Focused and Synthetic Nested Sequents
 3 Citations
 416 Downloads
Abstract
Focusing is a general technique for transforming a sequent proof system into one with a syntactic separation of nondeterministic choices without sacrificing completeness. This not only improves proof search, but also has the representational benefit of distilling sequent proofs into synthetic normal forms. We show how to apply the focusing technique to nested sequent calculi, a generalization of ordinary sequent calculi to treelike instead of listlike structures. We thus improve the reach of focusing to the most commonly studied modal logics, the logics of the modal S5 cube. Among our key contributions is a focused cutelimination theorem for focused nested sequents.
Keywords
Modal Logic Proof System Linear Logic Sequent Calculus Nest Sequent1 Introduction
The focusing technique has its origin in the foundations of logic programming [1, 22] and is now increasingly relevant in structural proof theory because it improves proof search procedures [11, 21] and because focused proofs have clearly identifiable and semantically meaningful synthetic normal forms [6, 8, 10, 31]. The essential idea of focusing is to identify and coalesce the nondeterministic choices in a proof, so that a proof can be seen as an alternation of negative phases, where invertible rules are applied eagerly, and positive phases, where applications of the other rules are confined and controlled. This, in turn, lets us abstract from the usual unary and binary logical connectives by collapsing whole phases into nary synthetic connectives. The full theory of focusing was initially developed for the sequent calculus for linear logic [1], but it has since been extended to a wide variety of logics [11, 19, 27] and proof systems [4, 7]. This generality suggests that the ability to transform a proof system into a focused form is a good indication of its syntactic quality, in a manner similar to how admissibility of cut shows that a proof system is syntactically consistent.
It is natural to ask whether the focusing technique works as well for modal logics. Traditionally, modal logics are specified in terms of Hilbertstyle axiomatic systems, but such systems are not particularly suitable since axioms reveal none of the structure of logical reasoning. It is well known that certain modal logics, \(\mathsf {S5}\) in particular, are not representable in a variant of Gentzen’s sequent calculus without sacrificing analyticity. There are two principal ways to overcome this problem. The first is based on labeled proof systems that reify the Kripke semantics—the frame conditions—directly as formulas in the sequents [24, 29]. These “semantic formulas” are not subformulas of the endsequent and they cause the interpretation of sequents to fall outside the class of propositional modal formulas, and for this reason, such calculi are also called external.
The second way is to use socalled internal calculi, that enrich the sequent structure such that analyticity is preserved and such that every sequent has an interpretation that stays inside the modal language. Wellknown examples are hypersequents [2] and display calculi [3]. A more recent development are nested sequents [5, 12, 14, 26], which generalize the notion of context from a listlike structure (familiar from Gentzen’s sequent calculus) to a treelike structure. Like ordinary sequents, nested sequents have a straightforward interpretation in the language of the logic, and enjoy cut admissibility (with a cutelimination proof that stays wholly internal to the system) and hence the usual subformula property. Moreover, nested proof systems can be built modularly for every modal logic in the \(\mathsf {S5}\) cube, in both classical and intuitionistic variants [5, 20].
In this paper, we build a focused variant, with its concomitant benefits, for all modal logics of the classical \(\mathsf {S5}\) cube. For simplicity we use a polarized syntax [15] consisting of two classes of positive and negative formulas and a pair of shift connectives to move back and forth between the classes. Crucially, we interpret \(\Diamond \) as positive and \(\Box \) as negative, which differs from the polarity that would be assigned to these connectives if they were interpreted in terms of ? and !, respectively, from linear logic [25]. Our key technical contributions are: (1) a purely internal proof of cutelimination for the focused nested calculus, given in terms of a traditional rewriting procedure to eliminate cuts (which shows that our system is compositional and suitably continuous), and (2) a proof of completeness of the focused system with respect to the nonfocused system (and hence to the Kripke semantics) by showing that the focused system admits the rules of the nonfocused system. It generalizes similar proofs of cutelimination and focusing completeness for (nonnested) sequent calculi [11, 19].
To our knowledge there have been only two other attempts to apply the focusing technique to modal logics. The first uses a labeled system [23], using the work in [24] on geometric axioms to obtain systems that extend the basic modal logic \(\mathsf {K}\). The cutelimination and completeness results in [23] are obtained externally as a reduction to LKF, a focused system for firstorder logic [19]. Therefore, the polarities of the modalities are inherited from the associated quantifiers, i.e., \(\Box \) is negative and \(\Diamond \) is positive, similar to our setting. The second approach [18] also uses nested sequents, but in a restricted form, in which the treestructure is reduced to a single branch. This efficiently simulates the standard sequent system for the modal logic K, but makes both modalities positive.
Our approach uses the full power of nested sequents. and is intended as a prototype for how similar focused systems may be built for other modal logic formalisms. After some preliminaries (Sect. 2), we start with a weakly focused proof system (Sect. 3), where negative rules may be applied everywhere, including in the middle of focused phases. From this system, we extract a strongly focused system (also Sect. 3) and a synthetic system (Sect. 4) where the logical content of the phases of focusing are abstracted from the level of formulas to the level of nested sequents. We also sketch the cutelimination theorem for this synthetic variant. The synthetic design generalizes similar designs for the sequent calculus [6, 31].
2 Modal Logics and the Nested Sequent Calculus \(\mathsf {KN}\)
Let us recall the notion of nested sequents, first defined by Kashima [14] and then independently rediscovered by Poggiolesi [26] (who called them treehypersequents) and Brünnler [5]. In Gentzen’s onesided sequent calculus, a sequent is just a multiset of formulas; nested sequents generalize this notion to a multiset of formulas and boxed sequents, resulting in a tree structure.
Definition 2.1
Definition 2.2
(Corresponding Formulas). For any nested sequent \(\varGamma \), a corresponding formula, written \(\mathrm {fm}(\varGamma )\), gives an interpretation of \(\varGamma \) as a modal logic formula. Corresponding formulas obey the following equivalences: \(\mathrm {fm}(\emptyset ) \equiv \bot \), \(\mathrm {fm}(A) \equiv A\), \(\mathrm {fm}([\varGamma ]) \equiv \Box \mathrm {fm}(\varGamma )\), and \(\mathrm {fm}(\varGamma _1, \varGamma _2) \equiv \mathrm {fm}(\varGamma _1) \vee \mathrm {fm}(\varGamma _2)\).
Definition 2.3
(Context). An nholed context is like a nested sequent but contains n pairwise distinct numbered holes of the form \(\{\}_i\) (for \(1 \le i \le n\)) in place of formulas. (No hole can occur inside a formula.) We depict such a context as \(\varGamma \{\}_1\cdots \{\}_n\). Given such a context and n nested sequents \(\varDelta _1, \ldots , \varDelta _n\), we write \(\varGamma \{\varDelta _1\}_1\cdots \{\varDelta _n\}_n\) to stand for the nested sequent where the hole \(\{\}_i\) (for \(1 \le i \le n\)) in the context has been replaced by \(\varDelta _i\), with the understanding that if \(\varDelta _i\) is empty then the hole is simply removed. Unless there is any ambiguity, we will omit the hole index subscripts in this paper to keep the notation light.
Definition 2.4
The depth of \(\varGamma \{\}\), written \(\mathrm {dp}(\varGamma \{\})\), is given inductively by: \( \mathrm {dp}(\{\}) = 0\), \(\mathrm {dp}(\varDelta , \varGamma \{\}) = \mathrm {dp}(\varGamma \{\})\), and \(\mathrm {dp}([\varGamma \{\}]) = \mathrm {dp}(\varGamma \{\}) + 1\).
Example 2.5
Let \(\varGamma \{\}\{\} = A, [B, \{ {} \}, [\{ {} \}], C]\). For the sequents \(\varDelta _1 = D\) and \(\varDelta _2 = A, [C]\), we get: \(\varGamma \{\varDelta _1\}\{\varDelta _2\} = A, [B, D , [A, [C]], C]\) and \(\varGamma \{\emptyset \}\{\varDelta _2\} = A, [B, [A, [C]], C]\). We also have that \(\mathrm {dp}(\varGamma \{\}\{\varDelta _1\}) = 1\) and \(\mathrm {dp}(\varGamma \{\varDelta _1\}\{\ \})=2\).
The basic modal logic \(\mathsf {K}\) (as presented in [5]) is captured using nested sequents as the cutfree proof system \(\mathsf {KN}\) shown in the first row in Fig. 2. The deductive system corresponding to each normal extension \(\mathsf {K}+ \mathsf {X}\), where \(\mathsf {X}\subseteq \{ {\mathsf {d}, \mathsf {t}, \mathsf {b}, \mathsf {4}, \mathsf {5}} \}\) is a set of modal axioms (Fig. 1), can be obtained by adding the corresponding diamond rules Open image in new window (final two rows of Fig. 2) to \(\mathsf {KN}\). The Open image in new window rule has a side condition that the context in which the principal \(\Diamond \)formula occurs has nonzero depth, i.e., that it does not occur at the root of the conclusion sequent.
To make the correspondence between extensions of \(\mathsf {K}\) and the proof systems precise, we need the following additional notion:
Definition 2.6

whenever \(\mathsf {4}\) is derivable in \(\mathsf {K}+ \mathsf {X}\), \(\mathsf {4}\in \mathsf {X}\)

whenever \(\mathsf {5}\) is derivable in \(\mathsf {K}+ \mathsf {X}\), \(\mathsf {5}\in \mathsf {X}\).
In this paper, we will always work with 45closed axiom sets. The reason is that, for example, the axiom \(\mathsf {4}\) is not provable (without cut) in Open image in new window even though \(\mathsf {4}\) is a theorem of the logic \(\mathsf {K}+\{ {\mathsf {t},\mathsf {5}} \}\) (which is \(\mathsf {S5}\)). Note that this is not a real restriction, since for every logic in the modal cube (Fig. 1) there is a 45closed set of axioms defining it (see [5] for details). We can now state the soundness, completeness, and cutadmissibility for \(\mathsf {KN}\) and its extensions.
Theorem 2.7
 1.
A is a theorem of \(\mathsf {K}+\mathsf {X}\).
 2.
A is provable in Open image in new window .
 3.
A is provable in Open image in new window .
The proof that \(1\Longrightarrow 2\Longrightarrow 3\Longrightarrow 1\) can be found in [5]. \(\square \)
3 The Focused Systems \(\mathsf {KNwF}\) and \(\mathsf {KNF}\)
The essence of the focusing technique [1] is to classify formulas into positive formulas, whose rules are not invertible, and negative whose rules are invertible. (As usual, we consider a rule to be invertible if whenever the conclusion of the rule is derivable then so are each of its premises.) Due to invertibility, when searching for a proof it is always safe to apply—reading from conclusion to premises—a rule for a negative formula, so these may be applied at any time. On the other hand, rules for positive formulas may require rules on other formulas to be applied first. For example, the \(\mathsf {KN}\) sequent \(\Box \bar{a}, \Diamond a\) can only be proved by first applying the \(\Box \) rule, showing that \(\Diamond \)formulas are positive. A focused proof is one where the decision to apply a rule to a positive formula has to be explicitly taken, which then commits the proof to continue applying rules to this focused positive formula and its immediate positive descendants (and no other formula in the sequent), which drastically reduces the search space. The main theorem of focusing is that this strategy is complete, i.e., every theorem has a focused proof.
We will now build such a focused version of \(\mathsf {KN}\). To simplify the metatheorems about this system, we will adopt a polarized syntax [15] where the positive and negative formulas are grouped together in different syntactic categories and explicitly mediated by shift connectives (\(\uparrow \) and \(\downarrow \)). As already mentioned, \(\Diamond \) is in the positive class, and its dual \(\Box \) is in the negative class, unlike what would be expected if they were interpreted in terms of the linear logic modalities ? and ! respectively. One way of explaining this phenomenon is that the standard shallow rule \(\mathsf k\) in sequent calculus (corresponding to the promotion rule of linear logic) realises two steps at once: accessing the formula under a \(\Box \) and moving the \(\Diamond \)formulas from the context next to that formula. These two steps are done by two different rules \(\Box \) and Open image in new window in nested sequent systems like \(\mathsf {KN}\).
Definition 3.1
(Neutral). A formula is said to be neutral if it is a positive formula or a negated atom. A nested sequent is neutral if it is built from multisets of neutral formulas and boxed neutral sequents. A context \(\varGamma \{\}\cdots \{\}\) is neutral if \(\varGamma \{\emptyset \}\cdots \{\emptyset \}\) is neutral.
Definition 3.2
A focused sequent is of the form \(\varGamma \{{\left\langle {P} \right\rangle }\}\) where \(\varGamma \{\}\) is a context and P is a positive formula. The formula P is called its focus. The notion of corresponding formula (Definition 2.2) is extended with \(\mathrm {fm}({\left\langle {P} \right\rangle }) \equiv \mathrm {fm}(P)\).
The inference rules of the focused system \(\mathsf {KNwF}\) (\(\mathsf {w}\) for “weak”) are shown in the first three rows in Fig. 3 (the basic system for \(\mathsf {K}\)). Then, for a set \(\mathsf {X}\subseteq \{ {\mathsf {d}, \mathsf {t}, \mathsf {b}, \mathsf {4}, \mathsf {5}} \}\), we write Open image in new window for the corresponding subset of the focused diamond rules in the last two rows. Observe that the rules for negative formulas are exactly the same as in the unfocused system, while the rules for positive formulas can only be applied if the principal formula is in focus. Mediating between ordinary and focused sequents are the rules \(\mathsf {dec}\) (“decide”), that chooses a positive formula in the conclusion and focuses on a copy of it in the premise, and \(\mathsf {rel}\) (“release”) that drops the focus on a shifted formula. Since \(\mathsf {dec}\) keeps the original positive formula in the context, there is no need to incorporate contraction in every positive rule, like in \(\mathsf {KN}\). The \(\mathsf {sto}\) (“store”) rule removes a shift in front of a positive formula and is used to produce neutral premises from nonneutral conclusions.
We define the system \(\mathsf {KNF}\) to be a restriction of \(\mathsf {KNwF}\) where the conclusion of the \(\mathsf {dec}\) rule is required to be neutral, as well as the contexts surrounding the focus in all rules involving focused sequents. Thus, in \(\mathsf {KNF}\), the \(\mathsf {dec}\) rule is only applicable when no other rule is applicable (no negative rule as there is no negative formula in a neutral sequent and no positive rule as there is no focus), and hence we sometimes call it strongly focused. We immediately have the following proposition:
Proposition 3.3
Let \(X\subseteq \{ {\mathsf {d}, \mathsf {t}, \mathsf {b}, \mathsf {4}, \mathsf {5}} \}\). A formula A is provable in Open image in new window if and only if it is provable in Open image in new window .
Proof
A derivation in Open image in new window is by definition also a derivation in Open image in new window . Conversely, to convert a derivation in Open image in new window into one in Open image in new window , we first have to replace all instances of \(\mathsf {id}\) with a sequence of applications of Open image in new window followed (reading from conclusion upwards) by \(\mathsf {id}\), to ensure that the conclusion of the \(\mathsf {id}\) rule is neutral. Then, the negative rules Open image in new window can be permuted down by straightforward rule permutations to ensure that \(\mathsf {dec}\) only applies to neutral sequents. \(\square \)
In order to establish the soundness and completeness of \(\mathsf {KNwF}\), we use the obvious forgetful injection of the polarized syntax into the unpolarized syntax.
Definition 3.4
(Depolarization). If A is a polarized formula, then we write \(\lfloor A \rfloor \) for the unpolarized formula obtained from A by erasing the shifts \(\mathord \uparrow \) and \(\mathord \downarrow \), collapsing Open image in new window and Open image in new window into \(\wedge \), and collapsing Open image in new window and Open image in new window into \(\vee \).
Theorem 3.5
(Soundness). Let \(X\subseteq \{ {\mathsf {d}, \mathsf {t}, \mathsf {b}, \mathsf {4}, \mathsf {5}} \}\). If a formula A is provable in Open image in new window , then \(\lfloor A \rfloor \) is provable in Open image in new window .
Proof
By forgetting the polarity information, every Open image in new window proof of A is transformed into a Open image in new window proof of \(\lfloor A \rfloor \). \(\square \)
3.1 Cut Elimination
In this section we will show that a collection of cuts is admissible for Open image in new window . As usual, a rule is said to be admissible if it is the case that whenever any instance of all its premises is derivable, so is the corresponding instance of the conclusion. In order to show the admissibility of the cut rules, it will be very useful to appeal to a collection of other admissible and invertible rules.
Lemma 3.6
Let \(\mathsf {X}\subseteq \{ {\mathsf {d}, \mathsf {t}, \mathsf {b}, \mathsf {4}, \mathsf {5}} \}\). The rules \(\mathsf {weak}\), \(\mathsf {weak}_{\mathsf {f}}\), \(\mathsf {cont}\), and Open image in new window (shown in Fig. 4) are admissible for Open image in new window . Moreover, the rules \(\mathsf {sto}\), \(\Box \), Open image in new window , and Open image in new window are invertible for Open image in new window .
Proof
By straightforward induction on the height of the derivation. \(\square \)
Note that here we use contraction only on negated atoms because that is all that is needed in the cutelimination proof below. One can indeed show that the general contraction rule on arbitrary sequents (and not just formulas) is admissible, but this requires a complicated argument for focused sequents. The corresponding result for \(\mathsf {KN}\) is shown in [5].
Lemma 3.7
Let \(\mathsf {X}\subseteq \{ {\mathsf {d}, \mathsf {t}, \mathsf {b}, \mathsf {4}, \mathsf {5}} \}\). If \(\mathsf {X}\) is 45closed, then any rule Open image in new window in Open image in new window (shown in Fig. 5) is admissible for Open image in new window .
Proof
Analogous to the corresponding lemma for \(\mathsf {KN}\) in [5, Lemma 9]. The full proof can be found in [9, Lemma 3.7]. \(\square \)
We are now ready to prove the admissibility of cuts. Specifically, we show all the cuts in Fig. 6 are simultaneously admissible. The \(\mathsf {cut}_1\) rule is our standard cut between ordinary nested sequents, while \(\mathsf {cut}_2\) defines a principal cut between a focus and its dual as cut formulas. Finally, \(\mathsf {cut}_3\) is a commutative cut for situations where the positive cut formula is not principal. Note that this collection of cuts is just sufficiently large to make the standard cuts admissible. It is easy to imagine many other cutlike rules, but it is not necessary—and may not even be possible—to admit them.
Definition 3.8
The height of a formula A, written \(\mathrm {ht}(A)\), is computed inductively as follows: \(\mathrm {ht}(a)=\mathrm {ht}(\bar{a})=1\), \(\mathrm {ht}(A\star B)=\max (\mathrm {ht}(A),\mathrm {ht}(B))+1\) where Open image in new window , and \(\mathrm {ht}(\bigcirc A)= \mathrm {ht}(A)+1\) where \(\bigcirc \in \{ {\Diamond ,\Box ,\mathord \uparrow ,\mathord \downarrow } \}\). The rank of an instance of one of the cut rules is the height of its cut formula (the P in Fig. 6).
Lemma 3.9
Proof
Let \(\mathcal {D}_1\) always stand for the derivation with the positive cut formula. We proceed by lexicographic induction: the induction hypothesis may be applied whenever (1) the rank of the cut decreases, or (2) the rank stays the same and a \(\mathsf {cut}_1\) is replaced by a \(\mathsf {cut}_2\), or (3) the rank stays the same and the height of \(\mathcal {D}_1\) decreases. The height of \(\mathcal {D}_2\) does not matter for the induction. The proof is then given in terms of a terminating rewrite sequence, written with Open image in new window .
Most cases of this rewrite are standard, so we show here only certain cases characteristic of focusing; the full list of cases can be found in [9, Lemma 3.9].
 The commutative cases are simple, relying on the invertibility of the negative rules (Lemma 3.7). Here we show the case of \(\mathsf {rel}\) above a \(\mathsf {cut}_3\): The resulting \(\mathsf {cut}_2\) can be reduced because it has a smaller height.
 The cases of \(\mathsf {dec}\) with the cut formula being principal: In the resulting derivation, we first reduce the upper \(\mathsf {cut}_3\), which is allowed because the height is smaller. Then, we reduce the lower cut, which is allowed because a \(\mathsf {cut}_2\) can be used to justify a \(\mathsf {cut}_1\) of the same rank.
 Finally, here is a characteristic case for the modal axioms, for the Open image in new window rule: The resulting \(\mathsf {cut}_2\) can be reduced because it has a smaller height. In the right branch, the \({\Box }^{1}\) rule is the admissible inverse of the \(\Box \) rule (Lemma 3.6) and the Open image in new window rule is given by Lemma 3.7. \(\square \)
Theorem 3.10
Let \(\mathsf {X}\subseteq \{ {\mathsf {d}, \mathsf {t}, \mathsf {b}, \mathsf {4}, \mathsf {5}} \}\) be 45closed. If a sequent \(\varGamma \) is provable in Open image in new window , then it is also provable in Open image in new window .
Proof
Apply Lemma 3.9 to all cut instances in the derivation, starting with a topmost one. This gives us a cutfree proof in Open image in new window , from which we get a cutfree proof in Open image in new window using Proposition 3.3. \(\square \)
3.2 Completeness
Lemma 3.11
A formula A is provable in Open image in new window if and only if \(\lfloor A \rfloor \) is provable in Open image in new window . \(\square \)
We are now going to simulate Open image in new window in Open image in new window . For this, we need another property of \(\mathsf {KNwF}\):
Lemma 3.12
Proof
By straightforward induction on the height of P. Details can be found in [9, Lemma 3.12]. \(\square \)
Lemma 3.13
(Simulation). Let \(\mathsf {X}\subseteq \{ {\mathsf {d}, \mathsf {t}, \mathsf {b}, \mathsf {4}, \mathsf {5}} \}\) and let A be a formula that is provable in Open image in new window . Then A is provable in Open image in new window .
Proof
Note that we make a crucial use of \(\mathsf {cut_1}\) and Lemma 3.12. \(\square \)
Theorem 3.14
(Completeness). Let \(\mathsf {X}\subseteq \{ {\mathsf {d}, \mathsf {t}, \mathsf {b}, \mathsf {4}, \mathsf {5}} \}\) be 45closed. For any A, if \(\lfloor A \rfloor \) is provable in Open image in new window , then A is provable in Open image in new window .
Proof
Suppose that we have a proof of \(\lfloor A \rfloor \) in Open image in new window . By Lemma 3.11, there is a proof of A in Open image in new window . Then, by Lemma 3.13, there is also a proof of A in Open image in new window . Finally, using Theorem 3.10, we get a proof in Open image in new window . \(\square \)
4 The Synthetic System
As already mentioned, the strongly focused system \(\mathsf {KNF}\) is given as a restriction of \(\mathsf {KNwF}\) where the \(\mathsf {dec}\) rule is restricted to neutral contexts. However, the cutelimination and admissibility theorems (3.9 and 3.10) were proved in the \(\mathsf {KNwF}\) system and made essential use of the admissibility of weakening by arbitrary formulas, including negative formulas, and of the possibility of applying negative rules even within a positive phase. This freedom simplifies the proofs of the metatheorems, and leaves them at least a recognizable variant of similar proofs in the nonfocused system \(\mathsf {KN}\). Of course, thanks to Proposition 3.3, we also have a cutelimination proof for Open image in new window , but this is not entirely satisfactory: it is not an internal proof, i.e., a sequence of cut reductions for Open image in new window that stays inside the system.
To avoid this complexity, it is better to consider the focused proof system in a synthetic form where the logical inference rules for the various connectives are composed as much as possible, so that the proof system itself contains exactly two logical rules: one for a positive and one for a negative synthetic connective [6, 31]. This synthetic view moreover improves the concept of focusing itself by showing that a focused proof consists of: (1) a selection of a certain substructure (the generalization of subformula) of the principal formula, and (2) the contextualization of that substructure. For positive principal formulas, this contextualization is in the form of a check in the surrounding context for other structures, such as dual atoms or nested sequents. For negative principal formulas, on the other hand, contextualization amounts to asserting the presence of additional structure in the surrounding context. This design will become clear in the explicit formulation of the synthetic version of \(\mathsf {KNF}\)—which we call \(\mathsf {KNS}\)—in the rest of this section.
4.1 Synthetic Substructures
For any negative formula, there is a collection of corresponding nested sequents that represents one of the possible branches taken in a sequence of negative rules applied to the formula. This correspondence is formally given below.
Definition 4.1
(Matching). The nested sequence \(\varGamma \)matches the negative formula N, written \(\varGamma \preccurlyeq N\), if it is derivable from the rules in Fig. 7.
For the system to follow, we will use two additional sequentlike structures that are not themselves neutral sequents.
Definition 4.2

An inversion sequent is a structure of the form \(\varGamma \{N\}\) where \(\varGamma \{\}\) is a neutral sequent context.

A focused sequent is a structure of the form \(\varGamma \{{\left\langle {\varDelta } \right\rangle }\}\) where \(\varGamma \{\}\) is a neutral sequent context and \(\varDelta \) is a neutral sequent.
Note that in \(\varGamma \{N\}\), there is exactly one occurrence of N as a toplevel formula anywhere; likewise, in \(\varGamma \{{\left\langle {\varDelta } \right\rangle }\}\), there is a single occurrence of the substructure \({\left\langle {\varDelta } \right\rangle }\). Hence, these extended sequent forms uniquely determine their decomposition into context (the \(\varGamma \{\}\)) and the extended entity (the N or the \({\left\langle {\varDelta } \right\rangle }\)).
Definition 4.3
Thus, the modal rules of Open image in new window are properly seen as structural rules rather than logical rules.
4.2 Cut Elimination
Cut elimination for Open image in new window is achieved in a similar way to that for Open image in new window .
Lemma 4.4
(Admissible Rules). The rules \(\mathsf {weak}\), \(\mathsf {weak}_{\mathsf {f}}\), \(\mathsf {cont}\), Open image in new window (shown in Fig. 4), restricted to neutral and extended sequents (as appropriate) are admissible in Open image in new window . Moreover, if \(\mathsf {X}\subseteq \{ {\mathsf {d}, \mathsf {t}, \mathsf {b}, \mathsf {4}, \mathsf {5}} \}\) is 45closed, then any rule Open image in new window (see Fig. 5) is admissible in \({\mathsf {KNS}}+{\mathsf {X}^{\left\langle {} \right\rangle }}\).
Proof
By induction on the height of the derivation, analogous to the proofs of Lemmas 3.6 and 3.7. \(\square \)
Theorem 4.5
Let \(\mathsf {X}\subseteq \{ {\mathsf {d}, \mathsf {t}, \mathsf {b}, \mathsf {4}, \mathsf {5}} \}\) be 45closed. Each rule of \(\{ {\mathsf {cut}^{\left\langle {} \right\rangle }_1, \mathsf {cut}^{\left\langle {} \right\rangle }_2, \mathsf {cut}^{\left\langle {} \right\rangle }_3} \}\) is admissible in \({\mathsf {KNS}}+{\mathsf {X}^{\left\langle {} \right\rangle }}\).
Proof
The instance of \(\mathsf {weak}\) is justified by Lemma 4.4. The remainder of the cases of the proof have to be adjusted similarly. The full proof can be found in [9, Theorem 4.5]. \(\square \)
It is worth remarking that this cutelimination proof did not have to mention any logical connectives, and was instead able to factorize all logical reasoning in terms of the matching. This means that the matching judgement can be modified at will without affecting the nature of the cut argument, as long as it leaves the structure of nested sequents untouched. This makes our result modular in yet another way, in addition to the modularity obtained by means of the structural rules for foci. Indeed, we can obtain a similarly synthetic version of identity reduction (Lemma 3.12).
Lemma 4.6
Proof
By induction on the structure of the focus, \({\left\langle {\varDelta } \right\rangle }\). The full proof can be found in [9, Lemma 4.6]. \(\square \)
Showing Open image in new window sound and complete with respect to Open image in new window is fairly straightforward and the details are omitted here. Soundness follows directly from replacing the \(\mathsf {KNS}\) sequent \(\varGamma \{{\left\langle {\varDelta } \right\rangle }\}\) with the \(\mathsf {KNF}\) sequent \(\varGamma \{{\left\langle {\overline{\mathrm {f\bar{m}}(\varDelta )}} \right\rangle }\}\) and then interpreting the Open image in new window proof in Open image in new window , using matching derivation (Fig. 7) to determine how to choose between the two Open image in new window rules. For completeness, we can follow the strategy of Lemma 3.13 nearly unchanged. However, like with the cutelimination proof, we have to avoid appeals to weakening with negative formulas by using the synthetic form of \(\mathsf {neg}^{\left\langle {} \right\rangle }\).
5 Perspectives
We have presented strongly focused and synthetic systems for all modal logics in the classical \(\mathsf {S5}\)cube. We used the formalism of nested sequents as carrier, but we are confident that something similar can be achieved for hypersequents, for example based on the work of Lellmann [17]. We chose nested sequents over hypersequents for two reasons. First, the formula interpretation of a nested sequent is the same for all logics in the \(\mathsf {S5}\)cube, which simplifies the presentation of the metatheory. Second, due to the close correspondence between nested sequents and prefixed tableaux [12] we can from our work directly extract focused tableau systems for modal logics. Furthermore, even though we spoke in this paper only about classical modal logic, we are confident that the same results can also be obtained for the intuitionistic variant of the modal \(\mathsf {S5}\)cube [29], if we start from the nonfocused systems in [30].
One extension that would be worth considering would be relaxing the restriction that there can be at most one focus in a \(\mathsf {KNF}\) or \(\mathsf {KNS}\) proof. Allowing multiple foci would take us from ordinary focusing to multifocusing, which is well known to reveal more parallelism in sequent proofs [10]. It has been shown that a certain well chosen multifocusing system can yield syntactically canonical representatives of equivalence classes of sequent proofs for classical predicate logic [8]. Extending this approach to the modal case seems promising.
References
 1.Andreoli, J.M.: Logic programming with focusing proofs in linear logic. J. Logic Comput. 2(3), 297–347 (1992)MathSciNetCrossRefzbMATHGoogle Scholar
 2.Avron, A.: The method of hypersequents in the proof theory of propositional nonclassical logics. In: Logic: From Foundations to Applications: European Logic Colloquium, pp. 1–32. Clarendon Press (1996)Google Scholar
 3.Belnap Jr., N.D.: Display logic. J. Philos. Logic 11, 375–417 (1982)MathSciNetzbMATHGoogle Scholar
 4.BrockNannestad, T., Schürmann, C.: Focused natural deduction. In: Fermüller, C.G., Voronkov, A. (eds.) LPAR17. LNCS, vol. 6397, pp. 157–171. Springer, Heidelberg (2010)CrossRefGoogle Scholar
 5.Brünnler, K.: Deep sequent systems for modal logic. Arch. Math. Logic 48(6), 551–577 (2009)MathSciNetCrossRefzbMATHGoogle Scholar
 6.Chaudhuri, K.: Focusing strategies in the sequent calculus of synthetic connectives. In: Cervesato, I., Veith, H., Voronkov, A. (eds.) LPAR 2008. LNCS (LNAI), vol. 5330, pp. 467–481. Springer, Heidelberg (2008)CrossRefGoogle Scholar
 7.Chaudhuri, K., Guenot, N., Straßburger, L.: The focused calculus of structures. In: Computer Science Logic: 20th Annual Conference of the EACSL. Leibniz International Proceedings in Informatics (LIPIcs), pp. 159–173. Schloss DagstuhlLeibnizZentrum für Informatik, Sept 2011Google Scholar
 8.Chaudhuri, K., Hetzl, S., Miller, D.: A multifocused proof system isomorphic to expansion proofs. J. Logic Comput., June 2014. doi: 10.1093/logcom/exu030 Google Scholar
 9.Chaudhuri, K., Marin, S., Straßburger, L.: Focused and Synthetic Nested Sequents. Technical report, Inria (2016). https://hal.inria.fr/hal01251722
 10.Chaudhuri, K., Miller, D., Saurin, A.: Canonical sequent proofs via multifocusing. In: Ausiello, G., Karhumäki, J., Mauri, G., Ong, L. (eds.) IFIPTCS 2008. IFIP, vol. 273, pp. 383–396. Springer, Heidelberg (2008)Google Scholar
 11.Chaudhuri, K., Pfenning, F., Price, G.: A logical characterization of forward and backward chaining in the inverse method. J. Autom. Reasoning 40(2–3), 133–177 (2008)MathSciNetCrossRefzbMATHGoogle Scholar
 12.Fitting, M.: Prefixed tableaus and nested sequents. Ann. Pure Appl. Logic 163, 291–313 (2012)MathSciNetCrossRefzbMATHGoogle Scholar
 13.Garson, J.: Modal logic. Stanford University, In The Stanford Encyclopedia of Philosophy (2008)Google Scholar
 14.Kashima, R.: Cutfree sequent calculi for some tense logics. Stud. Logica. 53(1), 119–136 (1994)MathSciNetCrossRefzbMATHGoogle Scholar
 15.Laurent, O.: Étude de la Polarisation en Logique. Ph.D. thesis, Univiversit AixMarseille II (2002)Google Scholar
 16.Laurent, O.: A proof of the focalization property in linear logic (2004) (Unpublished note)Google Scholar
 17.Lellmann, B.: Axioms vs hypersequent rules with context restrictions: theory and applications. In: Demri, S., Kapur, D., Weidenbach, C. (eds.) IJCAR 2014. LNCS, vol. 8562, pp. 307–321. Springer, Heidelberg (2014)Google Scholar
 18.Lellmann, B., Pimentel, E.: Proof search in nested sequent calculi. In: Davis, M., Fehnker, A., McIver, A., Voronkov, A. (eds.) LPAR20 2015. LNCS, vol. 9450, pp. 558–574. Springer, Heidelberg (2015). doi: 10.1007/9783662488997_39 CrossRefGoogle Scholar
 19.Liang, C., Miller, D.: Focusing and polarization in linear, intuitionistic, and classical logics. Theoret. Comput. Sci. 410(46), 4747–4768 (2009)MathSciNetCrossRefzbMATHGoogle Scholar
 20.Marin, S., Straßburger, L.: Labelfree modular systems for classical and intuitionistic modal logics. In: Advances in Modal Logic (AIML10) (2014)Google Scholar
 21.McLaughlin, S., Pfenning, F.: Imogen: focusing the polarized inverse method for intuitionistic propositional logic. In: Cervesato, I., Veith, H., Voronkov, A. (eds.) LPAR 2008. LNCS (LNAI), vol. 5330, pp. 174–181. Springer, Heidelberg (2008)CrossRefGoogle Scholar
 22.Miller, D., Nadathur, G., Pfenning, F., Scedrov, A.: Uniform proofs as a foundation for logic programming. Ann. Pure Appl. Logic 51, 125–157 (1991)MathSciNetCrossRefzbMATHGoogle Scholar
 23.Miller, D., Volpe, M.: Focused labeled proof systems for modal logic. In: Davis, M., Fehnker, A., McIver, A., Voronkov, A. (eds.) LPAR20 2015. LNCS, vol. 9450, pp. 266–280. Springer, Heidelberg (2015). doi: 10.1007/9783662488997_19 CrossRefGoogle Scholar
 24.Negri, S.: Proof analysis in modal logic. J. Philos. Logic 34(5–6), 507–544 (2005)MathSciNetCrossRefzbMATHGoogle Scholar
 25.Pfenning, F., Davies, R.: A judgmental reconstruction of modal logic. Math. Struct. Comput. Sci. 11, 511–540 (2001). Notes to an invited talk at the Workshop on Intuitionistic Modal Logics and Applications (IMLA 1999)MathSciNetCrossRefzbMATHGoogle Scholar
 26.Poggiolesi, F.: The method of treehypersequents for modal propositional logic. In: Makinson, D., Malinowski, J., Wansing, H. (eds.) Towards Mathematical Philosophy. Trends in Logic, vol. 28, pp. 31–51. Springer, New York (2009)CrossRefGoogle Scholar
 27.Reed, J., Pfenning, F.: Focuspreserving embeddings of substructural logics in intuitionistic logic. Draft manuscript, January 2010Google Scholar
 28.Simmons, R.J.: Structural focalization. ACM Trans. Comput. Log. 15(3), 21:1–21:33 (2014)MathSciNetCrossRefzbMATHGoogle Scholar
 29.Simpson, A.: The proof theory and semantics of intuitionistic modal logic. Ph.D thesis, University of Edinburgh (1994)Google Scholar
 30.Straßburger, L.: Cut elimination in nested sequents for intuitionistic modal logics. In: Pfenning, F. (ed.) FOSSACS 2013 (ETAPS 2013). LNCS, vol. 7794, pp. 209–224. Springer, Heidelberg (2013)CrossRefGoogle Scholar
 31.Zeilberger, N.: Focusing and higherorder abstract syntax. In: Proceedings of the 35th ACM SIGPLANSIGACT Symposium on Principles of Programming Languages, POPL, San Francisco, California, USA, January 7–12, pp. 359–369. ACM (2008)Google Scholar