$\def\l{|\!|}$ Given the elastic net regression
$$\min_b \frac{1}{2}\l y - Xb \l^2 + \alpha\lambda \l b\l_2^2 + (1 - \alpha) \lambda \l b\l_1$$
how can an appropriate range of $\lambda$ be chosen for cross-validation?
In the $\alpha=1$ case (ridge regression) the formula
$$\textrm{dof} = \sum_j \frac{s_j^2}{s_j^2+\lambda}$$
can be used to give an equivalent degrees of freedom for each lambda (where $s_j$ are the singular values of $X$), and degrees of freedom can be chosen in a sensible range.
In the $\alpha=0$ case (lasso) we know that
$$\lambda > \lambda_{\textrm{max}} = \max_j|\sum_t y_t X_{tj}|$$
will result in all $b_j$ being zero, and $\lambda$ can be chosen in some range $(0, \lambda_\textrm{max})$.
But how to handle the mixed case?