A spectral notion of Gromov–Wasserstein distance and related methods

We introduce a spectral notion of distance between objects and study its theoretical properties. Our distance satisfies the properties of a metric on the class of isometric shapes, which means, in particular, that two shapes are at 0 distance if and only if they are isometric when endowed with geodesic distances. Our construction is similar to the Gromov–Wasserstein distance, but rather than viewing shapes merely as metric spaces, we define our distance via the comparison of heat kernels. This allows us to establish precise relationships of our distance to previously proposed spectral invariants used for data analysis and shape comparison, such as the spectrum of the Laplace–Beltrami operator, the diagonal of the heat kernel, and certain constructions based on diffusion distances. In addition, the heat kernel encodes a natural notion of scale, which is useful for multi-scale shape comparison. We prove a hierarchy of lower bounds for our distance, which provide increasing discriminative power at the cost of an increase in computational complexity. We also explore the definition of other spectral metrics on collections of shapes and study their theoretical properties.