Algebra and probability in Lukasiewicz...
Transcript of Algebra and probability in Lukasiewicz...
Algebra and probability in Lukasiewicz logic
Ioana Leustean
Faculty of Mathematics and Computer ScienceUniversity of Bucharest
Probability, Uncertainty and RationalityCertosa di Pontignano (Siena), 1-3 November, 2009
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 1 / 32
LAP
Logic
Algebra
Probability
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 2 / 32
LAP
Logic
Algebra
Probability
lap = movement once arround a course
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 2 / 32
LAP
Logic
Algebra
Probability
lap = movement once arround a course
Classical logic
Boolean algebras
Classical probability theory: the set of events is a Boolean algebra
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 2 / 32
LAP
Logic
Algebra
Probability
lap = movement once arround a course
Classical logic
Boolean algebras
Classical probability theory: the set of events is a Boolean algebra
LAP interaction in Lukasiewicz logic
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 2 / 32
Lukasiewicz logic
3-valued (2-valued) Lukasiewicz logic classical logic
l l0, 1
2 , 1 0, 1
‖ ‖
L3 L2
J. Lukasiewicz, 1920.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 3 / 32
Lukasiewicz logic
n-valued (2-valued) Lukasiewicz logic classical logic
l l0, 1
n−1 ,2
n−1 , . . . ,n−2n−1 , 1 0, 1
‖ ‖
Ln L2
J. Lukasiewicz, 1920.
J. Lukasiewicz, 1929.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 3 / 32
Lukasiewicz logic
∞-valued n-valued (2-valued) Lukasiewicz logic Lukasiewicz logic classical logic
l l l[0, 1] 0, 1
n−1 ,2
n−1 , . . . ,n−2n−1 , 1 0, 1
‖ ‖
Ln L2
J. Lukasiewicz, 1920.
J. Lukasiewicz, 1929.
J. Lukasiewicz, A. Tarski, 1930.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 3 / 32
(∞-valued) Lukasiewicz logic:
Connectives
¬ and −→
• Lukasiewicz considered also Mp for ”p is possible”• Tarski defined Mp = ¬p −→ p
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 4 / 32
(∞-valued) Lukasiewicz logic:
Connectives
¬ and −→
• Lukasiewicz considered also Mp for ”p is possible”• Tarski defined Mp = ¬p −→ p
”truth-tables”
¬p := 1− p, p −→ q := min(1− p + q, 1) (p, q ∈ [0, 1] )
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 4 / 32
(∞-valued) Lukasiewicz logic:
Connectives
¬ and −→
• Lukasiewicz considered also Mp for ”p is possible”• Tarski defined Mp = ¬p −→ p
”truth-tables”
¬p := 1− p, p −→ q := min(1− p + q, 1) (p, q ∈ [0, 1] )
Lukasiewicz logic is truth-functional
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 4 / 32
(∞-valued) Lukasiewicz logic L:
Axioms
(L1) ϕ −→ (ψ −→ ϕ);(L2) (ϕ −→ ψ) −→ ((ψ −→ χ) −→ (ϕ −→ χ));(L3) ((ϕ −→ ψ) −→ ψ) −→ ((ψ −→ ϕ) −→ ϕ);(L4) (¬ψ −→ ¬ϕ) −→ (ϕ −→ ψ).
the deduction rule is modus ponens:ϕ,ϕ −→ ψ ⊢ ψ
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 5 / 32
(∞-valued) Lukasiewicz logic L:
Axioms
(L1) ϕ −→ (ψ −→ ϕ);(L2) (ϕ −→ ψ) −→ ((ψ −→ χ) −→ (ϕ −→ χ));(L3) ((ϕ −→ ψ) −→ ψ) −→ ((ψ −→ ϕ) −→ ϕ);(L4) (¬ψ −→ ¬ϕ) −→ (ϕ −→ ψ).
the deduction rule is modus ponens:ϕ,ϕ −→ ψ ⊢ ψ
L + ((ϕ −→ ¬ϕ) −→ ¬ϕ) ⇒ classical logic
L + An + Ak | k ∈ 2, . . . , (n − 2), k 6| (n − 1) ⇒ n-valued logic
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 5 / 32
(∞-valued) Lukasiewicz logic L:
McNaughton Theorem, 1951 (formulas as functions)
If f : [0, 1]n → [0, 1], TFAE:(a) f = ϕ[0,1] for some formula ϕ of Lukasiewicz logic,(b) f is continuous such that
∃q1, . . ., qk : Rn → R ∀(a1, . . . , an) ∈ [0, 1]n ∃i ∈ 1, kf (a1, . . . , an) = qi (a1, . . . , an),
where qi (a1, . . . , an) = m0i + m1ia1 + · · ·+ mnian, mji ∈ Z ∀i , j .
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 6 / 32
(∞-valued) Lukasiewicz logic L:
McNaughton Theorem, 1951 (formulas as functions)
If f : [0, 1]n → [0, 1], TFAE:(a) f = ϕ[0,1] for some formula ϕ of Lukasiewicz logic,(b) f is continuous such that
∃q1, . . ., qk : Rn → R ∀(a1, . . . , an) ∈ [0, 1]n ∃i ∈ 1, kf (a1, . . . , an) = qi (a1, . . . , an),
where qi (a1, . . . , an) = m0i + m1ia1 + · · ·+ mnian, mji ∈ Z ∀i , j .
Normal form representation theorem
A. Di Nola, A.Lettieri, 2004
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 6 / 32
(∞-valued) Lukasiewicz logic L:
McNaughton Theorem, 1951 (formulas as functions)
If f : [0, 1]n → [0, 1], TFAE:(a) f = ϕ[0,1] for some formula ϕ of Lukasiewicz logic,(b) f is continuous such that
∃q1, . . ., qk : Rn → R ∀(a1, . . . , an) ∈ [0, 1]n ∃i ∈ 1, kf (a1, . . . , an) = qi (a1, . . . , an),
where qi (a1, . . . , an) = m0i + m1ia1 + · · ·+ mnian, mji ∈ Z ∀i , j .
Normal form representation theorem
A. Di Nola, A.Lettieri, 2004
Rose and Rosser (1958)
A formula ϕ is a [0, 1]-tautology of L iff it can be derived from the axiomsusing modus ponens and substitution.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 6 / 32
Algebra: historical overview
Gr.C. Moisil: 3-valued and 4-valued Lukasiewicz algebras (1940)n-valued Lukasiewicz algebras (1941)
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 7 / 32
Algebra: historical overview
Gr.C. Moisil: 3-valued and 4-valued Lukasiewicz algebras (1940)n-valued Lukasiewicz algebras (1941)
A. Rose (1965): for n ≥ 5 the n-valued Lukasiewicz algebraKn = 0, 1
n−1 ,n−2n−1 , 1 is not closed to Lukasiewicz implication
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 7 / 32
Algebra: historical overview
Gr.C. Moisil: 3-valued and 4-valued Lukasiewicz algebras (1940)n-valued Lukasiewicz algebras (1941)
A. Rose (1965): for n ≥ 5 the n-valued Lukasiewicz algebraKn = 0, 1
n−1 ,n−2n−1 , 1 is not closed to Lukasiewicz implication
Gr.C. Moisil: ϑ-valued Lukasiewicz algebras (1968)
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 7 / 32
Algebra: historical overview
Gr.C. Moisil: 3-valued and 4-valued Lukasiewicz algebras (1940)n-valued Lukasiewicz algebras (1941)
A. Rose (1965): for n ≥ 5 the n-valued Lukasiewicz algebraKn = 0, 1
n−1 ,n−2n−1 , 1 is not closed to Lukasiewicz implication
Gr.C. Moisil: ϑ-valued Lukasiewicz algebras (1968)
Argentininan school: A. Monteiro, R.Cignoli, L. Monteiro, M.Abad, etc.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
C.C. Chang: MV-algebras (1958)
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 7 / 32
Algebra: historical overview
Gr.C. Moisil: 3-valued and 4-valued Lukasiewicz algebras (1940)n-valued Lukasiewicz algebras (1941)
A. Rose (1965): for n ≥ 5 the n-valued Lukasiewicz algebraKn = 0, 1
n−1 ,n−2n−1 , 1 is not closed to Lukasiewicz implication
Gr.C. Moisil: ϑ-valued Lukasiewicz algebras (1968)
Argentininan school: A. Monteiro, R.Cignoli, L. Monteiro, M.Abad, etc.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
C.C. Chang: MV-algebras (1958)
R. Grigolia: MVn-algebras (1977)
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 7 / 32
Algebra: historical overview
Gr.C. Moisil: 3-valued and 4-valued Lukasiewicz algebras (1940)n-valued Lukasiewicz algebras (1941)
A. Rose (1965): for n ≥ 5 the n-valued Lukasiewicz algebraKn = 0, 1
n−1 ,n−2n−1 , 1 is not closed to Lukasiewicz implication
Gr.C. Moisil: ϑ-valued Lukasiewicz algebras (1968)
Argentininan school: A. Monteiro, R.Cignoli, L. Monteiro, M.Abad, etc.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
C.C. Chang: MV-algebras (1958)
R. Grigolia: MVn-algebras (1977)
R. Cignoli: proper n-valued Lukasiewicz algebras (1982)
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 7 / 32
Algebra: historical overview
Gr.C. Moisil: 3-valued and 4-valued Lukasiewicz algebras (1940)n-valued Lukasiewicz algebras (1941)
A. Rose (1965): for n ≥ 5 the n-valued Lukasiewicz algebraKn = 0, 1
n−1 ,n−2n−1 , 1 is not closed to Lukasiewicz implication
Gr.C. Moisil: ϑ-valued Lukasiewicz algebras (1968)
Argentininan school: A. Monteiro, R.Cignoli, L. Monteiro, M.Abad, etc.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
C.C. Chang: MV-algebras (1958)
R. Grigolia: MVn-algebras (1977)
R. Cignoli: proper n-valued Lukasiewicz algebras (1982)
J.M. Font, A.J. Rodriguez, A. Torrens: Wajsberg algebras (1984)
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 7 / 32
Algebra: historical overview
Gr.C. Moisil: 3-valued and 4-valued Lukasiewicz algebras (1940)n-valued Lukasiewicz algebras (1941)
A. Rose (1965): for n ≥ 5 the n-valued Lukasiewicz algebraKn = 0, 1
n−1 ,n−2n−1 , 1 is not closed to Lukasiewicz implication
Gr.C. Moisil: ϑ-valued Lukasiewicz algebras (1968)
Argentininan school: A. Monteiro, R.Cignoli, L. Monteiro, M.Abad, etc.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
C.C. Chang: MV-algebras (1958)
R. Grigolia: MVn-algebras (1977)
R. Cignoli: proper n-valued Lukasiewicz algebras (1982)
J.M. Font, A.J. Rodriguez, A. Torrens: Wajsberg algebras (1984)
MV-algebras and Wajsberg algebrasMVn-algebras and proper n-valued Lukasiewicz algebras
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 7 / 32
Algebra: historical overview
Gr.C. Moisil: 3-valued and 4-valued Lukasiewicz algebras (1940)n-valued Lukasiewicz algebras (1941)
A. Rose (1965): for n ≥ 5 the n-valued Lukasiewicz algebraKn = 0, 1
n−1 ,n−2n−1 , 1 is not closed to Lukasiewicz implication
Gr.C. Moisil: ϑ-valued Lukasiewicz algebras (1968)
Argentininan school: A. Monteiro, R.Cignoli, L. Monteiro, M.Abad, etc.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
C.C. Chang: MV-algebras (1958)
R. Grigolia: MVn-algebras (1977)
R. Cignoli: proper n-valued Lukasiewicz algebras (1982)
J.M. Font, A.J. Rodriguez, A. Torrens: Wajsberg algebras (1984)
MV-algebras and Wajsberg algebrasMVn-algebras and proper n-valued Lukasiewicz algebras
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 7 / 32
the n-valued case
Boolean logicւ ց
Lukasiewiczn-valued logic
Postn-valued logic
ւ ց ↓ L-proper LMn-algebras ≃ MVn-algebras ⊃ Postn-algebras(Cignoli,1982) (Grigolia,1977) (Rosenbloom,1942)
∩LMn-algebras
(Moisil,1941)
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 8 / 32
n-valued Lukasiewicz-Moisil algebras
LMn-algebra
(L,∨,∧,∗ , ϕ1, . . . , ϕn−1, 1)(LM0)(L,∨,∧,∗ ) De Morgan algebra (LM1) ϕi (ϕj(x)) = ϕj(x)(LM2) ϕi (x ∨ y) = ϕi (x) ∨ ϕi (y) (LM3) ϕi (x) ∨ (ϕi (x))∗ = 1(LM4) ϕi (x∗) = (ϕn−i (x))∗ (LM5) ϕ1(x) ≤ · · · ≤ ϕn−1(x)
(LM6) ϕi (x) = ϕi (y) for any i ∈ 1, n − 1⇒ x = y
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 9 / 32
n-valued Lukasiewicz-Moisil algebras
LMn-algebra
(L,∨,∧,∗ , ϕ1, . . . , ϕn−1, 1)(LM0)(L,∨,∧,∗ ) De Morgan algebra (LM1) ϕi (ϕj(x)) = ϕj(x)(LM2) ϕi (x ∨ y) = ϕi (x) ∨ ϕi (y) (LM3) ϕi (x) ∨ (ϕi (x))∗ = 1(LM4) ϕi (x∗) = (ϕn−i (x))∗ (LM5) ϕ1(x) ≤ · · · ≤ ϕn−1(x)
(LM6) ϕi (x) = ϕi (y) for any i ∈ 1, n − 1⇒ x = y
L-proper LMn-algebra = LMn-algebra + Fiki ,k + axioms
Postn-algebra = LMn-algebra +cii=1,n−2+ axioms
LMn-algebras, MVn-algebras, Postn-algebras are equational classes.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 9 / 32
Examples:
Example 1:
0 1n−1
2n−1 . . . n−2
n−1 1
ϕ1 0 0 0 . . . 0 1ϕ2 0 0 0 . . . 1 1...
......
......
......
ϕn−2 0 0 1 . . . 1 1ϕn−1 0 1 1 . . . 1 1
Example 2:
L = LXn = f | f : X −→ Ln
ϕi (f )(x) =
1, f (x) ≥ n−in−1
0, otherwise
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 10 / 32
Moisil’s determination principle
L is LMn-algebra, MVn-algebra or Pn-algebraL ⊃ B(L) = x | x ∨ x∗ = 1 (the Boolean reduct)
L ∋ x 7→ ϕ1(x),. . .,ϕn−1(x) ∈ B(L)x = y iff ϕi (x) = ϕi (y) for any i ∈ 1, n − 1
Any element is characterized by (n − 1) Boolean nuances.
The functor B : LukMoisilm → Bool has a right adjoint.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 11 / 32
Moisil’s determination principle
L is LMn-algebra, MVn-algebra or Pn-algebraL ⊃ B(L) = x | x ∨ x∗ = 1 (the Boolean reduct)
L ∋ x 7→ ϕ1(x),. . .,ϕn−1(x) ∈ B(L)x = y iff ϕi (x) = ϕi (y) for any i ∈ 1, n − 1
Any element is characterized by (n − 1) Boolean nuances.
The functor B : LukMoisilm → Bool has a right adjoint.
Any element of L can be ”recovered” from its Boolean nuancesiff L is a Postn-algebra
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 11 / 32
Determination principle for subalgebras
Is it true that for S , T ⊆ A
ϕi (S) = ϕi (T ) for any i ∈ 1, n − 1⇒ S = T ?
Yes, if S and T are proper ideals.Not in general, if S and T are subalgebras.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 12 / 32
Determination principle for subalgebras
I.L.,2008.
L - LMn-algebra
S ≤ LI→ J1(L), . . ., Jn−2(L), Jn−1(L) = B(S)
x ∈ L : Ji (x) ∈ Ii∀iS← I1, . . ., In−2, In−1 ≤ B(L)
I injective, S surjective
I(S) = I(T )⇒ S = T
S is L-proper (MVn-algebra) ⇔ I(S) satisfies (MV) (n ≥ 5)(MV): Ii ∩ Ik ⊆ In−i+k−1, 3 ≤ i ≤ n − 2, 1 ≤ k ≤ n − 4, k < i
S is Pn-algebra ⇔ I1 = · · · = In−1 = B(S)
The Boolean nuances of a subalgebra are (n − 1) Boolean ideals
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 12 / 32
Nuances of truth vs truth degrees
G. Georgescu, A. Popescu, 2006
Moisil logic is derived from the classical logic by the idea of nuancing,mathematically expressed by a categorical adjunction
Starting from a logical system and using the idea of nuance, it ispossible to construct an n-nuanced logical system on the top of thegiven one.
Nuancing the Lukasiewicz logic, they defined the n-nuanced
MV-algebras and they proved that there is a pair of adjoint functorsbetween the two categories.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 13 / 32
MV-algebras ↔ Lukasiewicz ∞-valued logic
C.C.Chang, 1958
An MV-algebra is a structure (A,⊕,∗ , 0) such that:
1 (A,⊕, 0) abelian monoid,
2 (x∗)∗ = x ,
3 (x∗ ⊕ y)∗ ⊕ y = (y∗ ⊕ x)∗ ⊕ x ,
4 0∗ ⊕ x = 0∗.
([0, 1],⊕,∗ , 0) MV-algebra, x ⊕ y = min(x + y , 0), x∗ = 1− x
MV-algebras are bounded distributive lattices with1 = 0∗, x ∨ y = (y∗ ⊕ x)∗ ⊕ x , x ∧ y = (x∗ ∨ y)∗
MV-algebras are reziduated lattices withx −→ y = (x∗ ⊕ y), x ⊙ y = (x∗ ⊕ y∗)∗.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 14 / 32
MV-algebras ↔ Lukasiewicz ∞-valued logic
C.C.Chang, 1958
An MV-algebra is a structure (A,⊕,∗ , 0) such that:
1 (A,⊕, 0) abelian monoid,
2 (x∗)∗ = x ,
3 (x∗ ⊕ y)∗ ⊕ y = (y∗ ⊕ x)∗ ⊕ x ,
4 0∗ ⊕ x = 0∗.
R. Cignoli, I.M.L. D’Ottaviano, D. Mundici,Algebraic foundations of many-valued reasoning, 2000.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 14 / 32
MV-algebras ↔ Lukasiewicz ∞-valued logic
C.C.Chang, 1958
An MV-algebra is a structure (A,⊕,∗ , 0) such that:
1 (A,⊕, 0) abelian monoid,
2 (x∗)∗ = x ,
3 (x∗ ⊕ y)∗ ⊕ y = (y∗ ⊕ x)∗ ⊕ x ,
4 0∗ ⊕ x = 0∗.
Chang’s completeness theorem
For a formula ϕ TFAE:
(a) ϕ provable, (d) ϕ holds in Ln for any n ≥ 2,(b) ϕ holds in any MV-algebra, (e) ϕ holds in [0, 1] ∩Q,(c) ϕ holds in any MV-chain, (f) ϕ holds in [0, 1].
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 14 / 32
Mundici’s categorical equivalence
Mundici’s categorical equivalence (1986)
The category of MV-algebras is equivalent with the category of abelianlattice-ordered groups with strong unit. For any MV-algebra A there existsan abelian lattice-ordered group with strong unit (G , u) such that
A ≃ [0, u]G .
u strong unit: u ≥ 0, for any x ∈ G there is n ≥ 1 s.t. x ≤ nu
Γ(G , u) = ([0, u]G ,⊕,∗ , 0): x ⊕ y = (x + y) ∧ 1, x∗ = 1− x
- G0 u
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 15 / 32
Mundici’s categorical equivalence
Mundici’s categorical equivalence (1986)
The category of MV-algebras is equivalent with the category of abelianlattice-ordered groups with strong unit. For any MV-algebra A there existsan abelian lattice-ordered group with strong unit (G , u) such that
A ≃ [0, u]G .
u strong unit: u ≥ 0, for any x ∈ G there is n ≥ 1 s.t. x ≤ nu
Γ(G , u) = ([0, u]G ,⊕,∗ , 0): x ⊕ y = (x + y) ∧ 1, x∗ = 1− x
strong unit ⇒ logical interpretation⇒ existence maximal ideals
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 15 / 32
Functional representation
A MV-algebra, ∅ 6= I ⊆ A is an ideal if:(x ∈ I , y ≤ x ⇒ y ∈ I ) and (x , y ∈ I ⇒ x ⊕ y ∈ I )
for any MV-algebra A, the maximal ideal space MaxA with thespectral topology is a compact Hausdorff space(open sets: r(I ) = M ∈ MaxA | I 6⊆ M for some ideal I ).
A is semisimple if⋂
M | M ∈ Max(A) = ∅
C (MaxA) = f : MaxA −→ [0, 1] | f continuous
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 16 / 32
Functional representation
A MV-algebra, ∅ 6= I ⊆ A is an ideal if:(x ∈ I , y ≤ x ⇒ y ∈ I ) and (x , y ∈ I ⇒ x ⊕ y ∈ I )
for any MV-algebra A, the maximal ideal space MaxA with thespectral topology is a compact Hausdorff space(open sets: r(I ) = M ∈ MaxA | I 6⊆ M for some ideal I ).
A is semisimple if⋂
M | M ∈ Max(A) = ∅
C (MaxA) = f : MaxA −→ [0, 1] | f continuous
L.P.Belluce, 1986
Any semisimple MV-algebra A is isomorphic with a separatingMV-subalgebra of C (MaxA) (with pointwise operations).
ι : A→ C (MaxA) embedding∀ M1 6= M2 ∃ f ∈ ι(A) (f (M1) = 0 and f (M2) 6= 0)
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 16 / 32
Functional representation
ι : A→ C (MaxA) embedding
A. Di Nola, S.Sessa, 1995
A σ-complete ⇒ MaxA basically disconnected
A complete ⇒ MaxA extremally disconnected
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 17 / 32
Functional representation
ι : A→ C (MaxA) embedding
A. Di Nola, S.Sessa, 1995
A σ-complete ⇒ MaxA basically disconnected
A complete ⇒ MaxA extremally disconnected
(V. Marra, I.L.)
We characterized those MV-algebras A with the property that A ≃ C (X )for some compact Hausdorff space X . We proved that the category ofcompact Hausdorff spaces and continuous maps is equivalent with a fullsubcategory of MV-algebras.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 17 / 32
Semantical and sintactical consequences in L
For a set Θ of formulas, defineΘ⊢ = sintactic consequences of ΘΘ|= = semantic consequences of Θ
Theorem
TFAE:
Θ⊢ = Θ|=
L(Θ) (the Lindenbaum-Tarski algebra of Θ) is semisimple.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 18 / 32
Semantical and sintactical consequences in L
For a set Θ of formulas, defineΘ⊢ = sintactic consequences of ΘΘ|= = semantic consequences of Θ
Theorem
TFAE:
Θ⊢ = Θ|=
L(Θ) (the Lindenbaum-Tarski algebra of Θ) is semisimple.
R. Cignoli, I.M.L. D’Ottaviano, D. Mundici, Algebraic foundations ofmany-valued reasoning, 2000.
P. Hajek, Metamathematics of fuzzy logic, 1998.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 18 / 32
Di Nola’s representation theorem, 1991
Theorem
Up to isomorphism, every MV-algebra A is an algebra of [0, 1]∗-valuedfunctions over some set, where [0, 1]⋆ is an ultrapower of [0, 1], onlydepending on th cardinatlity of A.
[0, 1]⋆ is the unit interval of R⋆ (non-standard reals)
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 19 / 32
MV-algebras are twofold structures
generalization of Boolean algebras
intervals in abelian lattice ordered groups with strong unit
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 20 / 32
MV-algebras are twofold structures
generalization of Boolean algebras
intervals in abelian lattice ordered groups with strong unit
The theory of MV-algebras is a possible answer to Birkhoff’s problem:develop a common abstraction which includes Boolean algebras andlattice-ordered groups as special cases.
G. Birkhoff, Lattice Theory, 1973.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 20 / 32
Probability
MV-algebrasւ ց
Boolean algebras Lattice-ordered groupswith strong unit
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 21 / 32
Probability
MV-algebrasւ ց
Boolean algebras Lattice-ordered groupswith strong unit
↓ ↓measures states↓ ↓
σ-continuity finite additivitym : B → [0,∞] s : G → R
m(0) = 0, s(1) = 1,m(
∨
n an) = Σ∞
1 m(an) s(x + y) = s(x) + s(y),
ak ∧ an = 0, k 6= n s(G+) = R+
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 21 / 32
Probability vs truth degree
P. Hajek, Metamathematics of fuzzy logic, 1998.
Q1: ”The patient is young”Q2: ”The patient will survive next week”
The sentence Q1 is true to some degree - the lower the age of thepatient, the more the sentence is true.
The sentence Q2 is a crisp sentence (T | F ), but we do not knowwhich is the case; we may have some probability (degree of belief)that the sentence is true.
Most many-valued logics are truth-functional, while the probabilitycalculus is not, since P(a ∧ b) cannot be determined using P(a) andP(b).
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 22 / 32
Probability vs truth degree
P. Hajek, Metamathematics of fuzzy logic, 1998.
Q1: ”The patient is young”Q2: ”The patient will survive next week”
The sentence Q1 is true to some degree - the lower the age of thepatient, the more the sentence is true.
The sentence Q2 is a crisp sentence (T | F ), but we do not knowwhich is the case; we may have some probability (degree of belief)that the sentence is true.
Most many-valued logics are truth-functional, while the probabilitycalculus is not, since P(a ∧ b) cannot be determined using P(a) andP(b).
Truth of a many-valued sentence is a matter of degree.Probability is not a degree of truth.
Probability is a degree of belief.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 22 / 32
States (finitely additive case)
D. Mundici, 1995
Definition
A state on (A,⊕,∗ , 0) is a map s : A→ [0, 1] s.t s(1) = 1 ands(x ⊕ y) = s(x) + s(y) whenever x ≤ y∗ (x ⊙ y = 0).The state s is faithfull if s(x) = 0 implies x = 0.
s(p) is the ”average degree of truth” of p
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 23 / 32
States (finitely additive case)
D. Mundici, 1995
Definition
A state on (A,⊕,∗ , 0) is a map s : A→ [0, 1] s.t s(1) = 1 ands(x ⊕ y) = s(x) + s(y) whenever x ≤ y∗ (x ⊙ y = 0).The state s is faithfull if s(x) = 0 implies x = 0.
s(p) is the ”average degree of truth” of p
extension results, characterization of the state space, ...
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 23 / 32
de Finetti’s coherence criterion for MV-algebras
D. Mundici, 2006,
ϕ1, . . ., ϕn are formulas of Lukasiewicz, α1, . . ., αn ∈ [0, 1]. TFAE:
for all λ1, . . ., λn ∈ R there exists a valuation V s.t.∑n
i=1 λi (αi − V (ϕi )) ≥ 0
(the probabilistic assessment (P(ϕi ) = αi )i is coherent),
there exists a state s defined on the Lindenbaum-Tarski algebra of Ls.t. s([ϕi ]) = αi for all i ∈ 1, . . . , n.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 24 / 32
de Finetti’s coherence criterion for MV-algebras
D. Mundici, 2006,
ϕ1, . . ., ϕn are formulas of Lukasiewicz, α1, . . ., αn ∈ [0, 1]. TFAE:
for all λ1, . . ., λn ∈ R there exists a valuation V s.t.∑n
i=1 λi (αi − V (ϕi )) ≥ 0
there exists a state s defined on the Lindenbaum-Tarski algebra of Ls.t. s([ϕi ]) = αi for all i ∈ 1, . . . , n.
MV-algebra with internal state(T. Flaminio, F. Montagna, ?)
An SMV-algebra is a structure (A, σ) such that A is an MV-algebra andσ : A −→ A s.t.:
σ(0) = 0, σ(x ⊕ y) = σ(x)⊕ σ(y ⊙ (x ⊙ y)∗),σ(x∗) = σ(x)∗, σ(σ(x)⊕ σ(y)) = σ(x)⊕ σ(y).
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 24 / 32
de Finetti’s coherence criterion for MV-algebras
D. Mundici, 2006,T. Flaminio, F.Montagna, 2009
ϕ1, . . ., ϕn are formulas of Lukasiewicz, α1, . . ., αn ∈ [0, 1]∩Q. TFAE:
for all λ1, . . ., λn ∈ R there exists a valuation V s.t.∑n
i=1 λi (αi − V (ϕi )) ≥ 0
there exists a state s defined on the Lindenbaum-Tarski algebra of Ls.t. s([ϕi ]) = αi for all i ∈ 1, . . . , n.
for any 1 ≤ i ≤ n, any non-trivial SMV-algebra satisfies the equations
(mi − 1)xi = x∗i and σ(ϕi ) = nixi ,
where V (ϕi ) = ni
mifor any i .
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 24 / 32
σ-states
Definition
Let A be a σ-complete MV-algebra (i.e. closed to countable suprema).The state s ∈ St(A) is a σ-state if xn ր x implies lim s(xn) = s(x) for allx , x1, x2, . . . ∈ A.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 25 / 32
σ-states
Definition
Let A be a σ-complete MV-algebra (i.e. closed to countable suprema).The state s ∈ St(A) is a σ-state if xn ր x implies lim s(xn) = s(x) for allx , x1, x2, . . . ∈ A.
⇓study of measures on abstract algebra in Lukasiewicz logic
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 25 / 32
σ-states
Definition
Let A be a σ-complete MV-algebra (i.e. closed to countable suprema).The state s ∈ St(A) is a σ-state if xn ր x implies lim s(xn) = s(x) for allx , x1, x2, . . . ∈ A.
⇓study of measures on abstract algebra in Lukasiewicz logic
B. Riecan and D. Mundici, Probability on MV-algebras,Handbook of Measure Theory, 2002
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 25 / 32
Finite additivity (linearity) vs σ-continuity?
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 26 / 32
Finite additivity (linearity) vs σ-continuity?
Riesz representation theorem
Let X be a compact Hausdorff space, θ : C (X ) −→ R a positive linearfunctional. Then there exists a unique regular measure µ defined on theBorel σ-algebra of X such that
θ(f ) =∫
Xfdµ for any f ∈ C (X ).
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 26 / 32
Finite additivity (linearity) vs σ-continuity?
Riesz representation theorem
Let X be a compact Hausdorff space, θ : C (X ) −→ R a positive linearfunctional. Then there exists a unique regular measure µ defined on theBorel σ-algebra of X such that
θ(f ) =∫
Xfdµ for any f ∈ C (X ).
Kroupa-Panti theorem
Let A be a semisimple MV-algebra. For any state s : A −→ [0, 1], thereexists a unique regular measure µ defined on the Borel σ-algebra ofX = Max(A) such that
s(f ) =∫
Xfdµ for any f ∈ A.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 26 / 32
Riesz spaces (vector lattices)
A Riesz space are lattice-ordered real vector spaces A s.t.:
x ≤ y implies x + z ≤ y + z for any z ∈ A,
r ≥ 0 and x ≥ 0 implies r•x ≥ 0,where •: R× A −→ A is the scalar multiplication.
Riesz, Kantorovich, Freundenthal, 1936-1940.
... most of the standard real function spaces are Riesz spaces, and in avery natural way.
G. Birkhoff, Lattice Theory, 1973.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 27 / 32
Riesz MV-algebras
L is a Riesz space, •: R× L −→ L scalar multiplication,u ∈ L is a strong unit
•: [0, 1]× [0, u]L −→ [0, u]L
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 28 / 32
Riesz MV-algebras
L is a Riesz space, •: R× L −→ L scalar multiplication,u ∈ L is a strong unit
•: [0, 1]× [0, u]L −→ [0, u]L
The structure Γ(L, u) = ([0, u]L, •) is called Riesz MV-algebra.
The class of Riesz MV-algebras is equational.
Riesz MV-algebras are categorically equivalent with Riesz spaces withstrong unit.
A. Di Nola, P.Flondor, I.L., MV-modules, 2004.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 28 / 32
The logic of Riesz MV-algebras
LRMV can be developed following the algebraic approach of H. RasiowaLLuk ∪ δr : r ∈ [0, 1]Axioms= Axioms of Lukasiewicz logic + Axioms reflecting the equationaldescription of Riesz MV-algebras
LRMV is complete w.r.t. [0, 1]-evaluations.
For any r ∈ [0, 1] the formula r = δr (ϕ −→ ϕ) has the value r underany [0, 1]-evaluation. Hence the truth-constants are represented inlogic, making possible the Pavelka style approach.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 29 / 32
The logic of Riesz MV-algebras
LRMV can be developed following the algebraic approach of H. RasiowaLLuk ∪ δr : r ∈ [0, 1]Axioms= Axioms of Lukasiewicz logic + Axioms reflecting the equationaldescription of Riesz MV-algebras
LRMV is complete w.r.t. [0, 1]-evaluations.
For any r ∈ [0, 1] the formula r = δr (ϕ −→ ϕ) has the value r underany [0, 1]-evaluation. Hence the truth-constants are represented inlogic, making possible the Pavelka style approach.
F. Esteva, J. Gispert, L. Godo, C. Noguera, 2007.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 29 / 32
The logic of Riesz MV-algebras
LRMV can be developed following the algebraic approach of H. RasiowaLLuk ∪ δr : r ∈ [0, 1]Axioms= Axioms of Lukasiewicz logic + Axioms reflecting the equationaldescription of Riesz MV-algebras
LRMV is complete w.r.t. [0, 1]-evaluations.
For any r ∈ [0, 1] the formula r = δr (ϕ −→ ϕ) has the value r underany [0, 1]-evaluation. Hence the truth-constants are represented inlogic, making possible the Pavelka style approach.
ϕ formula, Θ theory
truth degree: ‖ ϕ ‖Θ= inf e(ϕ) : e [0, 1]-evaluation, e(Θ) = 1,
provability degree: |ϕ|Θ = supr ∈ [0, 1] : Θ ⊢ r −→ ϕ.
Pavelka completeness. |ϕ|Θ =‖ ϕ ‖Θ
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 29 / 32
Riesz MV-algebras with faithful states
A Riesz MV-algebra, s : A −→ [0, 1] a faithful state (s(x) = 0⇒ x = 0)ρs(x , y) = s(d(x , y)) = s((x ⊙ y∗) ∨ (y ⊙ x∗))
(A, ρs) metric space
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 30 / 32
Riesz MV-algebras with faithful states
A Riesz MV-algebra, s : A −→ [0, 1] a faithful state (s(x) = 0⇒ x = 0)ρs(x , y) = s(d(x , y)) = s((x ⊙ y∗) ∨ (y ⊙ x∗))
(A, ρs) metric space
Example
If (Y ,Ω, µ) a measure space then denote L1(µ) the Riesz space of allintegrable functions. The constant function 1 is a strong unit of L1(µ), soΓ(L1(µ), 1) is a Ries MV-algebra. Define the state
s : Γ(L1(µ), 1) −→ [0, 1], s(f ) =∫
Yfdµ.
Then (Γ(L1(µ), 1), ρs) is a complete metric space.
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 30 / 32
A logical approach to metric spaces?
Theorem
For any pair (A, s) s.t. A is a Riesz MV-algebra, s : A −→ [0, 1] is afaithful state and (A, ρs) is a complete metric space, there exists ameasure space (Y ,Ω, µ) s.t.:(a) Y is an extremally disconnected compact Hausdorff space,(b) Ω is the Borel σ-algebra of Y ,(c) A ≃ Γ(L1(µ), 1).
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 31 / 32
A logical approach to metric spaces?
Theorem
For any pair (A, s) s.t. A is a Riesz MV-algebra, s : A −→ [0, 1] is afaithful state and (A, ρs) is a complete metric space, there exists ameasure space (Y ,Ω, µ) s.t.:(a) Y is an extremally disconnected compact Hausdorff space,(b) Ω is the Borel σ-algebra of Y ,(c) A ≃ Γ(L1(µ), 1).
Y is the maximal ideal space of the complete Boolean algebra of allclosed ideals of A
MV-algebraic expresssion of Kakutani’s representation for abstractL-spaces (1942)
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 31 / 32
A logical approach to metric spaces?
Theorem
For any pair (A, s) s.t. A is a Riesz MV-algebra, s : A −→ [0, 1] is afaithful state and (A, ρs) is a complete metric space, there exists ameasure space (Y ,Ω, µ) s.t.:(a) Y is an extremally disconnected compact Hausdorff space,(b) Ω is the Borel σ-algebra of Y ,(c) A ≃ Γ(L1(µ), 1).
Y is the maximal ideal space of the complete Boolean algebra of allclosed ideals of A
MV-algebraic expresssion of Kakutani’s representation for abstractL-spaces (1942)
Problem
Is it possible to extend the above connection to a categorical equivalence ?
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 31 / 32
Thank you!
Ioana Leustean (UB) Algebra and Probability in Lukasiewicz logic PURa 09 32 / 32