The happy end problem, also called the "happy ending problem," is the problem of determining for
the smallest number of points
in general position
in the plane (i.e., no three of which are collinear),
such that every possible arrangement of
points will always contain at least one set of
points that are the vertices of a convex
polygon of
sides. The problem was so-named by Erdős when two investigators who first worked
on the problem, Ester Klein and George Szekeres, became engaged and subsequently
married (Hoffman 1998, p. 76).
Since three noncollinear points always determine a triangle, .
Random arrangements of
points are illustrated above. Note that no convex quadrilaterals are possible for
the arrangements shown in the fifth and eighth figures above, so
must be greater than 4. E. Klein proved that
by showing that any arrangement
of five points must fall into one of the three cases (left top figure; Hoffman 1998,
pp. 75-76).
Random arrangements of points are illustrated above. Note that no convex pentagons
are possible for the arrangement shown in the fifth figure above, so
must be greater than 8. E. Makai proved
after demonstrating that a counterexample could be found
for eight points (right top figure; Hoffman 1998, pp. 75-76).
As the number of points increases, the number of
-subsets of
that must be examined to see if they form convex
-gons increases as
, so combinatorial explosion prevents cases much bigger
than
from being easily studied. Furthermore, the parameter space become so large that
searching for a counterexample at random even for the case
with
points takes an extremely long time. For these reasons,
the general problem remains open.
was demonstrated by Szekeres and
Peters (2006) using a 1500 CPU-hour computer search which eliminated all possible
configurations of 17 points which lacked convex hexagons while examining only a tiny
fraction of all configurations. Marić (2019) and Scheucher (2020) independently
verified
using satisfiability (SAT) solving in a few CPU hours, a time later reduced to 10
CPU-minutes by Scheucher (2023) and to 8.53 CPU-seconds by Heule and Scheucher (2024).
The first few values of for
, 4, 5, and 6 are therefore 3, 5, 9, 17, which happen to
be exactly
.
However, the values of
for
are unknown.
Erdős and Szekeres (1935) showed that always exists and derived the bound
(1)
|
where
is a binomial coefficient. For
, this has since been reduced to
for
(2)
|
by Chung and Graham (1998), for
(3)
|
by Kleitman and Pachter (1998), and for
(4)
|
by Tóth and Valtr (1998).