Matching distance along positive affine slices
Let \(M,N\) be persistence modules on \(\mathbb R^n\), with \(n\ge1\), and let \(L(t)=b+ta\) have every \(a_i>0\). Put \(w(L)=\min_i a_i\). The positive minimum slope controls how a diagonal coordinate shift fits along the line:
Consequently, an actual coordinate \(\varepsilon\)-interleaving induces an actual real \(\varepsilon/w(L)\)-interleaving of the two restrictions. This construction works over arbitrary coefficient rings and uses the original module structure maps.
The restricted interleaving maps
Let \(\alpha_x:M(x)\to N(x+\varepsilon\mathbf1)\) be a supplied forward interleaving map and put \(\delta=\varepsilon/w(L)\). The restricted component is the composite
The second arrow is the actual structure map for the displayed coordinate inequality. The reverse component is defined in the same way. Naturality and the original double-shift equations prove both restricted interleaving equations. With unequal slopes, the second arrow can be nontrivial.
For example, take \(L(t)=(1+2t,-1+t)\), so \(w(L)=1\). With \(\varepsilon=1\), the slice shift is also \(\delta=1\). At \(t=0\) the three grades in the component are
Over a field, let \(H\) be the hook supported on the points \(x\) with
\((1,-1)\le x\) and \((3,0)\nleq x\). Use its coordinate self-interleaving
CoordinateInterleaving.refl 1 H. The forward map sends the element \(1\)
at \(L(0)\) to a nonzero element at \((2,0)\). The additional structure map
sends it to zero at \((3,0)\), which is outside the hook's support.
The existing
diagonal_transition_survives_but_padded_slice_dies fixture
checks both conclusions. Its line and endpoints are private fixture data.
This is a per-line witness example; its slopes are not normalized for the
supremum below.
AffineSliceInterleaving constructs these maps and exposes their component
equations. It passes from witnesses to the extended infimum defining
interleaving distance, obtaining
The proof includes zero interleaving radius and infinite extended distances; it assumes no minimizing interleaving.
Ordinary barcodes and the normalized supremum
Now take coefficients in an arbitrary field and suppose that \(M,N\) have original finite graded-free presentations. The slice presentation theorem derives the same property for their actual restrictions. Let \(B_L(M)\) and \(B_L(N)\) be their existing ordinary finite barcodes. The one-dimensional barcode isometry gives
PositiveAffineLine.ordinarySliceBarcode constructs the barcode from the
original presentation and exposes its positivity and actual realization.
weightedSliceBottleneck is the weighted ordinary bottleneck value above.
Normalize the lines by
This family is nonempty: the diagonal line with slopes one and offsets zero is normalized. Define matching distance by
sliceMatchingDistance_le_coordinateInterleavingDistance proves that this
extended supremum is at most coordinate interleaving distance. Self-distance
is zero, symmetry and the extended triangle inequality hold, and natural
isomorphisms preserve the value. These are extended pseudometric laws;
no separation or completeness theorem for module isomorphism classes is
asserted. The definition uses actual ordinary slice barcodes.
The witness construction corresponds to Landi, Lemma 1, and the normalized bound to Theorem 1 of the same version. The source states the barcode result for pointwise finite-dimensional modules; this library consumer uses original finite presentations, from which the finite slice barcodes are derived. The per-line estimate permits arbitrary positive slopes. Equivalence with a supremum over an unnormalized family and finite attainment of the supremum are outside these statements.