You are on page 1of 12

4.

THEORETICAL INFORMATICS
4.1 Relations. Structures
Let A be a set. Notation: ℘(A) or 2A – set of the parts in A.
Let M be a set: A binary relation on M is R ⊂ M × M. P (M) - M has the property P.
Notations: 1) xRy is (x,y)∈R 2) (M, R) – set M is structured by the relation R
Properties: ∀x,y,z∈M, reflexivity: xRx; transitivity: xRy, yRz⇒xRz;
symmetry: xRy⇒yRx;
antisymmetry: xRy, yRx⇒x=z
Equivalence: reflexive, transitive, symmetrical relation. Not. ~ or ≡
Preorder: reflexive, transitive relation.
Order: antisymmetrical preorder. Not. ≤
Total (linear) order: ∀x,y∈M (x ≤ y ∨ y ≤ x).
• Equivalence corresponding to a preorder: x ~ y if x ≤ y ∧ y ≤ x
• (℘(A), ⊂ ) - (partial) order
Let be (M, ≤), M ≠ ∅, X ⊂ M.
• Least element of X: a∈X ∀x∈X a ≤ x; if ∃ then ∃! (min X); min X is minimal and inferior bound.
• Minimal element of X: a∈X ∀x∈X ¬(x<a); can ∃ more or none.
• Minorant of X: a∈M a ≤ x ∀x∈X; if ∃ minorant then X is inferior bounded.
• Inferior bound of X (inf X) = the greatest minorant, if ∃ minorants in X.
Definitions for max X, maximal element, majorant, sup X are analogous.
Bounded set - ∃ (minorant, majorant).
Lattice: (M, ≤), ∀x,y∈M ∃ (sup{x,y}, inf{x,y}). Complete lattice: (M, ≤), ∀A⊂ M has inf and sup.
Well ordered set M: ∀A ∅≠A⊂ M ∃ min A.
Examples: (IN, ≤) - totally and well ordered (IN, |) - lattice
(ZI, , ≤) - totally ordered (ZI, , |) - preorder
(℘ (M), ⊂) - complete lattice :
inf X =  M' , sup X =  M' ; in particular: sup℘(M)=M, inf ℘(M)=∅.
M '∈X M '∈X
Binary relation on sets X, Y: R ⊂ X × Y.
Function is a binary relation on X × Y with the properties:
1. defined everywhere: ∀x∈X ∃y∈Y (x,y)∈f
2. univoque to right: (x,y)∈f ∧ (x,z)∈f ⇒ y = z
Notation: f∈X→Y; y=f(x) (or y= fx) for (x,y)∈f; f(X) = {y∈Y| ∃x∈X y=f(x)}; f-1(Y) = {x∈X| f(x)∈Y};
Properties: ∀{Ah}h∈H ⊂ ℘(X) ∀{Bh}h∈H⊂ ℘(Y) f(∪h∈HAh) = ∪h∈Hf(Ah),
f(∩h∈HAh) ∈ ∩h∈Hf(Ah), f-1(∪h∈HAh) = ∪h∈Hf-1(Ah), f-1(∩h∈HAh) = ∩h∈Hf-1(Ah).
Partial function: not everywhere defined (informatics – singularities related to computability) f∈X →p Y.
Possible properties: Bijection (121): injection ∧ surjection
Surjection: ∀y∈Y ∃x∈X y = f(x) Injection: ∀y∈Y (y = f(x) ∧ y = f(x’) ⇒ x = x’)
Characteristic function: Let A ⊂ T: chA∈T→ {0,1}, chA (x) = (x∈A) ?1 : 0.
Let ϕ ∈ M → M’, (M, ≤), (M’, ≤) , x,y∈M
• monotonous function: x ≤ y ⇒ ϕ(x) ≤ ϕ(y)
• antimonotonous function: x ≤ y ⇒ ϕ(x) ≥ ϕ(y)
Finite sequence of elements of A is a function s ∈i∈IN| 1≤ i ≤ n∈IN}→A. // Not. (a1,..,an)
Sequence of elements of A is a function s ∈IN → A. // Not. (an)n∈N
Notation: ∪A is the union of all sets in A, A being a set of sets.
Algebraic structures
Operation (internal binary): function op∈A×A→A // Not. op(a,b) = a op b
• associative: ∀a,b,c∈A (a op b) op c = a op (b op c)
• commutative: ∀a,b∈A a op b = b op a
nd
• distributive (the 2 binary operation defined on A is op’):
∀a,b,c∈A a op’ (b op c) = (a op’ b) op (a op’ c) ∧ (a op b) op’ c = (a op’ c) op (b op’ c)
• neutral element e∈A: ∀a∈A ∃e∈A a op e = e op a = a
• invertible element a∈A: ∃a’∈A a op a’ = a’ op a = e
Semigroup: (A, op), A ≠ ∅, op – binary associative operation.
Monoid: semigroup with neutral element.

59
Subsemigroup: (S, op’) - semigroup, (A, op) - semigroup, S ⊂ A, op’- induced operation of op on S.
Generated semigroup by an unvoid subset X of a semigroup A: // Not. <X>
The least subsemigroup of A that contains X. It equals ∩S ⊂ A, X ⊂ S - semigroup..
Free generated monoid/ semigroup by a set A: // Not. A* (A+)
Set of all unvoid sequences of elements in A, with the operation of concatenation (we consider A ⊂ A+);
Not. x = (x1 , .. , xk ), y = (y1 , .. , yl ):
A* = t A , concat : A* × A*→ A*, concat (x, y) = (x , .. , x y , .. , y ).
k ∈N
k
1 k 1 l

Group: monoid with invertible element.


Ring: (R, +, *) - (additive commutative group, multiplicative semigroup/ monoid, distributivity (*)).
Field: (F, +, *) - (additive commutative group, multiplicative group without 0, distributivity (*)).
Module: (M, R, γ) - (additive commutative group, ring, γ: R×M →M - associative - distributive (+, *)).
Vectorial space: module (field ← ring).
Equivalence relations
1. Let f ∈ M → M’; f defines an equivalence relation Rf on M: x Rf y ⇔ f(x) = f(y), x,y∈ M
2. M ≠ ∅, R – equivalence relation on M; we can build:
• M / R – set of equivalence classes C of M respective to R - factor (quotient) set;
C ∈ M/R if (∀x,y ∈C ⇒ xRy) ∧ ((z∈M ∃x∈C zRx) ⇒ z∈C)
• p ∈ M → M / R surjective, Rp = R - canonical surjection, p(x) - equivalence class of x relative to R
Partition on M: P ⊂ ℘(M) (C1, C2 ∈ P ⇒ C1 = C2 ∨ C1 ∩ C2 = ∅) ∧ M = tC
C∈P
Lemma: M’ = {p(x)| x ∈ M} is a partition/ M. Properties:
1. System of representatives: M’⊂ M in ∀ equivalence class ∃! element of M’
2. Factor set M/R can be described by the subset M’⊂M and the surjection ∈ M→ M’
3. Universality of the factor set: Let (M’, p) be a factor set of M and f ∈ M→ Q
i. (∃u: M’→ Q u°p = f) ⇔ Rp ≤ Rf (Rp ⊂ Rf as subsets of M × M); ∃u ⇒ ∃!u
ii. If ∃u with (i), then (u is injective ⇔ Rp = Rf)
iii. If ∃u with (i), then (u is surjective ⇔ f is surjective)
Corollary: Let (M’, p’), (M”, p”) be factor sets of M: Rp’= Rp”: ∃! bijection u∈M’→ M” u°p’=p”.
Closure: Let D be a set, n∈IN, R a (n+1)-relation on D: R⊂ Dn+1;
A ⊂ D is closed to R if: a1 , .. , an∈A, (a1 , .. , an , an+1)∈R ⇒ an+1∈A
Generalization: closure property of A ⊂ D relative to the relations R1,.., Rm of different parities.
Theorem: A ⊂ D, P – closure property on D; ∃!B - minimal A ⊂ B ⊂ D satisfying P(B).
Solution: B = {x∈D| x∈C, ∀C∈S } =  C , where S = {X  P(X) ∧ A∈X}.
C∈S
Proof: S ≠ ∅ ⇐ D∈S; A ⊂ B ⇐ A ⊂ C, ∀C∈S, P(B) ⇐ B is closed to any relation R to that P refers. Let
R-(n+1)-nary: a1,..,an∈B⇒a1,..,an∈C ∀C∈S; (a1,..,an∈C, (a1,..,an,an+1)∈R⇒an+1∈C), ∀C∈S⇒ an+1∈B
• B minimal ⇐ (A ⊂ B’, P (B’) ⇒ B’∈ S ⇒ B ⊂ B’, so: B’⊂ B ⇒ B = B’)
• Uniqueness ⇐ ( P(D), ⊂ ) - complete lattice; inf S = C
C∈S
qed

B is the closure of A in D, relative to the relations in property P.


Example: Closure of IN relative to subtraction (ternary relation) is IZ.
Relations are sets, so closure to relation can be realized.
Example: Reflexive transitive closure of a binary relation R (R*):
1. Top-view (behavioral): R ⊂ A × A:
- reflexivity (unary): Q = {((a,a))| a ∈ A}
- transitivity relation (ternary): Q’ = {((a,b), (b,c),(a,c)): a,b,c∈ A}
Proceeding from definition, ∃! minimal relation RB R⊂ RB reflexive and transitive
2. Bottom-view (structural): RL = R∪{(a,b): ∃ chain in R from a to b}:
chain in R a → b = (a, a1, .. , an, b), (ai-1, ai)∈R, ∀i = 1, .. , n+1, a0 = a, an+1 = b.
RB = RL Proof: RL is reflexive, transitive, and R ⊂ RL ⇒ RB ⊂ RL. Let (a,b)∈RL,
3 cases: 1. a=b (reflexivity) ⇒ (a,b)∈RB
2. (a,b)∈R ⇒ (a,b)∈RB
3. ∃ chain ≥ 2 links∈R ⇒ in RB
repeated apply of transitivity ⇒ (a,b)∈RB. Henceforth: RL ⊂ RB ⇒ RL = RB qed
Example: Transitive closure of a binary relation R (R+): R ∪ {(a,b)| ∃ acyclic chain in R form a to b}.

60
4.2 Interlevel relations in hierarchies
Let the objects be: e - prototype, e1, .., en - examples (instances), and their spheres: D(e), D(e1), .., D(en).
Example: objects can represent variables and spheres – data types.
Abstraction functions A∈ D(e1)×..×D(en)→D(e), and concretization functions K∈ D(e)→D(e1)×..×D(en) are
compatible, if A°K = IdD(e). In principle, K°A ≠ IdD(e1)×..×D(en), if the levels represent different abstraction
degrees, as the intention is the structure on D(e) to be simpler than the one on D(e1)×..×D(en). Abstraction
induces on its definition domain an equivalence relation, and reciprocally: a∼Ab, iff A(a)=A(b) – on the more
abstract level, a and b become indistinguishable one of other. Let A∈E→F be an abstraction function that is
induced by an equivalence relation ∼: x∈E (A)→ [x]A = {y∈Ey ∼x}∈F.
Let (E, rel ) and (F, rel1) be structured sets.
(i) X ⊂ E is an independent set if ∀x∈E-X ∀z∈X (∃y∈X x rel y⇒x rel z)∧(∃y∈X y rel x⇒z rel x).
(ii) Function A∈E→F is rel-homeomorphism, if ∀x, y∈E (x rel y ⇔A(x) rel1 A(y)) ∨ (A(x) = A(y).
Theorem (relation independence - homeomorphism):
(1) If A rel-homeomorphism, then ∀x∈E the set [x]A= {y∈EA(x)=A(y)}is independent.
(2) If F={Xii∈I} partition of set E in independent sets, then
(∃rel1⊂F Xi rel1 Xj ⇔ ∃(x,y)∈Xi×Xj x rel y) ∧ (A∈E→F, x∈Xi(A)→Xi = A(x), rel-homeomorphism).
Dem: (1) Supposing z∉[x]A, z rel x ∧ y∈[x]A⇒ A(z) rel1 A(x) ⇒ z rel y.
(2) Relation rel1 is well defined due to the independence of the sets. For x∈Xi, y∈Xj if i=j
A(x)=A(y), else x rel y ⇒ A(x) =Xi rel1Xj=A(y), and A(x)=Xirel1Xj=A(y)
⇒∃(x’,y’)∈Xi×Xj x’ rel y’ (Xi independent)⇒x rel y’ (Xj independent)⇒x rel y. qed
Example: If the relation rel is transitive/ +antireflexive, i.e., partial order / +strict, then the induced relation rel1
on the superior level is also transitive/ +antireflexive. The interpretation of such a relation can be: technological
succession/ +strict → design; temporal or spatial succession / +strict → verification. Morphism A can be
specified as total/ partial function defined by an algorithm of passing between components of different levels.
Conclusion: Hierarchic structure on different abstraction levels can effect in:
• modularization - grouping the elements in autonomous blocks;
• simplification - quantitative → qualitative representation (description + operation);
• specialization – limitation to some aspects;
• passing from the imprecise (vague) representations to categorical (firmly) representations.

4.3 Proof techniques


Box principle - finite sets
Let A, B be finite sets, A > B > 0: ¬∃ bijective function f∈A→B. // cycle = chain with equal extremities
Example: R – binary relation on a finite set A that contains chains of any length ⇒ ∃cycle in R // ⇐ induction
Induction principle - equivalent to an axiom for at most countable sets
If A⊂ N (0∈A) ∧ ∀n∈N ({0,1,..,n}∈A ⇒ n+1∈A) then A = N.
Example: If A finite set then 2A = 2 A . Dem: n = 0: 2A ={∅}, 20 = 1;
"n → n+1": A = n+1, n > 0 ⇒ ∃a∈A B = A \ {a}; B = n ⇒ 2B = 2n;
2A = 2B ∪{C∪{a}| C∈2B}, exterior disjunctive union ⇒ 2A= 2n + 2n = 2n+1 qed
Diagonal principle – arbitrary sets
Let R be a binary relation on A: its diagonal set is D={a∈A| (a,a)∉R}. Let Ra={b∈A| (a,b)∈R}⇒ ∀a∈A D≠Ra.
Dem: (RAA) Supposing ∃a∈A D = Ra we have: either a∈D ⇒ a∉Ra or a∉D ⇒ a∈Ra qed
Example (Cantor): 2IN is uncountable.
Dem: (RAA) Supposing ∃ bijection f ∈IN → 2IN ⇒ 2IN = {S0, S1,..} ∀i∈IN Si = f(i);
let D = {n∈IN| n∉Sn}: f bijective⇒ ∃k∈IN D = Sk ⇒ ((k∈Sk ⇒ k∉D) ∧ (k∉Sk ⇒ k∈D)) qed

4.4 Deductive theories


Let F be a set and let us call its elements phrases. An unvoid sequence of phrases is called text. The free
generated semigroup by F by concatenation is (F+, concatenation).
We consider F ⊂ F+ and note f1 , f2 ,.. , fn ⇒ f for f1 f2 .. fn f , suggesting the interpretation:
« f is deducted of {f1 , f2 ,.. , fn} ».
Deductive theory: T = (F, R), R ⊂ F+. The elements of the set of texts R are rules; a rule of length 1 is a axiom.
A is the set of axioms; A = F ∩ R. The other rules are deduction rules.
Demonstration: text α = f1 f2..fn ∀i∈IN∩[1,n] ∃r(i), j1 ,..,jr(i)∈IN∩[1,n] fj1 fj2 .. fjr(i) fi∈R.
Exercise: What order relation links r(i) to i?

61
Theorem: f∈F for that ∃ demonstration in T. Theor(T)=set of theorems of the theory T.
X ⊂ F is R-closed if ( f1 f2 .. fn f∈R, { f1 , f2 ,.. , fn }⊂X ) ⇒ f ∈ X.
Proposition: Let T = (F, R). Theor (T) is the least R-closed part of F,
i.e., Theor(T) R-closed ∧ (∀X ⊂ F R-closed ⇒ Theor(T) ⊂ X).
Consequence on F: function on the set of parts of F, C∈℘(F)→℘(F) being:
1. extensive: X ⊂ C(X), ∀X ⊂ F
2. monotonous: C(X) ⊂ C(Y), ∀X,Y ⊂ F X ⊂ Y
3. idempotent: C(X) = C(C(X)), ∀X ⊂ F
4. finitary: C(X) = ∪{ C(Y) | Y ⊂ X finite}, ∀X ⊂ F
Proposition: Let R⊂ F+. CR∈℘(F) → ℘(F), CR(X) = Theor (F, X ∪ R), ∀X⊂ F, consequence.
Proposition: ∀X⊂ F, CR(X) the least R-closed part of F including X.
Proposition: Let R⊂ F+, X⊂ F. The sequence of parts of F {Xn}n∈N,
X0= X, Xn+1= X∪{ f∈F| f1 f2 .. fm f ∈ R,{ f1 , f2 ,.., fm } ⊂ Xn }:
a) {Xn}n∈N ascendant, i.e., ∀n∈N, Xn⊂ Xn+1 
b) CR (X) = Xn
n∈ N
Proposition: ∀C consequence/ F ∃R⊂F+ C=CR. Solution: Let f1 f2 ..fn, f∈F: f1 f2 .. fn f∈R ⇔ f∈C({ f1, f2,.. , fn }).
Observation: Deductive theories sketch mathematically the concept of formal axiomatic system.

4.5 Universal algebra


Let n∈IN, A - set, Σ - set (of operation symbols); n-ary operation: σ∈An→A, a∈Σ→IN, a(σ) – arity of σ.
Σ-algebra: A=(A,{σ}σ∈Σ), A-support set. Stable part: B⊂A, A=(A,{σ}σ∈Σ) σ∈Σ∧b1,..,ba(σ)∈B⇒σ(b1,..,ba(σ))∈B.
R-closure - subset R⊂ A+ R={x1,..,xa(σ) σ(x1,..,xa(σ)): σ∈Σ, xi∈A, ∀i=1,..,a(σ)}associated to the Σ-algebra A.
Observation: A subset of A is stable part in A iff A is R-closed in A.
Stable part of A generated by B⊂A, noted CR(B)), - the least R-closed part of A including B.
Properties: A = (A, {σ}σ∈Σ), ∀B ⊂ A: a) B ⊂ CR(B) b) B ⊂ C ⊂ A ⇒ CR(B) ⊂ CR(C)
c) CR (CR(B)) = CR(B) d) CR(B) = ∪{CR(C)| C finite ⊂ B}
e) CR (B) = t
n ∈N
Bn , B0=B, Bn+1=B∪{σ(b1 , .., ba(σ))| σ∈Σ, bi∈Bn ∀i=1,.., a(σ)}

f) B⊂A, B - countable and Σ - at most countable ⇒ CR (B) - countable.


Generation: A = (A, {σ}σ∈Σ), B ⊂ A, CR (B) = A // B generates A
Morphism: A=(A, {σA}σ∈Σ), B=(B, {σB}σ∈Σ) - Σ-algebra h - morphism of de Σ-algebra ∈A → B:
h∈A→B, h(σA(x1, .. , xa(σ))) = σB(h(x1), .. , h(xa(σ))),∀σ∈Σ,∀x1, .. , xa(σ)∈A
Endomorphism: A =B, e.g., idA.
Proposition: The composition of 2 Σ-algebra morphisms is Σ-algebra morphism.
Isomorphism: f∈A→B - Σ-algebra morphism ∈A→B, ∃g Σ-algebra morphism ∈ B→A, fg=idA∧gf=idB.
Subalgebra: A - Σ-algebra, B ⊂ A stable part of support B:
B = (B, {σB}σ∈Σ), σB(b1 , .. , ba(σ))) = σA(b1) , .. , ba(σ))),∀σ∈Σ,∀b1 , .. , ba(σ)∈B
Proposition: Let A = (A, {σ}σ∈Σ). ∀B⊂A generates A - subalgebra of support CR(B).
Structural induction: reasoning method to prove a property of an algebra support.
Proposition: Let a) A = (A, {σ}σ∈Σ) - Σ-algebra b) B ⊂ A CR (B) ⊂ A c) P – property on A.
If ([ 1]∀b∈B P(b) ∧ [2]∀σ∈Σ (P(x1) ,.. ,P(xa(σ))⇒ P(σ(x1 , .. , xa(σ))) then ∀x∈A P(x).
Dem: Let ∀x∈C⊂A P(x): [1]⇒B⊂C, [2]⇒C stable part of A, B generates A⇒C=A qed
Free algebra
Free generated Σ-algebra by a subset M⊂A is a Σ-algebra A = (A, {σA}σ∈Σ), with the property:
∀Σ-algebra B = (B, {σB}σ∈Σ), ∀f:∈M→B ∃! Σ-algebra morphism h∈A →B hM = f.
Proposition: A - free generated Σ-algebra by M ⊂ A ⇒ A generated by M.
Dem: Let f = idA; if B = A then idA is the unique morphism whose restriction to M is f;
if B = CR (M) then ∃! Σ-algebra morphism h∈A →B hM = f; therefore, if B = A⇒
∃ 2 morphisms with the demanded property (RAA) ∨ the second is also idA ⇒ CR (M) = A qed
Theorem: ∀M ⊂ A ∃ free generated Σ-algebra by M (with support A). The proof bases on the equivalence of free
generated Σ-algebra by a set with a constructive concept: Peano Σ-algebra over the same set:
Peano Σ-algebra over a subset M ⊂ A is a Σ-algebra A = (A, {σA}σ∈Σ) with the properties:
a) M generates A b) ∀σ∈Σ ∀x1 ,.., xa(σ)∈A σ(x1,..,xa(σ))∉M
c) ∀σ,τ∈Σ x1,..,xa(σ)∈A y1,..,ya(τ)∈A (σ(x1,..,xa(σ))=τ(y1,..,ya(τ)) ⇒ σ = τ ∧ xi=yi∀i=1,..., a(σ))
Theorem: If A = (A, {σA}σ∈Σ), B = (B, {σB}σ∈Σ) - Σ-algebra generated by the same set M,
then ∃Σ-algebra isomorphism h∈A → B, ∀m∈M h(m) = m.

62
4.6 Qualitative theories (Johan De Kleer)
Let R be the definition domain of the abstracting function (lower level), and Q the codomain. The inferior level is
called quantitative while the superior one - qualitative. Qualitative simulation is applied especially for
processing, i.e., to solve a system of equations in Q. Noting qualitative simulation - simQual, respectively,
quantitative simulation – simQuant, and considering the hypotheses:
1. Given function a., k is e compatible, i.e., simQuant = k ° simQual ° a
2. Given ordered set (Q, ≤), x ≤ y⇒a(x) ≤ a(y).
• Completeness: ∀ can be represented as k ° simQual ° a
• Correctness: ∀ qualitative simulation ∃k adequate quantitative simulation k ° simQual ° a
• Compositionality: global behavior ⇐ (structure, local behavior)
• Classification – quasi-stationary: to consider context free the model of the parts, they are considered objects
of fixed classes. Classification concerns also individual behavior: ∀object, and behavior - states.
• Causality - locality: local interactions are modeled as linked structure, starting with cause-effect relations.
• Explicit functionality: the function of the simulated system can be extracted from the qualitative behavior, or
directly from the system’s structure. The set of attainable scopes by the systems of the class is fixed.

Q Qualitative qualitative Qualitative


problem simulation answer

a k

Quantitative quantitative Quantitative


R problem simulation answer

Q u a lita tiv e q u a lita tiv e B e h a v io r Q u a lita tiv e


d e s c rip tio n sim u la tio n d e s c r ip tio n fu n c tio n

a k

D iffe re n tia l n u m e ric a l in te rp re ta tio n A b s tra c t


e q u a tio n so lv in g S o lu tio n fu n c tio n
f∈ IR → IR

C o n c re te fu n c tio n in g C o n c r e te F u n c tio n
s y s te m b e h a v io r (ro le )
Figure 4.1: Qualitative simulation: idea + development

Correctness/ completeness conditions refer to passing between construction hierarchy levels. Desiderata are
linked to interference between the construction hierarchy and the other types of hierarchy:
Structure – global behavior ⇐ local behavior
Symbolization - discrete behavior ⇐ continuous behavior
Classification - causality ⇐ implicit behavior
Knowledge – system’s scope/ function ⇐ system’s behavior / structure
Qualitative simulation = (structure modeling, behavior prediction, explanation)
The stepwise fulfillment of desiderata by qualitative simulation has different hypostases:
a) static ⇒ time-invariant parameters + constraints for other parameters;
b) incremental (causative, intra-state) ⇒ constraint-compatible dynamic parameters;
c) inter-state ⇒ transformation of state (neglecting constraints that define a state);
d) teleological ⇒ system’s function (role).
Example: equant∈ U → IR, real description variable (continuous),
Equal∈ U → Q, qualitative description variable (discrete), Q ={+,0.,1},
+, x > 0
 ⊕ 0 + - ⊗ 0 + -
A(x) = 0, x = 0
−, x < 0
0 0 + - 0 0 0 0
 + + + ? + 0 + -
- - ? - - 0 - +

63
Basic operations are defined: Adding ⊕ is a partial operation (not everywhere defined) ⇐ equivalence relation
defining Q is not a congruence relation for ⊕; multiplication ⊗ is a homomorphism for the equivalence relation
defining Q. For S – a system of equations in Q, a solution of S for the operations in Q is named hard; this is
needed to define soft solutions for that operations are only partially defined on the appearing values –
incompleteness, otherwise, if the correspondent system in R is unsolvable and S admits hard solution -
incorrectness. Elimination of these disadvantages needs specific knowledge. For physical processes of simulated
domains described by differential equations, a qualitative precision monotonous description hierarchy is defined
for differentiable functions f∈[a,b]→ R, having multiple roots. Each degree step bases on dividing the interval
[a,b]: a = a0 a1...an-1 an = b, defining a (qualitative) set of arguments: D = {ai| i=0, .., n-1)}; ai are roots of f, i.e., f
in ai contains the zeroes of df/dt: description ← triplets, adding the qualitative values of the derivation, to
represent, by rules or constraints, knowledge about monotony, convexity, etc. In (ai, ai+1) f does not change sign; f
is described:
Arguments ai a0 an+1 (ai, ai+1)
Qualitative values 0 q0 qn+1 qi
Structure model
Whole behavior reduces to those of the components through structure; this is modeled by connections and
materials: behavior results of processing materials by the components and transporting materials through
connections; processing modifies material’s parameters, while transport does not. The model reflects these
aspects in structure of material + transport and constitutive equations of components (structural/ behavioral).
Behavior is described depending on the material attributes, generally, dual: voltage-current, load-flux, etc; they
represent sets of variable entities referred in equations, e.g., variable, differential and integral relative to time.
Qualitative variables take values of a restrained set – support of space Q, e.g., {0, +, -}, describing the 1st order
dynamics of the qualitatively studied entities. [x]Q= A(x)∈Q, A - abstraction function; ∂x = [dx/dt]Q.
Qualitative calculus bases on time invariant laws derived from differential calculus, passing to qualitative
values (⇒[x]Q) and to discrete time (⇒T). Time abstraction conserves only the events order, fixing discrete time
values depending on the crucial moments in system’s evolution ⇐ evolution equations. Qualitative calculus is
described by qualitative equations – confluences – of different forms depending on the region of functioning –
state – of the modeled system. For linear models of the intrastate behavior, a piecewise-linear model results for
the global behavior. General form of qualitative model: {state: [specifications-state], confluences}
Laws for connections: a) of continuity; b) of compatibility (when loops exist), e.g.,
a) Qualitative values: A0 ... Ai ... Az
IR values: (-∞, a0] ... (ai-1, ai] ... (az-1, ∞).
b) Continuity: [x(T1)]Q= Ak and [x(T2)]Q = Am ⇒ ∀An between Ak and Am, ∃T between T1 and T2 [x(T)]Q= An.
c) Average value: if [x(T1)]Q = [x(T2)]Q, T1 < T2, then ∀Ti , T1 ≤ Ti ≤ T2 ∂x(T i) = 0;
if T1 < T2, then ∃T i ∂x(T i) = [x(T2)]Q - [x(T1)]Q .
T, T'- discrete time successive moments, [x(T)]Q = [x(T')]Q ⇒ ∂x(T) = 0.
Linear qualitative behavior: [x(T')]Q = [x(T)]Q + ∂x(T).
Behavior prediction
Qualitative behavior is decomposed into intra- and interstate. Intrastate, changing is determined by tendencies
that can be represented by the derivatives of the qualitative variables. Episodes of discrete time are short enough
that tendencies and state do not change. Intrastate behavior, i.e., values of qualitative variables derivatives, is
described by a system of confluences, whose solving needs constraints propagation (relaxation) and algebraic
manipulation; multiple solutions can appear, interpretable as equivalent states.
State transition is determined by surpassing of borders by some qualitative variables. Possible transitions
between system’s states – belonging to the Cartesian product of possible states of the subsystem – mean the
qualitative interstate behavior; they are characterized by a set of conditions derived by qualitative computation:
- identification of all components’ states ⇒ {confluences}
- confluence solving for all components’ states ⇒{tendencies}
- comparing state-pairs to tendencies ⇒ {possible transitions} = general solution.
Behavior = (general solution, initial state)
The general solution can be constructed following various solving strategies, each of them being specific to a
problem class and basing on an essential property of qualitative computation:
Causality: a component’s state does not modify without an external influence;
Limitations: the only pairs (limit condition, tendency) ready for transition are those for that
tendency leads to state limit; e.g., (X < A, ∂X = +) and (X > A, ∂X = -);
Ordering: components modify as the correspondent limits are attained;
Momentary state: a state defined by equality modifies instantaneously.
The action of the anterior rules is constrained by rules that apply to transition between two states:

64
• ε-ordering: changing a momentary state is faster than any other state change;
• avoiding contradictions: transitions to contradictory states are not executed;
• continuity: states and tendencies vary continuously (transition happens only between neighbor states in Q);
• average: any transition has to respect an average theorem;
• reaction: in a negative reaction loop - containing no inertial component - do not appear overrisings; a
positive reaction loop does not allow all transitions.
Behavior explanation
De Kleer proposes a physical approach to explain the qualitative behavior, based on causality and locality of the
interactions: system’s behavior results of cause-effect interactions that take place in time between neighbor
components. State concept is essential, and interstate behavior revealed by the general solution fulfills the
causality and locality desiderata. The physical approach applies inside a state, where confluences appear as
constraints, and solving is a constraints satisfaction. The advantage of causal explanation is the revealing of
function-structure relation, e.g., reaction in a system can be recognized explaining causally its behavior, i.e.,
understanding its functioning - reaction is an attribute of a system’s function, not of its behavior. Intuitively,
causal explanation infers the system’s behavior from the interactions among the individual processors
(components) with: limited memory and information processing capacity, possibility to communicate/ interact
only with neighbor processors, specific contribution to general behavior for each perturbation influence. Such an
interstate behavior characteristic representation is lowered to intrastate level - otherwise characterized by the
quasistatic approximation: the existence of a mythical time – not observable in real time – is postulated, imposing
a partial order on the events that take place under the quasistatic level. The components’ laws, rendered by
confluences, are respected in mythical time only globally - not instantaneously: in mythical moments, laws that
are not respected cause activity. An explanation consists of a sequence of assertions E1, .., En, each En being
explained by {Ej}1<=j<i; this can be expressed as a proof in a deductive system, with axioms reflecting the
confluences corresponding to component models and input signals, theorems representing predictions of the
behavior, and deduction rules reflecting the logical + qualitative calculus.

function

interpretation cause

behavior structure
composition

Figura 3.2: Relations in qualitative theories

Disadvantages of demonstrative explanation can be synthesized by the observation: it is shown why, and not how,
the system behaves; more precisely:
a) unmotivated introduction of premises in explanation: often temporary hypotheses;
b) indirect demonstration (RAA) plays an important role, being intuitively unacceptable – not causal;
generally, causal reasoning is inversed - inverse chaining;
c) proofs are not unique, even on the same abstraction level and the same models.
Intuitionistic reasoning proves constructively, enabling to extract either behavior or structure - axioms/ theorems
are functional or structural models.
Simulation can result of constructive formalization.
4.7 Uncertainty
Passing to inferior abstraction levels means adding information - additional decisions, or formalization of a
flexibile approach on a nuanced universe, exclusively higher-level (incomplete) information-based, when
formalization problem is transmission of uncertainity by operations:
f∈A→B, A(a) = candidates for a, {y∈B| ∃x∈A(a) f(x) = y} = candidates for f(a).
A first distinction between the uncertainty types is:
a) subjective uncertainty (lack of knowledge) ⇐ human intervention → fuzzy concepts; //vague
b) objective uncertainty (lack of determination) ⇐ lack of information → statistical concepts.
The measure of uncertainty for a fuzzy/ statistical approach is evaluated by functions describing:
• propagation: E' (x)→ E (y)→ H; E (x)→ H //hypothesis H is deduced of evidence E with certitude x
• evidence combining (disjunction, conjunction): (E' (x)→ E1) ∧ (E' (y)→ E2) (z)→ H
• correlation of derivations: (E1' (x1)→ E1(x2)→ H) ∧ (E2' (y1)→ E2 (y2)→ H).

65
Indefinite sets // subjective/ objective uncertainty
Let ≈ be a reflexive symmetrical binary relation on universe U: // a≈b means they can not be distinguished
a) superior approximation of set T: Ts = {x∈U| ∃y∈U x≈y}
b) inferior approximation of set T:Ti = {x∈U| ∀y∈U x≈y ⇒ y∈T}. Tsup T
Tolerance of set T: ∆T = Ts - Ti.
Indefinite sets can formalize uncertain knowledge, subjective as objective. Set Tinf
operations extend to indefinite sets by superior/ inferior approximations. If ≈ is an
equivalence relation, Ti=(Ti)i, Ts=(Ts)s; the function expressing the correspondence
an element to the class in the partition ≈ divides U becomes an abstraction..
Probabilities // objective uncertainty
The central concept to describe verifiction of hypotheses in the context of certain
observations, is conditioned probability: P(A|B) = P(A∩B) / P(B).
Let h be a hypothesis and e an evidence (observation/ measuring); supposing P(e)>0, theorem of (Thomas) Bayes
relates a posteriori probability P(h|e) to a priori probability P(h), by the conditions P(e|h) for reaching to
evidence e and the total probability of evidence P(e): P(h|e) = P(e|h) × P(h) / P(e).
Hypotheses {hi}i=1,..,n∈Ω are: a) mutually-exclusive, if: hi ∩hj = ∅, i≠j, 1≤ i, j≤ n; (*)
n
b) collectively-exhaustive, if: th
i =1
i =Ω;

c) conditional-independent reported to h∈Ω, if:


P(hi1∩..∩hik|h) = P(hi1|h)..P(hik|h) ∀i1,..,i k⊂{1,..,n}.
If {hi}i=1,..,n realizes a partition of the event space Ω (mutual-exclusive and collectively-exhaustive), then:
n
P(hi|e) = P(hi) × P(e|hi) / ∑
j =1
P(hj) × P(e|hj).

In order that the necessary data to apply the prior formula do not increase exponentially, a further hypothesis of
conditional independence of evidences ej (1≤ j ≤m) relative to each hypothesis hi (1≤ i ≤n) is considered:

⇒ P (ej1 | hi )...P (ejk | hi ) P (hi )


P (hi | ej1 ∩ ... ∩ ejk ) = n

∑ P(h )P(e
j =1
j j1 | hj )...P (ejk | hj )

Certainty factors // subjective uncertainty


Expert systems usually attach numerical values to assertions and their propagation mechanisms along the
reasoning, even if probabilistic formalization is not adequate, i.e., conditions (*) are not fulfilled. Inspired by the
probabilities theory, measures are defined (both positive subunitary real numbers):
µB(H|E) = plausibility measure of H when evidence E; µC(H|E) = doubt measure of hypothesis H when E.
Certainty factor of hypothesis H when evidence E: CF(H|E)= µB(H|E)- µC(H|E) express the influence of evidence
E on the trust in H. Therefore, -1 ≤ CF(H|E) ≤ 1, with the interpretation:
CF(H|E) = 0 : no information from E
CF(H|E) > 0 : arguments rather favorable from E
CF(H|E) < 0 : arguments rather unfavorable from E.
Example: Subjective Bayes method: (P(H|E) - P(H)) / P(H|E) / (1 - P(H)) , if P(H|E) > P(H)

CF(H|E) = (P(H|E) - P(H)) / P(H) / (1 - P(H|E)) , if P(H|E) ≤ P(H).
Conditions to realize certainty factors:
1) commutativity and associativity relative to evidences (CF is order/ observation composition free)
2) if CF(H| E) = 0, then CF(H| E∧E1) = CF(H| E1)
3) if 1≠CF(H|E1) = -CF(H|E2)≠1, then CF(H|E1∧E2)=0 (CF(H|E1)=1 ∧ CF(H|E2)≠1⇒CF(H|E1∧E2) undefined).
For a particular realization the propagation/ combination/ correlation functions result:
• CF(H|E1) = F (CF(E|E1), CF(H|E), CF(H|¬E)), F continuous, monotonous in first variable, with:
CF(E|E1) CF(H|E), CF(H|¬E)
E1 ---------------> E ------------------------>H;
• conjunctive combination:
I ( CF(E1|E'), CF(E2|E') )
E' ------> E1∧E2 ------------------------------> H;
• correlating the derivations adapts Bayes formula to certainty factors: G (CF(H|E1), CF(H|E2))
E1∧E2 ---------------------------->H.

66
Evidence theory ((Arthur) Dempster-(Glenn) Shafer)
This mathematical extension of the probabilities theory aims to distinguish between uncertain and unknown.
Partially known situations, represented by evidences, are the basis for knowledge induction. The resulted
knowledge reflects both the uncertainty of real situations, as the (subjective) lack of knowledge about reality.
Starting with a complete set of disjunctive alternatives, A = {A1,..,An}, basic measures - numerical represented
evidences - are attached to the alternatives set. Basic measure for A is a function m∈2A→[0,1], m(∅) = 0 and
∑ m(X) = 1. Lack of information m : m (A) = 1, m (X) = 0 (X ⊂≠A).
X ⊂A
0 0 0

Belief function (B) attached to basic measure m reflects the accumulated evidence of knowing the correct
possibility in a subset X: Bm∈2A→[0,1], Bm(X) = ∑ m(Y) // the measure without information corresponds to a
Y ⊂X
function B without information: Bm=m0. B(∅)=0, B(A)=1; B(X∪Y)≥B(X)+B(Y)- B(X∩Y) , X,Y⊂A.
Doubt on X: D(x) = B(A - X). Maximum probability of X: B*(X) = 1 - D(X). X is focal if m(X) > 0.
Kernel of B: Ker(B) = t X . Difference between B(X) and B*(X) characterizes the lack of knowledge
m(X )>0

regarding X; B(X) = 1 ⇔ Ker(B) ⊂ X (B is Bayes function).


If B(X)=B*(X), ∀X⊂ A, then B(X∪Y) = B(X) + B(Y) - B(X∩Y), X,Y⊂A.
Evidence accumulation follows Dempster’s rule:
m1⊕m2 (∅) = 0; m1⊕m2 (H) = ( ∑ m (X )
H =X∩Y
1 * m 2( Y ) ) / ( 1 - ∑m(X)
X∩Y=∅
1 * m2(Y) ) , H≠∅
considering the conflicts, i.e., m1(X) > 0 and m2(Y) > 0, although X∩Y=∅.
Incompatibility (maximal conflict) between basic measures appears when: ∑ m ( X)
X ∩Y = ∅
1 * m 2(Y) = 1.
Evidence propagation is formalized by production rules (sure/ evidence degree) of the form (premise,
conclusion). For two sets of alternatives A, B, a basic measure m on A can be extended on A×B, by:
m0(X) = m(Π1(X)), X⊂A×B, with Π1(X)=prAX. Propagation idea: evidence on A accumulates by
Dempster’s rule extended with the evidence on the rule, and projects the result on B:
IF premises with evidence E1 AND rule with evidence E2 THEN conclusion with evidence E3.
Fuzzy sets - vague information affected by subjective uncertainty evaluated by a function with values in [0,1].
Fuzzy logic sets: P(x) - (unary) predicate interpreted in a universe U; set A={a∈U| P(a)}:
A fuzzy subset of U is a function µ∈U →[0,1]. Any concept of the set theory can be fuzzyfied, by passing to
fuzzy sets, e.g., fuzzy relation: R ⊂ X×Y ↔ µIR∈ X×Y→[0,1].
A fuzzy interpretation of predicate P on U, relates to P a fuzzy subset in U: variable x evaluation with a∈U
fulfills P(x) with degree µ(a) – valability of P(a) is accepted with µ(a).
Functions µ∈T⊂ IR →[0,1] are selected to fulfill regularitaty conditions:
(i) piece-wise continuity; (ii) x,y∈T , λ∈[0,1]⇒ µ(λx+ (1-λ)y)≥min(µ(x), µy)); (iii) ∃x∈T, µ(x)=1.
Fuzzy calculus needs representation functions of logical connectors; selection depends on the intended
application domain. For appropriate selection:
Theorem: If binary operations k,d:[0,1]2→[0,1] fulfill conditions (1-7), then k = min, d = max.
(1) commutativity, (2) associativity, (3) reciprocal distributivity,
(4) ky(x) = k(x,y), dy(x) = d(x,y), y ∈[0,1], ky and dy continuous and monotonous,
(5) k(x,x), d(x,x) strictly monotonous,
(6) k(x,y) ≤ min(x,y), d(x,y) ≥ max(x,y),
(7) k(1,1)=1; d(0,0)=0. // k = min, d = max and µ, µ (x) = 1-x, would return to classical logical calculus.
Fuzzy statistics
Subjective and objective uncertainty offently appear mixed, attracting the problem to distinguish between them in
the computation result. The concept of fuzzy alleatory variable extends that of alleatory variable, i.e., measurable
function ξ∈ Ω→IR, Ω - event space, using as values fuzzy subsets in IR. A fuzzy alleatory discrete variable x is
a regular function ω∈Ω(X)→f∈IR→[0,1].
Descriptive Statistics
Linguistic variable = (N, U, G, M): N - name; U – support set ; M(x) fuzzy ⊂ U; M – semantics;
rules: x∈L(G)→M(x); G-grammar, L(G) – set of possible values for linguistic variable.
Linguistic model attributes fuzzy sets to the words in L(G), enabling description of qualitative information.
Linguistic approximation starts from a fuzzy set, looking for linguistic expression x∈L(G), M(x) = µ; function
M is not univoquely invertible hence it needs search: {x∈L(G)| M(x) ≈ µ}, ≈ topological equivalence relation.

67
4.8 Propositional calculus
Calculus can be considered a language (vocabulary, syntax, semantics), associated to a formal system
(vocabulary, formula, axioms, deduction rules).
Language of propositional calculus:
• Vocabulary: countable set of propositional variables {x, y, z, .. }, together with a finite operation set, e.g.,
implication (a(→)=2) and false (a(f)=0); other operations are derivates:
negation: ~a = a→f disjunction: a∨b = (~a)→b conjunction: a∧b = ~((~a)∨( ~b))
• Syntax: formula are elements of the free universal f algebra F = (F, →, f) free generated of by the
vocabulary and the selected operations; the set of formulae F = {a, b, c, .. } results to be countable.
• Semantics: Algebra Bool = (Z2 , →,0) is the same type to F. Auxiliary operations are derived:
if h: F → Bool morphism, then h morphism for the derived operations.
Boolean calculus can be avoided: IZ2 –ring of rest-classes: a→b “=“ 1+a(1+b) and ~a “=“ 1+a.
Semantic deduction: X╞ a, for X ⊂ F, a∈F: ∀morphism h∈F→Bool X⊂h-1({1})⇒h(a)=1.
X=∅ , is noted ╞ a, and a is a tautology.
Deduction theorem: // general form can be applied to any propositional calculus containing operation →
Consider deductive theory over formula set F with axioms A ⊂ F and MP:
Modus Ponens (deduction rule): MP = { a a→b b| a,b∈F }
Deduction theorem (TD): Let A ⊂ F and consequence C = CA∪MP. If C has the properties:
1. a → a ∈ C(∅), 2. a → b ∈ C({b}), 3. a → b ∈ C({a → c, a → (c → b)}), ∀a,b,c∈F,
then: ∀X⊂ F ∀a,b∈F equivalent assertions: a→b∈C(X) and b∈C(X∪{a}).
If ∀a,b,c∈F {a→(b→a), (a→(b→c)→((a→b)→(a→c))}⊂A⊂F, hypotheses in TD are verified.
Syntactical deduction: X ⊂ F, a∈F, X├ a (a deducible of X) means a∈CA∪MP. X = ∅ is noted ├ a, a is theorem.
Deduction theorem (hypotheses assured by A∪MP): X├ a→b iff ∀X⊂ F ∀a,b∈F X∪{a}├ b.
Classic propositional calculus (F, →, f) with (schemes of) axioms:
A = {a→(b→a)a,b∈F}∪{(a→(b→c)→((a→b)→(a →c))a,b,c∈F}∪{~~a→aa∈F}; deduction rule: R = MP.
Correctness (syntax-semantics link): If X⊂ F, a∈F, X ├ a, then X╞ a.
// a) if a∈F then {f}├ a b) if a,b∈F then {a, ~a}├ b.
Consistency: X⊂ F is consistent if f∉CA∪MP(X).
Example: Consistency of propositional calculus: X = ∅.
a) If X⊂ F consistent then f∉X.
b) If ∃a∈F {a, ~a}⊂ X⊂ F then X inconsistent.
c) If X consistent then ∀Y⊂ X, Y e consistent.
d) If X inconsistent then ∀Y ⊃ X, Y e inconsistent.
e) If X⊂ F inconsistent then CA∪MP(X) = F.
Lemma: The union of a ⊂-rising sequence of consistent sets is consistent.
Proposition: Any consistent set is included in a maximal consistent set.
Lemma: If X⊂ F consistent maximal then: a) X = C(X); b) ∀a∈F: {a, ~a}∩X ≠ ∅.
Proposition: If X⊂ F e maximally consistent then h∈F→IZ2 h = IX morphism F→Bool.
Corollary: If X⊂ F consistent then ∃ morphism h∈F → Bool, X⊂ h-1 ({1}).
Completeness of propositional calculus (link syntax-semantics): X⊂ F a∈F: X├ a iff X╞ a.
Compactness of propositional calculus: Let X⊂ F a∈F; the assertions are equivalent:
(completeness) X├ a iff X╞ a and ∃Y finite ⊂ X, Y├ a iff ∃Y finite ⊂ X, Y╞ a.
Proposition: Let X⊂ F; equivalent are: X consistent, ∀Y finite ⊂ X, Y consistent,
∃ morphism h∈F→Bool, X⊂ h-1 ({1}),
∀Y finite ⊂ X, ∃ morphism h∈F → Bool, Y⊂ h-1 ({1}).
Intuitionistic propositional calculus (Arend Heyting) - complete calculus that fundaments constructive
mathematics - existence is proved only by construction. All axiom schemes are theorems of the classic
propositional calculus. Adding to the axioms one of the following equivalent formulae, results an equivalent
system to the axiom system of classic propositional calculus:
Excluded Tierce: p∨¬p Noncontradiction: ¬(p∧¬p) Double negation: ¬(¬p) →p
X – set of atomic propositions. I(X) = closure of X relative to logical operations: ¬, ∧, ∨, →.
Axiom schemes: p, q, r∈I(X)
A1: p→(p∧p) ; A2: p→(p∨q); A3: (p∧q)→(q∧p); A4: (p∨q)→(q∨p); A5: (p→q)→((p∧r)→(q∧r)); A6:
(p→q)→((p→r)→(q→r)); A7: (p∧(p→q))→q; A8: (p→q)→((p→r)→(q→r));
A9: ((p→r)∧(q→r))→((p∨q) →r); A10: ¬p→(p→q); A11: ((p→q)∧(p→¬q))→¬p.
Deduction rule (Modus Ponens): Of a and a→b is deduced b.

68
Excluded tierce formula is not deductible in intuitionistic propositional calculus ⇐ completeness of intuitionistic
calculus, i.e., theorem set coincides with the set of satisfied propositions in any (Saul) Kripke model
(intuitionistic semantics) = M – finite set of classical valuations v∈X→{0,1} partially ordered by v1≤v2, i.e.,
// p, q∈I(X)
∀x∈X v1(x)≤v2(x). ∀v∈M is inductively extended to I(X) by:
v(p∧q)=v(p)v(q), v(p∨q)=v(p)+v(q)-v(p)v(q), v(¬p)=1-max{u(p) u∈M u≥v}.
p∈I(X) is satisfied in M if ∀v∈M v(p) = 1.
Considering M={u,v}, u≤v: ∀p∈X u(p)=0, v(p)=1 ⇒ u(¬p)=0 ⇒ u(p∨¬p)=0, i.e., ET∉Theor(X,M). // u,v can
// be conceived as knowledge states(moment): initially, p is not satisfied, ¬p can not be satisfied.

Intelligence is the faculty to make [abstract] objects, especially tools to make tools.
[metathoughts, i.e., constructive conscious thoughts] Henri Bergson

4.9 Predicate calculus


Predicate logic of order ≥1
PL1 is at least equivalent to other logical formalisms; it builds the basis of problem approach in intelligence
simulation, e.g., What is true? What is a proof? How is a proof found/ constructed? What needs proof? Why?
PL1 refines the propositional calculus toward formalization of the internal structure of phrases. The calculus is
hierarchic by recursive variation and quantification of predicates in PL ⇒higher-order predicative logics - HOL:
Let T = set of types: 1. 0∈T 2. t1, .., tn∈T ⇒ (t1, .., tn)∈T 3. ∀type ⇐ (1) iterating (2).
∀type t ∃variables/ constants of type t, e.g., for t = (0, ..,0) tuple of variabeles/ constants in PL1.
The logical construction is the same as for PL1; atoms are to be constructed consistently to the type:
if P variable/ constant of type t = (t1, ..., tn), then P(a1, .., an) is atomic expression of type t,
iff ∀ai is variable/ constant of type ti ∀i∈IN 0≤i≤n. Functions are analogously formed.
The set hierarchy of H(A) over A: (1) type 0: x∈A ⇒ x∈H(A) (3) ∀X∈H(A)⇐ (1) iterating (2).
(2) type ti: ∀i∈IN 0≤i≤n Ai= x of type ti⇒ ∀R⊂A1×..×An is of type (t1, .., tn) in H(A)
The balance between expressivity and deductibility of PL1 is destructed. Otherwise, the calculus is accelerated,
what motivates discovering parts of a HOL, i.e., PLn, n≥2, for that competence and performance are rebalanced
⇐ theorems of Gödel (incompleteness, acceleration): ∀HOL deriving only tautologies ∃ unprovable tautology,
and ∀PLn, n≥0 ∃ formulae of length that can be shortened indefinitely in a calculus of superior order.
// as in superior logics, metastatements exist about statements/ proves of inferior order
PL1 syntax
Syntax bases on three symbol sets: V - variables, O - operations, T – terms; arity relation r∈O→IN expresses the
operands number of the operation symbol, e.g., constants: 0. Terms ∈ O-algebra T free-generated by V.
P =not predicate set, r∈P→IN =not arity function of the predicates. Proposition becomes an atomic expression, and
propositional calculus variables, atomic expressions constructed of a 0-ary predicate bulding the set AE.
Atomic expressions -----( propositional operations, quantification)----> compound expressions
Formally: PL-expressions form an AE-algebra E free generated by the set of atomic expressions with the binary
operation →, the 0-ary operation f=notfalse, and the family of unary operations {∀x}x. Other operations are
defined by those used for the free generation of the expression algebra. Variable/ operation symbols are
countable ⇒ terms/ predicates/ atoms are countable.
Free variable: Let Pf(V) = {M⊂V| M finite}.
A = (Pf(V), {oA}o∈O) - O-algebra, ∀o∈O ∀X1, .., Xa(O)∈Pf(V) oA(X1, ..., Xa(O)) = ∪1≤i≤a(O)Xi;
free algebra ⇒ ∃! O-algebra morphism ft∈T→A ∀x∈V ft(x)={x∈V| x free variable in term t∈T iff x∈ft(t)}.
Analogously is defined a free variable of an expression, constructing algebra B,
B = (Pf(V), ∪, ∅,{dx}x∈V) ∀x∈V ∀X∈Pf(V) dx(x)=X-{x}, with the unique algebra morphism fe∈E→B,
∀p(t1,...,tr(p))∈ AE fe(p(t1,...,tr(p))) = ∪1≤i≤r(p)ft(ti).
The set of free variables of an expression ⊂ set of variables that appear in expression: x∈V appears in e∈E
iff x∈a(e), a∈E→B' a(p(t1,...,tr(p))) = ∪1≤i≤r(p)ft(ti), ∀p(t1,...,tr(p))∈AE
⇐ B'= ((Pf(V), ∪, ∅,{d'x}x∈V) ∀x∈V ∀X∈Pf(V) d'x(x)=X∪{x}.
Closed expression: e∈E fe(e)=∅.
Substitution: 1. (of x with t in terms)∀x∈V ∀t∈T ∃!endomorphism sxt∈T→T sxt(y) = (y=x) ? t : y, ∀y∈V;
if t'∈T then sxt(t') results by substituting x with t in t'
2. (of x with t in expressions) ⇐ universal algebrae: ∃! sxt∈E→E ∀p(t1,..,tr(p))∈EA sxt (p(t1,..,tr(p))) = p(sxt (t1),..,
sxt (tr(p))) ∧ ∀e1,e2∈E sxt (e1→e2) = sxt (e1)→sxt (e2)
3. sxt (∀ye) = (y=x) ? ∀ye : ∀sxt (ye)
4. sxt (f) = f; if e∈E then sxt(e) results of the substitution of x with t in e.

69
PL1 semantics
Interpretation: I = (D, {σD}σ∈O,{pD}p∈P), i.e., domain: set ≠ ∅, operations: ∀o∈O σD∈Da(o)→D function, ∀p∈P
pD⊂Dr(p) predicate. Signification in D of operation symbols extends to terms and expressions; interpretation of
expression e in D is Ie⊂DV, Ie= {v∈V→D| e becomes true by interpretation I}.
I-true expression: Ie=DV.
Tautology: ∀I expression is I- true. // TI contains I-true expressions
Model: interpretation I for a set of I- true expressions.
Complete / contradictory set: X⊂E ∀e∈E closed (e∈X ∨¬e∈X) / ∃e∈E (e∈X ∧¬e∈X).
Modus Ponens (rule): MP = { e e→e' | e,e'∈E } // disorder : substitutes either | or ∈
Generalization (rule): Gen = { e ∀xe| e∈E∧x∈V} // {proxim genre| specific difference}
TI = { e∈E : Ie=DV}- complete, noncontradictory, closed at MP∪Gen // f∈D→C
Gödel set: G⊂E with the properties of TI, containing the expressions of A = ∪i=1,...,5Ai:
A1 = { e→(e'→e)}, A2 = { e→(e'→e")→((e→e') →(e→e")), A3 = {(¬¬e)→e},
A4 = { ∀x(e→e')→(e→∀xe')}, A5 = {∀xe→sxt (e)}, with: e, e', e"∈E, x∈V-fe(E),
t free to x in e - inductive definition as at substitution in expressions.
⇒ ∀I TI is gödelian; ∀ gödelian set is of the form TI, I - interpretation.
PL1 pragmatics can base on resolution and unification:
Resolution: complete logical deduction method for true expressions, e.g., Prolog (Programmation Logique).
Unification: systematic procedure to instantiate variables in expressions ⇐ substitution.
Unification of two terms, p, r, is a substitution σ = sxt, σ(p) = σ(r). The most general unifyier, mgu(p,r), is
a unifyier (p,r) σ, ∀other unifyier (p,r) τ ∃ρ substitution τ = ρ°σ ⇐ decision algorithm, i.e., constructive,
of the possibility to unify the terms p, r, e.g., LISP (List Processing).

70

You might also like