Why is a contraction not defined as a function from one metric space $M$ to another one $N$ with the usual property, i.e.: $ \exists k \in (0,1): $
$$ d_{N}(f(x), f(y)) \le k \, d_M(x,y), \quad \forall x,y \in M $$
Why is a contraction not defined as a function from one metric space $M$ to another one $N$ with the usual property, i.e.: $ \exists k \in (0,1): $
$$ d_{N}(f(x), f(y)) \le k \, d_M(x,y), \quad \forall x,y \in M $$
Primary use of contraction is in Banach fixed point theorem. Formulation of the theorem assumes, that contraction composition is well defined. Function can be composed with itself only when image is subset of domain. Note, that if you let $k$ in your definition to be arbitrary large, this will give definition of Lipschitz continuity between metric spaces $M$ and $N$: $$\exists k\in [0,+\infty) \ \forall x,y \in M: \ d_N(f(x),f(y))\leq kd_M(x,y) $$ According to @mTur11 your definition is mentioned on Wikipedia, but it is uncommon. When speaking about contraction in general, $N=M$ is given unless otherwise stated.