The momentum operator $\hat{p}$ can be written as $$ \hat{p} = i\hbar \lim_{a \rightarrow 0} \frac{\hat{T}(a\hat{r})- \hat{\mathbb{I}}}{a} $$ where $\mathbb{\hat{I}}$ is the identity operator and $\hat{r}$ being the unit vector in a given direction. For the case of a single pasrticle with wavefunction $\psi(r)$ it can be written as $$ ( \hat{p} \psi )(r) = i\hbar \lim_{a \rightarrow 0} \frac{(\hat{T}\psi)(r)- \psi(r)}{a} = i\hbar \lim_{a \rightarrow 0} \frac{\psi(r-a)- \psi(r)}{a} \, . $$
Now I am not sure how to proceed. My thought was to use L'Hôpital's rule: $$ = i\hbar \lim_{a \rightarrow 0} \frac{\frac{d}{da}[\psi(r-a) - \psi(r)]}{\frac{d}{da}a} = i\hbar \lim_{a \rightarrow 0 } \frac{\partial \psi(r-a)}{\partial a} = i \hbar \frac{\partial \psi(r)}{\partial a} = 0 \, . $$ However, it is stated on Wikipedia that the result should be $$ (\hat{p}\psi )(r) = -i \hbar \frac{\partial \psi(r)}{\partial r} \, . $$ I am not sure as to why the wavefuction is being derived with respect to $r$ and not $a$.