The concept of "magnitude" is a way to measure the effective size or complexity of a metric space, capturing how spread out its points are. This paper studies what happens to a generalized version of magnitude as you zoom in with increasing resolution, essentially asking: as you scale up the distances between points, what value does the magnitude approach? The setup is quite general, covering spaces where the distances need not satisfy the usual triangle inequality and where the mathematical machinery behind magnitude can break down in awkward ways.
The main result is that this high-resolution limit always exists and, whenever it is a finite number, it must be a rational number. This is surprising because the formula for magnitude involves matrix inverses and exponentials, which could in principle produce messy irrational values. The paper shows that the geometry of which limits are possible is controlled by a structured combinatorial object called a polyhedral fan, meaning only finitely many different finite limits can arise for spaces with a fixed number of points. The authors also show that any achievable limit can be realized by a space using only simple, integer-valued distances, which makes the theory more concrete and constructive.
The paper also explores what happens when you allow the number of points to vary, classifying which combinations of limit values are possible using remarkably simple distance values like 0, 1, 2, or 3. A complete classification is given for spaces with at most four points. Additionally, the authors handle a tricky technical situation where the key matrix remains singular at all scales, deriving an exact algebraic decomposition for that case. They close with a clean geometric characterization: the magnitude matrix behaves nicely (in a positive semidefiniteness sense) exactly when points at zero distance from each other are treated as indistinguishable, in which case the limiting magnitude simply counts how many distinct clusters of zero-distance points exist.