1: \begin{abstract}

2:   A basic issue in optimization, inverse theory,

3:   neural networks, computational chemistry and many other problems is

4:   the geometrical characterization of high dimensional functions.  

5:   In inverse

6:   calculations one aims to characterize the set of models that fit the

7:   data (among other constraints).  If the data misfit function is

8:   unimodal then one can find its peak by local optimization methods

9:   and characterize its width (related to the range of data-fitting

10:   models) by estimating derivatives at this peak.  On the other hand,

11:   if there are local extrema, then a number of interesting and

12:   difficult problems arise.  Are the local extrema important compared

13:   to the global or can they be eliminated (e.g., by smoothing) without

14:   significant loss of information?  Is there a sufficiently small

15:   number of local extrema that they can be enumerated via local

16:   optimization?  What are the basins of attraction of these local

17:   extrama?  Can two extrema be joined by a path that never goes

18:   uphill?  Can the whole problem be reduced to one of enumerating the

19:   local extrema and their basins of attraction?  For locally

20:   ill-conditioned functions, premature convergence of local

21:   optimization can be confused with the presense of local

22:   extrema. Addressing any of these issues requires topographic

23:   information about the functions under study.  But in many

24:   applications these functions may have hundreds or thousands of

25:   variables and can only be evaluated pointwise (by some numerical

26:   method for instance).  In this paper we describe systematic 

27:   (but generic) methods

28:   of analysing the topography of high dimensional functions using local

29:   optimization methods applied to randomly chosen starting models.  We

30:   provide a number of quantitative measures of function topography

31:   that have proven to be useful in practical problems along with error

32:   estimates.

33: \end{abstract}

34: