Information Technology Reference
In-Depth Information
there until the sweeper that starts from the nearest end point is
ε
-close. Then
run to the the middle until
ε
-close, turn, and run back until the end.
It is possible to show that there exists an optimal response in
M
1
∪M
2
against
any mixed strategy that is based on
. The conjecture in [
2
] is that conversely
against any mixed strategy that is based on
P
M
1
∪M
2
there exists an optimal
response that is in
. If this conjecture is true, then the princess and monster game
on an interval reduces to a standard optimization problem that can be solved numer-
ically by discretization and linear programming. An initial computation of the value
of the game under the assumption of the conjecture was carried out in [
2
], but it
contained some inaccuracies that have been corrected in [
10
]. The table below gives
the value of the game
V
n
against the number of grid points
n
of the discretization.
In [
2
] it was stated that the value of the game could perhaps be 11
P
8, but this table
demonstrates that the value must be slightly lower: the value
V
n
for
n
/
=
1
,
024 in
Tab le
5.1
is within 10
−
3
of the actual value of the game.
n
V
n
1
1
2
1
.
2667
4
1
.
3303
8
1
.
3547
16
1
.
3647
32
1
.
3689
64
1
.
3709
128
1
.
3719
.
256
1
3724
.
512
1
3726
3727
Tabl e 5. 1
Number of grid points versus value of the game
1
,
024 1
.
Since the solution of the game on an interval is already a hard problem, it may
seem that the solution of the game on an arbitrary graph is impossible. However,
one should observe the following nice confluency property of the princess' strategy
: suppose that
p
1
,
p
2
∈ P
and that
p
1
(
t
0
)=
p
2
(
t
0
)
for some
t
0
.Then
p
1
(
t
)=
p
2
(
t
)
for all
t
0
≥
t
(in Chap.
14
, Steve Alpern calls this 'sticky'). Even more so, a mixed
strategy based on
)
is the probability that
P
is in
A
at time
t
. Perhaps it is possible to derive the optimal
response of the monster from this property, and verify the conjecture. We conjecture
more generally, that for the game on a tree there exists an optimal princess strategy
that satisfies the confluency property.
P
can be described by a probability measure
μ
t
such that
μ
t
(
A
5.3 High-Low Search Games
Dresher's guessing game has recently been solved asymptotically [
8
], but there is a
very similar high-low search game that remains unsolved. It was proposed around
the same time as Dresher's game, by Ed Gilbert [
12
]. As in Dresher's game, Blue
Search WWH ::
Custom Search