Projecting games on hypercoherences
polarized bordered games
Pierre Boudes
IML, Marseille, France
Projecting games on hypercoherences – p.1/12
Games and hypercoherences
Denotational semantics:
if π →∗ π0 then [π] = [π0]
Games and hypercoherences
Denotational semantics:
if π →∗ π0 then [π] = [π0]
Games Hypercoherences
traces of computation results of interaction (two-players plays) (points/vertices)
Projecting games on hypercoherences – p.2/12
Games and hypercoherences
Denotational semantics:
if π →∗ π0 then [π] = [π0]
Games Hypercoherences
traces of computation results of interaction (two-players plays) (points/vertices) + coherence (hyperedges)
Games and hypercoherences
Denotational semantics:
if π →∗ π0 then [π] = [π0]
Games Hypercoherences
traces of computation results of interaction (two-players plays) (points/vertices) + coherence (hyperedges) Agents (proofs or programs) are:
strategies cliques
Projecting games on hypercoherences – p.2/12
Games and hypercoherences
Denotational semantics:
if π →∗ π0 then [π] = [π0]
Games Hypercoherences
traces of computation results of interaction (two-players plays) (points/vertices) + coherence (hyperedges) Agents (proofs or programs) are:
strategies cliques
First answer: extensional collapse
Sequential algorithms Hypercoherences strategies extensional collapse cliques
Ehrhard 1999
Projecting games on hypercoherences – p.3/12
First answer: extensional collapse
Sequential algorithms Hypercoherences strategies extensional collapse cliques
Ehrhard 1999
algorithm function
First answer: extensional collapse
Sequential algorithms Hypercoherences strategies extensional collapse cliques
Ehrhard 1999
algorithm function
Quotient by partial equivalence relations (Kreisel 50’):
x ∼ι x and f ∼σ→τ g iff x ∼σ y =⇒ f (x) ∼τ g(y)
Projecting games on hypercoherences – p.3/12
First answer: extensional collapse
Sequential algorithms Hypercoherences strategies extensional collapse cliques
Ehrhard 1999
algorithm function
Quotient by partial equivalence relations (Kreisel 50’):
x ∼ι x and f ∼σ→τ g iff x ∼σ y =⇒ f (x) ∼τ g(y)
➤ Extended to others games semantics by Melliès 2004 and Laird (200?).
First answer: extensional collapse
Sequential algorithms Hypercoherences strategies extensional collapse cliques
Ehrhard 1999
algorithm function
Quotient by partial equivalence relations (Kreisel 50’):
x ∼ι x and f ∼σ→τ g iff x ∼σ y =⇒ f (x) ∼τ g(y)
➤ Extended to others games semantics by Melliès 2004 and Laird (200?).
➤ Better detailed by Melliès 2003.
Projecting games on hypercoherences – p.3/12
First answer: extensional collapse
Sequential algorithms Hypercoherences strategies extensional collapse cliques
Ehrhard 1999
algorithm function
Quotient by partial equivalence relations (Kreisel 50’):
x ∼ι x and f ∼σ→τ g iff x ∼σ y =⇒ f (x) ∼τ g(y)
➤ Extended to others games semantics by Melliès 2004 and Laird (200?).
➤ Better detailed by Melliès 2003.
➤ Limited to simple types (σ := ι | σ → σ).
Another bridge: direct projection
Games plays
Hypercoherences points
time-forgetting projection
Projecting games on hypercoherences – p.4/12
Another bridge: direct projection
Games strategies
Hypercoherences sets of points in general, the image of a strategy is not a clique
time-forgetting projection
Another bridge: direct projection
Games strategies
Relational model relations
relational model ≈ hypercoherences without coherence time-forgetting
projection
Projecting games on hypercoherences – p.4/12
Another bridge: direct projection
Games strategies
Relational model relations
time-forgetting projection
➤ Timeless games, Baillot, Danos, Ehrhard & Regnier 1997: lax functor.
Another bridge: direct projection
Logic proofs
Games strategies
Relational model relations
time-forgetting projection
➤ Timeless games, Baillot, Danos, Ehrhard & Regnier 1997: lax functor.
Projecting games on hypercoherences – p.4/12
Another bridge: direct projection
Logic (linear and polarized) proofs
Games strategies
Relational model relations
time-forgetting projection
➤ Timeless games, Baillot, Danos, Ehrhard & Regnier 1997: lax functor.
➤ Here: commutation by use of polarized bordered games for the linear subsystem of linear logic with polarities (MALLPol).
Another bridge: direct projection
Logic (linear and polarized) proofs
Games strategies
Hypercoherences cliques
time-forgetting projection
➤ Timeless games, Baillot, Danos, Ehrhard & Regnier 1997: lax functor.
➤ Here: commutation by use of polarized bordered games for the linear subsystem of linear logic with polarities (MALLPol).
Projecting games on hypercoherences – p.4/12
Linear logic with polarities (Laurent)
positives P := 0 | 1 | α⊥ | P ⊕ P | P ⊗ P | !N negatives N := > | ⊥ | α | N & N | N N | ?P
(ax.)
`α, α⊥ (one)
` 1
` Γ
(bot)
`⊥,Γ (top)
`>,Γ
` Γ,N ` Γ,N0
(with)
`Γ,N & N0
` N,Pi (i = 1,2)
(plus)
`N,P1 ⊕ P2
` N,N0,Γ
(par)
`N N0,Γ
` N,P ` N0,P0
(tens.)
`N,N0,P ⊗ P0
` N,P
(der.)
`N,?P
` ?P1, . . . , ?Pn,N
(prom.)
`?P1, . . . ,?Pn,!N
` Γ
(weak.)
`?P,Γ
` ?P,?P,Γ
(cont.)
`?P,Γ
Linear logic with polarities (Laurent)
positives P := 0 | 1 | α⊥ | P ⊕ P | P ⊗ P | ↓N negatives N := > | ⊥ | α | N & N | N N | ↑P
(ax.)
`α, α⊥ (one)
` 1
` Γ
(bot)
`⊥,Γ (top)
`>,Γ
` Γ,N ` Γ,N0
(with)
`Γ,N & N0
` N,Pi (i = 1,2)
(plus)
`N,P1 ⊕ P2
` N,N0,Γ
(par)
`N N0,Γ
` N,P ` N0,P0
(tens.)
`N,N0,P ⊗ P0
` N,P
(der.)
`N,↑P
` ↑P1, . . . , ↑Pn,N
(prom.)
`↑P1, . . . ,↑Pn,↓N
` Γ
(weak.)
`?P,Γ
` ?P,?P,Γ
(cont.)
`?P,Γ
` N,N⊥ ` N,Γ
(cut)
`N,Γ
Projecting games on hypercoherences – p.5/12
Polarized bordered games
Games:
positive A = (+, Ao, Ap, SA) SA ⊆ (Ap · Ao)∗ negative A = (−, Ao, Ap, SA) SA ⊆ (Ao · Ap)∗
Polarized bordered games
Games:
positive A = (+, Ao, Ap, SA) SA ⊆ (Ap · Ao)∗ negative A = (−, Ao, Ap, SA) SA ⊆ (Ao · Ap)∗ Well-termination:
s, s0 ∈ S, s ≤ s0 =⇒ s = s0
Projecting games on hypercoherences – p.6/12
Polarized bordered games
Games:
positive A = (+, Ao, Ap, SA) SA ⊆ (Ap · Ao)∗ negative A = (−, Ao, Ap, SA) SA ⊆ (Ao · Ap)∗ Well-termination:
s, s0 ∈ S, s ≤ s0 =⇒ s = s0
Strategy σ ⊆ SA s.t.:
s, s0 ∈ σ, s , s0 =⇒ length(s ∧ s0) is
even, if A < 0 odd, if A > 0
Interleaving of words
➤ A ∩ B = ∅, u ∈ A∗, v ∈ B∗
u •A,B v = {w | w A = u, w B = v}
Projecting games on hypercoherences – p.7/12
Interleaving of words
➤ A ∩ B = ∅, u ∈ A∗, v ∈ B∗
u •A,B v = {w | w A = u, w B = v}
➤ E = A · A, F = B · B, u ∈ E∗, v ∈ F∗
u A,B v = u •E,F v
Interleaving of words
➤ A ∩ B = ∅, u ∈ A∗, v ∈ B∗
u •A,B v = {w | w A = u, w B = v}
➤ E = A · A, F = B · B, u ∈ E∗, v ∈ F∗
u A,B v = u •E,F v
a · u · a0 ⊗A,B b · v · b0 = {(a, b) · w · (a0, b0) | w ∈ u A,B v}
Projecting games on hypercoherences – p.7/12
Interpretation of MALLpol/LLpol
➤ Follows the pattern of Laurent’s polarized games 2002
➤ Orthogonal: exchanges opponant and proponent
➤ Additives: disjoint unions (SA⊕B = SA ] SB)
➤ Multiplicatives: SA⊗B = {s ⊗A,B s0 | s ∈ SA, s ∈ SB}
➤ shifts: S↓A = {∗ · s · ∗0 | s ∈ SA}
Interpretation of MALLpol/LLpol
➤ Follows the pattern of Laurent’s polarized games 2002
➤ Orthogonal: exchanges opponant and proponent
➤ Additives: disjoint unions (SA⊕B = SA ] SB)
➤ Multiplicatives: SA⊗B = {s ⊗A,B s0 | s ∈ SA, s ∈ SB}
➤ shifts: S↓A = {∗ · s · ∗0 | s ∈ SA}
➤ exponentials: !N = ↓ ]N (some choice for ])
Projecting games on hypercoherences – p.8/12
Projection
MALLpol proofs
PBG strategies
Hypercoherences cliques
projection
p([A]PBG) = [A]Rel = |[A]Hc| p([π]PBG) = [π]Rel = [π]Hc
Projection
MALLpol proofs
PBG strategies
Hypercoherences cliques
projection
p([A]PBG) = [A]Rel = |[A]Hc| p([π]PBG) = [π]Rel = [π]Hc
Reversibility:
reverse(SA) SA
reverse([π]PBG) = [π]PBG
Projecting games on hypercoherences – p.9/12
Projection
LLpol proofs
PBG strategies
Hypercoherences cliques
projection
p([A]PBG) ⊆ [A]Rel ⊇ |[A]Hc| p([π]PBG) ⊆ [π]Rel ⊇ [π]Hc
Projection
LLpol proofs
PBG strategies
Non uniform hypercoherences
cliques projection
p([A]PBG) ⊆ [A]Rel = |[A]NHc| p([π]PBG) ⊆ [π]Rel = [π]NHc
Projecting games on hypercoherences – p.9/12
Hypercoherences unfolding
LLpol
PBG Hypercoherences
unfolding
unfold(hypergraph) = tree
Hypercoherences unfolding
LLpol
PBG Hypercoherences
unfolding
unfold(hypergraph) = tree
➤ Under some hypothesis:
Projecting games on hypercoherences – p.10/12
Hypercoherences unfolding
LLpol
PBG Hypercoherences
unfolding
unfold(hypergraph) = tree
➤ Under some hypothesis:
➤ unfold(X ⊗ Y) unfold(X) ⊗ unfold(Y)
Hypercoherences unfolding
LLpol
PBG Hypercoherences
unfolding
unfold(hypergraph) = tree
➤ Under some hypothesis:
➤ unfold(X ⊗ Y) unfold(X) ⊗ unfold(Y)
➤ Sequential algorithms exponential unfold(!X) ↓ ]s unfold(X)
Projecting games on hypercoherences – p.10/12
Hypercoherences unfolding
LLpol
PBG Hypercoherences
unfolding
unfold(hypergraph) = tree
➤ Under some hypothesis:
➤ unfold(X ⊗ Y) unfold(X) ⊗ unfold(Y)
➤ Sequential algorithms exponential unfold(!X) ↓ ]s unfold(X)
Hypercoherences unfolding
hypercoherence a
b
c d
Projecting games on hypercoherences – p.11/12
Hypercoherences unfolding
hypercoherence a
b
c d
Hypercoherences unfolding
hypercoherence a
b
c d
{a, b, c, d} {a, b, c}
{a, b} a b
{a, c} a c
d
Projecting games on hypercoherences – p.11/12
Future works
➤ Merging the two approaches (collapse and projection)
Future works
➤ Merging the two approaches (collapse and projection)
➤ Bordered games on DAGs and unfolding
Projecting games on hypercoherences – p.12/12
Future works
➤ Merging the two approaches (collapse and projection)
➤ Bordered games on DAGs and unfolding
➤ Going back to Laurent’s polarized games
Future works
➤ Merging the two approaches (collapse and projection)
➤ Bordered games on DAGs and unfolding
➤ Going back to Laurent’s polarized games
➤ relation with bordered games
Projecting games on hypercoherences – p.12/12
Future works
➤ Merging the two approaches (collapse and projection)
➤ Bordered games on DAGs and unfolding
➤ Going back to Laurent’s polarized games
➤ relation with bordered games
➤ direct projection (tech. report)
Future works
➤ Merging the two approaches (collapse and projection)
➤ Bordered games on DAGs and unfolding
➤ Going back to Laurent’s polarized games
➤ relation with bordered games
➤ direct projection (tech. report)
Projecting games on hypercoherences – p.12/12
Future works
➤ Merging the two approaches (collapse and projection)
➤ Bordered games on DAGs and unfolding
➤ Going back to Laurent’s polarized games
➤ relation with bordered games
➤ direct projection (tech. report)