A proper k-coloring of a graph is acyclic if every 2-chromatic subgrap
h is acyclic. Borodin showed that every planar graph has an acyclic 5-
coloring. This paper shows that the acyclic chromatic number of the pr
ojective plane is at most 7. The acyclic chromatic number of an arbitr
ary surface with Euler characteristic chi = -gamma is at most O(gamma(
4/7)). This is nearly tight; for every gamma > 0 there are graphs embe
ddable on surfaces of Euler characteristic -gamma whose acyclic chroma
tic number is at least Omega(gamma(4/7)/(log gamma)(1/7)). Therefore,
the conjecture of Borodin that the acyclic chromatic of any surface bu
t the plane is the same as its chromatic number is false for all surfa
ces with large gamma (and may very well be false for all surfaces).