an application of the finite theory of revision
TRANSCRIPT
An application of the finite theory of revision
Riccardo Bruni
Dept. of Literature and Philosophy
University of Florence
Institut für Informatik
University of Bern19 November 2015
The revision–theoretic approach
circular predicate P(x) =def ϕ(x ,P)
hypothesis H = y | P(y)
revision step H 7→ H ′ = z | (M,H) |= ϕ(z ,P)
The revision–theoretic approach
circular predicate P(x) =def ϕ(x ,P)
hypothesis H = y | P(y)
revision step H 7→ H ′ = z | (M,H) |= ϕ(z ,P)
regularities H 7→ H ′ 7→ H ′′ 7→ . . . 7→ H 7→ . . .
A
Quasi–inductive sequence of sets
Let ∆ : P(N) → P(N). Define the sequence 〈∆α | α ∈ ON〉 by
∆0 = ∅∆α+1 = ∆(∆α)
∆λ = lim infβ→λ∆β , λ limit
(where lim infβ→λ∆β := n | ∃α < λ∀β < λ(α ≤ β → n ∈ ∆β)).
B
Properties of QID
the sequence admits a stability pair (∆+∞,∆−
∞) and(countably–many) countable ordinals representing it;
Properties of QID
the sequence admits a stability pair (∆+∞,∆−
∞) and(countably–many) countable ordinals representing it;
high definitional complexity;
Properties of QID
the sequence admits a stability pair (∆+∞,∆−
∞) and(countably–many) countable ordinals representing it;
high definitional complexity;
axioms QID(K) for any class K of operators (with unsharpproof–theoretic bounds);
C
The finite revision theory
Let δ : P(N) → P(N), H ⊆ N. Define 〈δn(H) | n ∈ N〉 by
δ0(H) = H
δn+1(H) = δ(δn(H))
Recurring hypotheses
Definition
H ⊆ N is m–reflexive, if δm(H) = H (m ∈ N+)
H ⊆ N is reflexive, if H is m–reflexive for some m ∈ N+
Sets definition
Definition
Z ⊆ N is definable iff Z =⋂
H | H is reflexive
Z ⊆ N is n–definable iff Z =⋂
H | H is n–reflexive
Semantics
Definition
1. A sentence of L+ is m–valid in a model M of LPA (m ∈ N+) if,
and only if A is true in (M,H) for every m–reflexive hypothesis H.2. A sentence of L+ is m–valid (m ∈ N) if, and only if A is m–valid
in M, for every model M of the base language.
Derivability
m–validity corresponds to derivability into system C0 by Guptaand Belnap (1993)
this system has Hilbert–style, as well as well–behavingGentzen–style counterparts HC0 and GC0
Derivability
m–validity corresponds to derivability into system C0 by Guptaand Belnap (1993)
this system has Hilbert–style, as well as well–behavingGentzen–style counterparts HC0 and GC0
indexed formulas Ai for “A is true at stage i of a revisionsequence”
Derivability
m–validity corresponds to derivability into system C0 by Guptaand Belnap (1993)
this system has Hilbert–style, as well as well–behavingGentzen–style counterparts HC0 and GC0
indexed formulas Ai for “A is true at stage i of a revisionsequence”
an extra implication connective Ai → B j for “if Ai holds, thenB j holds”
D
Rational choice
Rational agents are expected to maximize their utilities
Agent A has to reason on what is rational for agent B , andviceversa
Finite games
Definition
A finite game is triple Γ = 〈P , (Si )i∈P , u〉 where
P is a finite set (the players of Γ)
(Si )i∈P is a family of finite lists (the strategies of each playerof Γ)
let S be the set of tuples s = si sj . . . (for every i , j , . . . ∈ P),u : P × S → Z is a function (the preferences of the players ofΓ)
Rational choice
Suppose Γ = 〈P , (Si )i∈P , u〉 where P = A,B and Si = i1, i2.
R(·), uA/B(·), · = ·, · < ·
Rational choice
Suppose Γ = 〈P , (Si )i∈P , u〉 where P = A,B and Si = i1, i2.
R(·), uA/B(·), · = ·, · < · ϕ1 :≡ (R(b1) ∧ uA(a2b1) < uA(a1b1)) ∨ (R(b2) ∧ uA(a2b2) < uA(a1b2))
Rational choice
Suppose Γ = 〈P , (Si )i∈P , u〉 where P = A,B and Si = i1, i2.
R(·), uA/B(·), · = ·, · < · ϕ1 :≡ (R(b1) ∧ uA(a2b1) < uA(a1b1)) ∨ (R(b2) ∧ uA(a2b2) < uA(a1b2))
R(x) =Def
(x = a1 ∧ϕ1)∨ (x = a2 ∧ϕ2)∨ (x = b1 ∧ψ1)∨ (x = b2 ∧ψ2)
Rational choice
Suppose Γ = 〈P , (Si )i∈P , u〉 where P = A,B and Si = i1, i2.
R(·), uA/B(·), · = ·, · < · ϕ1 :≡ (R(b1) ∧ uA(a2b1) < uA(a1b1)) ∨ (R(b2) ∧ uA(a2b2) < uA(a1b2))
R(x) =Def
(x = a1 ∧ϕ1)∨ (x = a2 ∧ϕ2)∨ (x = b1 ∧ψ1)∨ (x = b2 ∧ψ2)
Use the operator δΓ to revise hypotheses for “x is rational in Γ”
Strict games
The previous approach generalizes to any strict game Γ owing tothe fact, that
Fact
1. For every strategy profile s in Γ, there exists k(s) ∈ N such
that δk(s)Γ (s) is reflexive
2. There exists k ∈ N such that for every strategy profile s in Γ,
δkΓ(s) is reflexive
Strict games
Definition
For any two strategy profiles s, s ′ and player i of Γ, say that s 6=i s ′
holds in case s and s ′ differs only in the action for i they feature.
Strict games
Definition
For any two strategy profiles s, s ′ and player i of Γ, say that s 6=i s ′
holds in case s and s ′ differs only in the action for i they feature.
Definition
A strategy profile s is Nash equilibrium in a strict game Γ if for noi ∈ P and no s ′ such that s 6=i s ′, ui (s) < ui (s
′) holds.
Strict games
Theorem
If Γ is strict, then a strategy profile s is fixed point of the revision
operator if and only if it is Nash equilibrium
Quasi–strict games
3,1
0,1
1,0
1,0
2,22,2
b1 b2
a2
a1
a3
if A plays a1, b1 and b2 yields one and the same payoff;
Quasi–strict games
3,1
0,1
1,0
1,0
2,22,2
b1 b2
a2
a1
a3
if A plays a1, b1 and b2 yields one and the same payoff;
B notices that, in case A played a2 or a3, b1 would be better;so, she drops b2;
Quasi–strict games
ψ∗
1 :≡ . . . ∨ (R(a1) ∧ uB(a1b1) = uA(a1b2) ∧ uB(a2b2) <uB(a2b1) ∧ uB(a3b2) ≤ uB(a3b1)) ∨ (R(a1) ∧ uB(a1b1) =uA(a1b2) ∧ uB(a2b2) ≤ uB(a2b1) ∧ uB(a3b2) < uB(a3b1)) ∨ . . .
Quasi–strict games
ψ∗
1 :≡ . . . ∨ (R(a1) ∧ uB(a1b1) = uA(a1b2) ∧ uB(a2b2) <uB(a2b1) ∧ uB(a3b2) ≤ uB(a3b1)) ∨ (R(a1) ∧ uB(a1b1) =uA(a1b2) ∧ uB(a2b2) ≤ uB(a2b1) ∧ uB(a3b2) < uB(a3b1)) ∨ . . .
R∗(x) =
Def(x = a1 ∧ϕ
∗
1)∨ (x = a2 ∧ϕ∗
2)∨ (x = b1 ∧ψ∗
1 )∨ (x = b2 ∧ψ∗
2 )
Quasi–strict games
Definition
For any two strategy profiles s, s ′ and player i of Γ, say that s =i s ′
holds in case s and s ′ are equal only in the action for i they feature.
Definition
A finite game Γ is quasi–strict if for every strategy profiles s, s ′ andplayer i ∈ P for which both s 6=i s ′ and ui (s1) = ui (s2) hold, theneither for every s∗, s† such that s∗ =i s and s† =i s ′ bothui (s
∗) ≤ ui (s†) and there exists s‡ such that s‡ = s ′ and
ui (s∗) < ui (s
‡) are the cases, or for every s∗, s† such that s∗ =i s
and s† =i s ′ both ui (s†) ≤ ui (s
∗) and there exists s‡ such thats‡ = s ′ and ui (s
‡) < ui (s∗) are the cases.
Quasi–strict games
Definition
A strategy profile s is “trembling–hand” equilibrium in a finite gameΓ if for no i ∈ P and s ′ such that s 6=i s ′, ui (s) < ui (s
′) holds, andfor no s∗, s† such that s∗ =i s and s∗ 6=i s†, ui (s
∗) < ui (s†).
Quasi–strict games
Theorem
If Γ is quasi–strict, then a strategy profile s is fixed point of the
revision operator if and only if it is “trembling–hand” equilibrium.
Non–strict games
3,1
0,-1
1,0
1,0
2,22,2
b1 b2
a2
a1
a3
B sticks to b1 as it carries the lowest index
Non–strict games
3,1
0,-1
1,0
1,0
2,22,2
b1 b2
a2
a1
a3
1,0
1,0
3,1
0,-1
2,22,2
b1 b2
a2
a1
a3
3,1
0,-1
1,0
1,0
2,22,2
b1 b2
a2
a1
a3
Non–strict games
3,1
0,-1
1,0
1,0
2,22,2
b1 b2
a2
a1
a3
1,0
1,0
3,1
0,-1
2,22,2
b1 b2
a2
a1
a3
3,1
0,-1
1,0
1,0
2,22,2
b1 b2
a2
a1
a3
Non–strict games
3,1
0,-1
1,0
1,0
2,22,2
b1 b2
a2
a1
a3
1,0
1,0
3,1
0,-1
2,22,2
b1 b2
a2
a1
a3
3,1
0,-1
1,0
1,0
2,22,2
b1 b2
a2
a1
a3
Non–strict games: the canonical ordering
Definition
A strategy profile s is a weak equilibrium in a non–strict game Γ ifffor no i ∈ P and for no s ′ such that s 6=i s ′ we have ui (s) < ui (s
′).
Definition
A weak equilibrium s is more efficient than a weak equilibrium s ′ inΓ if for no i ∈ P it is ui (s) < ui (s
′).
Non–strict games: the canonical ordering
Definition
A finite game Γ′ = 〈P ′, (S ′i )i∈P′ , u′〉 is a reordering of
Γ = 〈P , (Si )i∈P , u〉 iff P = P ′ and Si = S ′i for all i ∈ P and there
are i , j ∈ P such that, for every strategy profile s, ui (s) = u′j(s) and
uj(s) = u′i (s).
Definition
Γc = 〈P , (Si )i∈P , uc〉 is the canonical version of Γ = 〈P , (Si )i∈P , u〉
iff Γc is the reordering of Γ that features the least number ofchanges in the order such that every most efficient weak equilibriumis fixed point of the revision operator.
Non–strict games: the canonical ordering
3,1
0,-1
1,0
1,0
2,22,2
b1 b2
a2
a1
a3
1,0
1,0
3,1
0,-1
2,22,2
b1 b2
a2
a1
a3
Non–strict games: the canonical ordering
3,1
0,-1
1,0
1,0
2,22,2
b1 b2
a2
a1
a3
1,0
1,0
3,1
0,-1
2,22,2
b1 b2
a2
a1
a3
Non–strict games: the canonical ordering
3,1
0,-1
1,0
1,0
2,22,2
b1 b2
a2
a1
a3
1,0
1,0
3,1
0,-1
2,22,2
b1 b2
a2
a1
a3
Limiting the lexicographic choice: a proposal
3,9
0,0
1,0
1,1
2,22,2
b1 b2
a2
a1
a3
B observes that playing b1 is more risky than playing b2
Limiting the lexicographic choice: a proposal
3,9
0,0
1,0
1,1
2,22,2
b1 b2
a2
a1
a3
B observes that playing b1 is more risky than playing b2
Limiting the lexicographic choice: a proposal
3,9
0,0
1,0
1,1
2,22,2
b1 b2
a2
a1
a3
B observes that playing b1 is more risky than playing b2
B also observes that playing b1 may better compensate the risk
Limiting the lexicographic choice: a proposal
the risk factor of an action b of B given A’s choice of a is
r(b)a := |maxuB(a′b) : a′ 6= a − minuB(a
′b) : a′ 6= a|
Limiting the lexicographic choice: a proposal
the risk factor of an action b of B given A’s choice of a is
r(b)a := |maxuB(a′b) : a′ 6= a − minuB(a
′b) : a′ 6= a|
the compensation by b of its own risk factor is
c(b)a := MeanuB(a′b) : a′ 6= a − r(b)a
Limiting the lexicographic choice: a proposal
Theorem
The definition of strategic rationality that contains the “risk and
compensation” clause is more robust than the previous one under
reordering of a game.
Some questions to fix
what are fixed–points in the most general case?
what about other regular hypotheses?
The limit rule and why we need it
Let P be T , and set
T (x) := (x = ps = tq ∧ val(s) = val(t)) ∨∨ (x = pT (s)q ∧ T (pT (s)q)) ∨∨ (x = p¬ϕq ∧ ¬T (pϕq)) ∨
. . .
ϕ(x ,T )
Namely: ϕ(x ,T ) is a formula defining the Tarskian truth predicate.
The limit rule and why we need it
Let F be the set:
F := ϕ0, ϕ1, . . . , ϕn, . . .
with:
ϕ0 := ⊤
ϕn+1 := Tϕn
The limit rule and why we need it
Start from H0 = ∅, then you have:
Input Output
∅ ⇒ ϕ0ϕ0 ⇒ ϕ0, ϕ1
ϕ0, ϕ1 ⇒ ϕ0, ϕ1, ϕ2ϕ0, ϕ1, ϕ2 ⇒ ϕ0, ϕ1, ϕ2, ϕ3
......
...
The limit rule and why we need it
Start from H0 = ∅, then you have:
Input Output
∅ ⇒ ϕ0ϕ0 ⇒ ϕ0, ϕ1
ϕ0, ϕ1 ⇒ ϕ0, ϕ1, ϕ2ϕ0, ϕ1, ϕ2 ⇒ ϕ0, ϕ1, ϕ2, ϕ3
......
...
PROBLEM: we cannot conclude F ⊆ T by finitary means.
QID: Alternative views
Herzberger’s
(the ‘pure’ theory)
∅
arithmetical
liminf
Gupta’s
(the ‘corrected’ th.)
Z ⊆ N
=
liminf ∪ h0
Belnap’s
(the ‘maximal’ th.)
Z ⊆ N
=
X |X coherent†
†X ⊆ N s. t., for λ limit
h+<λ ⊆ X X ∩ h−
<λ = ∅
with h+/−<λ sets of positive/negative stable elements (below λ).
Axioms: logical, arithmetical, ordinal–theoretical
For a fixed K = ∆n,Πn,Σn, the axioms of QID(K) amount at:
a complete axiomatization of first–order classical logic with
equality
the axioms of arithmetic (with CI)
standard assumptions on the ordering <Ω, on ordinal individual
constants, the defining equations of the stock of primitiveordinal functions as well as axioms on their basic properties
(monotonicity, inverses), plus a schema of transfinite induction
Ordinal axioms
(Ω.1) ∀αβ(α = β ∨ α < β ∨ β ∨ α)
(Ω.2) ∀α(¬α < α)
(Ω.3) ∀αβγ(α < β ∧ β < γ → α < γ)
(Ω.4) ∀α(0Ω ≤ α) [where α ≤ β := (α < β ∨ α = β)]
(Ω.5) ∀α(α < α′) [with α′ = succΩ(α)]
(Ω.6) ∀αβ(α < β → α′ ≤ β)
(Ω.7) 0Ω < ω ∧ ∀α < ω(α′ < ω)
(Ω.8) ∀λ(Lim(λ) → ω ≤ λ)
[where Lim(α) := (0 < α ∧ ∀β < α(β′ < α))]
Ordinal axioms
(Ω.9) ∀α(α+ 0Ω = α)
(Ω.10) ∀αβ(α+ β′ = (α+ β)′)
(Ω.11) ∀αβγ(α < β → γ + α < γ + β)
(Ω.12) ∀αβγ(α ≤ β → α+ γ ≤ β + γ)
(Ω.13) ∀α(α0Ω = 0Ωα = 0Ω)
(Ω.14) ∀αβ(αβ′ = αβ + α)
(Ω.15) ∀αβγ(0Ω < γ ∧ α < β → γα < γβ)
(Ω.16) ∀αβγ(α ≤ β → αγ ≤ βγ)
(Ω.17) ∀αβ(α < β → ∃γ ≤ β(α+ γ = β))
(Ω.18) ∀αβ(0Ω < β → ∃γ ≤ α∃δ < β(α = βγ + δ))
Ordinal axioms
(L(K)− IN) A(0) ∧ ∀x(A(x) → A(x ′)) → ∀xA(x)
(L(K)− IΩ) ∀α((∀β < α)A(β) → A(α)) → ∀αA(α)
Axioms: QID
(QID.1) x ∈ H0
A → x 6= x
(QID.2) x ∈ Hα+1
A ↔ A(x ,HαA)
(QID.3) Lim(λ) → [x ∈ HλA ↔ (∃α < λ)(∀β < λ)(α ≤ β → x ∈ Hβ
A)]
Axioms: QID
(QID.1) x ∈ H0
A → x 6= x
(QID.2) x ∈ Hα+1
A ↔ A(x ,HαA)
(QID.3) Lim(λ) → [x ∈ HλA ↔ (∃α < λ)(∀β < λ)(α ≤ β → x ∈ Hβ
A)]
(QID.4) ∀α∃λ(Lim(λ)∧α < λ∧(H+A (λ) ≡ H+
A (∞))∧(H−A (λ) ≡ H−
A (∞)))
Alphabet and syntax
Let L+ = L∪ G (·). We introduce a language L(I ) extending L+
by means of indices for formulas, and a new implication connective→.
Alphabet and syntax
Let I be the set of index terms, which contains terms p forevery p ∈ Z;
Terms of L(I ) are the terms of L+ and index terms;
Alphabet and syntax
Let I be the set of index terms, which contains terms p forevery p ∈ Z;
Terms of L(I ) are the terms of L+ and index terms;
Formulas of L(I ) are obtained by indexing formulas of L+, andby composing them by means of the new implicationconnective, according to the clauses:
Alphabet and syntax
Let I be the set of index terms, which contains terms p forevery p ∈ Z;
Terms of L(I ) are the terms of L+ and index terms;
Formulas of L(I ) are obtained by indexing formulas of L+, andby composing them by means of the new implicationconnective, according to the clauses:
Ai is a formula of L(I ), where A is a formula of L+ and i ∈ I ;
Alphabet and syntax
Let I be the set of index terms, which contains terms p forevery p ∈ Z;
Terms of L(I ) are the terms of L+ and index terms;
Formulas of L(I ) are obtained by indexing formulas of L+, andby composing them by means of the new implicationconnective, according to the clauses:
Ai is a formula of L(I ), where A is a formula of L+ and i ∈ I ; B i → C j is a formula of L(I ), where B i,C j are formulas of
L(I ), with i = i1, . . . , ik , j = j1, . . . , jq(i1, . . . , ik , j1, . . . , jq ∈ I ).
Axioms: logical
Ai → (B j → Ai), (Ai → (Bh → C j)) → ((Ai → Bh) → (Ai → C j));
Ai → (A ∨ B)i , B i → (A ∨ B)i ;
(Ai → C i ) → ((B i → C i ) → ((A ∨ B)i → C i ));
(A ∧ B)i → Ai , (A ∧ B)i → B i , Ai → (B i → (A ∧ B)i );
(B ∧ ¬B)i → Ai, (¬¬A)i → Ai ;
(∀xA)i → A[x/t]i , A[x/t]i → (∃xA)i ;
Ai → B(x)i , x 6∈ FV (A)Ai → (∀xB)i
(GEN),A(x)i → B i, x 6∈ FV (B)
(∃xA)i → B i (PAR)
B i → C j B i
C j (MP)
(where A,B,C are formulas of L(I )n wherever they carry sequences ofindices, formulas of L+ elsewhere).
Axioms: logical
Ai → (B j → Ai), (Ai → (Bh → C j)) → ((Ai → Bh) → (Ai → C j));
Ai → (A ∨ B)i , B i → (A ∨ B)i ;
(Ai → C i ) → ((B i → C i ) → ((A ∨ B)i → C i ));
(A ∧ B)i → Ai , (A ∧ B)i → B i , Ai → (B i → (A ∧ B)i );
(B ∧ ¬B)i → Ai, (¬¬A)i → Ai ;
(∀xA)i → A[x/t]i , A[x/t]i → (∃xA)i ;
Ai → B(x)i , x 6∈ FV (A)Ai → (∀xB)i
(GEN),A(x)i → B i, x 6∈ FV (B)
(∃xA)i → B i (PAR)
B i → C j B i
C j (MP)
(where A,B,C are formulas of L(I )n wherever they carry sequences ofindices, formulas of L+ elsewhere).
System HC0: special axioms
The system HC0 features axioms for G–definition and index shift:
(DEF) AG (t,G )i ↔ G (t)i+1
System HC0: special axioms
The system HC0 features axioms for G–definition and index shift:
(DEF) AG (t,G )i ↔ G (t)i+1
(IS) B i → B j , B formula of L, i , j ∈ I
Systems HCn: special axioms
In addition, systems HCn (n > 0) feature the followinggeneralization of index shift:
(ISn) B i ↔ B i+n, B formula of L+, i ∈ I
Rules: the system GC0
Let Γ,∆, ... range over multisets of formulas of L(I )n, and A,Bover formulas of L+.
Rules: the system GC0
Let Γ,∆, ... range over multisets of formulas of L(I )n, and A,Bover formulas of L+.
A. AxiomsP i , Γ ⇒ ∆,P j (Ax)
[where P is an atomic formula of L and possibly i 6= j ]
G (t)i , Γ ⇒ ∆,G (t)i (Ax)
Rules: the system GC0
B. Logical Rules
Ai , Γ ⇒ ∆Γ ⇒ ∆, (¬A)i
(L¬) Γ ⇒ ∆,Ai
(¬A)i , Γ ⇒ ∆(R¬)
Aim, Γ ⇒ ∆
(A1 ∧ A2)i , Γ ⇒ ∆
(L∧m)Γ ⇒ ∆,Ai Γ ⇒ ∆,B i
Γ ⇒ ∆, (A ∧ B)i(R∧)
Γ ⇒ ∆,Ai B j , Γ ⇒ ∆Ai → B j , Γ ⇒ ∆
(L →)Ai , Γ ⇒ ∆,B j
Γ ⇒ ∆,Ai → B j (R →)
[where, in (L/R →), i 6= j is possible (NB: A,B are formulas ofL+)]
A[v := t]i , Γ ⇒ ∆(∀vA)i , Γ ⇒ ∆
(L∀)Γ ⇒ ∆,A[v := v ′]i
Γ ⇒ ∆, (∀vA)i(R∀)
[v ′ is not free in the conclusion of (R∀)]
Rules: the system GC0
B. Logical Rules
Ai , Γ ⇒ ∆Γ ⇒ ∆, (¬A)i
(L¬) Γ ⇒ ∆,Ai
(¬A)i , Γ ⇒ ∆(R¬)
Aim, Γ ⇒ ∆
(A1 ∧ A2)i , Γ ⇒ ∆
(L∧m)Γ ⇒ ∆,Ai Γ ⇒ ∆,B i
Γ ⇒ ∆, (A ∧ B)i(R∧)
Γ ⇒ ∆,Ai B j , Γ ⇒ ∆Ai → B j , Γ ⇒ ∆
(L →)Ai , Γ ⇒ ∆,B j
Γ ⇒ ∆,Ai → B j (R →)
[where, in (L/R →), i 6= j is possible (NB: A,B are formulas ofL+)]
A[v := t]i , Γ ⇒ ∆(∀vA)i , Γ ⇒ ∆
(L∀)Γ ⇒ ∆,A[v := v ′]i
Γ ⇒ ∆, (∀vA)i(R∀)
[v ′ is not free in the conclusion of (R∀)]
Rules: the system GC0
C. Revision Rules
AG (t,G )i , Γ ⇒ ∆G (t)i+1, Γ ⇒ ∆
(L1)Γ ⇒ ∆,AG (t,G )i
Γ ⇒ ∆,G (t)i+1 (R1)
Rules: the system GC0
E. Structural Rules
Ai ,Ai , Γ ⇒ ∆Ai , Γ ⇒ ∆
(LC )Γ ⇒ ∆,Ai ,Ai
Γ ⇒ ∆,Ai (RC )
B j → Ch,B j → Ch, Γ ⇒ ∆B j → Ch, Γ ⇒ ∆
(LC→)Γ ⇒ ∆,B j → Ch,B j → Ch
Γ ⇒ ∆,B j → Ch (RC→)