The discriminating power of the generalized rank invariant

The rank invariant (RI), one of the best known invariants of persistence modules $M$ over a given poset P, is defined as the map sending each comparable pair $p\leq q$ in P to the rank of the linear map $M(p\leq q)$. The recently introduced notion of generalized rank invariant (GRI) acquires more discriminating power than the RI at the expense of enlarging the domain of RI to the set Int(P) of intervals of P or to an even larger set. Given that the size of Int(P) can be much larger than that of the domain of the RI, restricting the domain of the GRI to smaller, more manageable subcollections $\mathcal{I}$ of Int(P) would be desirable to reduce the total cost of computing the GRI. This work studies the tension which exists between computational efficiency and strength when restricting the domain of the GRI to different choices of $\mathcal{I}$. In particular, we prove that the discriminating power of the GRI over restricted collections $\mathcal{I}$ strictly increases as $\mathcal{I}$ interpolates between the domain of RI and Int(P). Along the way, some well-known results regarding the RI or GRI from the literature are contextualized within the framework of the M\"obius inversion formula and we obtain a notion of generalize persistence diagram that does not require local finiteness of the indexing poset for persistence modules. Lastly, motivated by a recent finding that zigzag persistence can be used to compute the GRI, we pay a special attention to comparing the discriminating power of the GRI for persistence modules $M$ over $\mathbb{Z}^2$ with the so-called Zigzag-path-Indexed Barcode (ZIB), a map sending each zigzag path $\Gamma$ in $\mathbb{Z}^2$ to the barcode of the restriction of $M$ to $\Gamma$. Clarifying the connection between the GRI and the ZIB is potentially important to understand to what extent zigzag persistence algorithms can be exploited for computing the GRI.