2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p.

Notre Dame Journal of Formal Logic

Advance publication}
1
2
3
4
5
6
7
8
9
10
Geometric Modal Logic
11
12
13 Brice Halimi
14
15
16 Abstract The purpose of this paper is to generalize Kripke semantics for
17 propositional modal logic by geometrizing it, that is, by considering the space
18 underlying the collection of all possible worlds as an important semantic feature
19 in its own right, so as to take the idea of accessibility seriously. The result-
20 ing new modal semantics is worked out in a setting coming from Riemannian
21
geometry, where Kripke semantics is shown to correspond to a particular case,
namely, the discrete one. Several correspondence results, established between
22
variants of well-known modal systems and corresponding geometric properties,
23
illustrate the import of this new framework.
24
25
Each Kripke structure for propositional modal logic consists of a graph whose
26
vertices are elements of a set W of “possible worlds” and whose edges indicate which
27
worlds are “accessible” from which. Such a structure is entirely discrete: the edges
28
of the graph are figurative and simply represent the purely set-theoretic extension of
29
an accessibility relation R. Could the idea of accessibility be taken more seriously,
30
as the existence of an actual path? It would imply the consideration of the space
31
underlying all possible worlds in a Kripke frame hW; Ri as a semantic feature in its
32
own right. Kripke modal semantics takes that space to be the set W itself, namely,
33
a discrete space. But what if W is introduced in a more general way as embedded
34
within a geometric structure right from the start—Kripke semantics corresponding
35
to the case where that structure is discrete?
36
What follows will be devoted to building up a semantics for propositional modal
37
logic along these lines. As will be seen presently, the corresponding geometric
38
framework naturally associates each possible world with a collection of higher-level
39
worlds specifying its infinitesimal neighborhood. The procedure iterates: for each
40
41
42 Received April 5, 2019; accepted July 11, 2023
}
43 Final volume, issue, and page numbers to be assigned.
44 2020 Mathematics Subject Classification: Primary 03A45, 53C21, 53C22; Secondary
45 53A99, 53C05, 70G45
46 Keywords: propositional modal logic, Kripke modal semantics, iterated modalities,
47
higher-order possible worlds, differential geometry, Riemannian geometry
48
© 2023 by University of Notre Dame 10.1215/00294527-2023-0012
49
1
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 2

2 Brice Halimi

1 addition of a new modal clause, new possible worlds will be considered at higher
2 and higher levels with respect to some already given possible world, the latter being
3 thought of as the index of a modal level in discourse. Thus, higher-order modalities
4 will be understood on the model of higher-order derivatives. The sense in which each
5 possible world constitutes, in Kripke semantics, a local point of view on the system
6 of all possible worlds will thereby be enriched with a fully geometric meaning, and
7 Kripke semantics generalized in a substantial way, through new connections between
8 modal logic and geometry.
9
10 1 Beyond Kripke Modal Semantics
11
1.1 Kripke modal semantics Although they are well known, let us recall first, for
12
the sake of completeness, the basic features of the semantics for propositional modal
13
logic introduced by Kripke [14] in 1963.1
14
15 Definition 1.1 Given a fixed set of propositional variables p, q, r; : : : , the set
16 of formulae of propositional modal logic is defined inductively as follows: (i) all
17 propositional variables are formulae; (ii) if ' is a formula, so is :'; (iii) if ' and
18 are formulae, so are .' ^ / and .' _ /; (iii0 ) .' ! / 2 F is defined as
19 .:' _ /; (iv) if ' is a formula, so are Þ' (to be read, “Possibly, '”) and ' (to
20 be read, “Necessarily, '”). (As usual, ‘Þ’ and ‘’ are considered as interdefinable
21 dual operators: ' WD : Þ :' and Þ' WD ::'. So only either one needs to be
22 mentioned in the language.)
23 Definition 1.2 The modal depth d.'/ of a formula ' is defined inductively as
24 well: (i) for any propositional variable p, d.p/ D 0; (ii) d.:'/ D d.'/; (iii)
25 d..' ^ // D d..' _ // D max.d.'/; d. //; (iv) d.Þ'/ D d.'/ D d.'/ C 1.
26
27
Definition 1.3 The axioms of the system K of modal logic include all propo-
28
sitional tautologies as well as the axiom K: .p ! q/ ! .p ! q/. The
29
inference rules of K are modus ponens, the necessitation rule (if ' is a theorem, so is
30
'), and the rule of uniform substitution (if a formula .p/ containing p is a theo-
31
rem, so is Œ'=p, for any formula '). A system of modal logic is said to be normal
32
if it is an extension of K. In particular, the addition of the axiom T: p ! p, defines
33
the system T D KT. The further addition of the axiom 4: p ! p (resp., of the
34
axiom 5: Þp !  Þ p) defines the system S4 D T4 (resp., S5 D T5).
35 Definition 1.4 A Kripke frame for propositional modal logic consists of a pair
36 F D hW; Ri, where W is a set, whose elements are called “possible worlds” (among
37 these, a given world w0 may be distinguished as being the “actual world”), and R is
38 a binary accessibility relation on W , the intended meaning of wRw 0 being that the
39 world w 0 represents a plausible possible world from the point of view of w.
40
41
Definition 1.5 Given a Kripke frame F D hW; Ri, a valuation V on F is the
42
assignment, to each propositional variable p, of a subset V .p/ of W , that indicates
43
the set of all possible worlds in which p is true. A Kripke model is a triple M D
44
hW; R; V i resulting from a frame being endowed with a valuation on it.
45 Definition 1.6 Given a Kripke model M D hW; R; V i and a formula ', the
46 satisfaction condition M; w  ' is defined inductively as follows:
47
(i) M; w  p iff w 2 V .p/,
48
(ii) M; w  :' iff M; w ² ',
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 3

Geometric Modal Logic 3

1 (iii) M; w  .' ^ / iff M; w  ' and M; w  ,

2 (iv) M; w  .' _ / iff M; w  ' or M; w  ,
3 (v) M; w  Þ' iff there exists w 0 2 W such that wRw 0 and M; w 0  ',
4 (vi) M; w  ' iff, for every w 0 2 W , wRw 0 implies M; w 0  '.
5
6 Definition 1.7 A formula ' is said to be valid in a Kripke model M, written
7 M  ', iff M; w  ' for every w 2 W . It is said to be valid in a Kripke frame
8 F , written F  ', if it is valid in every Kripke model whose underlying frame is
9 F ; and valid in a class K of Kripke frames, written K  ', if it is valid in every
10 F 2 K.
11 A modal system ƒ is said to be complete with respect to a class K of Kripke
12 frames when, for any modal formula ', ƒ ` ' iff K  '.
13
It is easy to see that the Kripke frames in which the axiom scheme 4 is valid are
14
exactly those whose accessibility relation is transitive. In the same way, the axiom T
15
corresponds to the class of all reflexive Kripke frames. Kripke semantics is partic-
16
ularly appropriate, by virtue of its algebraic simplicity, for obtaining such “corre-
17
spondence results” (see [4, Sections 4.3–4.4]). This is certainly one of the reasons
18
why Kripke semantics has been elected as a central framework, by logicians, and by
19
philosophers as well.
20
One related major interest of Kripke semantics is its ability to interpret modal
21
iteration, that is, the superposition of modal clauses, and to tame it at the same time,
22
by linearizing it:
23
24 M; w  p iff, for every v; u 2 W such that wRv and vRu, M; u  p.
25
The unfolding of the extension (i.e., of the graph) of the accessibility relation dis-
26
charges the interpretation of modal iteration: the higher the modal depth of the for-
27
mula, the longer the paths in the graph to consider. Thus, modal iteration trans-
28
lates into a ramification: each possible world w gives rise to the range E 1 .w/ WD
29
¹w 0 2 W W wRw 0 º of all the accessible worlds hereof, then to the range E 2 .w/ WD
30
¹E 1 .w 0 / W w 0 2 E 1 .w/º of ranges of possible worlds, and so on, and hence to local
31
accessibility systems hE n .w/ W n 2 Niw2W .
32
Modal iteration has a very strong meaning: for instance, saying that a proposition,
33
although impossibly true, is possibly possibly true (i.e., could have been possibly
34
true) amounts intuitively to saying that the whole range and configuration of the pos-
35
sible itself could have been such as to include that proposition as possibly true. In
36
other words, describing a state of affairs as merely “possibly possible,” one allows a
37
much wider variation than in the case of possibility simpliciter, because one implic-
38
itly varies the variation system itself to which the notion of possible refers: one shifts
39
from a given system of possibility into a context in which that system is relocated as
40
being only one among others, and states that with respect to some other system, such
41
and such state of affairs becomes possible.
42
Let us call such a widening, a modal change of scale: in terms of possible worlds,
43
it prompts one to consider different possible specifications of the collection of all
44
possible worlds, in other words different second-order possible worlds, and more
45
generally higher-order possible worlds. Kripke semantics deals with possible worlds
46
lying all at the same level and formalizes higher-order possible worlds in that setting.
47
Is that enough? For both philosophical and technical reasons to be adduced soon, this
48
paper will opt for a geometric modal semantics meant to both give a more radical
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 4

4 Brice Halimi

1 account of modal iteration and generalize Kripke semantics, by modeling higher-

2 order possible worlds on higher-order derivatives, so as to introduce an open-ended
3 collection of nested ranges of possible worlds lying at higher and higher levels.
4
5 1.2 Previous developments Before getting down to that semantics, a look at previous
6 formalizations of modal change of scale is in order. Indeed, multileveled modal struc-
7 tures have already been explored, in particular in the context of attempts at “gentz-
8 enizing” modal logic. Thus, Andrea Masini [16] introduced a derivability relation
9 between “2-sequents” represented as vertical lists of ordinary sequents. As Masini
10 puts it, modalities “internalize” the level structure of 2-sequents: a 2-sequent can
11 be seen as a setting of different levels, each occurrence of a formula A being con-
12 vertible into a modalization A at the next upper level—which suggests intuitively
13 identifying level lines inside a 2-sequent with modal iteration levels.2
14 The semantic counterpart of that impetus coming from proof theory is a modal
15 semantics in which each introduction of a modal operator sets the evaluation to take
16 place at a higher level: starting from possible worlds of level 0, the evaluation of a for-
17 mula of modal depth n (for n large enough) would involve possible worlds of level n,
18 each of which being relative to some world of level n 1, itself relative to some world
19 of level n 2, and so forth. There are at least two areas in the field of modal logic
20 which can be traced back to such idea of increasing levels: the “hypermodal logics”
21 designed by Dov Gabbay [8], and the “structural theory of sets” proposed by Alexan-
22 dru Baltag [1]. But Gabbay’s enterprise seeks to combine different interpretations of
23 the necessity operator (switching from one model to another), rather than to iterate
24 the same interpretation along some scale-changing process. As to Baltag’s article, it
25 mainly uses modal logic as a way to describe the graph of the membership relation
26 that characterizes the structure of a (not necessarily well-founded) set—which has
27 little to do with the present issue.
28 A straightforward analogue for a Kripke frame hW; Ri where the accessibility
29 relation, for the sake of simplicity, is taken to be universal (R D W  W ) would
30 be as follows. Set } 0 .W / D W and } nC1 .W / D }.} n .W // (for every n 2 N).
31 Then define a satisfaction relation by setting, for X 2 } n .W /: M; X   iff,
32 for all Y 2 } nC1 .W /, X 2 Y implies M; Y  . But this does not as such yield
33 a semantics before it is specified how to evaluate atomic formulae at elements of
34 an arbitrary high level } n .X/. So one should find a way of defining the truth of
35 p at a set X 2 } n .W / in terms of the truth of p at some or all elements of X.
36 One conceivable option to provide such a definition might be saying that p is true
37 at X 2 } n .W / iff p is true at some Z 2 X. This clause, however, looks like what
38 the truth conditions of Þp, not of p, should be, which trivializes the interpretation
39 of modal operators. So the use of the set-theoretic hierarchy .} n .W //n2N , although
40 natural and simple, is not rich enough to provide one with the means of lifting the
41 interpretations of propositions to higher levels in a canonical way. Other uses might
42 be found, but they remain to be found and, at any cost, would certainly be less natural
43 and less simple.
44 A more complicated set-theoretic option could be to interpret modal iteration
45 along a chain U0 2 U1 2 U2 : : : such that, for n  1, each x 2 Un is a set endowed
with a binary relation Rx , and such Q that, for every n  0, each z 2 Un is associ-
46
47 ated with a choice function fz 2 w2UnC1 w. The elements of each Un are called
48
n-worlds; possible worlds in the usual sense are the 0-worlds; and, for z 2 Un , fz
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 5

Geometric Modal Logic 5

1 gives the counterpart of z in each .n C 1/-world. Let us suppose that a valuation on

2 each member of each Un , for every n  1, is given: U1 becomes a family of models,
3 U2 a family of models as well as a family of families of models, and so on. One then
4 defines:
5
U0 ; z  Þp iff there is w 2 U0 such that zRU0 w and U0 ; w  p.
6
7
As a consequence:
8 U0 ; z  Þ Þ p iff there is V 2 U1 such that U0 RU1 V and V; fz .V /  Þp
9 iff there is V 2 U1 and w 2 V such that U0 RU1 V , fz .V /RV w and V; w  p.
10 But the main problem is that there is no naturalSway to specify the sets Un (for all
11 n 2 N) and the choice functions fz (for all z 2 n2N Un ).
12 Actually, the closest development to the idea of modal change of scale is Kit Fine’s
13 analysis of higher-order vagueness (in the sense that some case being vague may
14 itself turn out to be vague, and indefinitely so).3 The key feature of Fine’s semantics
15 is that higher-order boundaries are not linear sequences of nodes lying all on the
16 same level, but sequences of nested points—which comes close to the idea of change
17 of scale. What follows can actually be understood as transposing Fine’s higher-order
18 boundaries to a modal context, by replacing them with higher-order possible worlds.
19
20
1.3 Toward another semantics But how could a mere set of possible worlds make
21
up a higher-order possible world? Is not a world made of individuals and properties,
22
instead of worlds? How could a structured unified world be compounded of multiple
23
isolated possible worlds? Lumping the worlds into some single “grand” world will
24
not do, because a higher-order possible world is supposed to be a disjunctive system
25
of alternative worlds, considered itself as a possible datum in its own right.
26
The default picture is that a higher-order possible world is a higher-level possible
27
world: a world which is composed of worlds on a lower level, and whose order
28
coincides with its level, or set-theoretic rank. This picture of nested sets of worlds,
29
each of which is a member of a world of the next upper level, is maybe a useful
30
start, but it runs right into the problem just mentioned that a higher-order world, thus
31
construed as a set of (lower-level) worlds, cannot constitute any world.
32
So let us try to turn it upside down: this means thinking of each possible world, not
33
as an element of some higher-level world, but as a set of higher-level possible worlds.
34
Then a possible world becomes a multiplicity of worlds insofar as the analysis of a
35
certain structure at that world uncovers such a multiplicity—exactly as mathematical
36
analysis uncovers, at a point, all the possible Taylor approximations of functions
37
having that point as value. Following this new scheme, upper-level possible worlds
38
should be conceived of as specifications of lower-level possible worlds, and each
39
lower-level world as being virtually the set of all its possible specifications at the
40
next level, thus becoming a virtual higher-order world. The resulting picture is more
41
convenient, because it is much easier to analyze an internal structure at some basic
42
world than to try to build up superstructures on top of basic worlds without having
43
any clue how to do it.
44
As a result, the level and the order of a possible world have to be sharply dis-
45
tinguished. A higher-order possible world is a basic possible world analyzed as a
46
range of possible worlds lying on higher levels. So, instead of making up a world
47
with a multiplicity of worlds, the idea is to reveal, attached to any possible world
48
given at the onset, a new space of worlds all lying on the next level. The problem of
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 6

6 Brice Halimi

1 explaining how a higher-order world could make up a world disappears: such a world
2 corresponds to a system of higher-level worlds, but remains a basic-level world.
3 In the new picture that has just been adumbrated, a higher-order system of pos-
4 sibility is not a possible world of possible worlds, but a possible world viewed as
5 the index of a collection of higher-level possible worlds. Given possible worlds of
6 different levels n D 0; 1; : : : , a world of order 1 is a world of some level n 2 N to
7 which worlds of level n C 1 are attached; a world of order 2 is a world of some level
8 n 2 N to which worlds of level n C 1 and order 1 are attached; and so on. Because
9 the order just corresponds to the depth of a succession of layers, the interpretation
10 of a formula of modal depth n  1 should involve nested worlds of respective levels
11 0; 1; : : : ; n and respective orders n; n 1; : : : ; 0.
12 But how to build up a stratified structure within which each possible world w gives
13 rise to a system of possible worlds intrinsically relative to w? Insofar as higher-level
14 possible worlds are supposed to specify some lower-level possible world, it seems
15 quite natural to draw on mathematical analysis and to compare the lower-level world
16 to some value in a space and the upper-level worlds to all the possible derivatives
17 of functions having that value (at 0, say); in other words, to model the order of a
18 possible world on the order of a Taylor expansion. Still, it remains to explain how to
19 introduce worlds of different levels and how a set of worlds of a given level n C 1
20 can represent a coherent system of possibility attached to some base world of level n
21 (for each n 2 N).
22 This is where differential geometry, extending the theory of Taylor expansions,
23 comes on the scene. The hypothesis to be pursued is that the tangent space, at a given
24 point, to a differentiable manifold fits the description of that system. A differentiable
25 manifold (or manifold, for short) of dimension n, say, is a geometric entity M (such
26 as the sphere) which locally looks like a piece of Rn : it can be covered by copies
27 Ui of pieces of Rn (i ranging some indexing set I ) along “charts” 'i W Ui  M !
28 'i .Ui /  Rn in such a way that two charts agree wherever they overlap, in the sense
29 that for any i; j 2 I , 'j ı'i 1 W 'i .Ui \Uj / ! 'j .Ui \Uj / is a diffeomorphism (see
30 Lang [15, pp. 22–23] or Gallot, Hulin, and Lafontaine [9, p. 6]). So a manifold has
31 to be thought of as a geometric locus embodying in its very definition its description
32 through an “atlas” of charts which simply generalizes what a planisphere is to the
33 2-dimensional sphere S . Let us keep thinking of S : it is easy to see that there
34 is a plane Px D Tx S tangent to S at each point x of S . The plane Px may be
35 thought of as the fiber above x made up of all the tangent vectors to S at x, and is
36 of the same dimension as S : any element of Px can be described as the velocity
37 vx at x of some curve passing through x, a curve on S being nothing else but a
38 smooth map W R ! S , and its velocity nothing else but its derivative
` (vx D 0 .0/
39 with .0/ D x) (see [15, pp. 27–28]). The envelope T S WD x2S Px of all these
40 tangent planes can be given a structure of manifold. That construction applies to any
41 manifold M : TM is called the tangent bundle of M and is naturally divided into
42 fibers, namely, into all the tangent spaces to M . The natural projection p W TM !
43 M , defined by p.vx / D x, sends each tangent vector to the point at which it is
44 tangent to M , and thus projects each fiber onto its root. Furthermore, since TM is
45 itself a manifold, the transition from M to TM can be iterated, which means getting
46 from TM to T TM DW T 2 M , and thus to T n M for any n 2 N (with T 0 M WD M
47 and T 1 M WD TM ).
48
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 7

Geometric Modal Logic 7

1 Hence the following basic scheme: points of a given manifold M are 0-level
2 possible worlds, and to each point x of M is attached a subspace of Tx M , and more
3 generally a subset of TM , composed of 1-level possible worlds relative to x. Taken
4 in itself, x 2 M is a basic possible world. Taken as the label of the corresponding
5 subset of TM , x becomes a possible world of order 1. Next, for any tangent vector
6 vx to M at x, .x; vx / is a point of TM , and the collection of all possible worlds of
7 level 2 relative to .x; vx / singles out a certain subspace of T TM . All those worlds
8 of level 2 turn .x; vx /, of level 1, into a world of order 1; and x, of level 0, into a
9 world of order 2; and so on.
10
11
Definition 1.8 Given a manifold M and its iterated tangent bundles T n M , n 2 N,
a possible world is a member of the disjoint union n2N T n M . The level of a
`
12
13
possible world w is the (unique) integer n such that w 2 T n M . The order of w,
14
in the course of evaluating a formula of modal degree n, is m n, where m is the
15
maximal level of the possible worlds that are considered relatively to w. The 0-
16
level possible worlds x 2 M will be referred either as “possible worlds” or—as is
17
customary—as “points” of M .
18
So the evaluation of a nonmodal formula ' takes place at any point of the manifold
19
M ; the evaluation of ' appeals to a set of subsets of TM ; similarly, the evaluation
20
of ' will bring T TM into play, and so on. The lift from M to TM , then to
21
T TM , and so on, follows the move from ' to ', then to ', and so on. Each
22
T nC1 M is twice as dimensional as T n M (see [9, pp. 15–16]), which accounts for
23
the geometric change of scale prompted by each modal iteration.
24
25
1.4 Tasks Within the setting of a given manifold M , it is quite natural to interpret
26
any propositional variable p by a family of curves W R ! M on M , and then
27
to state that p is true at x if one of those curves passes through x.4 Such a course,
28
however, needs some tasks to be fulfilled.
29
First of all, the formulae :p, .p ^ q/ and .p _ q/ need to be interpreted by
30
families of curves, starting from the respective interpretations of p and q. The inter-
31
pretation of :p has to consist of curves disjoint from those interpreting p, so that
32
a formula such as .p ^ :p/ never holds. Moreover, if ‘Þ’ is taken as the primitive
33
modal operator, p being interpreted by a family of curves on M , the interpretation
34
of Þp has to be a family of curves, not on M anymore, but on TM : the shift from
35
the curves interpreting p to the curves interpreting Þp shall be called a modal lift.
36
Another kind of lift has to be defined, however, due to the discrepancies in modal
37
depth among the subformulae of a given formula. Indeed, let us consider a formula
38
such as .Þp ^ q/: Þp being interpreted by a family of curves on TM , the interpre-
39
tation of q, consisting of a family of curves on M , will have to be lifted onto TM so
40
that the interpretation of q can be combined with that of Þp. Otherwise, how could
41
.Þp ^ q/ be interpreted? Still, that lift of the interpretation of q is clearly distinct
42
from a modal lift, since Þq does not occur in the formula under consideration. So
43
a second kind of lift, which shall be called a nonmodal lift, has to be defined. Such
44
a problem does not exist in Kripke semantics, since any formula, of whatever modal
45
depth, gets interpreted by a subset of the same set of all possible worlds. On the
46
contrary, our new semantics works on principle with sets of possible worlds lying
47
at different levels so as to take modal iteration more seriously. As a consequence,
48
bridges have to be secured: modal and nonmodal lifts.
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 8

8 Brice Halimi

1 One ends up with the following four tasks:

2
1. The first task (T1) consists in defining the modal lift of any family € of curves
3
on T n M into a family .€/ of curves on T nC1 M , for each n 2 N, in order
4
to interpret the modal operator taken as primitive.
5
2. The second task (T2) consists in defining the nonmodal lift of any family €
6
of curves on T n M into a family L.€/ of curves on T nC1 M , for each n 2 N,
7
in order to interpret a formula built from subformulae having different modal
8
depths.
9
3. The third task (T3) derives from (T1) and (T2). Indeed, the natural semantic
10
clause for the satisfaction of a formula ' of modal depth n  1 should be
11
that ' is true at x 2 M if some of the curves on T n M that are part of the
12
interpretation of ' pass, not through x (which is impossible since x … T n M ),
13
but through a point of T n M that is related to x in a principled way. Such a
14
clause supposes eventually that each point x 2 M gives rise to a sequence
15
 0 .x/ D ¹xº;  1 .x/  TM; : : : ;  n .x/  T n M; : : : of counterparts of x.
16
4. The fourth and last task (T4) consists in checking the validity of the axioms
17
of normal modal logic and in obtaining correspondence results.
18
19 Definition 1.9 A geometric modal frame is a sequence .pnC1 W MnC1 !
20 Mn /n2N of maps between manifolds, making it possible to define: (1) the modal lift
21 .€/ onto MnC1 of any family € of curves on Mn (n 2 N); (2) the nonmodal lift
22 L.€/ onto MnC1 of any family € of curves on Mn (n 2 N); and (3) a sequence
23 h n .x/ W n 2 Ni above any x 2 M0 , with  0 .x/ D ¹xº. A valuation on such a
24 frame is the assignment, to every propositional variable, of a family of curves on M .
25 A frame endowed with a valuation is a geometric modal model.
26
27 1.5 A simpler semantics (In what follows, ‘’ is taken as the primitive modal oper-
28 ator.)
29 Let us consider a fixed given manifold M , together with a valuation V assigning
30 to each propositional variable a family of curves on M . A natural way to address
31 (T1) consists in setting a condition on 0 .t/ as the condition under which p is
32 true at x D .t/, for some curve in the interpretation V .p/ of p. The most
33 natural condition of that kind corresponds to the selection of a continuous family of
34 subspaces Dx  Tx M , one for each x 2 M . Indeed, as a working hypothesis, the
35 evaluation of p at x is supposed to take place in Tx M , which induces a connection
36 between p and the set V 0 .p/ D ¹ 0 W 2 V .p/º. As 0 .t/ belongs by definition,
37 for any t, to T .t/ M , a more restrictive condition is required for the truth of p at x,
38 namely the fact that 0 .t/ lies in some proper subspace D .t/ of T .t/ M (otherwise
39 the truth of p at x would automatically entail the truth of p at x). Each Dx 
40 Tx M , for x 2 M , can be conceived of as a constraint on the velocity vector of a
41 curve at x. The whole family D D ¹Dx W x 2 M º brings in some further structure
42 on M , called a distribution. Given a distribution D on M , one can state:
0
43 x  p iff 8 2 V .p/ 8t 2 R; x D .t/ ) .t/ 2 Dx :
44
45
As a distribution on M is needed for the evaluation of formulae of modal depth 1,
46
a distribution on TM is also needed for the evaluation of formulae of modal depth 2,
.2/
47 namely, a distinguished subspace D.x;v/ of T.x;v/ TM for each .x 2 M; v 2 Dx /:
48
x  p iff 8v 2 Dx ; TM; v  p
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 9

Geometric Modal Logic 9

0 00 .2/
1 iff 8v 2 Dx 8 2 V .p/ 8t 2 R; v D .t/ ) .t/ 2 D.x;v/ :
2
There is no canonical way to determine an infinite sequence of distributions on
3
M , TM; : : :, that is, an infinite sequence of distinguished subspaces Dx  Tx M ,
4
.2/
5
D.x;v/  T.x;v/ TM; : : : (for all x 2 M; v 2 Tx M; : : :). Yet it is worth mentioning
6
that the investigation of most common modal logics, such as S5, may restrict itself,
7
by virtue of equivalence results, to low modal depth formulae. Notably (see Hughes
8
and Cresswell [12, pp. 98–101]), any modal formula is deductively equivalent in S5
9
to a formula whose modal depth is at most 1. So let us consider the modal system S5,
10
and let us suppose that any formula has been replaced by an equivalent formula of
11
modal depth at most 1. A geometric frame for S5, then, needs only to be a manifold
12
M endowed with a distribution D.
13 Definition 1.10 Given a curve W R ! M on a manifold M , denotes the set
14 ¹ .t/ S
W t 2 Rº of all the points of . Given a set € of curves on a manifold M ,
15 € WD 2€ .
16 x
Definition 1.11 Given a manifold M and x 2 M , denotes the curve on M
17 x
reduced to x (i.e., .t/ D x for all t 2 R).
18
19 Definition 1.12 Given a manifold M , C .M / denotes the set of all curves on M .
20 A valuation on M is the assignment, to each propositional variable p, of a family
21 V .p/  C .M / of curves on M .
22
Definition 1.13 Given a manifold M endowed with a distribution D and a valu-
23
ation V , the interpretation of any formula is defined inductively as follows:
24
(a) V .:'/ D ¹ 2 C .M / W \ V .'/ D ;º,
25
(b) V ..' ^ // D V .'/ \ V . /,
26
(c) V ..' _ // D V .'/ [ V . /,
27
(d) V .'/ D ¹ 2 V .'/ W 8t 2 R .t/ D x ) 0 .t/ 2 Dx º.
28
29
One then defines:
30 (e) hM; D; V i; x  ' iff x 2 V .'/.
31
Any distribution D on M constitutes a field of infinitesimal constraints set upon
32
any move in M . In that respect, each subspace Dx should be conceived of as a
33
local restriction of the range of possibilities rather than as an open array of possible
34
worlds relative to each x—hence the choice of the necessity operator ‘’ instead of
35
‘Þ’ as the primitive modal operator. Putting all the Dx together, the distribution
36
D determines all the admissible trajectories for the movement of an ideal physical
37
body. This comes very close to the modern picture of a physical law as a “foliation,”
38
that is, as a division of the “configuration space” (the set of all possible states of
39
a physical body) into disjoint “leaves:” the trajectory of the body is bound to keep
40
inside a single one of all the leaves. In the case of S5, geometric modal semantics
41
thus draws on the conception of necessity that can be gathered from the mathematical
42
formalization of a law in abstract mechanics. More generally, it establishes a parallel
43
between the satisfaction clause for p and the restriction to a subset of V 0 .p/, so
44
between modal operators and differential operators: ultimately, between the validity
45
of an axiom schema seen as an equation whose variable is p, and the satisfaction, by
46
at least one of the curves in V .p/, of some corresponding differential equation.
47
The simplest example of a distribution on M is the one given by a nonzero vector
48
field on M . A vector field X on a manifold M is the continuous assignment, to each
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 10

10 Brice Halimi

1 point x of M , of a particular tangent vector Xx in Tx M . A nonzero vector field is

2 a vector field X such that Xx ¤ 0 for all x 2 M . Taking Dx D RXx (the real
3 line whose direction is given by Xx ), one gets a foliation: the integral curves of the
4 vector field X —that is, all the curves whose derivative, at any point x, abides by
5 the direction indicated by Xx —divide the manifold M into lines which are the 1-
6 dimensional leaves of the foliation. Consequently, a natural variant (d0 ) of the clause
7 (d) would interpret the shift from ' to ' by the restriction, among the curves in
8 V .'/, to those which are integral curves of X :
9
(d0 ) V .'/ D ¹ 2 V .'/ W is an integral curve of X º.
10
As a consequence, hM; X; V i; x  ' iff there is an integral curve of X in V .'/
11
that passes through x. The condition (d0 ) is stronger than (d), because it involves the
12
behavior of a curve, not only at x, but in a whole neighborhood of x.
13
14 Definition 1.14 A vector field frame, for the modal semantics defined by (a)–(c),
15 (d0 ) and (e), is a manifold M endowed with a nonzero vector field X .
16
17
Proposition 1 The axioms K and T are valid in the class of all vector field frames.
18 Proof Let hM; X i be a vector field frame. It validates the axiom T by construction.
19 Indeed, for any formula ' and any x 2 M , if an integral curve 2 V .'/ passes
20 through x, then x 2  V .'/. The axiom K is valid as well: given two formulae
21 ' and and x 2 M , if M; x  .' ! / and M; x  ', then M; x 
22  . Indeed, there are, by hypothesis, two integral curves ı and  in V . / and in
23 V ..' ! // D V ..:' _ //, respectively, both passing through x. But, owing
24 to the Cauchy–Lipschitz theorem, which states the existence of a unique maximal
25 integral curve passing through any given point of M , either two integral curves are
26 disjoint or one is a restriction of the other. So  is in V . /, and so M; x   .
27 For the same reason, suppose that M; x  ' and M; x   . Then there are two
28 integral curves of X, 2 V .'/ and ı 2 V . /, both passing through x. Again, the
29 Cauchy–Lipschitz theorem allows one to conclude that is either a restriction or an
30 extension of ı, so that M; x  .'^ /. As a consequence, .'^ / $ .'^ /
31 is also valid in the class of all vector field frames.
32
33 Yet the more general case of formulae of modal depth at least 2 has still to be
34 accounted for, otherwise modal iteration cannot be properly examined. Another,
35 more complex, framework has to be introduced.
36
37 2 A Semantics for Higher Modal Depth
38
(In what follows, until the end, ‘Þ’ is taken as the primitive modal operator.)
39
Addressing the tasks (T1)–(T3) requires some further structure on a manifold M
40
so that the modal and the nonmodal lift can be defined as canonically as possible.
41
Instead of adding an infinite sequence of distributions on M , then TM , and so on,
42
it is much more natural to start with a manifold already endowed with a sufficient
43
amount of internal structure. For reasons which will presently appear, Riemannian
44
geometry is a natural area to look at. It is a very rich source of concepts and tech-
45
niques, and a very central area in mathematics, whose application to logic has still to
46
be explored.5
47
The developments below, up to the conclusion, are organized as follows. First,
48
some preliminary notions in Riemannian geometry will be introduced.6 Then, in
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 11

Geometric Modal Logic 11

1 consideration of the first three tasks that have been brought out earlier, a general
2 illustration of geometric modal semantics will be laid out in detail and shown to
3 generalize Kripke semantics. Finally, basic axioms will be shown to be valid and
4 several correspondence results will be presented so as to meet the fourth task.
5
6 2.1 Preliminaries A Riemannian manifold is a (differentiable) manifold endowed
7 with a “metric” g, that is, with a way of measuring the length gx .v; v/ of any tangent
8 vector v 2 Tx M (for any x 2 M ), thus the velocity of any movement on M , and so
9 finally the distance between any two points along a given curve on M . Any metric
10 on a manifold M induces the so-called Levi-Civita connection r associated with g
11 (see [9, pp. 70–72]). Intuitively, r is the operator that assigns, to any couple .X; Y /
12 of vector fields defined in the neighborhood of some point x0 of M , the infinitesimal
13 deviation of Y , at x0 , with respect to the direction indicated by Xx0 . A vector field
14 X defined along a curve is said to be parallel if rX X D 0 at any point of .
15 A curve on M whose derivative 0 is parallel is called a geodesic of hM; gi: its
16 velocity vector never deviates from itself. In case hM; gi is the Euclidean space
17 Rn , a geodesic is nothing but a straight line. In the general case, a geodesic can
18 be characterized as a curve that locally minimizes the distance between any two of
19 its points (taken close enough to each other). Another example is provided by the
20 meridians on a sphere.
21 The connection r associated with g allows one to define, for any given curve
22 on M and for any two parameters t and t 0 , the parallel transport along of any
23 vector v 2 T .t / M from the point .t/ to the point .t 0 /. The result of this parallel
24 transport, written Jt;t 0 .v/, is the vector in T .t 0 / M obtained by moving v along
25 from .t/ to .t 0 / while avoiding any artificial rotation. Thus, parallel transport
26 yields a natural way of relating two distinct tangent spaces T .t/ M and T .t 0 / M .
27 Besides, any metric g previously defined on a manifold M induces a natural metric
28 gT on TM , called the Sasaki metric of TM (see [9, p. 80]). It has the following
29 property: the derivative 0 (“velocity field”) of any geodesic of hM; gi is a geodesic
30 of hTM; gT i.
31 We now work with a fixed given Riemannian manifold hM; gi. Using iterated
32 Sasaki metrics, one inductively defines .pnC1 W hMnC1 ; gnC1 i ! hMn ; gn i/n2N as
33 follows: M0 WD M; g0 WD g; MnC1 WD TMn ; gnC1 WD .gn /T ; and pnC1 is the
34 natural projection. Each Mn D T n M (n 2 N) is always implicitly endowed with the
35 metric gn .
36 Given a vector field X defined along a curve on M , if X is parallel along ,
37 then X 0 .0/ (an element of TX.0/ TM ) is said to be horizontal. A curve ı W R !
38 TM on TM is horizontal if ı 0 .t/ is horizontal for every t 2 R. This means that
39 ı coincides locally with a parallel vector field X along p1 .ı/, X being viewed as a
40 curve X W t 7! X .t/ on TM . Thus, horizontal curves on the tangent bundle TM
41 correspond to parallel vector fields on the manifold hM; gi. As a consequence, for
42 each geodesic of hM; gi and each fixed vector v 2 T .0/ M , there exists locally a
43 unique horizontal curve ev on TM lying above (i.e., p1 ı ev D ) and starting
44 at v (i.e., ev .0/ D v) (see [9, pp. 63, 80]). The curve ev is called the (local)
45 horizontal lift of passing through v at t D 0, and is a geodesic of hTM; gT i. In
0
46
particular, e .0/ D 0 . Conversely, it is sufficient, for any geodesic ı of hTM; gT i to
47
be horizontal, that ı 0 .0/ be horizontal, and in that case p1 ı ı is a geodesic of hM; gi
48
(see [9, Proposition 2.109, pp. 100–102]). The construction of the horizontal lift of a
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 12

12 Brice Halimi

1 geodesic to the tangent bundle applies not only to our original Riemannian manifold
2 M , but more generally to any T n M (n 2 N).
3
4
2.2 (T3) It is now required, for each given possible world x 2 M , to introduce
5
possible worlds at higher levels as the counterparts of that world. To that end, let
6
us suppose that a selection A D ¹ i W i 2 I º of curves on M , which shall be
7
called accessibility curves, has been fixed: these curves represent accessibility paths
8
between possible worlds of level 0. For any given point x 2 M of a curve , let
9
tx 2 R be the parameter corresponding to x, that is, such that .tx / D x.7 Besides,
10
recall that, given any curve W R ! M on M , denotes the set ¹ .t/ S W t 2 Rº
11
of all its points (see Definition 1.10). Now, for any x 2 i , A.x/ WD x2 i i
12
will represent the set of all the possible worlds that are accessible from x. The
13
latter have to be distinguished from the possible worlds relative to x, namely, from
14
all the elements of TM contained in the modal lift of a curve passing through x
15
(about possible worlds relative to x, see Definition 2.3 below). The idea is that
16
the counterparts of x in TM come from the infinitesimal data at x sending x to a
17
possible world accessible from x. Instead of looking at the worlds accessible from
18
x, one looks at the velocity vectors at x of geodesics starting from x and reaching
19
a world accessible from x at time t D 1. The actual path followed to reach a world
20
accessible from x is taken seriously, as an important geometric datum.
21
To be specific, for any v 2 Tx M , let cv be the geodesic of M defined by cv .0/ D
22
x and cv0 .0/ D v.8 The map v 2 Tx M 7! cv .1/ 2 M is a diffeomorphism between
23
a maximal neighborhood Vx of the null vector 0Ex in Tx M and some neighborhood
24
of x in M : it makes it possible to code the points lying in some neighborhood of
25
x 2 M by tangent vectors in Tx M (v codes cv .1/ and the null vector codes x itself).
26
The biggest real number rx > 0 such that the ball Bx WD B.0Ex ; rx / is included
27
in Vx is called the injectivity radius of M at x, and Bx itself, the injectivity ball
28
of M at x. So, for any two distinct vectors v and w in Bx , cv .1/ ¤ cw .1/ is
29
always true. The set P .x/ WD ¹v 2 Bx W cv .1/ 2 A.x/º can then be introduced
30
as a set of proxies of x in Tx M . Its extension to TM as a whole, beyond Tx M , is
31
.x/ WD ¹cv0 .t/ W v 2 P .x/; t 2 Rº.9 The elements of .x/ will be called the 1-
32
counterparts of x. The shift from P .x/ to .x/ leads from isolated possible worlds
33
of level 1 to curves of level 1.
34
How does one extend the definition of  to any level in the sequence hM; gi
35
hTM; g1 i hT 2 M; g2 i


? For n D 0,  0 .x/ D ¹xº and, for n D 1,
36
 1 .x/ D .x/ has just been defined. But how does one build  2 .x/? Every curve
37 i
i on M gives rise to a parallel transport Jt;t 0 W T i .t/ M ! T i .t 0 / M and thus, for
38
39
any v 2 T i .t/ M (for some t 2 R), to a curve Jv i W t 0 7! Jt;ti 0 .v/ on TM . So one
40 can introduce A.˛/, P .˛/ and .˛/, for any ˛ 2 TM , exactly as A.x/, P .x/ and
41 .x/ have been introduced for any x 2 M : it suffices to replace the geodesic curves
42 c of hM; gi with those c T of hTM; gT i, and the accessibility curves i from x 2 M
43 with all the 2-accessibility curves J˛ i from each element ˛ 2 .x/  TM . More
44 precisely, ˛ being in some Tx0 M , I0 being the set of all indices j of I such that
45 x0 2 j and tj being the parameter such that x0 D j .tj /, the set of all possible
worlds that are accessible from ˛ is A.˛/ WD j 2I0 Jtjj; .˛/. The set of the proxies
S
46
47 of ˛ is then defined as P .˛/ WD ¹u 2 B˛ W cuT .1/ 2 A.˛/º, where cuT is the geodesic
48
of TM such that cuT .0/ D ˛ and .cuT /0 .0/ D u, and the set of all counterparts of ˛
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 13

Geometric Modal Logic 13

1 as .˛/ WD ¹.cuT /0 .t/

SW u 2 P .˛/; t 2 Rº. Finally, the set of the 2-counterparts of
2 x 2 M is  2 .x/ WD ˛2 1 .x/ .˛/. The definition of  n .x/ for n  3 is analogous.
3
4
Definition 2.1 For each n 2 N, the accessibility curves involved in the construc-
5
tion of .˛/, for any x 2 M and any ˛ 2  n .x/, will be called .n C 1/-accessibility
6
curves, and the elements of  n .x/ will be called the n-counterparts of x.
7
2.3 (T1)–(T2) It remains to define the modal and the nonmodal lift of any curve
8
on M to TM , then to T TM , and so on. In that perspective, a selection Cn of
9
curves on T n M is supposed to be specified at the beginning, for each n 2 N. Those
10
curves, called admissible curves, are all the curves from which the interpretation of
11
any formula (of modal depth n) is bound down to being drawn. Only three conditions
12
have to hold: (i) for any v 2 T n M , there is at least one 2 Cn passing through v;
13
(ii) for any ı 2 CnC1 , pnC1 .ı/ 2 Cn ; and (iii) for any 2 Cn , there is at least one
14
ı 2 CnC1 that lifts , that is, satisfies pnC1 .ı/ D .
15 0
The family of all horizontal lifts ek .t/ (t 2 R; k 2 R ) is the most natural
16 nC1
extension to T M of any curve on T n M and can be taken as the definition
17
of the modal lift of (the nonzero coefficients k are meant to disregard the actual
18
speed and to retain only the direction).10 Besides, it is equally natural to define the
19
nonmodal lift of , for any curve on T n M , as being simply the derivative 0 , which
20
can be read off directly from and is actually known as the canonical lifting of
21
(see [15, p. 96]). Hence, we have the following definitions.
22
23 Definition 2.2 For each n 2 N, given a set € of curves on T n M , the two follow-
24 ing sets of curves on T nC1 M are the modal and nonmodal lifts of €, respectively:
0 .t/
25
.€/ WD ¹ek 2 CnC1 W 2 €; k 2 R ; t 2 Rº;
26
27 LnC1
n .€/ WD ¹ 0
2 CnC1 W 2 ۼ:
28 As the gap in modal depth between two principal subformulae of a single formula
29 can be strictly greater than 1, the iteration of the nonmodal lift operator has to be
30 considered: for any n; m 2 N,
31
Lnn WD id; LnCm
n WD LnCm
nCm 1 ı


ı LnC1
n :
32
33 Definition 2.3 Given a Riemannian manifold hM; gi and x 2 M , a possible
34 world of level 1 relative to x is any point in TM that belongs to the modal lift of a
35 curve passing through x. For n  2, a possible world of level n relative to x is a
36 possible world relative to some possible world of level n 1 relative to x.
37
Definition 2.4 A metric modal frame (or metric frame, for short) is a quadruple
38 ˝ ˛
39 F D M; g; A; .Cn /n2N ;
40 where hM; gi is a Riemannian manifold, A is a set of curves on M and each Cn is
41 a set of curves on T n M . All the constructions above show that any metric modal
42 frame amounts to a geometric modal frame (in the sense of Definition 1.9).
43 A valuation V on F is the assignment, to any propositional variable p, of a subset
44 V .p/ of C0 closed under connected portions of its members.11 As a consequence, one
45
has that V .p/ \ V .q/ D V .p/ \ V .q/ and that C0 n V .p/ D C0 n V .p/.
46
A metric modal model (or metric model, for short) is a metric modal frame
47
endowed with a valuation. Any metric modal model amounts to a geometric modal
48
model.
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 14

14 Brice Halimi

1 Definition 2.5 Given a metric model hF; V i (with F as above), the interpretation
2 V .'/ of any formula ' is defined inductively as follows:
3
(i) V .:'/ D Cn n V .'/ (where d.'/ D n);
4
(ii) V ..' ^ // D V .'/ \ Lnm .V . // (if d.'/ D n and d. / D m  n);
5
(iii) V ..' _ // D V .'/ [ Lnm .V . // (under the same hypothesis as above);
6
(iv) V ..' ^ // D Lm n .V .'// \ V . / (if d.'/ D n and d. / D m  n);
7
(v) V ..' _ // D Lm n .V .'// [ V . / (under the same hypothesis as above);
8
(vi) V .Þ'/ D .V .'//.
9
10
Definition 2.6 Given a metric model hF; V i (with F as above), the final semantic
11
clause is, for any formula ' of modal depth n and for any x 2 M :
12 hF; V i; x  ' iff  n .x/ \ V .'/ ¤ ;:
13
In other words, a formula ' is true at x if one of the possible worlds relative to x that
14
are prompted by the construction of V .'/ coincides with some counterpart of x at
15
the level defined by the modal depth of '.
16
A simple illustration of that semantic apparatus is provided by the Euclidean plane
17
E endowed with an orthonormal frame, with all the horizontal lines as the accessi-
18
bility curves. Let O D .0; 0/ be the point taken as the origin of E. The unique
19
accessibility curve passing through O, by definition, is the x-axis, so A.O/ D
20
¹.a; 0/ W a 2 Rº. Thus P .O/ D ¹..0; 0/; ˛0 W ˛ 2 Rº, since a vector whose


21
tail is applied at O has its head on
the x-axis only if it is a horizontal vector. Con-
22
sequently,  1 .O/ D ¹..a; 0/; ˛0 / W a 2 R; ˛ 2 Rº. Now, let be a curve on E in
23
V .p/. By definition, . / is a set of curves on T E of the form t 7! . .t/; k 0 .t0 //
24
(for some k 2 R and t0 2 R). For E; O  Þp to obtain, it is necessary and suf-
25
ficient that there be a curve p 2 V .p/ satisfying . p .t/; k p0 .t0 // D ..a; 0/; ˛0 /


26
27
for at least one quintuple .t; k; t0 ; a; ˛/ 2 R5 with k ¤ 0. Since p can always be
28
replaced with the curve A reduced to A D .0; a/, and since . A /0  0, one finally
29
gets: E; O  Þp iff there exists a curve in V .p/ that crosses the x-axis at least
30
once (not necessarily at O).
31
2.4 Kripke semantics reconsidered The latter clause is reminiscent of Kripke
32
semantics. Still, the whole apparatus, in the general case, differs from Kripke
33
semantics in a qualitative way, insofar as it determines the accessibility relations and
34
the evaluation to proceed in an essentially local way, in accordance with the metric.
35
It also captures a much wider array of modal models than Kripke semantics. On that
36
score, Kripke semantics amounts to disregarding the space that underpins the web of
37
accessibility relations between the possible worlds, and Kripke models correspond
38
to a particular case of metric models. Before getting to it, a few notions need to be
39
introduced about graphs.
40
41 Definition 2.7 Given a graph G, the topological realization jGj of G is obtained
42 by assigning a closed interval to each edge of G and then identifying endpoints of
43 those intervals according to the coincidences of the vertices of the corresponding
44 edges of G.
45
Definition 2.8 An embedding of a graph G in some space (see Gross and Yellen
46
[10, p. 618]), for instance, in some surface S , assigns points of S to the vertices of
47
G and arcs (homeomorphic images of Œ0; 1) to the edges of G in such a way that: (i)
48
the endpoints of the arc corresponding to some edge are the points corresponding to
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 15

Geometric Modal Logic 15

1 the vertices of that edge, (ii) no arc includes points associated to other vertices, and
2 (iii) two intersecting arcs intersect only at some common endpoint.
3
4
Fact 1 ([10, Chapter 7]) The realization of any graph can be embedded in some
5
Euclidean space Rd (for some d  1) in such a way that the image of every edge is
6
a straight-line segment, so as to obtain what is called a straight-line drawing.
7
Topological graph theory is the study of graph embeddings, such as the drawing of a
8
graph on a torus. Any image of the embedding of a Kripke model defines a certain
9
metric frame (but the converse is not true).
10
11 Definition 2.9 Each Kripke frame K D hW; Ri induces the metric frame
F K D M K ; g K ; AK ; .CnK /n2N
12
˝ ˛
13
14 defined as follows:
15 (i) M K WD .GR /, where GR is the graph of R, d is the least integer k such
16 that the graph GR of R has a straight-line drawing in Rk , and  W GR ! Rd
17 is any such straight-line drawing;
18 (ii) g K is the metric on M K induced from the Euclidean metric on the ambient
19 Rd ;
20 (iii) AK is the set of all self-R-related vertices (considered as constant curves)
21 and of all straight-line segments in the straight-line drawing of GR ;
22 (iv) C0K WD AK and, for n  1, CnK is the set of all straight-line segments l of
n
23 T n Rd ' R2 d such that .p1 ı p2 ı


ı pn /.l/ 2 AK .
24
25
Definition 2.10 Let K D hW; Ri be a Kripke frame and F K the metric frame
26
induced by K. A natural correspondence between valuations on K and valuations
27
on F K can be established, namely:
28 (i) Given a valuation V on F K , VL is the valuation on K given by
29 VL .p/ D  1 .V .p// \ W .
30 (ii) Given a valuation V on K, VO is the valuation on F K given by
VO .p/ D w2W Vw .p/, where:
S
31
32 – if w … V .p/, then Vw .p/ D ;;
33 – if w 2 V .p/ and no possible world is accessible from w, then Vw .p/ D
34 ¹ .w/ º;
35 – if w 2 V .p/S and w is not isolated,
S then
36 Vw .p/ D e2E.w/;e2V .p/ e [ e2E.w/;e…V .p/ Œ.w/; .e/Œ,
37 E.w/ being the set of edges (of GR ) of which w is an end, “e 2 V .p/”
38 meaning that both endpoints of e belong to V .p/ and, for each e 2
39 E.w/, .e/ being the midpoint of .e/. (Here, e and Œ.w/; .e/Œ are
40 considered as curves of speed 1.)12
41
42
Lemma 2.11 Let K D hW; Ri be a Kripke frame and V a valuation on it. Then
43 VLO D V .
44
Proof For any propositional variable p,
45
VO .p/ D w2V .p/ ¹.w/º [ e2E.w/;e2V .p/ e [ e2E.w/;e…V .p/ Œ.w/; .e/Œ,
S S S
46
47
e and Œ.w/; .e/Œ being now considered as solid segments, not as curves any more.
48
Consequently, m.e/ being the midpoint of each edge e of (the realization of) GR :
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 16

16 Brice Halimi

VLO .p/ D  1 VO .p/ \ W

1


2  [   [ 
3 D ¹wº [ e
4 w2V .p/ w2V .p/;e2E.w/\V .p/
 [ 
5
[ Œw; m.e/Œ \W
6
w2V .p/;e2E.w/nV .p/
7 [
8 D ¹wº \ W
9 w2V .p/
10 D V .p/:
11
12
Lemma 2.12 Let K D hW; Ri be a Kripke frame, F K the induced metric frame
13
and V a valuation on K. Then, for any w 2 W and any formula ':
14 hF K ; VO i; .w/ MM ' iff hK; V i; w Kripke ';
15
where ‘MM ’ (resp. ‘Kripke ’) refers to satisfaction according to metric modal
16
semantics as defined in Definition 2.6 (resp. according to Kripke semantics as
17
defined in Definition 1.6).
18
19 Proof The proof is by induction on '.
20
21 hF K ; VO i; .w/ MM p iff .w/ 2 VO .p/
22
iff w 2 VLO .p/
23
iff w 2 V .p/ (by Lemma 2.11)
24
25 iff hK; V i; w Kripke p:
26
27 hF K ; VO i; .w/ MM :' iff hF K ; VO i; .w/ ²MM '
28 iff hK; V i; w ²Kripke ' (by induction hypothesis)
29 iff hK; V i; w Kripke :':
30
31
The cases of .' ^ / and .' _ / are straightforward as well. Furthermore, let us
32
suppose that hF K ; VO i; .w/ MM Þ', with d.'/ D n, say. By definition, there
33 exists ˛ 2  n ..w// such that .˛/ \ .VO .'// ¤ ;. For any N 2 N, any straight
34 line l in RN , any t 2 R and any vE 2 Tl.t/ RN ' RN , JvEl W t 0 7! Jt;t
l
v / D .l.t 0 /; vE/
0 .E
2N
35 is a straight line in R , so by induction all n-accessibility curves are straight-line
S
36 segments. One has A.˛/ D j 2J lj , where the lj ’s (indexed by some set J ) are all
37 the n-accessibility curves passing through ˛. So ˛ D lj .tj / for some tj 2 R, for
38
each j 2 J . Accordingly, P .˛/ D ¹.˛; klj0 .tj // W j 2 J; k 2 Rº  T nC1 M K and
39 0 0
.˛/ D ¹c.˛;kl 0 .t // .u/ W j 2 J I k; u 2 Rº D ¹.lj .tj C u/; klj .tj C u// W j 2
40 j j

41 J I k; u 2 Rº  RN  RN . In particular, ˛ 2 pr 1 ..˛//, where pr 1 W .X; Y / 2

42 RN  RN 7! X 2 RN is the projection onto the first factor. By induction, ˛ 2
43 prn . n .˛// for each n  1, where pr n W RN  .RN /n ! RN is again the projection
0
44 onto the first factor. On the other hand, .VO .'// D ¹e l kl .t/ W l 2 VO .'/; k 2 R ; t 2
45 Rº. Any l 2 VO .'/ is a straight-line segment (by definition of CnK ), so its horizontal
0
46 l kl .t/ W u 7! .l.t C u/; kl 0 .t C u//. So the
lift coincides with its canonical lift .l; l 0 /: e
47
condition .˛/ \ .VO .'// ¤ ; entails the existence of j; k1 ; k2 ¤ 0; t; u1 and u2
48
such that .lj .tj C u1 /; k1 lj0 .tj C u1 // D .l.t C u2 /; k2 l 0 .t C u2 //. Since lj and l are
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 17

Geometric Modal Logic 17

n
1 straight-line segments in R2 d with one point in common and the same direction,
n
2 the line .lj / generated by lj and the line .l/ generated by l in R2 d are identical.
3 Now, because lj is a straight-line segment whose projection contains .w/, there is
4 e 2 E.w/ such that .e/ D prn .lj / and w 0 2 W such that e D .w; w 0 /. From
5
the above, one has .w 0 / 2 pr n .l/  pr n .VO .'//, so hF K ; VO i; .w 0 / MM '. By
6
the induction hypothesis, hK; V i; w 0 Kripke ', and thus (because w 0 2 e 2 E.w/
7
amounts to wRw 0 ) hK; V i; w Kripke Þ'.
8
Conversely, let us suppose that hK; V i; w Kripke Þ', with d.'/ D n, say. Then
9
there exists w 0 2 W such that wRw 0 and hK; V i; w 0 Kripke ', hence (by induc-
10
tion hypothesis) hF K ; VO i; .w 0 / Kripke '. By construction, hF K ; VO i; x MM ' for
11
12
some point x 2 M K lying between .w 0 / and .e/, where e D .w; w 0 /. So there
13
are ˛ 2  n .x/ and l 2 VO .'/ such that ˛ 2 l, say ˛ D l.t˛ /.13 Now, as above,
14
the straight-line segment l and some n-accessibility segment lj overlap in the neigh-
15
borhood of ˛, with, say, ˛ D lj .u˛ /, so lj .u˛ / D l.t˛ / and lj0 .u˛ / D kl 0 .t˛ / for
16 some k 2 R . As lj is derived from the 1-accessibility segment to which both x
17 and .w/ belong, for any u 2 R there exists k 0 2 R such that .lj .u/; k 0 lj .u// 2
18  nC1 ..w//. In particular, .lj .u˛ /; k 0 lj0 .u˛ // 2  nC1 ..w// for some k 0 2 R .
19 As a consequence, .l.t˛ /; k 0 kl 0 .t˛ // 2  nC1 ..w//. But .l.t˛ /; k 00 l 0 .t˛ // 2 .l/ 
20
.VO .'// for any k 00 2 R , so in the end  nC1 ..w// \ .VO .'// ¤ ;, and so
21
hF K ; VO i; .w/ MM Þ'.
22
23 Lemma 2.13 Let K D hW; Ri be a Kripke frame, F K the induced metric frame
24 and ' any formula. Then, for any point x of M K and any valuation V on F K ,
25 there exist a point x 0 of M K and a valuation V 0 on K such that hF K ; V i; x MM '
26 implies hF K ; VO 0 i; x 0 MM '.
27
28 Proof The proof is by induction on d.'/. For d.'/ D 0, one may suppose that
29 x D .w/ for some w 2 W , since it requires at most “translating” V and x along
30 one of the straight lines of .GR /. Then taking x 0 D x and V 0 D VL will do. For
31 d.'/ D n  1, hF K ; V i; x MM ' is equivalent to .˛/ \ V .'/ ¤ ; for some
32 ˛ 2  n 1 .x/. As in the proof above, any member of .˛/ belongs to .lj ; lj0 / for
33 some n-accessibility straight line lj , and any curve in V .'/ is the direct lift .l/
34 or Lnn 1 .l/ of some straight line l 2 CnK 1 . So one gets that .lj .t1 /; k1 lj0 .t1 // D
35
.l.t2 /; l 0 .t2 // for some t1 ; k1 ; t2 2 R or that .lj .t1 /; k1 lj0 .t1 // D .l.t2 /; k2 l 0 .t2 // for
36
some t1 ; k1 ; t2 2 R and some k2 2 R . So hF K ; V i; x MM ' finally amounts, up
37
to reparameterization, to a local overlapping of the straight-line segments lj and l.
38
By induction hypothesis, this overlapping is still the case when V is replaced with
39
40
VO 0 , so the conclusion follows.14
41
Lemma 2.14 Let K D hW; Ri be a Kripke frame, F K the induced metric frame,
42
V any valuation on K and ' any formula. Then, for any x 2 M K , there exists
43
w 2 W such that hF K ; VO i; x MM ' implies hF K ; VO i; .w/ MM '.
44
Since M K D e edge of GR .e/ (vertices being identified with single-point
S
45 Proof
46 edges), there are, for each x 2 M K , two possible cases: either x D .w/ for some
47 w 2 W , and the problem is solved; or there is an edge e of GR such that x 2 .e/, x
48
being neither endpoint of .e/. In the latter case, by construction, ¹' W x 2 VO .'/º D
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 18

18 Brice Halimi

1 ¹' W .w/ 2 VO .'/º for the endpoint .w/ to which x is closer on .e/. There could
2 seem to be a problem only in case x is the midpoint .e/ of the image of some edge
3 e D .w; w 0 / of GR . But then, by definition of VO , ¹p W x 2 VO .p/º D ¹p W e 2
4 V .p/º D ¹p W .w/ 2 VO .p/º \ ¹p W .w 0 / 2 VO .p/º, which (by induction) is
5 readily seen to be true, not only for any propositional variable p, but for any formula
6
' in general. So hF K ; VO i; x MM ' implies both hF K ; VO i; .w/ MM ' and
7
hF K ; VO i; .w 0 / MM '.
8
9 Theorem 2.15 For any Kripke frame K and any modal formula ',
10
F K MM ' iff K Kripke ':
11
12 Proof Let us suppose that F K MM '. By hypothesis, for any valuation V on K
13 and any w 2 W , hF K ; VO i; .w/ MM ', so (by Lemma 2.12) hK; V i; w Kripke '.
14 Thus, K Kripke '. Conversely, let us suppose that F K ²MM '. Then there are a
15 valuation V on F K and x 2 M K such that hF K ; V i; x MM :'. By Lemma 2.13,
16 there are a valuation V 0 on K and x 0 2 M K such that hF K ; VO 0 i; x 0 MM :'. By
17
Lemma 2.14, there is w 0 2 W such that hF K ; VO 0 i; .w 0 / MM :'. But then (by
18
Lemma 2.12 again) hK; V 0 i; w 0 Kripke :', so K ²Kripke '.
19
20 It is not difficult to see that many metric frames are not similar to any metric frame
21 induced by a Kripke frame. As a result, general validity with respect to all met-
22 ric frames does not amount to validity to all metric frames induced by a Kripke
23 frame. However, metric modal semantics departs from Kripke semantics only as far
24 as modal iteration goes. Indeed, one has the following.
25
Proposition 2 For any modal formula ' of modal depth at most 1, Kripke '
26
implies MM '.
27
28 Proof Let us suppose that ²MM '. Then there are a metric frame F D hM; g; A;
29 .Cn /n2N i, a valuation V on F and x 2 M such that hF; V i; x MM :'. By Whit-
30 ney’s theorem (see [9, p. 49]), M can be embedded in R2d C1 , where d is the dimen-
31 sion of M . Given such an embedding e, one defines:
32 (i) M  as the set of all points of all straight lines .y  z  / in R2d C1 with y  WD
33 e.y/, z  WD e.z/, for all points y and z of M belonging to the same geodesic
34 of hM; gi;
35 (ii) g  as the metric induced on M  by the Euclidean metric on R2d C1 ;
36 (iii) A as the set of all straight lines .y  z  / with z 2 A.y/, considered as curves;
37 (iv) C0 as the set of all straight-line segments contained in some member of A ,
38 
!
and CnC1 (for each n 2 N) as the set of all straight lines R 0 .z/ (in T nC1 M  )
39
with 2 Cn and z 2 dom. /.
40

41 The metric frame F  WD hM  ; g  ; A ; .Cn /n2N i coincides with F K for the
42 Kripke frame K  WD hR2d C1 ; R i, where R is the binary relation displayed by
43 the straight lines in A . Given 2 V and x 2 , let d ;x be the straight line
44 in R2d C1 to which e. / is tangent at x  WD e.x/: d ;x is the line that passes
!
45
through x  and whose direction isS.e. /0 .tx /, with tx given by e. /.tx / D x  .
46
One then introduces V  .p/ WD ¹d ;x W 2 V .p/; x 2 º. Since noth-
47
ing is changed up to modal depth at most 1 in comparison to hF; V i, and since
48
:' is of modal depth at most 1, hF  ; V  i; x  MM :'. So (by Lemma 2.13)
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 19

Geometric Modal Logic 19


1 hF K ; VO1 i; x1 MM :' for some valuation V1 on K  and some point x1 of M  ,

2 and so (by Lemma 2.14) hF K ; VO1 i; .w1 / MM :' for some w1 2 M  . As a
3 consequence (by Lemma 2.12), hK  ; V1 i; w1 Kripke :', so ²Kripke '.
4
5 Thus, Kripke modal semantics can be considered as the 1-truncation of the modal
6 semantics based on metric models, to the extent that its metric counterpart disregards
7 any datum related to the higher-order relationships between the base possible worlds.
8 In contrast, the main motivation of geometric modal logic is to geometrize the graphs
9 of accessibility relations: whereas Kripke-style possible worlds remain in themselves
10 without any connection to each other, geometric modal logic introduces the higher-
11 order structure which lumps them together and gives them the texture of a particular
12 geometric universe.
13 This allows the notion of accessibility to take on the specifically geometric mean-
14 ing that it otherwise only hints at: that of a path on a manifold and, at the upper
15 levels, that of a connection (parallel transport between tangent and higher tangent
16 spaces). Let us explain this point a bit. Given a set W of possible worlds and a val-
17 uation V on it, let us call “propositions” the valuations V .'/ of formulae in W . For
18 the sake of simplicity, let us use ‘'’ to actually refer to the corresponding proposition
19 V .'/. Any accessibility relation R on W induces an operator ‘ÞR ’ on propositions:
20 ÞR ' D ¹w 0 W wRw 0 for some w 2 'º. Conversely, any operator ‘Þ’ on propo-
21 sitions induces an accessibility relation RÞ on W : wRÞ w 0 iff Þp 2 V .w 0 / for
22 any p 2 V .w/. It is straightforward to check that R D RÞR . In other words, an
23 accessibility relation R amounts to a device to lift a proposition ' into a proposi-
24 tion Þ' of the next higher modal depth. Now, the link between Kripke models and
25 metric models can be described by saying that, exactly in the same way, but in the
26 geometrically much richer environment of a manifold, a connection r on a bundle
27 pn W T nC1 M ! T n M (n 2 N) amounts to a device to lift curves on T n M into
28 curves on T nC1 M .
29 The comparison with Kripke semantics leads to a more general description of the
30 semantic upshot of the possibility operator. Indeed, in any framework, the number
31 of matches between an evaluation world w and the valuation V .Þ'/ is supposed to
32 be no less and generally greater than the number of matches between w and V .'/.
33 There are three options to ensure that phenomenon: (i) either V .Þ'/ enlarges V .'/,
34 while the evaluation world w is left unchanged; (ii) or V .'/ is left unchanged, but
35 the shift to Þ' results in the evaluation world w being enlarged into a whole set of
36 worlds; (iii) or both V .'/ and w get enlarged. The first option corresponds to the
37 topological interpretation of modal logic initiated by Tarski–McKinsey semantics
38 (see McKinsey and Tarski [17]), where each formula ' is interpreted by a subset
39 V .'/ of a topological space, and Þ' by the topological closure of V .'/. The second
40 option corresponds to Kripke semantics, where assessing V .Þ'/ instead of V .'/ at
41 w consists in “dilating” w 2 W into the set E 1 .w/ D ¹w 0 2 W W wRw 0 º of all
42 the possible worlds accessible from w.15 The third and last option corresponds to the
43 present framework and combines both ideas while transferring the evaluation process
44
to the next level: V .Þ'/ enlarges V .'/, but in a radically new way, and  d.'/ .w/
45
is enlarged into  d.'/C1 .w/ as well, to take into account the fact that the new space
46
T d.'/C1 M is twice as dimensional as T d.'/ M . (For the sake of comparison, we
47
write here ‘w’ for a 0-level possible world in a metric model, instead of the customary
48
notation ‘x’ for a point on a manifold that we have used up to now.)
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 20

20 Brice Halimi

1 An important correlative fact is that two formulae ' and , interpreted as sets of
2 curves, may have coextensional valuations in a given metric model hF; V i (i.e., ¹x W
3 hF; V i; x MM 'º D ¹x W hF; V i; x MM º), yet have different possibilizations
4 (i.e., ¹x W hF; V i; x MM Þ'º ¤ ¹x W hF; V i; x MM Þ º). This is because
5 each curve takes into account the way it passes through each possible world that it
6 contains. This intensional feature sets apart our geometric semantics from all above-
7 mentioned semantic frameworks.
8 A final remark is about the operator LnC1 n . Indeed, the interpretation of q within
9 a formula such as .Þp ^ q/ (namely, L10 .V .q//) differs from the interpretation of
10 q considered in isolation (namely, V .q/). So one could object that this discrepancy
11 entails a systematic ambiguity in the interpretation of modal formulae. Against such a
12 possible objection, it should be replied that .Þp^q/ constitutes a whole, and a modal
13 context for q even though q does not occur there in the scope of a modal operator.
14 That is why the interpretation of the same formula q actually varies from one context
15 to another and takes place at different levels (in M in one case, in TM in the other).
16 Besides, the definition (in some metric model) of LnC1 n .V .'// (for n D d.'/) makes
17 sure that, for any point x of the manifold underlying that model,  n .x/ \ V .'/ ¤ ;
18
implies  nC1 .x/\LnC1 n .V .'// ¤ ;. Indeed, let z 2  n .x/\V .'/. Then z D .tz /
19
20
for some 2 V .'/ and some tz 2 R, and thus . z /0 .tz / D 0Ez 2 LnC1 n .V .'// by
definition of LnC1
n . Furthermore, c Ez .1/ D z 2 A.z/, so c 0
.0/ D E
0 z 2 P .z/ 
21 0 0zE
22
.z/   nC1
.x/. So 0Ez 2  nC1 .x/ \ LnC1
n .V .'//. As a consequence, a formula is
23
true at some level only if its nonmodal lift is true at the next upper level, which fits
24
the intended meaning of LnC1n : the interpretation of the formula is only transferred,
25
with no loss.
26
27 3 (T4)
28
It remains to reach certain correspondence results, so as to complete the fourth and
29
last of the tasks delineated in Section 1.4.
30
31 Lemma 3.1 The axiom T, equivalent to p ! Þp, and the axiom K are both valid
32 in all metric frames.
33
Proof For simplicity’s sake, one confines oneself here to the case of a formula p of
34
modal depth 0. Let us suppose, in a given metric model, that, for some point x of the
35
underlying manifold,  1 .x/ \ 0 ¤ ;, in other words that 0 .tx / 2  1 .x/ for some
36
curve 2 V .p/ and some tx 2 R. Without loss of generality, it can be supposed
37
38
that tx D 0. Then . x /0 .0/ D 0Ex , so e0E x .0/ D 0Ex 2 V .Þp/. Since 0Ex 2  1 .x/,
39  1 .x/\V .Þp/ ¤ ;. Hence  1 .x/\L10 .V .p// ¤ ; implies  1 .x/\.V .p// ¤ ;,
40 so hF; V i; x  p ! Þp.
41
Moreover, the definition of ' implies that a curve (on T nC1 M if ' is of
42 modal depth n) belongs to V .'/ iff if is disjoint from any curve (on T nC1 M )
43 that lifts a curve (on T n M ) disjoint from V .'/. One thus checks that the axiom K,
44 .p ! q/ ! .p ! q/, is valid as well.
45
The axiom 4 is not generally valid. Indeed, given some metric model, hF; V i; x 
46
Þp obtains if .V .p// passes through at least one point in .x/. On the contrary,
47
hF; V i; x  Þ Þ p obtains simply if ..V .p/// passes through at least one point
48
in .w/ for some element w of .x/ which can perfectly be other than x. However,
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 21

Geometric Modal Logic 21

1 the axiom 4 becomes valid if one restricts oneself to a very special class of metric
2 frames.
3
4
Definition 3.2 A metric frame F D hM; g; A; .Cn /n2N i is sparse if the following
5
conditions hold: (i) hM; gi is a simple Riemannian manifold, which means that any
6
two points of M are related by at most one geodesic; (ii) A and C0 are composed
7
of geodesics only; and (iii) for each n 2 N, CnC1 is composed of horizontal lifts of
8
curves in Cn .
9
Proposition 3 The axiom 4 is valid in all sparse metric frames.
10
11 Proof Let F D hM; g; A; .Cn /n2N i be a sparse metric frame and V a valuation on
12 it. Since, for each n 2 N, the horizontal lifts of geodesics of hT n M; gn i are geodesics
13 of hT nC1 ; gnC1 i, Cn is composed of geodesics only. Furthermore, because of the
14 local uniqueness of horizontal lifts, if two points ˛ and ˇ of TM were related by two
15 distinct horizontal geodesics of hTM; gT i, then p1 .˛/ and p1 .ˇ/ would be related
16 by two distinct geodesics of hM; gi. Thus, the simplicity of hM; gi guarantees that
17 of hTM; gT i (and thus that of hT n M; gn i for each n 2 N) when the geodesics of the
18 latter are restricted to the horizontal lifts of the geodesics of the former. So, in what
19 follows, one may consider the axiom 4 restricted to a propositional variable p: to
20 get the proof about a formula ' of higher modal depth, one just needs to replace M
21 with T d.'/ M and to consider V .'/ as the valuation of a propositional variable on
22 the latter manifold.
23 Given some x 2 M , let us suppose that there exists a curve of A passing through
24 x, and let ¹ci W i 2 Ix º be the set of all curves of ASpassing through x. The
25 points accessible from x are all the points contained in i2Ix ci and, for any such
26 point y distinct from x, the geodesic ci relatingS x and y is unique, since hM; gi is
27 supposed to be simple. Thus, one has P 1 .x/ D i 2Ix ¹k0 ci0 .0/ 2 Bx W k0 2 Rº and
 1 .x/ D i 2Ix ¹k0 ci0 .t0 / W k0 ci0 .0/ 2 Bx ; k0 ; t0 2 Rº. Therefore,
S
28
29
c
[®
30  2 .x/ D .cuT /0 .t2 / W cuT .1/ 2 Jt0i; .w/I
31 i 2Ix

w D k0 ci0 .t0 / 2 Bx ; u 2 Bw ; k0 ; t0 ; t2 2 R
¯
32
33 [®
34
D .cuT /0 .t2 / W cuT .1/ 2 t 7! k0 ci0 .t0 C t/I
i 2Ix
35
w D k0 ci0 .t0 / 2 Bx ; u 2 Bw ; k0 ; t0 ; t2 2 R
¯
36
[®
37
D .cuT /0 .t2 / W cuT .1/ D k0 ci0 .t1 /I
38
i 2Ix
39
w D k0 ci0 .t0 / 2 Bx ; u 2 Bw ; k0 ; t0 ; t1 ; t2 2 R :
¯
40
41 But, for fixed k0 2 R and i 2 Ix , cuT and k0 ci0 are, up to linear reparameterization,16
42
two geodesics of TM with two intersection points, namely, cuT .0/ D w D k0 ci0 .t0 /
43
and cuT .1/ D k0 ci0 .t1 /. Those two points are actually distinct; otherwise c ET .1/ D
44 0w

45 w D cuT .1/, which is excluded because 0Ew and u are both in the injectivity ball Bw .
46 Since hM; gi is simple, cuT D k0 ci0 (up to linear reparameterization), so  2 .x/ D
00 1
S
47 k0 2R;i2Ix k0 ci . If no geodesic in A passes through x (i.e., Ix D ;), then  .x/ D
48
A.x/ D ¹0Ex º and  2 .x/ D ¹0E E º, so the formula above remains valid: it suffices
0x
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 22

22 Brice Halimi

1 to choose k0 D 0. If only one geodesic ci in A contains x (i.e., Ix D ¹iº) and

2 ci is reduced to ¹xº, then ci0 D 0Ex and ci00 D 0E0E x on the one hand; on the other,
3
 1 .x/ D ¹0Ex º and  2 .x/ D ¹0E0E x º, so the formula is still valid. Moreover, for any
4
5 p 2 V .p/, . p / contains exactly (up to linear reparameterization) all the curves

6
ep u with u D k1 p0 .t1 / (for some k1 2 R and t1 2 R), since those curves are
all geodesics belonging to C1 . Likewise, .. p //  C2 contains exactly (up to
7
8
e
linear reparameterization) all the curves . ep u /v with v D k2 . ep u /0 .t2 / (for some u
9 as above, k2 2 R and t2 2 R).
Now, let us suppose that  2 .x/ \ ..V .p/// ¤ ;. This implies that there
10
e
are p 2 V .p/, i 2 Ix and k; k0 ; t; t 0 2 R such that k0 ci00 .t/ D . ep u /v .t 0 /.
e
11
12 It follows that p2 .k0 ci00 .t// D p2 .. ep u /v .t 0 //, that is, k0 ci0 .t/ D ep u .t 0 /. But
13
then k0 ci0 .t/ 2 .V .p//, and so k0 ci00 .t/ 2 L21 ..V .p///, so finally k0 ci00 .t/ 2
14
15
 2 .x/ \ L21 ..V .p///. Hence  2 .x/ \ ..V .p/// ¤ ; implies  2 .x/ \
16 L21 ..V .p/// ¤ ;, for every x 2 M . Consequently, hF; V i  Þ Þ p ! Þp, for
17 every valuation V on F .
18
19 Modus ponens obviously preserves validity in sparse metric frames. So does the
20 necessitation rule, but on the condition that sparse metric frames are replaced with
21 a narrower class of metric frames (see Theorem 3.7 below). The rule of uniform
22 substitution does not preserve validity, but that is an expected feature, given our
23 project of interpreting modal iteration as being a stronger operation than it is usually
24 made out to be. Indeed, when one is committed to some modal claim (such as Þp !
25  Þ p), one should not be thereby committed to the analogous claim about formulae
26 of higher modal depth than that of p (such as Þ Þ p !  Þ Þp).
27 Closure under substitution is motivated by the belief that if one is able to establish
28 something about simple necessity, one is by the same token entitled to admit the
29 same thing about necessary necessity. This is exactly what this paper challenges.
30 The claim that the meaning of necessity is uniform is not self-evident at all. In fact,
31 the meaning of the operator ‘Þ’ depends on the place in a formula where it occurs.17
32 (The same context principle has been encountered earlier when the interpretation of a
33 subformula has been shown to depend on the modal degree of the formula of which
34 it is a subformula.) So let S4* be the modal system S4 with the rule of uniform
35 substitution restricted to the case where the substituens is of modal depth not higher
36 than that of the substituendum, and let S4** be S4* without the necessitation rule.
37 By the previous proposition, the system S4** is valid in all sparse metric frames.
38 There is more.
39
Theorem 3.3 The system S4** is complete with respect to the class of all sparse
40
metric frames.
41
42
Proof The proof is an adaptation of the “canonical models” technique. Let us con-
43
sider the set W  of all maximal S4**-consistent sets of formulae and the binary
44
relation R on W  given by: wR w 0 iff for every formula ', ' 2 w implies
45
' 2 w 0 . One defines a canonical valuation V  in the Kripke frame hW  ; R i
46
by stating that w 2 V  .p/ iff p 2 w. The model hhW  ; R i; V  i is called the
47
canonical model of S4**. Let G  be the contraction of the graph of R obtained by
48
removing from the latter all its self-loops. The realization jG  j of G  , considered
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 23

Geometric Modal Logic 23

1 as a 1-dimensional manifold embedded in some Rd by means of straight-line seg-

2 ments only, and equipped with the Euclidean metric inherited from Rd , constitutes a
3 sparse metric frame F  , whose accessibility curves are all its segments and whose
4 admissible curves at level 0 are all the connected portions of those segments. The
5 canonical valuation V  , which is defined only for endpoints of segments of jG  j,
6 can be extended to jG  j in the following way. To any valuation V on propositional
7
variables is associated a unique converse valuation V on possible worlds, defined by
8 _
V .w/ WD ¹p W w 2 V .p/º. Given any point x contained in an arc x0 x1 D f .Œ0; 1/
9
for some homeomorphism f , there is a unique t 2 Œ0; 1 such that x D f .t/. Then,
10
11 stating  V .x/ WD V  .x0 / if 0  t  1=2 and  V .x/ WD V  .x1 / if 1=2 < t  1,
12 one defines an extension  V of V  taking all points of jG  j into account, and
13 hF  ;  V i  ' iff S4** ` '. The conclusion follows: if a formula ' is valid in all
14 sparse metric frames, it is valid in particular in F  , and thus derivable in S4**.
15
A completeness result can also be reached for S4*.
16
17 Lemma 3.4 Let F D hM; g; A; .Cn /n2N i be a metric frame and V a valuation
18 on it. Then, for any formula of modal depth n and any x 2 M ,  nC1 .x/  .V .'//
19 implies hF; V i; x  '.
20
21 Proof By definition, hF; V i; x  ' iff hF; V i; x ² Þ:' iff  nC1 .x/ \
22 V .Þ:'/ D ; iff  nC1 .x/ \ .V .:'// D ;. Now, let us suppose that .V .'// \
23 .V .:'// ¤ ;, in other words that there are 2 V .'/ and ı 2 V .:'/
k1 0 .t1 / 0 k2 0 .t2 / 0
24 such that e .t1 / D ı e .t2 / for some k1 ; k2 ; t1 ; t2 ; t10 ; t20 2 R. Then
k1 0 .t1 / 0 0
25 u WD pnC1 .e .t1 // D pnC1 .e ı k2 .t2 / .t20 // DW v. But u 2 whereas
26 v 2 ı, so u D v is excluded since \ ı  V .'/ \ V .:'/ D ;. Hence
27 .V .'// \ .V .:'// D ;, so hF; V i; x  '.
28
29 Definition 3.5 A metric frame F D hM; g; A; .Cn /n2N i is flat if: (i) hM; gi is
30 a simply connected, geodesically complete Riemannian manifold which is flat, that
31 is, whose sectional curvature is constant and identically zero;18 (ii) A and C0 are
32 composed of geodesics only; (iii) for each n 2 N, CnC1 is composed of horizontal
33 lifts of curves in Cn only.
34
Lemma 3.6 Let F D hM; g; A; .Cn /n2N i be a flat metric frame, V a valuation
35
V
on it, F D hTM; gT ; A;V .CV n /n2N i a flat metric frame with underlying manifold TM
36
37 endowed with the Sasaki metric and such that, for each n 2 N, CV nC1 D .CV n /, and
38 let V  be the valuation on FV defined by V  .p/ D .V .p// for any propositional
39 variable p. Then, for any formula ', V  .'/ D .V .'//.
40
41
Proof The proof is by induction on '. For ' D : (with d. / D n, say),
42 V  .: / D CV n n V  . / D CV n n .V . // (by induction hypothesis), which is
43 .V .: // by construction of CV n . For ' D . ^ / (with d. / D n and d./ D
44 n C 1, say), V  .. ^ // D L V nC1
n .V  . // \ V  ./ D LnC2
nC1 ..V . /// \ .V .//
45 (once again by induction hypothesis) is the set of all curves c that are identical both
46 with .f v /0 (for some 2 V . /, v D k 0 .t/, k 2 R and t 2 R) and with e w
47 (for some  2 V ./, w D l 0 .s/, l 2 R and s 2 R). This implies that w D
48
.f v /0 .0/, and actually that e w D .f v /0 , since f v is a horizontal geodesic. But
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 24

24 Brice Halimi

1 e v 0
then c is both .f v /0 D . f v /. e / .0/ and e w , so c 2 .V .. ^ ///. Conversely,
2 .V .. ^ /// D .LnC1 n .V . // \ V .//. Let c 2 .LnC1 n .V . // \ V .//:
0
3 there exists 2 Ln .V . // \ V ./ such that c D ek .t/ for some k 2 R and
nC1
4 t 2 R, and as a result c 2 .LnC1 .V . /// \ .V .//. But, for any 2 V . /,
5
6
.LnC1
n .¹ º// D .¹ 0
º/ D ¹
n
. e
0 k 00 .t/
/ W k 2 R
; t 2 Rº and L nC2
nC1 ..¹ º// D
k 0 .t/ 0
7 LnC2
nC1 .¹f W k 2 R ; t 2 Rº/ D ¹.f k .t/ /0 W k 2 R ; t 2 Rº. Even


e
8 if it implies to consider a linear reparameterization of , one may suppose that
0 k 00 .0/ 00 0 .0/
9 k D 1 as well as t D 0, thus . / D D .f /0 . So .LnC1
n .V . ///
10 nC2 V nC1  
D L nC1 ..V . /// D Ln .V . //. Moreover, .V .// D V ./. So finally
11
12
c 2 LV nC1
n .V  . // \ V  ./ D V  .. ^ //. The proof about . _ / is analogous.
13 Finally, for ' D Þ , V  .Þ / D .V V  . // D .V  . // D ..V . /// D
14 .V .Þ //.
15
16
Theorem 3.7 The system S4* is complete with respect to the class of all flat metric
17
frames.
18
Proof The Cartan–Hadamard theorem (see [15, pp. 252–53], [9, pp. 163–64])
19
states that any simply connected, complete Riemannian manifold whose sectional
20
curvature is everywhere nonpositive, is simple. As a consequence, flat metric frames
21
make up a subclass of the class of all sparse metric frames, so the system S4** is
22
valid in all flat metric frames. Since the model hF  ;  V i of S4** is a flat metric
23
model, it only remains to prove that the necessitation rule preserves validity in all flat
24
metric frames.
25
To that end, it suffices to show that a flat metric frame with underlying manifold
26
TM can always be built up from a flat metric frame with underlying manifold M .
27
But, given a flat metric frame F D hM; g; A; .Cn /n2N i, the Sasaki metric on TM
28
is flat (see Gudmundsson and Kappos [11, p. 19]) and, even if it requires replacing
29
TM with its completion, hTM; gT i can be supposed to be complete. Furthermore,
30
the Cartan–Hadamard theorem ensures that M is diffeomorphic to Rd (where d
31
is the dimension of M ), and thus that TM is simply connected. Finally, AV and
32
33 .CV n /n2N can be chosen so that FV WD hTM; gT ; A; V .CV n /n2N i is a flat metric frame
34 with CV nC1 D .CV n / for each n 2 N.
35 One may then reason as follows. Let us suppose that ' is a valid modal formula
36 in all flat metric frames: for any flat metric frame F D hM; g; A; .Cn /n2N i and any
37 valuation V on F , hF; V i  '. But, by hypothesis, ' is also valid in any flat metric
38 frame with underlying manifold TM , and at least one such frame is ensured to exist,
39 namely, FV . For any valuation V T on FV , ' is valid in hFV ; V T i, so V T .'/ D TM .
40 In particular, for any x 2 M , any v 2 P 1 .x/ and any t 2 R, cv0 .t/ 2 V T .'/;
41
hence .x/  V T .'/. This is true in particular for any valuation V T such that
42
V T .'/ D .V .'//, the existence of which is guaranteed by Lemma 3.6. Therefore
43
.x/  .V .'//, so (by Lemma 3.4) hF; V i  '. This is true for any F and any
44
V , so ' is valid in all flat metric frames, too.
45
46
As one can see, the substantial neutralization of modal iteration expressed by the
47
axiom 4 corresponds to one’s confining to quite a special class of metric frames
48
(much more so than the class of transitive Kripke frames is within all Kripke frames
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 25

Geometric Modal Logic 25

1 in general), as befits the fact that geometric modal logic originates with the recog-
2 nition of the strong meaning of modal iteration, which cannot be mitigated without
3 strong restrictions.
4 A completeness result can be reached for S5* as well—S5* being S5 with the rule
5 of uniform substitution restricted as in the case of S4*.
6
7
Definition 3.8 A metric frame F D hM; g; A; .Cn /n2N i is minimal if it is sparse
8
and, moreover, any two geodesics in A are either disjoint or identical (up to linear
9
reparameterization).
10 Proposition 4 The axiom 5 is valid in all minimal metric frames.
11
12
Proof Let F D hM; g; A; .Cn /n2N i be a minimal metric frame and V some valu-
13 ation on it. Let us suppose further that, for some x 2 M ,  2 .x/\L21 ..V .p/// ¤ ;.
14 Let cx be a (geodesic) curve in A passing through x. By hypothesis, cx is unique
15 (up to linear reparameterization). So A1 .x/ D cx and, as a result,  1 .x/ D cx0 :
16 indeed, cv0 .t/ 2  1 .x/ (for some t 2 R) iff cv .0/ D x 2 cx and cv .1/ 2 cx , but
17 then cv and cx are two geodesics with two points in common, so cv D cx and thus
18 cv0 .t/ 2 cx0 . Hence  2 .x/ D cx00 . So the hypothesis  2 .x/ \ L21 ..V .p/// ¤ ;
0
19 becomes cx00 .t0 / D . ep k1 p .t1 / /0 .t2 / for some p 2 V .p/, k1 2 R and t0 ; t1 ; t2 2 R.
0
20 As a consequence, as in the proof of Proposition 3, cx0 .t0 / D ep k1 p .t1 / .t2 /, and
21 again cx .t0 / D p .t2 /. But p belongs to some Cn and thus is a geodesic, and so is
22 cx by hypothesis. So cx D p (up to linear reparameterization). Then, let us con-
0
23
sider dx W t 7! cx0 .t C t0 / and ı W t 7! ep k1 p .t1 / .t C t2 /. The curve cx0 is a geodesic
0
24
of hTM; g T i, and thus so is dx . Besides, dx0 .0/ D cx00 .t0 / D . ep k1 p .t1 / /0 .t2 /, which
25 0
is horizontal by definition of ep k1 p .t1 / ; so the geodesic dx is the horizontal lift of
26 0
p1 ı dx W t 7! cx .t C t0 / passing through dx .0/ D cx0 .t0 / D ep k1 p .t1 / .t2 / DW v0
27
28
at t D 0. In the same way, ı is a horizontal geodesic of hTM; g T i; thus the hori-
0
29
zontal lift of p1 ı ı W t 7! p .t C t2 / passing through ı.0/ D ep k1 p .t1 / .t2 / D v0
30
at t D 0. But t 7! cx .t C t0 / is the same curve as t 7! p .t C t2 /, so one must
31
conclude that dx D ı, and so that cx0 D ı (up to linear reparameterization). Hence,
0
32 for any t 2 R, cx00 .t/ D e ı ı .t/ .0/ 2 .ı/. Now, up to linear reparameterization,
k1 p 0 .t / 00 0
33 ı D ep 1 , and so ı 2 .
p /. Therefore, cx .t/ D ı .t/ 2 .. p //. From
34  2 .x/ D cx00 , it follows that  2 .x/  .. p //  .V .Þp//. To sum up, it has been
35 shown that  2 .x/ \ L21 ..V .p/// ¤ ; implies  2 .x/  .V .Þp//. By Lemma 3.4,
36 hF; V i; x  Þp !  Þ p, for every x 2 M and every valuation V on F .
37
38
Theorem 3.9 The system S5* is complete with respect to the class of all minimal
39
metric frames.
40 Proof Since any minimal metric frame is sparse, the previous proposition ensures
41 that S5** (defined analogously to S4**) is valid in all minimal metric frames. Fur-
42 thermore, the necessitation rule preserves validity in all minimal metric frames.
43 Indeed, let F D hM; g; A; .Cn /n2N i be a minimal metric frame and V a valua-
44 tion on it. For each x 2 M , let cx be the unique geodesic (considered up to linear
45 reparameterization) passing through x, so that .x/ D ¹kcx0 .t/ W k 2 R ; t 2 Rº.
46 Any formula ' which is valid in F satisfies V .'/ D ¹cx W x 2 M º, so .V .'// D
0
47 ¹cex kcx .t/ W x 2 M; k 2 R ; t 2 Rº. Hence, .x/ D ¹kcx0 .t/ W k 2 R ; t 2 Rº D
0
48
¹cex kcx .t/ W k 2 R ; t 2 Rº  .V .'//, so (by Lemma 3.4) hF; V i; x  '. This is
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 26

26 Brice Halimi

1 true for any x 2 M and any valuation V , so ' is valid in F , which concludes the
2 proof that the necessitation rule preserves validity in any minimal metric frame, and
3 thus that S5* is valid in all minimal metric frames.
4 Conversely, the accessibility relation R0  of the canonical model of S5* (defined
5 analogously to that of S4**) is an equivalence relation. Let us contract the real-
6 ization of the graph of R0  by turning every triangle w1 w2 w3 (for a fixed arbitrary
7 ordering of its vertices) into a single segment whose endpoints are w1 and w3 and
` point is w2 . The full contraction gives rise actually to a disjoint union
8 whose middle
9 M 0  WD i 2I Di of straight lines (for some indexing set I ), each of which contains
10 all the points wij corresponding to the vertices of some connected component (R0  -
11 equivalence class) of the original graph of R0  . The disjoint union M 0  , considered
12 as embedded in the Euclidean plane, is a 1-dimensional Riemannian manifold that
13 one can readily enrich into a metric frame F 0  by taking all Di ’s to be the accessi-
14 bility curves and all the connected portions of the Di ’s to be the admissible curves at
15 level 0. As in the case of S4**, the canonical valuation V 0  that is part of the canon-
16 ical model of S5 can be extended into some valuation  V 0 so as to take all points
17 of M 0  into account, and hF 0  ;  V 0 i constitutes a minimal metric model such that
18 hF 0  ;  V 0 i  ' iff S5* ` '. The conclusion follows as above: if a formula ' is valid
19 in all minimal metric frames, then it is valid in particular in F 0  , and thus derivable
20 in S5*, which completes the proof.
21
22
Conclusion The motivation of this paper has been to generalize Kripke semantics by
23
unfolding the geometric content of the notion of accessibility. The resulting semantic
24
framework involves an internally structured collection of possible worlds, based on
25
a Riemannian manifold. Resorting to differential geometry is quite natural since it
26
affords rich tools to understand modal iteration as a change of scale, namely, as the
27
shift from a manifold to its tangent bundle. Hence the introduction of possible worlds
28
of increasing level. In that context, Riemannian geometry, through the concept of
29
lift, makes it possible to define possible worlds of higher order, following an analysis
30
rather than a synthesis—thus distinguishing between level and order. A key feature
31
is the fact that each possible world of higher order corresponds to a genuinely new
32
structure incorporating all the higher-level, lower-order possible worlds that depend
33
on it. Possible worlds of the nth level only deliver a verdict as to the truth or falsity
34
of formulae of modal depth n: to interpret a formula of modal depth n C 1, one has
35
to ascend to the .n C 1/th level, which is determined in a natural way from the nth
36
level.
37
The resulting semantics actually generalizes Kripke semantics (Theorem 2.15),
38
but also carries an important geometric twist. Indeed, Riemannian geometry helps,
39
through the concept of connection, to fully articulate the geometric intuition implied
40
by the notion of accessibility. In Kripke modal semantics, the structure called up
41
by the interpretation of higher modalities remains purely external: it is the structure
42
of the sole accessibility relation, not of the possible worlds that this relation relates
43
to one another. In metric modal semantics, on the contrary, iterated modalities are
44
matched by internally structured possible worlds, whose structure is provided by the
45
geometric texture of the manifold and the differential accessibility systems attached
46
to the metric. Accessibility at the base level corresponds to path connection, and
47
accessibility at higher levels corresponds to the connection induced by the metric,
48
that is, to the parallel transport of local data along a curve via horizontal lifts.
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 27

Geometric Modal Logic 27

1 Compared to that new framework, Kripke semantics corresponds to the degen-

2 erate case where the structure of the space underlying the possible worlds is dis-
3 crete and the connection along the sequence of iterated tangent bundles is induced
4 by a relation on the base manifold. Metric modal semantics generalizes and local-
5 izes Kripke semantics, by taking into account the way in which a certain proposi-
6 tion is true at a given possible world. Whereas locality of modal truth, in Kripke
7 semantics, corresponds only to the partiality of the access to the whole extension
8 of the accessibility relation that each possible world offers, metric modal semantics,
9 whose workings rely on derivatives of increasing order, deepens the sense in which
10 modalities can be described as local. In particular, the “local” accessibility system
11 hE n .w/ W n 2 Ni attached to any world w in a Kripke frame is replaced with the
12 accessibility system h n .x/ W n 2 Ni attached to any point x of a Riemannian mani-
13 fold M : a system whose members  n .x/ live in spaces of increasing dimension—an
14 expression of modal change of scale—and constitute the specification at all orders of
15 the infinitesimal neighborhood of x, as shaped by the accessibility curves on M .
16 Correspondence results about variants of S4 and S5 have been obtained (Theo-
17 rems 3.3, 3.7 and 3.9). Thus, geometric modal logic turns out to be an operational
18 semantic framework, which actually extends Kripke modal semantics and enriches
19 it with new tools. These correspondence results also attest, not only to the geomet-
20 ric meaning that modalities can be given, but also maybe, conversely, to the modal
21 meaning that can be bestowed on geometric operations such as parallel transport and
22 the shift to the tangent bundle of a manifold. Other reasons concur in turning toward
23 geometry, in particular the similarity, that could be pursued, between a modal oper-
24 ator and a differential operator. A whole avenue of interaction between modal logic
25 and geometry thus opens up, holding out the prospect of further correspondence
26 results between modal axioms and geometric properties.19
27
28
Notes
29
30
31
1. See Blackburn, de Rijke, and Venema [4] for further details.
32
2. The idea can be found also in Dǒsen [6] and Benevides and Maibaum [3].
33
34
3. In [7], Fine defines a specification as a partial assignment of truth values to the atomic
35
sentences of a language containing vague predicates (such as “bald”). The relation of
36
extension between two specifications induces a partial order within any given set of all
37
“admissible” specifications, called a “specification space.” Fine’s option to account for
38 higher-order vagueness is to consider a specification space as being itself a point of a
39 higher-order specification space: “Suppose we pick upon an admissible complete spec-
40 ification for the language L. If the language suffers from first-order vagueness, this
41 specification is not unique and we may pick upon an admissible set of complete specifi-
42 cations. If the language also suffers from second-order vagueness, this set is not unique
43 and we may pick upon an admissible set of sets, and so on. After .n C 1/ such choices,
44
we obtain what might be called an n-th-order boundary.
Let us be more precise. A 0-th order space is a complete specification and a .n C 1/-
45
th order space is a set of n-th order spaces. A n-th order boundary is then a sequence
46
s 0 s 1


s n such that s i is an i -th order space, i  n, and s j 2 s j C1 , j < n; and a
47
!-order boundary is an infinite sequence s 0 s 1


such that each s 0 s 1


s i is an i-order
48
boundary, i D 1; 2; : : :” [7, p. 293].
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 28

28 Brice Halimi

1 4. This supposes that the interpretation of a proposition contains in general uncountably

2 many possible worlds, but there are reasons not to be bothered by that. A combinatorialist
3 justification will do: it is reasonable to suppose that the actual world is made up out of a
4 countable set of elementary states of affairs (represented by mutually independent atomic
5 propositions), and hence to admit that there are uncountably many possible worlds (each
6
possible world being represented by a maximally consistent set of negatomic proposi-
7
tions). Since there are only countably many propositions, some propositions have to
hold in uncountably many different possible worlds.
8
9
5. Connections between modal logic and geometry have already been proposed (most
10
notably: Kremer and Mints [13], Venema [18], and Beklemishev [2]), but each time
11
from a completely different angle.
12
13 6. All the relevant definitions can be found in full detail in [9].
14
15 7. Each accessibility curve will be supposed not to overlay itself, so that it passes only once
16 through each of its points. The parameter tx is thus defined unequivocally.
17
18 8. The existence and unicity of cv can be established (cf. [9, p. 81]).
19
20 9. The map .v; t/ 7! cv0 .t/ is called the geodesic flow of the Riemannian manifold hM; gi.
21
22 10. The existence and unicity of a (local) horizontal lift of are secured only if is a
23
geodesic of hT n M; gn i. This is a reason to restrict each Cn to a subset of geodesics.
24
25
11. A connected portion of a curve 2 V .p/ is simply the restriction of to a connected
subset of . Any connected portion of a member of V .p/ will be taken to be a member
26
of V .p/. In particular, for any 2 V .p/ and for any x 2 , the curve x reduced to
27
x belongs to V .p/ as well. From now on, admissible curves will be understood in that
28
looser sense, as connected portions of admissible curves in the stricter sense.
29
30 12. A more direct definition for VO could seem to be: VO .p/ D .w/ .
S
w2V .p/ However,
31 .w/ /0
for any w 2 W , .  0, so one would have, for any valuation V on K and any
32 E
formula ' of modal depth n: .V .'// D ¹e0x 2 CnC1 W 2 V .'/; x 2 º. As a
33
consequence, one would get, for any propositional variable p: hF K ; VO i; x  Þp iff
34
 1 .x/ \ .V .'// ¤ ; iff there is ˛ 2  1 .x/ such that ˛ D 0Ey for some y 2 V .p/ iff
35
x D y 2 V .p/ iff hF K ; VO i; x  p. In other words, the interpretation of the possibility
36
operator would be trivialized.
37
38
13. Although all accessible points from x are all points of .e/ and are thus accessible
39 from .w/, neither .x/  ..w// nor  n .x/   n ..w// (for n  1) is true.
40 Indeed, given that .e/ is oriented (as a curve) from .w/ to .w 0 /, k..e//0 .t/ 2 P .x/
41 for some k < 0 and t 2 R, whereas k..e//0 .t/ 2 P ..w// for no k < 0. So
42 the existence of ˛ 2  n .x/ such that ˛ 2 VO .'/ does not ensure the existence of
43
˛ 2  n ..w// such that ˛ 2 VO .'/—which otherwise would equate the satisfaction
44
conditions of hF K ; VO i; .w/ MM Þ' with those of hF K ; VO i; .w/ MM '.
45
46 14. A more constructive explanation is the following. By induction on d.'/, one may replace
47 M K with T d.'/ 1 M K and consider V .'/ as the valuation of a formula of modal
48 depth 1. For each 1-accessibility curve lj and each propositional variable p, three cases
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 29

Geometric Modal Logic 29

1 can be distinguished: (1) either x … V .p/ and, for any y 2 lj , y … V .p/; (2) x … V .p/
2 and there exists y 2 lj such that y 2 V .p/; (3) x 2 V .p/ and, for any y 2 lj ,
3 y 2 V .p/. Let us say that x 2 Œ.w/; .w 0 /. One defines V 0 such that w … V 0 .p/
4 and w 0 … V 0 .p/ in the first case; w 2 V 0 .p/ and w 0 … V 0 .p/ in the second case;
5 w 2 V 0 .p/ and w 0 2 V 0 .p/ in the third case. Up to replacement of x by some other
6 point in Œ.w/; .w 0 /, the semantic situation of x is the same whether with respect to VO 0
7 or with respect to V .
8
9 15. Chellas–Segerberg semantics is a mixed case, where the semantic clause to be applied
10
depends on whether the point w at which ' is evaluated is “normal” (then the truth of
Þ' at w is defined as in Kripke semantics) or “queer” (then the truth of Þ' at w is
11
defined as in Tarski-McKinsey semantics); see Chellas and Segerberg [5, p. 7].
12
13
16. Given a curve W R ! M on a manifold M , a linear reparameterization of is any
14
curve on M of the form ı f , with f W t 2 R 7! at C b 2 R for some a; b 2 R.
15
16 17. Although it has in fact almost nothing to do with the framework developed here, Dov
17 Gabbay’s “hypermodalities” and “fibred” modal semantics also stem from this remark.
18
19 18. See [9, pp. 96, 132 and 133, respectively] for definitions of the geodesic completeness,
20 sectional curvature and flatness of a Riemannian manifold.
21
22 19. Natural developments of geometric modal logic would be its extension to first-order
23 modal logic and the exploration of its proof-theoretic counterpart. This is a matter for
24 further work.
25
26
27
28
References
29 [1] Baltag, A., “STS: A structural theory of sets,” pp. 1–34 in Advances in Modal Logic,
30 Vol. 2 (Uppsala, 1998), edited by M. Zakharyaschev, K. Segerberg, M. de Rijke,
31 and H. Wansing, vol. 119 of CSLI Lecture Notes, CSLI Publications, Stanford, 2001.
32 Zbl 0993.03065. MR 1838245. 4
33
[2] Beklemishev, L., and D. Gabelaia, “Topological completeness of the provability logic
34
GLP,” Annals of Pure and Applied Logic, vol. 164 (2013), pp. 1201–23. Zbl 1320.03088.
MR 3093387. DOI 10.1016/j.apal.2013.06.008. 28
35
[3] Benevides, M. R. F., and T. S. E. Maibaum, “A constructive presentation for the
36
modal connective of necessity ./,” Journal of Logic and Computation, vol. 2 (1992),
37
pp. 31–50. Zbl 0763.03014. MR 1166503. DOI 10.1093/logcom/2.1.31. 27
38 [4] Blackburn, P., M. de Rijke, and Y. Venema, Modal Logic, vol. 53 of Cambridge
39 Tracts in Theoretical Computer Science, Cambridge University Press, Cambridge, 2001.
40 Zbl 0988.03006. MR 1837791. DOI 10.1023/A:1010358017657. 3, 27
41 [5] Chellas, B. F., and K. Segerberg, “Modal logics in the vicinity of S1,” Notre Dame
42 Journal of Formal Logic, vol. 37 (1996), pp. 1–24. Zbl 0859.03011. MR 1379545. DOI
43 10.1305/ndjfl/1040067312. 29
44
[6] Dǒsen, K., “Sequent-systems for modal logic,” Journal of Symbolic Logic, vol. 50
45
(1985), pp. 149–68. Zbl 0562.03009. MR 0780533. DOI 10.2307/2273797. 27
[7] Fine, K., “Vagueness, truth and logic,” Synthese, vol. 30 (1975), pp. 265–300.
46
Zbl 0311.02011. DOI 10.1007/BF00485047. 27
47
[8] Gabbay, D. M., “A theory of hypermodal logics: Mode shifting in modal logic,” Journal
48
of Philosophical Logic, vol. 31 (2002), pp. 211–43. Zbl 1009.03012. MR 1917266. DOI
49
 2023/02/22 v1.5.3 NDJFL article class F:ndj-19-0405.tex; p. 30

30 Brice Halimi

1 10.1023/A:1015782031903. 4
2 [9] Gallot, S., D. Hulin, and J. Lafontaine, Riemannian Geometry, Third Edition, Springer,
3 Berlin, 2004. Zbl 1068.53001. MR 2088027. DOI 10.1007/978-3-642-18855-8. 6, 7,
4 11, 18, 24, 28, 29
5 [10] Gross, J. L., and J. Yellen, eds., Handbook of Graph Theory, CRC Press, Boca Raton,
6
2004. MR 2035186. 14, 15
7
[11] Gudmundsson, S., and E. Kappos, “On the geometry of tangent bundles,” Expositiones
Mathematicae, vol. 20 (2002), pp. 1–41. Zbl 1007.53027. MR 1888866. DOI 10.1016/
8
S0723-0869(02)80027-5. 24
9
[12] Hughes, G. E., and M. J. Cresswell, A New Introduction to Modal Logic, Routledge,
10
London, 1996. Zbl 0855.03002. MR 1666445. 9
11 [13] Kremer, P., and G. Mints, “Dynamic topological logic,” Annals of Pure and Applied
12 Logic, vol. 131 (2005), pp. 133–58. Zbl 1067.03028. MR 2097225. DOI 10.1016/
13 j.apal.2004.06.004. 28
14 [14] Kripke, S. A., “Semantical considerations on modal logic,” Acta Philosophica Fennica,
15 vol. 16 (1963), pp. 83–94. Zbl 0131.00602. MR 0170800. 2
16 [15] Lang, S., Fundamentals of Differential Geometry, vol. 191 of Graduate Texts in Math-
17
ematics, Springer, New York, 1999. Zbl 0932.53001. MR 1666820. DOI 10.1007/
18
978-1-4612-0541-8. 6, 13, 24
[16] Masini, A., “2-Sequent calculus: A proof-theory of modalities,” Annals of Pure and
19
Applied Logic, vol. 58 (1992), pp. 229–46. Zbl 0819.03045. MR 1191942. DOI 10.1016/
20
0168-0072(92)90029-Y. 4
21
[17] McKinsey, J. C. C., and A. Tarski, “The algebra of topology,” Annals of Mathematics
22 (2), vol. 45 (1944), pp. 141–91. Zbl 0060.06206. MR 0009842. DOI 10.2307/1969080.
23 19
24 [18] Venema, Y., “Points, lines and diamonds: A two-sorted modal logic for projective
25 planes,” Journal of Logic and Computation, vol. 9 (1999), pp. 601–21. Zbl 0941.03020.
26 MR 1727211. DOI 10.1093/logcom/9.5.601. 28
27
28
29
30
Department of History and Philosophy of Science, Université Paris Cité, Paris, France;
[email protected]; https://sites.google.com/view/brice-halimi
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49