full ranking

q1 q1 q1 q1 q1 q3 q3 q3 q3 q3 q5 q5 q5 q2 q2 q2 q2 q2 q4 q4 q4 q4 Vertices with qs not shown for clarity 3 3 1 4 4 2 2 3 5 6 3 3 6 4 4 ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞

Minimal full ranking

q1 q1 q1 q1 q1 q3 q3 q3 q3 q3 q5 q5 q5 q2 q2 q2 q2 q2 q4 q4 q4 q4 Vertices with qs not shown for clarity 1 1 1 2 2 2 2 3 3 3 3 3 4 4 4 ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ 4 4 4 4

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-42
SLIDE 42

Properties of minimal full rankings

Theorem For every run DAG, there exists a unique minimal full ranking. A word α is accepted by A iff the minimal full ranking of the run DAG assigns ∞ as the rank of the root vertex.

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-43
SLIDE 43

Properties of minimal full rankings

Theorem For every run DAG, there exists a unique minimal full ranking. A word α is accepted by A iff the minimal full ranking of the run DAG assigns ∞ as the rank of the root vertex. F-vertex: Vertex in run DAG for which the state is final.

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-44
SLIDE 44

Properties of minimal full rankings

Theorem For every run DAG, there exists a unique minimal full ranking. A word α is accepted by A iff the minimal full ranking of the run DAG assigns ∞ as the rank of the root vertex. F-vertex: Vertex in run DAG for which the state is final. Local properties (successors)

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-45
SLIDE 45

Properties of minimal full rankings

Theorem For every run DAG, there exists a unique minimal full ranking. A word α is accepted by A iff the minimal full ranking of the run DAG assigns ∞ as the rank of the root vertex. F-vertex: Vertex in run DAG for which the state is final. Local properties (successors) Every vertex that is not a F-vertex has a successor with the same rank

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-46
SLIDE 46

Properties of minimal full rankings

Theorem For every run DAG, there exists a unique minimal full ranking. A word α is accepted by A iff the minimal full ranking of the run DAG assigns ∞ as the rank of the root vertex. F-vertex: Vertex in run DAG for which the state is final. Local properties (successors) Every vertex that is not a F-vertex has a successor with the same rank Every even ranked vertex either has a successor with the same rank or one with the next lower odd rank

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-47
SLIDE 47

Properties of minimal full rankings

Global properties (descendants)

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-48
SLIDE 48

Properties of minimal full rankings

Global properties (descendants) Every even ranked vertex has at least one descendant with the next lower odd rank

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-49
SLIDE 49

Properties of minimal full rankings

Global properties (descendants) Every even ranked vertex has at least one descendant with the next lower odd rank Every odd ranked (> 1) vertex has at least one F-vertex descendant with the next lower even rank

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-50
SLIDE 50

Properties of minimal full rankings

Global properties (descendants) Every even ranked vertex has at least one descendant with the next lower odd rank Every odd ranked (> 1) vertex has at least one F-vertex descendant with the next lower even rank Every path from every even ranked vertex eventually encounters a vertex with a lower rank

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-51
SLIDE 51

Properties of minimal full rankings

Global properties (descendants) Every even ranked vertex has at least one descendant with the next lower odd rank Every odd ranked (> 1) vertex has at least one F-vertex descendant with the next lower even rank Every path from every even ranked vertex eventually encounters a vertex with a lower rank Every ∞ ranked vertex has at least one ∞ ranked F-vertex descendant

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-52
SLIDE 52

Properties of minimal full rankings

Global properties (descendants) Every even ranked vertex has at least one descendant with the next lower odd rank Every odd ranked (> 1) vertex has at least one F-vertex descendant with the next lower even rank Every path from every even ranked vertex eventually encounters a vertex with a lower rank Every ∞ ranked vertex has at least one ∞ ranked F-vertex descendant Every ∞ ranked vertex has at least one descendant with the largest non-infinity rank in range of the ranking function.

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-53
SLIDE 53

Intuition of disambiguation construction

Level 3 Level 2 Level 1 Level 0 q1 q1 q1 q1 q1 q3 q3 q3 q3 q3 q5 q5 q5 q2 q2 q2 q2 q2 q4 q4 q4 q4 Vertices with qs not shown for clarity 1 1 1 2 2 2 2 3 3 3 3 3 4 4 4

∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ 4 4 4 4

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-54
SLIDE 54

Intuition of disambiguation construction

Level 3 Level 2 Level 1 Level 0 q1 q1 q1 q1 q1 q3 q3 q3 q3 q3 q5 q5 q5 q2 q2 q2 q2 q2 q4 q4 q4 q4 Vertices with qs not shown for clarity 1 1 1 2 2 2 2 3 3 3 3 3 4 4 4

∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ 4 4 4 4

Level 3 Level 2 Level 1 Level 0 q1 q1 q1 q1 q1 q3 q3 q3 q3 q3 q5 q5 q5 q2 q2 q2 q2 q2 q4 q4 q4 q4 Vertices with qs not shown for clarity 1 1 1 2 2 2 2 3 3 3 3 3 4 4 4

∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ ∞ 4 4 4 4

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-55
SLIDE 55

Disambiguation construction

Construct an automaton whose states are full-ranked levels of run DAG

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-56
SLIDE 56

Disambiguation construction

Construct an automaton whose states are full-ranked levels of run DAG

Goal: Only minimally full-ranked levels must be accepted

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-57
SLIDE 57

Disambiguation construction

Construct an automaton whose states are full-ranked levels of run DAG

Goal: Only minimally full-ranked levels must be accepted

Local properties of minimal full-ranking easy to enforce in transition relation

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-58
SLIDE 58

Disambiguation construction

Construct an automaton whose states are full-ranked levels of run DAG

Goal: Only minimally full-ranked levels must be accepted

Local properties of minimal full-ranking easy to enforce in transition relation Enforcing global properties requires maintaining additional book-keeping information

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-59
SLIDE 59

Disambiguation construction

Construct an automaton whose states are full-ranked levels of run DAG

Goal: Only minimally full-ranked levels must be accepted

Local properties of minimal full-ranking easy to enforce in transition relation Enforcing global properties requires maintaining additional book-keeping information

Global properties checked one vertex (and also one rank) at a time Decompose every global property of an infinite run into properties of finite segments of the run, which can then be concatenated. Ensure that each finite segment satisfies relevant property checkable over finite steps

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-60
SLIDE 60

Disambiguation construction

Construct an automaton whose states are full-ranked levels of run DAG

Goal: Only minimally full-ranked levels must be accepted

Local properties of minimal full-ranking easy to enforce in transition relation Enforcing global properties requires maintaining additional book-keeping information

Global properties checked one vertex (and also one rank) at a time Decompose every global property of an infinite run into properties of finite segments of the run, which can then be concatenated. Ensure that each finite segment satisfies relevant property checkable over finite steps

Acceptance condition simply ensures that every finite segment

  • f an infinite run satisfies relevant properties and root vertex

is ranked ∞

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-61
SLIDE 61

State representation

State of resulting automaton: (S, O, X, f , i), where S : subset of states of NBW in current level f : ranking function at current level i : rank of vertices for which (decomposed) global properties are currently being checked O ⊆ S : subset of states with rank i for which global properties yet to be checked X ⊆ S : subset of states being used to check global property

  • f one state with rank i

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-62
SLIDE 62

State representation

State of resulting automaton: (S, O, X, f , i), where S : subset of states of NBW in current level f : ranking function at current level i : rank of vertices for which (decomposed) global properties are currently being checked O ⊆ S : subset of states with rank i for which global properties yet to be checked X ⊆ S : subset of states being used to check global property

  • f one state with rank i

Total count of states is O(n2.(0.76n)n) Uses a modification of a counting argument used by Schewe (2009) for NBW complementation

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-63
SLIDE 63

Why is it unambiguous?

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-64
SLIDE 64

Why is it unambiguous?

Recall minimal full-ranking for every run DAG is unique.

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-65
SLIDE 65

Why is it unambiguous?

Recall minimal full-ranking for every run DAG is unique. Our construction accepts only those runs that enforce both local and global properties of minimal full-ranking

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-66
SLIDE 66

Why is it unambiguous?

Recall minimal full-ranking for every run DAG is unique. Our construction accepts only those runs that enforce both local and global properties of minimal full-ranking

Accepted full-ranking is minimal

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-67
SLIDE 67

Why is it unambiguous?

Recall minimal full-ranking for every run DAG is unique. Our construction accepts only those runs that enforce both local and global properties of minimal full-ranking

Accepted full-ranking is minimal

Any two accepting runs must differ in the ranking of at least

  • ne level

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-68
SLIDE 68

Why is it unambiguous?

Recall minimal full-ranking for every run DAG is unique. Our construction accepts only those runs that enforce both local and global properties of minimal full-ranking

Accepted full-ranking is minimal

Any two accepting runs must differ in the ranking of at least

  • ne level

Since minimal ranking is unique, only one accepting run possible

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata

slide-69
SLIDE 69

Conclusion

Using a variant of KV-ranking (similar to that used by Carton and Michel), we obtain a UNBW (not SUNBW) with better bound than reported in the literature We conjecture that this matches the lower bound for disambiguation Shows potential close connection between disambiguation and complementation

Hrishikesh Karmarkar Supratik Chakraborty Ranking based Techniques for Disambiguating B¨ uchi Automata