It should be clear that there is a bijective correspondence between MLL proofs and the proofsof sequentialization. Let us check one direction. We show that if a proof structure with conclusions
A
1
,...,A
n
is sequentializable, then a proof that it is so yields an MLL proof of the sequent
⊢
A
1
,...,A
n
. This is easily seen by induction: the proof associated with
{
C
{
ǫ
}
,
(
C
⊥
)
{
ǫ
}
}
[
{{
C,C
⊥
}}
]is the axiom
⊢
C,C
⊥
, while the last steps of the proofs associated with
{
A
U
1
1
,...,A
U
n
n
,
(
B
C
)
U
}
[
X
]and
{
A
U
1
1
,...,A
U
n
n
,A
′
U
′
1
1
,...,A
′
U
′
n
n
,
(
B
⊗
C
)
U
}
[
X
] are, respectively:
⊢
A
1
,...,A
n
,B,C
⊢
A
1
,...,A
n
,B
C
⊢
A
1
,...,A
n
,B
⊢
A
′
1
,...A
′
n
′
,C
⊢
A
1
,...,A
n
,A
′
1
,...,A
′
n
′
,B
⊗
C
The question we shall address in the rest of the section is the following: given a proof structure,when is it the case that it is sequentializable? It turns out that the right level of generality for thisquestion is to lift it to paraproof nets, which are deﬁned next.
Deﬁnition 1.4
The
sequentializable
paraproof structures for (cutfree) MLL are the paraproof structures obtained by the following rules:
{
A
{
ǫ
}
1
,...,A
{
ǫ
}
n
}
[
{{
A
1
,...,A
n
}}
]
{
A
U
1
1
,...,A
U
n
n
,B
V
,C
W
}
[
X
]
{
A
U
1
1
,...,A
U
n
n
,
(
B
C
)
U
}
[
X
]
{
A
U
1
1
,...,A
U
n
n
,B
V
}
[
X
]
{
A
′
U
′
1
1
,...,A
′
U
′
n
′
n
′
,C
W
}
[
Y
]
{
A
U
1
1
,...,A
U
n
n
,A
′
U
′
1
1
,...,A
′
U
′
n
′
n
′
,
(
B
⊗
C
)
U
}
[
X
∪
Y
]
where
U
=
{
1
v

v
∈
V
}∪{
2
w

w
∈
W
}
. Sequentializable paraproof structures are called
paraproof nets
.
What is the prooftheoretic meaning of a paraproof net? Well, the above bijective correspondence extends to a correspondence between sequentialization proofs of paraproof structures and MLL
paraproofs
, which are deﬁned by the following rules:
⊢
Γ
⊢
Γ
1
,B
⊢
Γ
2
,C
⊢
Γ
1
,
Γ
2
,B
⊗
C
⊢
Γ
,B,C
⊢
Γ
,B
C
where in the ﬁrst rule Γ is an arbitrary multiset of formulas. The ﬁrst rule is called
generalized axiom
,or
daimon
rule. Starting in a proof search mode from an MLL formula
A
one may build absolutelyfreely a paraproof of
A
, making arbitrary decisions when splitting the context in a
⊗
rule. Anychoice is as good as another, in the sense that the process will be successful at the end, i.e., we shalleventually reach generalized axioms. There is an interesting subset of paraproofs, which we call the
extreme
ones. An extreme paraproof is a paraproof in which in each application of the
⊗
rule we haveΓ
1
=
∅
or Γ
2
=
∅
. The following deﬁnition formalizes this notion.
Deﬁnition 1.5
The
extreme
paraproof nets for (cutfree) MLL are the proof structures obtained by the following rules:
3