You are right. Here is an appropriate potential function, which unfortunately is not easy to describe.
Let
and for every
define
as the set of points closest to
.
Define a graph
whose vertices are the elements of
and every vertex
has incident edges
for every
. For each edge, we assign length equal the distance between its endpoints.
Notice that graph
does not depend on a particular partition of
into sets
and
.
Let
be a fixed ordering of the edges of
such that the length of edges does not increase.
Now, define the potential function associated with partition
as
where the coefficient
if the endpoints of edge
are in distinct parts (i.e., one in
and the other in
); otherwise
.
For example, suppose that
and
so that we move
from
to
.
The inequality
implies that
.
The potential here changes as follows:
(i) the coefficient of every edge
for
turns from 1 to 0 (there is at least one such edge);
(ii) the coefficients of edges
(if they present in
) for
turn from 0 to 1.
However, the edges of type (i) are strictly shorter than the edges of type (ii) (since
), and thus edges of type (i) have higher indices, making the potential decrease.