1: \begin{abstract}

2: We consider the problem of numerically identifying roots of a target function

3: -- under the constraint that we can only measure the derivatives of the

4: function at a given point, not the function itself. We describe and

5: characterize two methods for doing this:  (1) a local-inversion ``inching

6: process", where we use local measurements to repeatedly identify approximately

7: how far we need to move to drop the target function by the initial value over

8: $N$, an input parameter, and (2) an approximate Newton's method, where we

9: estimate the current function value at a given iteration via estimation of the

10: integral of the function's derivative, using $N$ samples.  When applicable,

11: both methods converge algebraically with $N$, with the power of convergence

12: increasing with the number of derivatives applied in the analysis.

13: \end{abstract}

14: