The stability of persistence diagrams is a cornerstone of Topological Data Analysis, guaranteeing that small perturbations in data lead to small changes in topological summaries. While the Isometry Theorem provides a foundational link between the Gromov-Hausdorff distance on input spaces and the bottleneck distance on diagrams, a deeper, more powerful understanding comes from an algebraic perspective. This view reframes stability not at the level of diagrams, but at the level of the underlying persistence modules themselves.

By focusing on the interleaving distance between persistence modules, we can establish a more robust framework. This allows us to directly connect function-level perturbations to changes in homological structure, providing a tighter and more generalizable theory. The classic bottleneck distance and its generalization, the p-Wasserstein distance, then emerge as consequences of this fundamental algebraic relationship. The key result is the Algebraic Stability Theorem, which states that the persistence map from the space of persistence modules (with the interleaving distance) to the space of persistence diagrams (with the bottleneck distance) is 1-Lipschitz.

This result is powerful because the interleaving distance is often easier to bound directly from data transformations. For example, if two functions $f, g: X \to \mathbb{R}$ satisfy $||f - g||_\infty \le \delta$, then their corresponding sublevel set persistence modules are $\delta$-interleaved. The stability of the bottleneck distance is a direct corollary: $d_B(\text{Dgm}(f), \text{Dgm}(g)) \le d_I(V_f, V_g) \le \delta$.

The bottleneck distance $d_B$ measures the cost of the optimal matching between points of two persistence diagrams, where cost is the supremum of distances between matched points. Points can also be matched to the diagonal at a cost related to their persistence. While fundamental, its reliance on a single, worst-case pairing makes it insensitive to the overall distribution of differences, especially in noisy, complex datasets.

This is where the p-Wasserstein distance $W_p$ provides a more nuanced alternative. Instead of focusing on the supremum, it sums the $p$-th powers of the distances between matched points. For $p=1$, it captures the total cost of transforming one diagram into another. As $p \to \infty$, $W_p$ converges to the bottleneck distance.

The stability theorem extends to these distances. For $p \in [1, \infty]$, the map from persistence modules to diagrams is also $(2/p)^{1/p}$-Lipschitz with respect to the $p$-Wasserstein distance, though this bound is not always tight. For practical purposes, $p$-Wasserstein stability ($p < \infty$) is often more desirable as it better reflects the similarity of diagrams when there are many small differences rather than one large one.

The proof of the Algebraic Stability Theorem relies on constructing an explicit matching between the bars of two interleaved persistence modules. Given a $\delta$-interleaving $(\phi, \psi)$ between modules $V$ and $W$, one can show that for any bar in the barcode of $V$ with persistence greater than $2\delta$, there must be a corresponding bar in the barcode of $W$ that overlaps with it significantly.

This matching argument forms the core of the proof for q-tame modules, which are the class of modules for which the barcode decomposition is guaranteed to exist. The argument proceeds by showing that if a bar $[b, d)$ in $\text{Dgm}(V)$ is matched to a bar $[b', d')$ in $\text{Dgm}(W)$, their distance $||(b, d) - (b', d')||_\infty$ is bounded by $\delta$. If a bar is matched to the diagonal, its own persistence must be less than or equal to $2\delta$. Taking the supremum over all such matches yields the bottleneck stability result: $d_B(\text{Dgm}(V), \text{Dgm}(W)) \le \delta$.

In essence, the algebraic framework provides a powerful machine for proving stability. By establishing an interleaving between modules derived from different data or functions, the stability of their persistence diagrams follows directly. This is a more abstract but ultimately more flexible and powerful approach than relying on geometric constructions alone.