apply => + => + => + => 6 / \ / \ / \ / \ λ apply apply 1 + 1 5 1 / \ / \ / \ / \ x + λ 3 λ 3 3 2 / \ / \ / \ x 1 y + y + / \ / \ y 2 y 2
apply => λ => λ λy.y / \ / \ / \ λ λ y apply y y / \ / \ / \ x λ z z λ y / \ / \ y apply z z / \ x y
apply / \ apply apply / \ / \ L L L apply / \ / \ / \ / \ x x x y x x L L / \ / \ x apply x apply / \ / \ x x x x
apply1 / \ / \ L apply3 / \ / \ x apply2 L a / \ / \ x x y yIn this example we have two redexes: apply1 and apply3; apply1 is outermost and apply3 is innermost.
If we do NOR (reduce apply1 first) we will get two copies of the tree rooted at apply3 (replacing the two instances of x in the subtree rooted at apply2). We'll then need to reduce each instance of apply3, for a total of 3 reductions.
If instead we do AOR (reduce apply3 first) we will replace the subtree rooted at apply3 with just the expression "a". Now when we reduce apply1 we'll get (a a), an expression in normal form. And we'll have used only 2 reductions to get there.