Modal and Temporal Logic, 2011-2012

James Lawson

1.(a) i. {1, 2, 3}. 1 because 1 sees 2, and ¬p is true at 2. 2 because 2 sees 4, and
¬p is true at 4. 3 because 3 sees 4, and ¬p is true at 4. Not 4 (it has no
accessible worlds).
ii. {2, 3}. p is true at world 4 only (all other worlds can see some world
where p is false.) So only the worlds that can access 4 satisfy ♦p. These
are worlds 2, 3.
iii. {2, 4}. Look for worlds where either ¬p is true, or p is true, or both.
¬p is true at t iff there is some u s.t. R(t, u) where p is false at u.
Not 1 - no u as above and p is false at 1. 2, because ¬p is true - (2
can access 4, and p is false at 4). Not 3 - no u as above and p is false
at 3. 4, because p is true.
(b) i. May be true.
ii. May be true.
iii. Definitely true.
iv. Definitely false.
(c) i.

p ∨ (p → ¬p) ≡ (p → ¬p) ∨ p A∨B ≡ B∨A

≡ ¬♦¬(p → ¬p) ∨ p ¬♦¬A ≡ A
≡ ¬♦¬(¬[p ∧ ¬¬p]) ∨ p A → B ≡ ¬(A ∧ ¬B)
≡ ¬♦[p ∧ p] ∨ p ¬¬A ≡ A
≡ ♦[p ∧ p] → p A → B ≡ ¬A ∨ B

ii. ¬(♦[p ∧ p] ∧ ¬p)

iii. Let A = ♦[p ∧ p] ∧ ¬p
We will apply Sahlqvist’s algorithm to find the Sahlqvist’s correspondant
of A, α[t] and thus have F, u |= ¬A iff F |= ∀t¬α(t). Boxed atoms: p.
Negative formulas: ¬p. We want to find a lazy assignment that makes
boxed atoms true without concern for negative formulas. Suppose arbt
world t sees some world u. We need to make p true at all worlds, v that u
can see (should it see any). So lazy assignment is ho (p) = {x | R(u, x)}.
Now we take the standard translation of A:
At = (♦[p ∧ p])t ∧ (¬p)t
= ∃u(R(t, u) ∧ [p ∧ p]u ) ∧ ¬pt
= ∃u(R(t, u) ∧ [(p)u ∧ pu ]) ∧ ¬pt
= ∃u(R(t, u) ∧ [(p)u ∧ P(u)]) ∧ ¬P(t)

1
 Our lazy assignment allows us to replace (p)u with >u = >, replace P(t)
with R(u,t), and P(u) with R(u, u). This gives:
∃u(R(t, u) ∧ [> ∧ R(u, u)]) ∧ ¬R(u,t).
We move ∃u to preserve equivalence:
∃u(R(t, u) ∧ [> ∧ R(u, u)] ∧ ¬R(u,t)) = α[t]
So ∀t¬α(t) = ∀t¬∃u(R(t, u)∧[>∧R(u, u)]∧¬R(u,t)) which is equivalent
to ∀t∀u([R(t, u) ∧ R(u, u)] → R(u,t)).
Since (F, h),t |= A iff F |= (F, ho ),t |= A iff F |= α[t].
we have: (F, h),t |= ¬A iff F |= ∀t¬α[t] for any F, h,t. That is, B is valid
in F iff F satisfies (using first-order semantics) the sentence ∀t¬α[t].

2.(a) i. F × N = (W × , R× ) where W × = {(x, y) | x ∈ W, y ∈ N}, and R× is a rela-

tion on W × where (x, u)R× (y, v) iff xRy and u < v.
ii. Take any world (x, u) ∈ W × . Assume (x, u)R× (x, u). Then by def of R× ,
u < u. Contradiction ((N, <) is irreflexive). Hence for all worlds w ∈ W × ,
wR× w is false.
iii. Take any worlds (x, u), (y, v), (z, w) ∈ W × .
Assume (x, u)R× (y, v) and (y, v)R× (z, w). Then by def of R× , we have
R(x, y), R(y, z), but F is transitive so we have R(x, z). Also, by def of R× ,
we have u < v and v < w. So u < w ((N, <) is transitive). Since R(x, z),
and u < w, we have (x, u)R× (z, w).
(b) Let A be any modal formula valid in flow of times. From (a)(ii) and (a)(iii),
F × N is a flow of time so A is valid in it. From lectures, there is a p-morphism
from F × N to F. So F is a p-morphic image of F × N . From lectures, p-
morphisms preserve validity forwards, that is, if F 0 is a p-morphic image of F,
then any modal formula valid in F is valid in F 0 . So A is valid in F.
(c) i. ⊥Uq → >Uq
ii. qS> → qU>
(d) Assume (AU[A ∧ (AUB)]) holds at t for some arbitrary model M = (N , h).
By semantics of Until, there is some time, f , in the future of t where M, f |=
A ∧ (AUB) and for all f 0 such that t < f 0 < f , we have M, f 0 |= A.
Because M, f |= A ∧ (AUB), there is some future time of f , l, M, l |= B and
for all l 0 such that f < l 0 < l, we have M, l 0 |= A.
So we must have M,t |= AUB. There is some time in the future of t, namely l,
where B is true. And for all times t 0 with t < t 0 < l, we have M,t 0 |= A - given:
t < t 0 < f satisfies A, t 0 = f satisfies A and f < t 0 < l satisfies A.

3.(a) i. The logic of a class C is the set of all modal formulas that are valid in all
the frames in C

2
 ii. L is sound if every modal formula proven by L is valid in all frames of C.
L is complete if it can prove every modal formula in the logic of C.

(b) If C1 ⊆ C2 then Log(C2 ) ⊆ Log(C1 ).

Proof. Take some formula A, in Log(C2 ). To show A in Log(C1 ).
If A ∈ Log(C2 ) then A is valid in all the frames of C2 . All of the frames in C1
are in C2 , so A is valid in all the frames of C1 . Hence A ∈ Log(C1 ).
(c) p → p is valid in reflexive frames. In lectures we proved this several times
e.g. using direct proof, using Sahqvist algorithm to get correspondant ∀tR(t,t).
However p → p is not valid in serial frames. To see this, find some model
M constructed from some serial frame F and verify that p → p is not valid
in M. Take serial frame F = (N, <) and form model M with h(p) = {w ∈
N|n > 1}. Then at world 1, we have p, but not p, hence M, 1 2 p → p. So
p → p not valid over serial frames.
(d) i. Assume without proof: M, w |= A → B iff [if M, w |= A then M, w |= B].
Assume M, w |= p ∧ ♦q. To show M, w |= ♦(p ∧ q).
We have M, w |= ♦q, so w has some accessible world, u, with M, u |= q.
But since M, w |= p, we must have M, u |= p. So w has some accessible
world that satisfies p and q. Hence M, w |= ♦(p ∧ q).
ii. 1. `K (p ∧ ¬[p ∧ q]) → ¬q taut.
2. `K [(p ∧ ¬[p ∧ q]) → ¬q] UG(1).
3. `K (A → B) → (A → B) (instance of normality).
4. `K ((p ∧ ¬[p ∧ q]) → ¬q) → ((p ∧ ¬[p ∧ q]) → ¬q) SUB(3).
5. `K (p ∧ ¬[p ∧ q]) → ¬q MP(2,4).
6. `K (p ∧ ¬[p ∧ q]) → ¬q RofE(5). (A ∧ B) ≡ A ∧ B
7. `K (p ∧ ¬¬q) → (¬¬(p ∧ q)) RofE(6). 6 ≡ 7
8. `K (p ∧ ♦q) → ♦(p ∧ q)) RofE(7). ♦A ≡ ¬¬A
4.(a) i. Φ is maximally consistent with respect to L iff
• it is L-consistent, there are no formulas A1 , ..., An ∈ Φ, n > 0, such that
`L (A1 ∧ ... ∧ An ) → ⊥.
• it is maximal, there is no larger set that is L-consistent.
ii. First we will show Φ is consistent.
Suppose it was not. Then there are formulas A1 , ..., An ∈ Φ such that `L
(A1 ∧ ... ∧ An ) → ⊥. Since K4 is sound, (A1 ∧ ... ∧ An ) → ⊥ is valid in the
class of transitive frames. Since M was based on a transitive frame,we
must have M, w |= (A1 ∧ ... ∧ An ) → ⊥.
But since A1 , ..., An ∈ Φ, we must have: M, w |= A1 , ..., M, w |= An , by def.
of Φ. So M, w |= A1 ∧ ... ∧ An . So M, w |= ⊥ which gives a contradiction.
Hence Φ is consistent.

3
 Now to show it is maximally consistent. Suppose some formula A is not
in Φ and Φ0 = Φ ∪ {A} gives a consistent set. Since A is not in Φ, we must
have M, w |= ¬A, by def of Φ. So ¬A is in Φ and so is also in Φ0 .
But, having A, ¬A ∈ Φ0 along with the proof:
(1). (p ∧ ¬p) → ⊥ (taut)
(2). (A ∧ ¬A) → ⊥ (sub)
shows that Φ0 is inconsistent. Contradiction.
Hence Φ is maximally consistent.
(b) i. Assume R0 (X,Y ). To show R0 (Y, X).
By def. of R0 (X,Y ), there is some x ∈ X, y ∈ Y such that R(x, y). But since
R is symmetric, we also have R(y, x). Since we have some y ∈ Y, x ∈ X
such that R(y, x), by def. of R0 , R0 (Y, X). Hence R0 is symmetric.
ii. Let M = ({1, 2, 3, 4}, {(2, 1), (4, 3)}, h) where h(p) = {1}. Let A = p.
Here R is transitive, but R0 resulting from the filtration of M wrt A is not.
The filtration is (W 0 , R0 , h0 ) = ({X,Y, Z}, (Z,Y ), (Y, X), h0 ) where h0 (p) =
{X}. Here R0 is not transitive because there are relations R0 (Z,Y ) and
R0 (Y, X) but not R0 (Z, X).
(c) i. L has the finite model property iff for any formula A where A ∈ / L there is
a finite model, M such that for all B ∈ L, B is valid in M, and for some
world t of M, we have M,t |= ¬A.
L is decidable iff there is some algorithm/program that given any modal
formula A as an input, outputs yes when A ∈ L and outputs no when A ∈ / L.
ii. Assume we have an algorithm to check a model validates a logic.
Then if modal logic has the finite model property, then it is decidable. We
run two algorithms in parallel. The first enumerates all possible theorems,
the second enumerates all possible finite models.
For each finite model enumerated, check it validates L. and see whether
M satisfies ¬A. If so, then it is a non-theorem. We halt and print no. If
L has the finite model property, then every non-theorem will eventually
be printed. Otherwise A is a theorem, it will never be found by the first
algorithm, and will eventually be printed by the first algorithm - at this
point we halt and print yes.