A projection game consists of a bipartite graph , finite label sets
and
, and, for every edge
, a function
. A labeling assigns a label to every
vertex and satisfies
when the label of
equals the image under
of the label of
.
The value of the game is the largest fraction of edges that can be satisfied simultaneously.
Projection games are also called label-cover instances. They are central constraint systems in hardness-of-approximation reductions and in formulations of the PCP theorem.