sqig - it, lisbon - portugal [email protected]/seminars/slides/baltazar.pdf ·...
TRANSCRIPT
![Page 1: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/1.jpg)
How to make a logic probabilistic?
Pedro Baltazar
SQIG - IT, Lisbon - Portugal
CMU, CMACS Seminar - January 14th, 2010
![Page 2: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/2.jpg)
Sources:
D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus,Probabilistic quantified linear temporal logic: Model checking,
SAT and complete Hilbert calculus.submitted for publication.
P. Baltazar and P. Mateus.Temporalization of probabilistic propositional logic.LFCS 2009, LNCS, 2009.
P. Baltazar, P. Mateus, R. Nagarajan, and N. Papanikolaou.Exogenous probabilistic computation tree logic.Electronic Notes in Theoretical Computer Science, 190(3) : 95–110,2007.
![Page 3: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/3.jpg)
CPS : Cyber-Physical Systems
@@
@@@
![Page 4: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/4.jpg)
CPS : Cyber-Physical Systems
'
&
$
%System_pc
• language; // syntax
- specification; // theory
or/and
- model(s) // semantics
@@
@@@
![Page 5: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/5.jpg)
CPS : Cyber-Physical Systems
'
&
$
%System_pc
• language; // syntax
- specification; // theory
or/and
- model(s) // semantics
@@
@@@
System_car · · ·
![Page 6: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/6.jpg)
CPS : Cyber-Physical Systems
'
&
$
%System_pc
• language; // syntax
- specification; // theory
or/and
- model(s) // semantics
@@
@@@
System_car · · ·
System_servers · · ·
![Page 7: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/7.jpg)
CPS : Cyber-Physical Systems
'
&
$
%System_pc
• language; // syntax
- specification; // theory
or/and
- model(s) // semantics
@@
@@@
System_car · · ·
System_servers · · ·
System_train · · ·
![Page 8: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/8.jpg)
CPS : Cyber-Physical Systems
'
&
$
%System_pc
• language; // syntax
- specification; // theory
or/and
- model(s) // semantics
property:
ϕ = “Always ( NOT car_train_crash )”
@@
@@@
System_car · · ·
System_servers · · ·
System_train · · ·
![Page 9: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/9.jpg)
CPS : Cyber-Physical Systems
'
&
$
%System_pc
• language; // syntax
- specification; // theory
or/and
- model(s) // semantics
property:
ϕ = “Always ( NOT car_train_crash )”
ϕ1
ϕ2
ϕ3
ϕ4
@@
@@@
System_car · · ·
System_servers · · ·
System_train · · ·
![Page 10: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/10.jpg)
CPS : Cyber-Physical Systems
'
&
$
%System_pc
• language; // syntax
- specification; // theory
or/and
- model(s) // semantics
property:
ϕ = “ALWAYS ( NOT car_train_crash )”
ϕ1
ϕ2
ϕ3
ϕ4
@@
@@@
System_car · · ·
System_servers · · ·
System_train · · ·
![Page 11: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/11.jpg)
CPS : Cyber-Physical Systems
'
&
$
%System_pc
• language; // syntax
- specification; // theory
or/and
- model(s) // semantics
?
YES or NO ϕ
@@
@@@
System_car · · ·
System_servers · · ·
System_train · · ·
![Page 12: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/12.jpg)
(some) Logics in Verification
non-probabilistic probabilistic
Propositional logic
Modal logic, CTL, LTL
First-order theories:
Presburger arithmeticPointer logic...
Separation logic
Duration calculus
Metric temporal logic
Differential dynamic logic...
PCTL and PCTL*
Continuous stochastic logic...
![Page 13: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/13.jpg)
Outline
1 Exogenous Combination of Logics
2 Probabilization of Logics:(generic) SATcompleteness
3 Examples:EPPL - Probabilistic propositional logicPTL - Probabilistic temporal logicCTPL - Temporal EPPL
![Page 14: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/14.jpg)
Exogenous Combination of Logics
Definition (Satisfaction system)
Let L be a set of formulas,M a class of models and ⊆M×La satisfaction relation.The tuple S = 〈L,M, 〉 is a satisfaction system.
![Page 15: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/15.jpg)
Exogenous Combination of Logics
Definition (Satisfaction system)
Let L be a set of formulas,M a class of models and ⊆M×La satisfaction relation.The tuple S = 〈L,M, 〉 is a satisfaction system.
Definition (Morphism and weak morphism)
A morphism h : S → S ′ is a pair 〈h, h〉, with
h : L → L′ and h :M′ → 2M
morphism: for all m ∈ h(m′), m ϕ iff m′ ′ h(ϕ)
![Page 16: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/16.jpg)
Exogenous Combination of Logics
Definition (Satisfaction system)
Let L be a set of formulas,M a class of models and ⊆M×La satisfaction relation.The tuple S = 〈L,M, 〉 is a satisfaction system.
Definition (Morphism and weak morphism)
A morphism h : S → S ′ is a pair 〈h, h〉, with
h : L → L′ and h :M′ → 2M
morphism: for all m ∈ h(m′), m ϕ iff m′ ′ h(ϕ)
weak morphism: exists m ∈ h(m′), m ϕ iff m′ ′ h(ϕ)
for all ϕ ∈ L and for all m′ ∈ Mhdef= m′ ∈ M′ : h(m′) 6= ∅.
![Page 17: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/17.jpg)
1 - Exogenous Combination of Logics
Definition ((Weak) equivalent systems)
S and S ′ are (resp. weak) equivalent if there are (resp. weak)total morphisms h : S → S ′ and h′ : S ′ → S such that
ϕ ′ h
′(h(ϕ)) and ψ h(h
′(ψ)), for ϕ ∈ L, ψ ∈ L′.
Denoted by
equivalent, S1 ≅S S2
weak equivalent, S1 ≅wS S2
![Page 18: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/18.jpg)
1 - Exogenous Combination of Logics
Definition ((Weak) equivalent systems)
S and S ′ are (resp. weak) equivalent if there are (resp. weak)total morphisms h : S → S ′ and h′ : S ′ → S such that
ϕ ′ h
′(h(ϕ)) and ψ h(h
′(ψ)), for ϕ ∈ L, ψ ∈ L′.
Denoted by
equivalent, S1 ≅S S2
weak equivalent, S1 ≅wS S2
Proposition ( 〈L,M1, 1〉 ≅S 〈L,M2, 2〉 )
Γ 1 ϕ iff Γ 2 ϕ.
Proposition ( 〈L,M1, 1〉 ≅wS 〈L,M2, 2〉 )
1 ϕ iff 2 ϕ.
![Page 19: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/19.jpg)
Exogenous Combination of Logics
Let h1 : S → S1 and h2 : S → S2 be morphisms.
S1
S
h1
OO
h2// S2
![Page 20: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/20.jpg)
Exogenous Combination of Logics
Let h1 : S → S1 and h2 : S → S2 be morphisms.
S1
S
h1
OO
h2// S2
Idea: S1 ⊗S2 = 〈L1 ⊗ L2,M′, ′〉, with M′ ⊆M1 ×M2
Example (Parametrization)
S(h1⇒h2) = 〈L1,M(h1⇒h2), 1〉,
where M(h1⇒h2) = m ∈ Mh1: h1(m) ⊆ h2(M2).
![Page 21: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/21.jpg)
2 - Exogenous Probabilization of Logics
Definition (probabilization + globalization)
The probabilization + globalization operator transforms〈L,M, 〉 into the system S (p+g) = 〈L(p+g),M(p+g), (p+g)〉:
L(p+g) is (with β ∈ L and r ∈ Alg(R))
t ::= r 8∫β 8 (t+ t) 8 (t.t)
ϕ ::= [β] 8 (t < t) 8 (∼ϕ) 8 (ϕ = ϕ);
![Page 22: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/22.jpg)
2 - Exogenous Probabilization of Logics
Definition (probabilization + globalization)
The probabilization + globalization operator transforms〈L,M, 〉 into the system S (p+g) = 〈L(p+g),M(p+g), (p+g)〉:
L(p+g) is (with β ∈ L and r ∈ Alg(R))
t ::= r 8∫β 8 (t+ t) 8 (t.t)
ϕ ::= [β] 8 (t < t) 8 (∼ϕ) 8 (ϕ = ϕ);
M(p+g) is the class of all m = 〈S,F ,P, V 〉, where 〈S,F ,P〉is a probability space, and V : S →M is a measurable
valuation, i.e. V −1[β]def= s ∈ S : V (s) β ∈ F ;
![Page 23: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/23.jpg)
2 - Exogenous Probabilization of Logics
Definition (probabilization + globalization)
The probabilization + globalization operator transforms〈L,M, 〉 into the system S (p+g) = 〈L(p+g),M(p+g), (p+g)〉:
L(p+g) is (with β ∈ L and r ∈ Alg(R))
t ::= r 8∫β 8 (t+ t) 8 (t.t)
ϕ ::= [β] 8 (t < t) 8 (∼ϕ) 8 (ϕ = ϕ);
M(p+g) is the class of all m = 〈S,F ,P, V 〉, where 〈S,F ,P〉is a probability space, and V : S →M is a measurable
valuation, i.e. V −1[β]def= s ∈ S : V (s) β ∈ F ;
the satisfaction relation (p+g) is given by
[[∫β]]m = P(V −1[β])
m (p+g) [β] iff V (S) β;
(. . . )
![Page 24: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/24.jpg)
2 - Exogenous Probabilization of Logics
weak morphism hp : S p → SRCF(xβ : β ∈ L ∪Xalg ∪X)
∆pS
- probabilistic (sub)theory of S in RCF
![Page 25: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/25.jpg)
2 - Exogenous Probabilization of Logics
weak morphism hp : S p → SRCF(xβ : β ∈ L ∪Xalg ∪X)
∆pS
- probabilistic (sub)theory of S in RCF
finite ∆Σϕ ⊆ LRCF, such that ∆p
SRCF ϕ iff ∆ϕ
Σ RCF ϕ
![Page 26: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/26.jpg)
2 - Exogenous Probabilization of Logics
weak morphism hp : S p → SRCF(xβ : β ∈ L ∪Xalg ∪X)
∆pS
- probabilistic (sub)theory of S in RCF
finite ∆Σϕ ⊆ LRCF, such that ∆p
SRCF ϕ iff ∆ϕ
Σ RCF ϕ
![Page 27: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/27.jpg)
2 - Exogenous Probabilization of Logics
weak morphism hp : S p → SRCF(xβ : β ∈ L ∪Xalg ∪X)
∆pS
- probabilistic (sub)theory of S in RCF
finite ∆Σϕ ⊆ LRCF, such that ∆p
SRCF ϕ iff ∆ϕ
Σ RCF ϕ
Proposition (Transference of SAT)
ϕ has a model in Mp iff hp(ϕ) ∧∆Σϕ has a model in RX .
![Page 28: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/28.jpg)
2 - Exogenous Probabilization of Logics
weak morphism hp : S p → SRCF(xβ : β ∈ L ∪Xalg ∪X)
∆pS
- probabilistic (sub)theory of S in RCF
finite ∆Σϕ ⊆ LRCF, such that ∆p
SRCF ϕ iff ∆ϕ
Σ RCF ϕ
Proposition (Transference of SAT)
ϕ has a model in Mp iff hp(ϕ) ∧∆Σϕ has a model in RX .
Theorem (SAT complexity lower-bound)
The SAT problem for S p is at least PSPACE and obtaining awitness is at least EXPSPACE.
Proposition (Transference of weak completeness)
The axiomatization AXpS
def= h−1
p (AXRCF +∆pS) is a sound and
weakly complete axiomatization for S p.
![Page 29: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/29.jpg)
2 - Exogenous Probabilization of Logics
Let ϕ ∈ L(p+g)
bf(ϕ) = β1, . . . , βk - base formulas in ϕ
![Page 30: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/30.jpg)
2 - Exogenous Probabilization of Logics
Let ϕ ∈ L(p+g)
bf(ϕ) = β1, . . . , βk - base formulas in ϕ
atb(ϕ) = (∧i∈Aβi)∧ (∧i 6∈A¬βi) : A ∈ 2k - atomic fml. for ϕ
![Page 31: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/31.jpg)
2 - Exogenous Probabilization of Logics
Let ϕ ∈ L(p+g)
bf(ϕ) = β1, . . . , βk - base formulas in ϕ
atb(ϕ) = (∧i∈Aβi)∧ (∧i 6∈A¬βi) : A ∈ 2k - atomic fml. for ϕ
Γϕ,N is the set of all β ∈ atb(ϕ) such that g (ϕ = [¬β])
![Page 32: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/32.jpg)
2 - Exogenous Probabilization of Logics
Let ϕ ∈ L(p+g)
bf(ϕ) = β1, . . . , βk - base formulas in ϕ
atb(ϕ) = (∧i∈Aβi)∧ (∧i 6∈A¬βi) : A ∈ 2k - atomic fml. for ϕ
Γϕ,N is the set of all β ∈ atb(ϕ) such that g (ϕ = [¬β])
let ψg = (⊓β∈Γϕ,N[¬β]) and ψp = (⊓β∈Γϕ,N
(∫β = 0))
![Page 33: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/33.jpg)
2 - Exogenous Probabilization of Logics
Let ϕ ∈ L(p+g)
bf(ϕ) = β1, . . . , βk - base formulas in ϕ
atb(ϕ) = (∧i∈Aβi)∧ (∧i 6∈A¬βi) : A ∈ 2k - atomic fml. for ϕ
Γϕ,N is the set of all β ∈ atb(ϕ) such that g (ϕ = [¬β])
let ψg = (⊓β∈Γϕ,N[¬β]) and ψp = (⊓β∈Γϕ,N
(∫β = 0))
![Page 34: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/34.jpg)
2 - Exogenous Probabilization of Logics
Let ϕ ∈ L(p+g)
bf(ϕ) = β1, . . . , βk - base formulas in ϕ
atb(ϕ) = (∧i∈Aβi)∧ (∧i 6∈A¬βi) : A ∈ 2k - atomic fml. for ϕ
Γϕ,N is the set of all β ∈ atb(ϕ) such that g (ϕ = [¬β])
let ψg = (⊓β∈Γϕ,N[¬β]) and ψp = (⊓β∈Γϕ,N
(∫β = 0))
Let ϕg ∈ Lg and ϕp ∈ Lp.
Proposition
A formula (ϕg ⊓ ϕp) is satisfiable iff ϕg and (ϕp ⊓ ψp) aresatisfiable.
![Page 35: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/35.jpg)
2 - Exogenous Probabilization of Logics
Let ϕ ∈ L(p+g)
bf(ϕ) = β1, . . . , βk - base formulas in ϕ
atb(ϕ) = (∧i∈Aβi)∧ (∧i 6∈A¬βi) : A ∈ 2k - atomic fml. for ϕ
Γϕ,N is the set of all β ∈ atb(ϕ) such that g (ϕ = [¬β])
let ψg = (⊓β∈Γϕ,N[¬β]) and ψp = (⊓β∈Γϕ,N
(∫β = 0))
Let ϕg ∈ Lg and ϕp ∈ Lp.
Proposition
A formula (ϕg ⊓ ϕp) is satisfiable iff ϕg and (ϕp ⊓ ψp) aresatisfiable.
Theorem (Transference of SAT)
If the SAT problem is solvable in S , then it is solvable in S (p+g).
![Page 36: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/36.jpg)
2 - Exogenous Probabilization of Logics
Schema axiom: IN ([β] = (∫β = 1))
![Page 37: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/37.jpg)
2 - Exogenous Probabilization of Logics
Schema axiom: IN ([β] = (∫β = 1))
Theorem (Transference of weak completeness)
If S has a weakly complete axiomatization AXS , then
AX(p+g)S
def= AX
pS
+ AXgS
+ IN
is a weakly complete for S (p+g).
Theorem (small-model theorem)
Every ϕ satisfiable has a model (probability dist.) of 2× size(ϕ).
Theorem (SAT complexity lower-bound)
The SAT problem for S (p+g) is at least PSPACE and obtaining awitness is at least EXPSPACE.
![Page 38: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/38.jpg)
2 - Exogenous Probabilization of Logics
Algorithm 1: Sat(p+g)S
(ϕ)
Input: formula ϕ ∈ L(p+g)
Output: m = 〈M,P〉 (m (p+g) ϕ) or ∅ (No Model)
1 foreach ϕi = (ϕi,g ⊓ ϕi,p) molecule of ϕ do
2 foreach Γ ⊆ atb(ϕ) of size ≤ 2× Size(ϕ) do
3 M = ∅;4 foreach β ∈ Γ do
5 mβ ←− SatS (β); M =M ∪ mβ;6 end
7 if M 6= ∅ and M g ϕi,g then
8 φ←− hp(ϕi,p ⊓ ψi,p);
9 δ ←− φ ∧∆Σφ (Γ);
10 η ←− SatRCF(δ);11 if η 6= ∅ then return m = 〈M,Pη〉;
12 end
13 end
14 end
15 return ∅ (No Model);
![Page 39: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/39.jpg)
EPPL - Probabilistic propositional logic
Let Λ be a countable set of propositional symbols.
Definition (EPPL)
SEPPL(Λ) = 〈LEPPL(Λ),MEPPL, EPPL〉:
set of formulas LEPPL(Λ) is
β ::= α 8 (¬β) 8 (β⇒ β)
t ::= r 8∫β 8 (t+ t) 8 (t.t)
ϕ ::= [β] 8 (t < t) 8 (∼ϕ) 8 (ϕ = ϕ)
with α ∈ Λ and r ∈ Alg(R);
Let Xα : Ω→ 2α∈Λ be a stochastic process over 〈Ω,F ,P〉.
X(¬β) = 1−Xβ ;
X(β1⇒β2) = max1−Xβ1,Xβ2
.
![Page 40: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/40.jpg)
EPPL - Semantics
Definition (EPPL (cont.))
the class of modelsMEPPL are the tuples m = 〈S,F ,P,X〉such that X := Xα : S → 2α∈Λ is a stochastic process over〈S,F ,P〉;
the satisfaction relation EPPL is defined by:
[[r]]m = r;[[∫β]]m = P(Xβ = 1)
[[t1 + t2]]m = [[t1]]m + [[t2]]m;[[t1.t2]]m = [[t1]]m.[[t2]]m;
m EPPL [β] iff Xβ(s) = 1 for all s ∈ S;m EPPL (t1 < t2) iff [[t1]]m < [[t2]]m;m EPPL (∼ϕ) iff m 6 EPPL ϕ;m EPPL (ϕ1 = ϕ2) iff m 6 EPPL ϕ1 or m EPPL ϕ2,
for m ∈ MEPPL and ϕ ∈ LEPPL(Λ).
![Page 41: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/41.jpg)
title
Theorem (equivalence)
SEPPL(Λ) ≅S S(p+g)CPL (Λ).
Corollary (weak completeness)
The axiomatization AX(p+g)CPL is weakly complete and sound for the
satisfaction system SEPPL(Λ).
Theorem (SAT complexity)
The SAT problem for EPPL is PSPACE, and providing a witness (amodel) is EXPSPACE.
Theorem (model-checking complexity)
It takes O(|ϕ| × |S|) time to decide if an EPPL modelm = 〈S,P,X〉 satisfies ϕ.
![Page 42: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/42.jpg)
EPPL - SAT
Algorithm 2: SAT (ϕ)
Input: formula ϕ ∈ L(p+g)(Λ)
Output: m = 〈M,P〉 (m (p+g)CPL ϕ) or ∅ (No Model)
1 foreach ϕi = (ϕi,g ⊓ ϕi,p) molecule of ϕ do
2 foreach M ⊆ 2Λ(ϕ) of size(M) ≤ 2× Size(ϕi) do
3 if M g ϕi,g then
4 φ←− hp(ϕi,p ⊓ ψi,p);5 ψ ←− φ ∧∆Σ
φ (M);
6 η ←− SatRCF(ψ);7 if η 6= ∅ then return m = 〈M,Pη〉;
8 end
9 end
10 end
11 return ∅ (No Model);
![Page 43: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/43.jpg)
EPPL - Axiomatization
AXEPPL is
G1 ⊢EPPL [β] for all valid β ∈ LCPL(Λ);
G2 ⊢EPPL ([β1⇒ β2] = ([β1] = [β2]));
IN ⊢EPPL ([β] = (∫β = 1)) ;
EqN ⊢EPPL (∫¬β = 1−
∫β);
EqP ⊢EPPL (∫β ≥ 0) ;
EqA ⊢EPPL (∫(β1 ∨ β2) =
∫β1 +
∫β2 −
∫(β1 ∧ β2));
RCF ⊢EPPL ϕ
if hp(ϕ) ∧ (∧r∈alg(ϕ)ϕr(xr)) is a valid formula in the realclosed fields - RCF;
MP ϕ1, (ϕ1 = ϕ2) ⊢EPPL ϕ2.
![Page 44: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/44.jpg)
EPPL - Application: Faulty Hardware
α1α2 α4
α3
α5α6
Figure: AND-OR-INVERTER (AOI21)
![Page 45: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/45.jpg)
EPPL - Application: Faulty Hardware
α1α2 α4
α3
α5α6
Figure: AND-OR-INVERTER (AOI21)
implementation:
(∫(α4⇔ α1 ∧ α2) > 0.97)⊓(
∫(α5⇔ α3 ∨ α4) > 0.99)⊓[(α6⇔¬α5)]
![Page 46: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/46.jpg)
EPPL - Application: Faulty Hardware
α1α2 α4
α3
α5α6
Figure: AND-OR-INVERTER (AOI21)
implementation:
(∫(α4⇔ α1 ∧ α2) > 0.97)⊓(
∫(α5⇔ α3 ∨ α4) > 0.99)⊓[(α6⇔¬α5)]
specification:
(∫α6⇔¬(α3 ∨ (α1 ∧ α2)) ≥ 0.98)
![Page 47: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/47.jpg)
EPPL - Application: Boolean Probabilistic Programs
1) x = rand();
2) y = rand();
3) y = x ∨ y;
4) if (x)
5) x = ¬ x;
6) else
7) x = x ∨ y;
ϕP = (∫αx1 = 0.5) ⊓ (
∫αy1 = 0.5)⊓
⊓[αy2⇔ αx1 ∨ αy1] ⊓ [αx3⇔¬αx2]⊓
⊓[αx4⇔ (αx2 ∨ αy2)]⊓
⊓[αx5⇔ (αx2?αx3 : αx4)]
Table: Translation to EPPL formula
ϕsaf = ((∫αx1 ≤ 0.5) ⊓ (
∫αx2 ≤ 0.5) ⊓ . . . ⊓ (
∫αx5 ≤ 0.5))
![Page 48: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/48.jpg)
EPPL - Application: Boolean Probabilistic Programs
1) x = rand();
2) y = rand();
3) y = x ∨ y;
4) if (x)
5) x = ¬ x;
6) else
7) x = x ∨ y;
ϕP = (∫αx1 = 0.5) ⊓ (
∫αy1 = 0.5)⊓
⊓[αy2⇔ αx1 ∨ αy1] ⊓ [αx3⇔¬αx2]⊓
⊓[αx4⇔ (αx2 ∨ αy2)]⊓
⊓[αx5⇔ (αx2?αx3 : αx4)]
Table: Translation to EPPL formula
ϕsaf = ((∫αx1 ≤ 0.5) ⊓ (
∫αx2 ≤ 0.5) ⊓ . . . ⊓ (
∫αx5 ≤ 0.5))
SAT ((ϕP ⊓ ∼ϕsaf )) =?
![Page 49: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/49.jpg)
PTL- Probabilistic LTL
Let Λ be a countable set of propositional symbols.
Definition (PTL)
The probabilistic temporal logic (PTL) over Λ, is the systemSPTL(Λ) = 〈LPTL(Λ),MPTL, PTL〉 where LPTL(Λ) is
β ::= α 8 (¬β) 8 (β⇒ β) 8 (Xβ) 8 (βUβ)
t ::= r 8 (∫β) 8 (t+ t) 8 (t.t)
ϕ ::= [β] 8 (t ≤ t) 8 (∼ϕ) 8 (ϕ = ϕ)
with α ∈ Λ, and r ∈ alg(R);
Xα : S → 2α∈Λ is extended to a stochastic process over〈Sω,F ,P〉 (sequence space of a Markov chain).
X(Xβ)(π) = Xβ(π(1))
X(β1Uβ2)(π) = Xβ2(π) +X(¬β2)(π).Xβ1
(π).X(β1Uβ2)(π(1))
![Page 50: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/50.jpg)
PTL- Semantics
Definition (PTL (cont.))
MPTL is the class of tuples m = 〈S,P, µ, V 〉 where 〈S,P, µ〉 isa Markov chain and V : S → 2Λ;
PTL is defined by
[[r]]m = r;[[∫β]]m = P(Xβ = 1);
[[t1 + t2]]m = [[t1]]m + [[t2]]m;[[t1.t2]]m = [[t1]]m.[[t2]]m;
m PTL [β] iff Km LTL β;m PTL (t1 < t2) iff [[t1]]m < [[t2]]m;m PTL (∼ϕ) iff m 6 PTL ϕ;m PTL (ϕ1 = ϕ2) iff m 6 PTL ϕ1 or m PTL ϕ2,
for m ∈ MPTL and ϕ ∈ LPTL(Λ).
![Page 51: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/51.jpg)
PTL- SAT
Proposition (Exogenous weak equivalent)
SPTL(Λ) ≅wS S
(p+g)LTL (Λ).
Corollary (Transference of weak completeness)
The axiomatization
AX(p+g)LTL
def= AX
gLTL + AX
pLTL + IN
is a sound and weakly complete axiomatization for SPTL(Λ).
Theorem (Transference of SAT)
The SAT problem for PTL is PSPACE and obtaining a witness(model) is EXPSPACE.
![Page 52: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/52.jpg)
Temporal EPPL
Definition (CTPL)
Consider the system
SCTPL(Λ) = 〈LCTPL(Λ),MCTPL, CTPL〉,
LCTPL(Λ) is
ϕ := β 8 (¬ϕ) 8 (ϕ⇒ ϕ) 8 (AXϕ) 8 (A(ϕUϕ)) 8 (AGϕ)
with β ∈ LEPPL(Λ);
MCTPL is the class of tuples m = 〈S,R, V : S →MEPPL〉,where 〈S,R〉 is a Kripke frame;
CTPL is defined by
m, s CTPL β iff V (s) EPPL β;... (as in CTL)
![Page 53: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/53.jpg)
Temporal EPPL
SCTL(Λ′)
SCPL(Λ′)
h1
OO
h2
// SEPPL(Λ)
Proposition (Equivalence)
S(h1⇒h2) ≅S SCTPL(Λ).
Theorem (Transference of weak completeness)
The axiomatization AXCTL + h1(h−12 (AXEPPL)) is weakly complete
and sound for SCTPL(Λ).
Theorem (SAT complexity)
The satisfaction problem for CTPL is 2EXPTIME.
![Page 54: SQIG - IT, Lisbon - Portugal pedro.baltazar@ist.utlcmacs.cs.cmu.edu/seminars/slides/baltazar.pdf · Sources: D. Henriques, M. Biscaia, P. Baltazar, and P. Mateus, Probabilistic quantified](https://reader034.vdocuments.net/reader034/viewer/2022050507/5f98d9067a6f683232427da6/html5/thumbnails/54.jpg)
Future work
Future Work:
study exogenous combination as a generic tool toanalyze heterogeneous systems (cyber-physical systems):
automatic methods to combine systems;generalize Nelson-Oppen combination procedure;reuse of SAT and model-checking procedures (tools).
investigate Craig’s interpolation on probabilistic logics;
developed non-Hilbert calculus for probabilistic logics(to applied in verification by rewriting)