Morse Theory

consider being able to "read" the entire topology of a mountain landscape simply by studying its peaks, passes, and valleys. That is exactly what Morse theory does: it connects the critical points of functions to the "holes" of a space.

• **Proof of the Poincaré conjecture:** Smale used Morse theory to prove the generalized Poincaré conjecture in dimensions ≥5 (Fields Medal, 1966)
• **3D visualization:** The Reeb graph of a height function compactly encodes the topology of a 3D object: used in medical imaging
• **Neural network robustness:** The loss landscape is analyzed via Morse theory to understand the number and type of critical points

Предварительные знания

• Manifolds

Critical Points and the Morse Index

Neural network loss landscapes are smooth manifolds - Morse theory counts saddle points and local minima. A 2019 OpenAI paper showed overparameterized networks (10M+ params) have no spurious local minima: all critical points are saddles, which SGD escapes efficiently. Morse index of a saddle = number of negative Hessian eigenvalues.

A smooth function is called a Morse function if all its critical points are nondegenerate. Key fact: Morse functions are dense in C∞(M): almost any smooth function is a Morse function.

The Morse Lemma and Topology Changes

The **Morse Lemma** states that in a neighborhood of a nondegenerate critical point p of index λ, there exist local coordinates in which f takes a standard form. This allows a precise description of how the topology of sublevel sets f⁻¹(-∞, c] changes as c passes through a critical value.

Intuition: consider a mountain range. As water rises from below, we see: islands appear (λ=0), islands merge (λ=1), holes fill in (λ=2). This is exactly what happens to sublevel sets.

Morse Inequalities

Let cλ be the number of critical points of index λ and bλ = rank Hλ(M; Z) the λ-th Betti number. The **Morse inequalities** relate these numbers, giving lower bounds on the number of critical points.

Edward Witten in 1982 rediscovered the Morse inequalities through supersymmetric quantum mechanics. The Morse-Witten complex is a chain complex built from critical points of a Morse function, computing the homology of the manifold.

Applications and the Morse Complex

Morse theory became the foundation for several breakthroughs in 20th-century topology. It unites analysis and topology through concrete computational algorithms.

Morse theory inspired TDA: the Reeb graph of a Morse function captures the topology of level sets. It is used in 3D shape analysis and scientific visualization to detect topological features of data.

Key Ideas

• **Morse index λ**: the number of negative eigenvalues of the Hessian at a critical point
• **Morse Lemma**: near a critical point the function takes a standard quadratic form
• **Morse inequalities:** cλ ≥ bλ, exact identity: Σ(-1)λcλ = χ(M)
• **Morse complex** computes manifold homology via critical points and gradient flow trajectories

Related Topics

Morse theory bridges analysis and topology:

• Euler Characteristic — The Morse-Euler identity: Σ(-1)λcλ = χ(M)
• Persistent Homology — Persistent homology is a parametric generalization of Morse theory to filtrations
• Manifolds — Morse theory studies smooth functions on manifolds and their global consequences

Вопросы для размышления

• Why are "almost all" smooth functions on a manifold Morse functions?
• How are saddle points of a function related to "handles" in the decomposition of a manifold?
• What does it mean to "kill" a cycle using a high-index critical point?

Связанные уроки

• calc-20-extrema-multi