For a Lipschitz function $f: X \rightarrow X$, Clarke's generalized Jacobian at $x$ is defined as the convex hull of the following set:

$$\delta f (x) = \text{convex hull} \left \{\lim_{x_i \rightarrow x} Jf(x_i) : x_i \rightarrow x, f \text{ is differentiable at } x_i \right\}.$$

Clarke's inverse function theorem says that if every matrix in $\delta f(x)$ is invertible, then there is a Lipschitz inverse function $f^{-1}$ on some neighborhood of $x$.

It seems natural to guess that the generalized Jacobian of the inverse function $f^{-1}$ is just the convex hull of the inverses of all the matrices in the generalized Jacobian of $f$. That is,

$$\delta (f^{-1}) (x) = \text{convex hull} \{M^{-1} : M \in \delta f \}.$$

Is this true? If not, are there some special cases where it holds? I can't find it in Clarke's papers or anywhere else.


It is true. It is easy to get a formula for Crlarke directional derivative and it should follow from there.

  • $\begingroup$ Could you add details and explain how do you prove it? $\endgroup$
    – Amir Sagiv
    Jun 12 '20 at 16:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.