Gromov-Hausdorff Distance

Summary of notes on the Hausdorff and Gromov-Hausdorff distance (see https://www2.math.upenn.edu/~brweber/Lectures/USTC2011/Lecture%205%20-%20Gromov-Hausdorff%20distance.pdf)

Let be a set in a metric space and . The r-thickening or r-neighborhood of is given by the cover of balls centered at each point of with radius , i.e.,

Hausdorff distance

For closed sets in , we define the Hausdorff distance by

One may show that is the farthest distance any point of is from , or the farthest distance any point of is from (take the larger one). Moreover, the set of closed sets in with forms a metric.

Note that if is not bounded, there may be closed sets and with .

The Gromov-Hausdorff distance generalizes the Hausdorff distance.

Gromov-Hausdorff distance

Given two closed metric spaces and , the Gromov-Hausdorff distance is given by

where denotes an isometric embedding of into some metric space . The infimum is taken over all such embeddings.

In other words, we embed the two metric spaces and into some larger metric space and find the Hausdorff distance. We then take the smallest value given across all possible embeddings.

Bottleneck Distance

Recall that a persistence diagram is simply a set of points in . Let and be two such persistence diagrams. We want to define a distance between the two diagrams. If and have the same cardinality, we can find bijections between the two. For each bijection we find the supremum of distances between a point in and its corresponding point in . We then want to the tahe smallest distance found. That is,

Bottleneck Distance

The bottleneck distance between two persistence diagrams is given by

One may easily see that is a metric.