Nonnegative four-rank difference without decomposition
Let M be a persistence module on Fin N over a field, with each fiber finite
dimensional. For 0 ≤ i ≤ j < N, let r(i,j) be the dimension of the image of
the structure map M(i → j). The auxiliary integer extension
moduleRankInteger is zero unless 0 ≤ i ≤ j < N. Thus r(-1,j) = 0 and
r(i,N) = 0; an invalid pair does not represent a reverse structure map.
For natural b < d ≤ N, moduleMultiplicityInteger is the integer difference
r(b,d-1) - r(b-1,d-1) - r(b,d) + r(b-1,d).
The predecessor calculations occur in integers. The terminal endpoint d=N
means survival through the last observed stage, not an infinite lifetime.
The formula is nonnegative for every such module, with no interval-decomposition
assumption.
To see why, take a ≤ b ≤ c ≤ d and put U = im M(a → c) and
W = im M(b → c). Functoriality gives U ⊆ W. Restrict g = M(c → d) to
each image. Its range is the corresponding image at d, so rank-nullity gives
r(a,c) = r(a,d) + dim ker(g|U),
r(b,c) = r(b,d) + dim ker(g|W).
The two kernels identify inside the common ambient fiber at c with
U ∩ ker g and W ∩ ker g. Their dimensions are ordered by U ⊆ W.
Subtracting the rank-drop equations proves the four-rank inequality. Interior
endpoints follow by taking a=b-1, c=d-1. At b=0, the negative-source
terms vanish and ordinary rank drop applies. At d=N, the terminal terms
vanish and monotonicity in the source applies. For N=0 no admissible pair
exists.
The genuine fixtures check zero and one-stage boundaries, identity and zero
two-stage transitions, and two three-stage modules
k → k × k → k. Both three-stage modules use x ↦ (x,0) first. Projection
onto the first coordinate yields μ(1,2)=1; projection onto the second gives
μ(1,2)=0 and μ(0,2)=1, because the long composite is zero. These
calculations exercise the functor maps over ℚ and ZMod 2.
The closed interval [p,q] four-rank formula in
Chazal–Crawley-Boevey–de Silva, §1.1
uses p=b, q=d-1 for this library's half-open [b,d) convention.
The image-and-kernel proof above is a finite-dimensional linear-algebra
argument; it does not use the paper's decomposition statement.