Next: M(k)-contract paths
Up: Lower Bounds on Path Length
Previous: O-contract Paths - Unrestricted
We conclude our results concerning -contracts by presenting a lower bound on
,
i.e. the length of paths when the utility functions are required to be monotone.
In principle one could attempt to construct appropriate monotone utility functions that would
have the desired properties with respect to the path used in Theorem 3. It is, however,
far from clear whether such a construction is possible. We do not attempt to resolve this
question here. Whether an exact translation could be
accomplished is, ultimately, a question of purely combinatorial interest: since
our aim is to demonstrate that exponential length contract paths are needed with monotone utility
functions we are not, primarily, concerned with obtaining an optimal bound.
Proof. We describe the details only for the case of being even: the result when is odd is obtained
by a simple modification which we shall merely provide in outline.
Let with . For any path
in
(where describes a subset of
by an -bit label),
the path
in
is defined by
(The reason for successive indices of increasing by will become clear subsequently)
Of course,
does not describe an -contract path6: it is, however, not difficult to interpolate appropriate allocations, , in
order to convert it to such a path. Consider the subsets (with ) defined
as follows:
If we now consider the path, , within
given by
then this satisfies,
- a.
- If has property (SC) of Theorem 3 in
then
has property (SC) in
.
- b.
- If is odd then .
- c.
- If is even then .
From (a) and the bounds proved in [Abbott & Katchalski, 1991]
we deduce that can be chosen
so that with denoting the allocation
- d.
- describes an -contract path from to .
- e.
- For each pair
with , the deal
is not an -contract.
- f.
- If is chosen as in the proof of Theorem 3 then the number of deals
in is as given in the statement of the present theorem.
We therefore fix as the path from Theorem 3 so that
in order to complete the proof we need to construct utility functions
that are monotone and with which defines the unique IR -contract path
realising the reallocation
.
The choice for is relatively simple. Given
,
In this is the number of allocations in . The behaviour of is clearly monotone.
The construction for is rather more complicated. Its main idea is to make use of the fact
that the size of each set occurring in is very tightly
constrained: is either or according to whether is odd or even.
We first demonstrate that
each set of size can have at most two strict subsets (of size ) occurring within :
thus, every of size has exactly or or subsets of size on .
To see this suppose the contrary.
Let , , , and be such that with
Noting that
and
that has the property (SC) it must be the case that (at least) two of the -bit
labels from
differ in at least two positions. Without loss of generality
suppose this is true of and . As a result we deduce that the sets
and have at most elements in common, i.e.
:
and
so in any position at which differs from ,
differs from
at exactly the same position. In total
, i.e. there are
(at least) two elements of that do not occur in ; and
in the same way
, i.e. there are
(at least) two elements of that do not occur in .
The set , however, has only members and so cannot have both
and as subsets: this would require
but, as we have just seen,
One immediate consequence of the argument just given is that for any set of size there
are exactly two strict subsets of occurring on if and only if
for some value of with .
We can now characterise each subset of
of size as falling into one
of three categories.
- C1.
- Good sets, given by
.
- C2.
- Digressions, consisting of
- C3.
- Inaccessible sets, consisting of
sets are those describing allocations to within the path defined by
; are the allocations that could
be reached using an -contract from a set of size on , i.e. , but
differ from the set that actually occurs in , i.e. .
Finally, sets are those that do not occur
on and cannot be reached via an -contract from any set on .
We note that we view any set of size that could be reached by an -contract from
as being inaccessible: in principle it is possible to extend the -contract path
beyond , however, we choose not complicate the construction in this way.
We now define as
It remains only to prove for these choices of
that the -contract path
defined from is the unique IR -contract path
realising the IR deal
and that is monotone.
To show that
is IR we need to demonstrate
We have via the definition of
Thus, via Definition 4, it follows that gives rise to an IR -contract path.
To see that this path is the unique IR -contract path implementing
,
consider any position
and allocation other than
or . It may be assumed that the deal
is an -contract.
If then
and . Hence
. In the former
case, and from which
and thus
is
not IR. In the latter case since is a from
and giving
. Again
fails to be IR since
fails to give any increase in the value of . We are left with the case
so that
and . Since
is
assumed to be an -contract this gives
. For the first possibility
could not be a set on : and are both subsets
of and there can be at most two such subsets occurring on . It follows,
therefore, that giving
so that
is not IR.
In the second possibility, but as so
the deal would result in an overall loss.
We deduce that for each the only IR -contract consistent with it is the
deal
.
The final stage is to prove that the utility function is indeed a monotone
function. Suppose and are subsets of
with . We need to show
that
. We may assume that , that occurs as some
set within , and that . If or but does not occur on we
have and the required inequality holds; if then in order for
to be possible we would need , which would give
and this is the maximum value that any subset is assigned by . We are left with
only , and on to consider.
It has already been shown that there are at most two subsets of that can occur on .
Consider the different possibilities:
- a.
- so that exactly two subsets of occur in : and .
Since
and this is at least
, should
be either of or then
as required.
- b.
- is a from
, so that and and, again,
.
We deduce that is monotone completing our lower bound proof for
for even values of .
We conclude by observing that a similar construction can be used if is odd: use the path
described above but modifying it so that one resource () is always
held by . Only minor modifications to the utility function definitions are needed.
Example 2
For
, we can choose
so that
. This gives
as
with the
-contract path being defined from
which is
Considering the 15 subsets of size
, gives
Notice that both of the sets in
are
: in principle we
could continue from
using either, however, in order to simplify the construction
the path is halted at
.
Following the construction presented in Theorem 4, gives the following
utility function definitions with
.
For
we obtain
The monotone utility functions,
, employed in proving Theorem 4
are defined so that the path arising from is IR: in the event of either agent
suffering a loss of utility the gain made by the other is sufficient to provide a compensatory
payment. A natural question that now arises is whether the bound obtained in Theorem 4
can be shown to apply when the rationality conditions preclude any monetary payment, e.g. for
cases where the concept of rationality is one of those given in Definition 7. Our next
result shows that if we set the rationality condition to enforce cooperatively rational
or equitable deals then the bound of Theorem 4 still holds.
Proof. We again illustrate the constructions only for the case of being even, noting the
modification to deal with odd values of outlined at the end of the proof of Theorem 4.
The path is used for both cases.
For (a), we require
to be defined as monotone functions with which
will be the unique cooperatively rational -contract path to realise the cooperatively rational deal
where
. In this
case we set
to be,
Since,
it is certainly the case that
and
all deals on the -contract path defined by are cooperatively rational.
Furthermore if
is any allocation other than then
the deal
will fail to be a cooperatively rational -contract. For suppose
the contrary letting
without loss of generality be an -contract, with
- we can rule out the former case since we have already shown such an
deal is not cooperatively rational. If so that
then
: the former
case leads to a loss in utility for ; the latter, (since is a from )
a loss in utility
for . Similarly, if so that
then
: for the first
leading to a loss of utility
for ; the second results in a loss of utility for .
It follows that the path defined by is the unique cooperatively rational -contract
path that realises
.
It remains only to show that these choices for
define monotone utility functions.
Consider and
suppose and are subsets of
with . If , or
does not occur on then . If or is
then which is the maximum value attainable by .
So we may assume that , occurs on , i.e.
for some , and that and is either a set or a .
From the definition of , : if
then
;
if is a from then
.
We deduce that if then
, i.e.
the utility function is monotone.
Now consider with and subsets of
having . If
or
does not occur in then its maximal value.
If or
is then . Thus we may
assume that
giving and ,
so that
is either a
or one of the sets
. If
is
a then ; if it is the set then
; if it
is the set then
. It follows that is monotone completing
the proof of part (a).
For (b) we use,
These choices give as the unique equitable -contract path
to realise the equitable deal
, since
each deal
is equitable.
If
is any allocation other than
then the deal
is not an equitable -contract. Assume that
is
an -contract, and that
.
If , so that
and
then
. In the first of these
;
in the second
since must be a .
This leaves only with
and
.
For this,
: if
then
(with equality when
); if
then
. In total these
establish that is the unique equitable -contract path realising the equitable deal
.
That the choices for
describe monotone utility
functions can be shown by a similar argument to that of part (a).
Example 3
For
and using the same
-contract path
as the previous example,
i.e.
For
in (a) we obtain
Similarly, in (b)
That we can demonstrate similar extremal behaviours for contract path length with rationality
constraints in both money-based (individual rationality) and money-free (cooperative rationality,
equitable) settings irrespective of whether monotonicity properties are assumed, has some interesting
parallels with other contexts in which monotonicity is relevant. In particular we can
observe that in common with the complexity results already noted from
dunne:2003 - deciding if an allocation is Pareto optimal, if an allocation maximises ,
or if an IR -contract path exists - requiring utility functions to be monotone does not result
in a setting which is computationally more tractable.
Next: M(k)-contract paths
Up: Lower Bounds on Path Length
Previous: O-contract Paths - Unrestricted
Paul Dunne
2004-11-26