Comparing Modal Sequent Systems

Greg Restall

abstract. This is an exploratory and expository paper, comparing display

logic formulations of normal modal logics with labelled sequent systems.
We provide a translation from display sequents into labelled sequents. The
comparison between different systems gives us a different way to under-
stand the difference between display systems and other sequent calculi as
a difference between local and global views of consequence. The mapping
between display and labelled systems also gives us a way to understand
labelled systems as properly structural and not just as systems encoding
modal logic into first-order logic.1

Keywords: Sequents, Display Logic, Labelled Proof System

Labelled systems and display systems are very different generalisations of
the pure sequent calculus, giving what appear to be quite different accounts
of modal deduction. Display sequents are equipped with a rich “structural”
vocabulary, allowing us to directly express modal facts in the punctuation
of a sequent. Labelled sequents encode into the proof theory the structure
of a Kripke model. Formulas are equipped with labels (effectively replacing
formulas by predicates of worlds) and the accessibility relation from the
model makes its appearance in the syntax of the sequent. In this paper, I
show how derivations in display logic may be converted into derivations in a
labelled sequent system, lending some support to the claim that a labelled
sequent system need be no more expressive than a display system. Using
this result, we may we may simplify a labelled proof theory further, so that
the labels disappear and we are left with a different, structural sequent
system for modal logics.

1 Display Logic
In Belnap’s Display Logic [1, 2], as in other sequent systems, we consider
structured collections of formulas, sequents. Here, sequents are of the form
X ` Y , where X and Y are structures, made up from formulas. Structures
are made up of structure-connectives, constructing structures from smaller
structures, in much the same way as formulas are constructed formula-
connectives. Structures and their connectives have a polarity. They can
be either positive or negative structures. In Belnap’s original formulation
1 This research is supported by the Australian Research Council, through grant

DP0343388. See http://consequently.org/writing/comparingmodal for the latest ver-

sion of the paper, to post comments and to read comments left by others.

Paper submitted to Advances in Modal Logic 2006

2 Greg Restall

of display logic, each structure-connective could appear both in negative

and positive position — but connectives could be interpreted differently ac-
cording to the positions in which they appeared. For example, Belnap’s
calculus features a binary structure-connective ◦, which is interpreted as
conjunction-like in an negative position, and disjunction-like in a positive
position. The unary structure-connective ∗ is negation-like in both posi-
tions. However, if ∗X is in negative position, then the X underneath is
positive position, and if ∗X is in positive position, the X underneath is in
negative position. If X is a negative structure, and Y is a positive structure,
then X ` Y is a sequent. The negative parts of the sequent are X and the
proper negative parts of X and of Y . The positive parts of the sequent are
Y and the proper positive parts of X and of Y .
The crucial feature of display logic is the display property. If X ` Y is a
sequent involving Z as a positive substructure, then X ` Y can be trans-
formed into W ` Z using display rules alone, and W depends only on the
position of Z in the original sequent. Similarly if Z is an negative substruc-
ture, then the sequent can be transformed into Z ` V for some appropriate
V . Here, the display rules are a particular class of rules involving only the
structure-connectives alone. In Belnap’s original formulation, the display
rules were as follows:
X ◦ Y ` Z ⇐⇒ X ` ∗Y ◦ Z
X ` Y ◦ Z ⇐⇒ X ◦ ∗Y ` Z ⇐⇒ X ` Z ◦ Y
X ` Y ⇐⇒ ∗Y ` ∗X ⇐⇒ ∗ ∗ X ` Y

In addition to rules that treat structure, there are rules that introduce
connectives. Because of the display property, there is no loss of generality
in assuming that the connective to be introduced can be either the entire
antecedent or the entire consequent of the sequent. For example, these are
Belnap’s original rules for conjunction.

X`A Y `B A◦B `X
X ◦Y `A∧B A∧B `X
In these rules, we can see the way that a ◦ in negative position acts like
conjunction. (Similarly, in positive position, it acts like disjunction.)
In this paper, I will use slightly different rules for conjunction and dis-
junction, to facilitate comparison with Negri’s labelled sequent system [8],
the topic of our next section.

A◦B `Y A`Y B`Y ∗A ` Y

L∧ L∨ L¬
A∧B `Y A∨B `Y ¬A ` Y
X`A X`B X `A◦B X ` ∗A
R∧ R∨ R¬
X `A∧B X `A∨B X ` ¬A
These modified rules are interderivable with Belnap’s rules, given the pres-
ence of the structural rules of weakening and contraction for ◦, which we
 Comparing Modal Sequent Systems 3

shall assume. In each case, the display property is satisfied and the intro-
duced connective is the entire antecedent or the entire consequent of the
sequent structure.
To model modal logics, we will need to consider another example of
structure-connectives, and connective rules. Wansing [13, 14] extended Bel-
nap’s original work by adding a unary structure • (in both antecedent and
consequent position) with display rules

•X ` Y ⇐⇒ X ` •Y

with corresponding modal rules

A`Y X ` •A
L  R 
A ` •Y X ` A
These rules make clear that a • in positive position does the work of . (A
• in negative position does the work of a dual  operator, looking backwards
down the accessbility relation used by .)
With these rules, we may derive valid sequents in the basic normal modal
logic K. Here is an example:

A`A B`B
L 
L
A ` •A B ` •B
weaken weaken
A ◦ B ` •A A ◦ B ` •B
display display
•(A ◦ B) ` A •(A ◦ B) ` B
R∧
•(A ◦ B) ` A ∧ B
display
A ◦ B ` •(A ∧ B)
R 
A ◦ B ` (A ∧ B)
L∧
A ∧ B ` (A ∧ B)

The display rules are used to choreograph the deduction — they place the
formula required as the main connective as the entire left or entire right
side of the consecution.
Belnap gave an account of eight conditions sufficient to guarantee the
admissibility of the rule Cut (from X ` A and A ` Y to infer X ` Y ).
That is, if there are deriviations of X ` A and A ` Y , then there is also a
derivation of X ` Y .
It is a straightforward result that every derivable sequent is valid on a
model — once you define what it is for a sequent to be valid on a model. It
generalises the notion for simple sequents A ` B, where we require that any
world where A is true is a world where B is true. Given a frame hW, Ri —
consisting of a set W of worlds and a binary accessibility relation R on W
— and a relation of truth at worlds, we may define for each structure X
the conditions Pw (X) (the structure X in positive position is true at world
4 Greg Restall

w) and Nw (X) (the structure X in negative position is true at world w). It

is defined inductively:

A ∗X X ◦Y •X
Nw w A ¬Pw (X) Nw (X) ∧ Nw (Y ) (∃v)(vRw ∧ Nv (X))
Pw w A ¬Nw (X) Pw (X) ∨ Pw (Y ) (∀v)(wRv ⊃ Pv (X))

Then a sequent X ` Y is said to be valid on a model if and only if, according

to that model (∀w)(Nw (X) ⊃ Pw (Y )). It is a straightforward induction on
the construction of a derivation that all derivable sequents are valid. For
example, the sequent X ` •Y is valid on a model if and only if on that
model we have

(∀w)(Nw (X) ⊃ (∀v)(wRv ⊃ Pv (Y )))

a straightforward quantifier shift converts this to

(∀v)((∃w)(wRv ∧ Nw (X)) ⊃ Pv (Y ))

which is (∀v)(Nv (•X) ⊃ Pv (Y )), the condition arising from the display
equivalent sequent •X ` Y .
One may model the behaviour of many different modal logics by imposing
new ‘structural rules’ governing the structural punctuation of •, ◦ and ∗.
For example the interchange of •∗ and ∗• underwrites the inference from A
to ♦A.
A`A
display
∗A ` ∗A
L¬
¬A ` ∗A
L 
¬A ` •∗A
sym
¬A ` ∗•A
display
•A ` ∗¬A
R¬
•A ` ¬¬A
display
A ` •¬¬A
R 
A ` ¬¬A
Very many more modal logics may be found by imposing purely structural
rules governing •, ◦ and ∗, preserving the admissibility of cut and the elegant
properties of the display calculus.
Despite these pleasing features, display logic has not been widely used.2
Part of this may be explained in terms of the unique features of display
calculi: systems for modal logics are not merely expansions of classical
Gentzen-style sequent systems, as proofs in the boolean fragment use the
exotic machinery of ∗, and ◦ instead of the familiar sequent structure X ` Y
2 This is not to disparage the work done in the area [4, 6, 10, 11, 14]. However, there

is no doubt that the work in this area has been driven by a small number of researchers.
 Comparing Modal Sequent Systems 5

where X and Y are multisets (or lists) of formulas. This new structure
does not simplify derivations: it complicates them with what seem to be
inessential and bureaucratic choreography which does nothing to expose
the essential deductive steps in a derivation.3 The essential work of the
display property seems to be to ensure that every position in a sequent is
uniform, in that it is available for a cut or for a connective rule. A formula
in a consecution may be displayed, and a displayed position is the site for
a cut or for a connective step. Actually going to the trouble of displaying
a formula in order to process it seems to indicate that we do not have the
most perspicuous mode of formulating our proof theory.
We may highlight another, related, feature of display logic which is the
cause of some dissatisfaction. The cut-elimination result for traditional
sequent systems provides a number of important corollaries, such as the
subformula property, and if we are lucky, decidability and interpolation. In
this case, the decidability result does not give us the same fine degree of
control as elsewhere, for even though we have the subformula property, we
do not have a substructure property, and we have quite a rich structural
vocabulary, instead of the slim vocabulary of the comma on the left and
right in a traditional sequent system. To use a cut elimination argument
to prove decidability of a display calculus is a difficult task [11]. No known
interpolation result has been proved by means of a display calculus.

2 Labelled Sequents
Labelled sequents are a different solution to the issue of giving a proof
theory for modal logics [8]. The core idea of a labelled sequent system is
to internalise into the proof system the relational structure of the Kripke
model. This allows us to construct derivations like this:
v:A`v:A v:B`v:B
L  L 
wRv, w : A ` v : A wRv, w : B, ` v : B
weaken weaken
wRv, w : A, w : B ` w : A wRv, w : A, w : B ` v : B
R∧
wRv, w : A, w : B ` v : A ∧ B
R 
w : A, w : B ` w : (A ∧ B)
L∧
w : A ∧ B ` w : (A ∧ B)

This derivation is, at the one time, simpler than the corresponding display
derivation of A ∧ B ` (A ∧ B), (it has fewer inference steps, leaving out
each display step) and more complex (it introduces labels and the explicit
relational symbol). It appears that have moved from a proof theory of modal
formulas to a proof theory as a tool for reasoning about for modal models.
To make the point we do not need to look at all of the details of la-
belled proof theories. Detail may be found elsewhere [8, 12]. It suffices to
3 A helpful characterisation of the costs and benefits of a display formulation of modal

logics may be found in Phiniki Stouppa’s Master’s thesis [?].

6 Greg Restall

understand a labelled system as one in which the sequents take the form
X ` Y where X and Y are multisets of labelled formulas (of the form x : A
where x is a label and A is a formula) together with relational statements
(of the form xRy). Axiomatic sequents take the form x : A ` x : A.4 For
connective rules we have
x : A, x : B, X ` Y x : A, X ` Y x : B, X ` Y X ` Y, x : A
L∧ L∨ L¬
x : A ∧ B, X ` Y x : A ∨ B, X ` Y x : ¬A, X ` Y

X ` Y, x : A X ` Y, x : B X ` Y, x : A, x : B x : A, X ` Y
R∧ R∨ R¬
X ` Y, x : A ∧ B X ` Y, x, A ∨ B X ` Y, x : ¬A
x : A, X ` Y xRy, X ` Y, y : A
L R 
yRx, y : A, X ` Y X ` Y, x : A
where the last rule has the side condition that y does not appear in X ` Y .
Notice that in these rules, relational statements appear only on the left of
the sequent. We may without loss of deductive power, restrict our attention
to sequents in X ` Y which relational statements appear only in X and not
in Y .
Just as with display logic, we may extend the system with rules governing
the distinctive modal machinery (here R) to encode different modal systems.
Negri [8] shows how different conditions on R may be added as rules without
breaking the admissibility of cut (or indeed the admissibility of contraction
and weakening in her G3-style system). We will not go through the detail
of these conditions here.
Just as with display logic, we can understand what it is for a sequent to
be valid on a model model. In this case, the translation is much simpler.
To translate the sequent X ` Y we replace each x : A by x A, we
replace the multiset X by its conjunction; Y by its disjunction; the ` by
a conditional, and you universally quantify over all world labels. So, the
sequent xRy, x : A ` y : B, x : C is valid on a frame if and only if

(∀x, y)((xRy ∧ x A) ⊃ ((y B) ∨ (x C)))

Notice that the rules here do not satisfy the subformula property if we
take relational facts to be formulas, as xRy appears in the premise of R
but not in the conclusion. We could repair this in two ways. One is to
take relational facts to not be formulas properly so-called, or to take R the
predicate to be present as a ‘part’ of the operator , as it would be if we
were to rewrite the modal rules as explicit special cases of the quantifier
rules in first-order logic:
x : A, X ` Y xRy, X ` Y, y : A
L 0 R 0
yRx, (∀z)(yRz ⊃ z : A), X ` Y X ` Y, (∀z)(xRz ⊃ z : A)
4 One could take them instead to have the form x : A, X ` Y, x : A if we wish to

eliminate weakening as an explicit structural rule.

 Comparing Modal Sequent Systems 7

where we translate x : A by (∀z)(xRz ⊃ z : A). Then, clearly, a kind of

subformula property is satisfied.
However, satisfying the subformula property in this way seems unsatis-
factory. For one thing, we still have no genuine subformula property since
we still have to deal with renaming variables. The formulas xRy and y : A
are not subformulas of the formula (∀x)(xRz ⊃ z : A) in any straightforward
sense. Reworking our proof-theoretical analysis to deal with variables and
quantification seems like a high price to play to deal with modal inference
which does not explicitly mention such things. The alternative, of course,
is to attempt to view the sequent system in such a way that the variables
and relational statements occur as the structure of the sequent, and not as
its content. After all, we have seen that we are not making use of the full
power of first-order logic. Relational statements need only appear on the
left side of the sequent. They are not compounded with other statements.
They are used only to control the application of L and R.
Another shortcoming of the labelled system is that it is still not a straight-
forward extension of classical propositional logic. A classical propositional
sequent with no modal operators is not derived with the traditional sequent
derivation, but with a sequent derivation littered with redundant world la-
bels. While the mismatch with traditional sequent systems is not as great,
it is still there.

3 From Display Sequents to Labelled Sequents

Before attempting to resolve issues with either display logic or labelled se-
quent systems, we will attempt to understand the relationship between
them. Consider the way that sequents in either system relate to mod-
els. A labelled sequent Rxy, x : A ` y : B is valid on a model just when
(∀x, y)((Rxy ∧ x A) ⊃ y B). This condition on a model is expressed
by (at least) two different display sequents: •A ` B and A ` •B. This
suggests that we may find a way to translate display sequents into labelled
sequents, and that this translation will be many-to-one. If we can do this,
then it may point to a way to do away with the redundancy in a display
system.
One way to go from a display sequent X ` Y to a labelled sequent is
to use the translation Nw (X) ` Pw (Y ). Unfortunately, N and P produce
statements in the first-order theory of models, and not sequences of labelled
formulas and relational facts. We will need to modify our tranlsation in
order to produce a labelled sequent. This takes some care, the antecedent
structure X might have substructures in postive position (those under an
odd number of asterisks), and the result of translating these will quite pos-
sibly need to go into the right of the labelled sequent. So, the translation
of X ` Y is the sequent
nlw (X), plw (Y ) ` nrw (X), prw (Y )
where nlw , plw , nrw , prw are ancillary translation functions, defined recursively.
The functions nlw and nrw are used to give us the significance of a display
8 Greg Restall

structure appearing in negative position in the display sequent. nlw (X) is

the contribution X makes on the left side of the labelled sequent, considered
as true at world w. nrw (X) is the contribution that this structure makes to
the right side of the labelled sequent, again taking the world of evaluation to
be w. Similarly, plw (Y ) and prw (Y ) are the contributions of the structure Y ,
appearing in positive position in the display sequent, in the left and right of
the labelled sequent, respectively. In each case, the output of each function
is a multiset (possibly empty) of formulas and relation statements. Now we
may present the definitions of the four functions:

A ∗X X ◦Y •X
nlw w : A plw (X) nlw (X), nlw (Y ) vRw, nlv (X)
nrw − prw (X) nrw (X), nrw (Y ) nrv (X)
plw − nw (X) pw (X), pw (Y ) wRv, plv (X)
l l l

prw w : A nrw (X) prw (X), prw (Y ) prv (X)

In the case for •X, the world variable v is fresh.5

The display sequent • ∗ (A ◦ ∗•B) ` ∗(D ◦ E) is translated as follows:

nlw (• ∗ (A ◦ ∗•B)), plw (∗(D ◦ E)) ` nrw (• ∗ (A ◦ ∗•B)), prw (∗(D ◦ E))

after you trace through the inductive definitions, you get to:

vRw, uRv, u : B, w : D, w : E ` v : A

Notice that the structure • ∗ (A ◦ ∗•B) deposits material on the right side of
the turnstile (v : A) and the left side (w : B), and the w and v are related
by a relational fact vRw.
Now we come to the first fact.
FACT 1. A display sequent is valid on a model if and only if its translation
as a labelled sequent is also valid on that frame.

Proof. A straightforward induction on the construction of the translation.

The simplest technique is to consider when a sequent fails in a model. For
the display sequent X ` Y itself, we need some w where Nw (X) ∧ ¬Pw (Y ).
For the translation as a labelled sequent, we need a w where nlw (X) ∧
¬nrw (X) ∧ plw (Y ) ∧ ¬prw (Y ) is true. To prove the equivalence, it suffices to
prove by induction that ¬Pw (Y ) holds if and only if plw (Y ) ∧ ¬prw (Y ) and
Nw (X) if and only if nlw (X)∧¬nrw (X). (The free variables in the evaluation
of n and p are to be treated as implicitly existentially quantified.) Proving
the equivalence is a simple induction on the construction of X and of Y . 
5 Stating this condition precisely is not straightforward. We need, in fact, to keep a

stack of world labels available for substitution throughout the translation, and pass the
stack from one stage of the translation to another. I presume that the reader enough
familiarity with translations into first-order logic to sidestep these fiddly details for the
sake of ease of exposition.
 Comparing Modal Sequent Systems 9

Notice that the two display sequents •A ` B and A ` •B are translated as

very similar labelled sequents:

vRw, v : A ` w : B wRv, w : A ` v : B

For the first translation, we evaluate B using our first world label w, and
then step backwards to v to find the point of evaluation for A. For A ` •B,
on the other hand, we evaluate A at w, and move forward to a new world
v to find B. The two display equivalent sequents are translated by labelled
sequents differing only in the identity of the labels and not the structure
of the sequent. If we think of the labelled sequent as ‘starting’ at w, then
•A ` B ‘says’ that if we have world that can access this world, where A is
true then B is true here. A ` •B says, on the other hand, that if A is true
here then for any world accesible from here, B is true.
Both “facts” are unproblematic ways of stating the same thing. Con-
sidered as validities on a model, there is nothing to stand between them, as
we did not pick out any particular point of evaluation. As a matter of fact,
no world mentioned in a labelled sequent is “here.” Labelled sequents take
a global view of a model, not picking out any particular point as a starting
point. Display sequents are no less general, but they express the validity of a
deduction in a local manner by distinguishing (as sequents) the fact •A ` B
(thinking of A worlds as ancestors of this B world) and A ` •B (thinking
of B worlds as descendents of this A world). The position of the turnstile
marks the location of the “you are here” marker in the modal model.
So, let us be a bit liberal concerning the identity of the labels in a labelled
calculus. It is clear that the fact vRw, v : A ` w : B as a modal sequent
is no different to the fact wRv, w : A ` v : B, as both are valid if and
only if they hold for each w and v. The world labels in labelled sequents
are universally quantified6 and the particular labels we use are no matter,
provided that we keep different labels different.
Now, consider what happens when we translate each of the display rules in
our system. The rules for conjunction and disjunction are quite trivial, and
become, unproblematically, the rules for conjunction and negation in the
labelled system. Negation might appear to be different, under translation
it, too, becomes the labelled rule. Consider L¬. ∗A ` Y is translated into
nlw (∗A), plw (Y ) ` nrw (∗A), prw (Y ) which is plw (A), plw (Y ) ` prw (A), prw (Y ).
plw (A) disappears, but prw (A) is w : A, so the premise becomes

plw (Y ) ` prw (Y ), w : A

which looks just like the premise of the labelled L¬ rule. The conclusion is
more straightforward, and it becomes

w : ¬A, plw (Y ) ` prw (Y )

6 At least, if they are read as expressing validity. If you are pessimistic and look for

invalidity you may think of them as existentially bound. The point is no difference
10 Greg Restall

Now, the structures plw (Y ) and prw (Y ) are arbitrary (we can choose Y how-
ever we like, to put any material in the left and right of this sequent7 ) so
this rule is just as general as the original labelled L¬ rule.
The same thing happens with the modal rules. Consider the display rule
L. The premise A ` Y is translated as

w : A, plw (Y ) ` prw (Y )

the conclusion A ` •Y becomes

w : A, wRv, plv (Y ) ` prv (Y )

which, is not exactly the labelled rule L, but comes close. If we do not care
about the identity of labels from premise to conclusion and hare happy to
relabel the conclusion as v : A, vRw, plw (Y ) ` prw (Y ) then we would have
exactly the rule required. Similarly, for R. The premise X ` •A becomes

wRv, nlw (X) ` nrw (X), v : A

where v is not present in nlw (X) and nrw (X) (it is a label introuduced in the
translation of •A). The conclusion X ` A is

nlw (X) ` nrw (X), w : A

which is precisely the conclusion of the R rule. In this explanation, we

have gone quite some way to proving the following fact.
FACT 2. Any modal display derivation may be transformed step by step
into a labelled derivation, using this translation, modulo some relabelling.
‘Display’ steps in a display inference are redundant in the labelled deriva-
tion.
To complete this proof we need to show that structural rules in the display
calculus may be treated in the labelled sequent calculus. Different choices
are available in each system, and provided that the effect of contraction
and weakening is provided for in each system, this will work. The details
are tedious and will be skipped here, as our topic is the significance of the
choices of different sequent structure, and not the detail of
In the same vein, we shall not tarry to consider the effect of different
modal rules in the display calculus and different sequent rules governing the
accessibility relation. This is a rich and interesting area, as there are very
many things one can do in display calculi [7].

4 Delabelling Labelled Sequents

We have seen that the labels in a labelled caclulus have their drawbacks. Not
only do the labels break the subformula property, they also make translation
from the display calculus less straightforward than they might be. In fact,
7 Though note, we put R statements only on the left, never on the right.
 Comparing Modal Sequent Systems 11

the display calculus and the labelled system both suffer from bureaucracy at
the same point. In display calculus, we have two ways of expressing the one
fact about a model, •A ` B and A ` •B. In the labelled system, we have
many more: wRv, w : A ` v : B, vRw, v : A ` w : B, vRx : v : A ` x : B,
and so on. It is limited only by the number of world labels available.
We could get around this needless multiplicitly by requiring that each
labelled sequent be canonically labelled. Take a labelled sequent, replace
the first label by w0 , everywhere it appears, the next one (other than w0 )
by w1 , the third by w2 , etc. Then this modal fact is expressed in only one
way: w0 Rw1 , w0 : A ` w1 : B.
An approach like this solves the multiplicity problem, but it does nothing
for the subformula property. Instead, let us see the behaviour of R and
the labels as a part of the structural furniture of a sequent rather than
its content. For each labelled sequent of the form R, X ` Y where R is a
collection of R statements, and X and Y are labelled formulas. The content
of the sequent is given by the formulas in X and in Y . For each formula we
have two pieces of information: the place on the network of “points” given
by R and the polarity, given by its position, in X (in negative position)
or Y (in positive position). Let us think of a different way of representing
this information, without requiring labels. R is a directed graph with a
node for each label occurring in the sequent, and an arc from w to v when
wRv is in R. Then, the label on a formula in X or in Y tells us where the
formula can occur on the directed graph. If we have w : A in X we put A
in “antecedent” position at the w node in the graph. If w : B is in Y we
put w : B in “consequent” position at the w node of the graph. How can
we represent this? A straightforward way is to represent a sequent at the
node, with antecedent formulas on the left and consequent formulas on the
right. Once we have sequents at each node of the graph, we may rub out the
labels. We have a directed graph of (traditional) sequents. For exmample,
the labelled sequent wRv, w : A ` v : B becomes the graph of sequents

A` `B

The sequent vRw, uRv, u : B, w : D, w : E ` v : A we saw before, coming

from the display sequent • ∗ (A ◦ ∗•B) ` ∗(D ◦ E) becomes

B` D, E ` `A

This is not a new sequent structure, it is merely a new way of representing

the labelled sequent structure, pushing the relational statements and labels
into the syntax of the proof theory, leaving the formulas to remain as the
content.8 The derivation of the sequent A ∧ B ` (A ∧ B), rendered in
8 As far as I can tell, the detail here is new, and the claim that this structure is simply

a de-labelling of labelled sequent structures. However, the idea of using a graph structure
on sequents is not new [5, 3].
12 Greg Restall

this format becomes

A`A B`B
L  L 
A ` `A B ` `B
weaken weaken
A, B ` `A A, B ` `B
R∧
A, B ` `A∧B
R 
A, B ` (A ∧ B)
L∧
A ∧ B ` (A ∧ B)
And the derivation using a structural rule becomes straightforward:
x:A`x:A A`A
L¬ L¬
x : ¬A, x : A ` ¬A, A `
L 
L
Ryx, y : ¬A, x : A ` ¬A ` A`
R¬ R¬
Ryx, x : A ` y : ¬¬A ` ¬¬A A`
sym sym
Rxy, Ryx, x : A ` y : ¬¬A ` ¬¬A A`
R  R 
x : A ` x : ¬¬A A ` ¬¬A
The subformula property is now straightforward, and we no longer have a
proliferation of ways to represent the one modal fact. The resulting sim-
plification of the labelled sequent system (either by eliminating labels or
by choosing a canonical labelling) is less bureaucratic than either display
logic or the traditional labelled sequent system. We can acknowledge the
behaviour of R as the behaviour of the structure of our modal deduction.
Another benefit of this approach is the relation with the nonmodal pro-
positional sequent calculus. If we identify a sequent with the one-point
graph of that sequent (with no arrow at all), then there is no difference
between a proof of a classical nonmodal sequent in this sequent system
when compared with a traditional classical sequent system with no modal
features. We do not need to modify anything of the nonmodal system. The
extensions model the new vocabulary and nothing else.
Furthermore, if we impose the constraints of reflexivity, symmetry and
transitivity on R to find S5, the graph structure becomes trivial. Whenever
a new node in our graph of sequents is introduced, it is implicitly and
automatically taken to be related to all of the other nodes in the graph
of sequents. Our graph is universal. Instead of a graph of sequents we
need keep track of only a multiset of sequents. The result is the simple
hypersequent calculus for S5. It is the simplest labelled calculus for S5,
without requiring any relational facts and without all labels. All we need
do is segregate the formulas in different nodes as required for the modal
inference [9]. We have the following modal rules

A, X ` Y | S X`Y | `A | S
L  R 
A ` | X ` Y | S X ` Y, A | S
 Comparing Modal Sequent Systems 13

where S is a multiset of sequents. Our deductions become quite simple.

A`A
A`A L¬
 L A, ¬A `
A ` | ` A 
L
R A ` | ¬A `
A ` | ` A R¬
R A ` | ` ¬¬A
A ` A R 
A ` ¬¬A
BIBLIOGRAPHY
[1] Nuel D. Belnap. Display logic. Journal of Philosophical Logic, 11:375–417, 1982.
[2] Nuel D. Belnap. Linear logic displayed. Notre Dame Journal of Formal Logic, 31:15–
25, 1990.
[3] Claudio Cerrato. Modal sequents. In Heinrich Wansing, editor, Proof Theory of Modal
Logic, pages 141–166. Kluwer Academic Publishers, Dordrecht, 1996.
[4] Stéphane Demri and Rajeev Goré. Display calculi for logics with relative accessibility
relations. Journal of Logic, Language and Information, 9:213–236, 2000.
[5] Kosta Došen. Sequent-systems for modal logic. Journal of Symbolic Logic, 50:149–168,
1985.
[6] Rajeev Goré. Substructural logics on display. Logic Journal of the IGPL, 6(3):451–
604, 1998.
[7] Marcus Kracht. Power and weakness of the modal display calculus. In Heinrich Wans-
ing, editor, Proof Theory of Modal Logic, pages 93–121. Kluwer Academic Publishers,
Dordrecht, 1996.
[8] Sara Negri. Proof analysis in modal logic. Journal of Philosophical Logic, 34:507–544,
2005.
[9] Greg Restall. Proofnets for s5: sequents and circuits for modal logic. To appear in
the Proceedings of the Logic Colloquium 2006.
[10] Greg Restall. Display logic and gaggle theory. Reports in Mathematical Logic, 29:133–
146, 1995.
[11] Greg Restall. Displaying and deciding substructural logics 1: Logics with contraposi-
tion. Journal of Philosophical Logic, 27:179–216, 1998.
[12] Phiniki Stouppa. The design of modal proof theories: The case of S5. Master’s thesis,
Technische Universität Dresden, 2004.
[13] Luca Viganò. Labelled Non-Classical Logics. Kluwer, 2000.
[14] Heinrich Wansing. Sequent calculi for normal propositional modal logics. Journal of
Logic and Computation, 4:125–142, 1994.
[15] Heinrich Wansing. Displaying Modal Logic. Kluwer Academic Publishers, Dordrecht,
1998.

Greg Restall
Philosophy Department
The University of Melbourne
Parkville, 3010, Australia
[email protected]