An Introduction to Morse Theory Gianmarco Molino UConn Sigma Seminar 27 July, 2017 Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 1 / 1024
An Introduction to Morse Theory
Gianmarco Molino
UConn Sigma Seminar
27 July, 2017
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 1 / 1024
A quick introduction to Differential Geometry
Geometry is the study of shape, size, relative position of figures, andthe properties of space, and has been historically one of the majormotivating reasons for the field of mathematics.
In applying the methods of calculus to this, we arrive at the modernfield of Differential Geometry.
Our primary objects of interest are manifolds.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 2 / 1024
A quick introduction to Differential Geometry
A manifold is a generalization of Euclidean geometry, defined asobjects that “locally look like” Euclidean Rn.
The first examples of a nontrivial manifold are surfaces in R3, such asspheres, tori, and surfaces of revolution.
Some 2-dimensional manifolds can’t be embedded in R3, like theKlein bottle.
We can also consider higher-dimensional manifolds, but it can be veryhard to visualize these.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 3 / 1024
A quick introduction to Differential Geometry
Moreover, since we are considering differential geometry, we want toconsider smooth manifolds.
Simply put, smooth manifolds are manifolds on which calculus can bedone; there can be no sharp corners.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 4 / 1024
A quick introduction to Differential Geometry
One idea to keep in mind: when working with manifolds, you have tojump back and forth between local and global properties.
Locally (that is, in a small area around any point,) manifolds look justlike Rn. We have coordinate systems, and most of the ideas frommultivariable calculus can be carried forward in the way you expect.
Globally (that is, on the whole manifold,) most of these ideas don’tusually work.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 5 / 1024
What is Morse Theory?
In the following, let M be a closed smooth manifold of dimension n.
Initiated by Marston Morse, 1920-1930.
Study of critical points of smooth functions f : M → R.
Attempts to recover topological (not dependent on calculus)information about M.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 6 / 1024
1 Definitions
2 Motivating Example
3 First Results
4 Morse Inequalities
5 Existence Results
6 Applications and Further Reading
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 7 / 1024
Definitions
A smooth manifold M is a topological manifold with compatiblesmooth atlas.
A critical point p ∈ M of a smooth function f : M → R is a zero ofthe differential df .
The Hessian Hp(f ) of f at a critical point p ∈ M is the matrix ofsecond derivatives. (Independent of coordinate system at criticalpoints.)
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 8 / 1024
Morse Functions
A smooth function f : M → R is called Morse if its critical points arenondegenerate (that is, the Hessian of f is nonsingular.)
I Remark: Nondegenerate critical points are necessarily isolated.
The index λ(p) of a critical point p is the dimension of the negativeeigenspace of Hp(f ).
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 9 / 1024
Torus with height function
Consider the 2-dimensional torus T2 embedded in R3 and a tangent plane:
Define f : T2 → R to be the height above the plane.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 10 / 1024
The function h has 4 critical points, a, b, c , d , withλ(a) = 0, λ(b) = λ(c) = 1, λ(d) = 2.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 11 / 1024
Morse Lemma
Nondegeneracy of critical points is a generalization of non-vanishingof the second derivative of functions f : R→ R.
I Remember, the 2nd derivative test lets you decide if a critical point is alocal maximum or minimum if the 2nd derivative is nonzero.
We thus expect to be able to describe M in relation to these points.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 12 / 1024
Morse Lemma
Theorem (Lemma of Morse)
Let f ∈ C∞(M,R), and let p ∈ M be a nondegenerate critical point of f .Then there exists a neighborhood U ⊂ M of p and a coordinate system(y1, . . . , yn) on U such that yi (p) = 0 for all 1 ≤ i ≤ n, and moreover
f = f (p)− (y1)2 − · · · − (yλ)2 + (yλ+1)2 + · · ·+ (yn)2
where λ = λ(p) is the index of p.
Corollary
If p ∈ M is a nondegenerate critical point of f , then it is isolated.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 13 / 1024
Given f : M → R, define the ‘half-space’
Ma = f −1(−∞, a] = x ∈ M : f (x) ≤ a.
Theorem (Milnor)
Let f : M → R be C∞. If f −1([a, b]) is compact and contains no criticalpoints of f , then Ma is diffeomorphic to Mb and furthermore Ma is adeformation retract of Mb.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 14 / 1024
The gradient of f induces a local 1-parameter family of diffeomorphismsφt : M → M away from critical points. Thus allowing the points of Ma toflow along these gives the desired deformation retract.Remark: The condition that f −1([a, b]) be compact cannot be relaxed.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 15 / 1024
Theorem (Milnor)
Let f : M → R be C∞ and let p ∈ M be a (nondegenerate, isolated)critical point of f . Set c = f (p) and λ = λ(p) to be the index of p.Suppose there exists ε > 0 such that f −1([c − ε, c + ε]) is compact andcontains no critical points of f other than p. Then for all sufficiently smallε, Mc+ε has the homotopy type of Mc−ε with a λ-cell attached.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 16 / 1024
The key observation is that when crossing a critical point, the MorseLemma is applicable. It can be shown that attaching a λ-cell eλ to Mc−ε
along the (y1, . . . , yλ) axis,
Mc−ε ∪ eλ ∼= Mc+ε.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 17 / 1024
Intuitively then, the manifold can be constructed from cells determined bythe indices of the critical points.
Theorem (Milnor)
If f : M → R is Morse and for all a ∈ R it holds that Ma is compact, thenM has the homotopy type of a CW complex with one cell of dimension λfor each critical point with index λ.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 18 / 1024
This is enough to get a few results. For example,
Theorem (Reeb)
Let M be a compact smooth manifold, and let f : M → R be Morse. If fhas exactly two critical points, then M is homeomorphic to a sphere.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 19 / 1024
Differential Forms
Recall the space Ωk(M) of differential k-forms over M, and the exteriorderivative d : Ωk → Ωk+1, which gives rise to the deRham co-chaincomplex
0→ · · · d−→ Ωk(M)d−→ Ωk+1(M)
d−→ · · · → 0
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 20 / 1024
Betti Numbers
The associated cohomology group is the deRham cohomology group
HkdR(M) =
ker d : Ωk → Ωk+1
im d : Ωk−1 → Ωk
and further we define the k-th Betti number of M,
βk = dimHkdR(M).
This cohomology encodes topological information about the manifoldalgebraically, and is the starting point for fields such as Hodge Theory andIndex Theory.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 21 / 1024
The Betti numbers are topological invariants. They are related to theclassical Euler characteristic χ(M) by
χ(M) =n∑
k=0
(−1)kβk .
Which is an explicit expression for the following lemma from Index Theory:
Lemma
Let D = d + δ be the Dirac operator for the Hodge Laplacian∆ = D2 = dδ + δd . Then
χ(M) = index(D)
where index(D) = dim ker(D)− dim coker(D) denotes the analytic index.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 22 / 1024
The Betti numbers can be interpreted directly:
β0 is the number of connected components of M.
β1 is the number of 1-dimensional “holes” (nontrivial loops) in M.
β2 is the number of 2-dimensional “cavities” (nontrivial spheres) inM.
and so on.Note, the largest nonzero Betti number in a n-dimensional manifold M isβn.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 23 / 1024
The Betti numbers have many other properties, and are of wideinterest for many applications. They can also be defined for generaltopological spaces, not just smooth manifolds.
Unfortunately, the Betti numbers can be remarkably difficult tocompute directly. This is where Morse Theory provides a solution.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 24 / 1024
Weak Morse Inequalities
Let f : M → R be Morse, and define the Morse numbers, Mk , by
Mk = #p ∈ M, df (p) = 0, λ(p) = k
Theorem (Weak Morse Inequalities)
Let M be compact, βi be the Betti numbers of M, f : M → R be Morse,and Mk be the Morse numbers of f . Then
βk ≤ Mk
and moreover
χ(M) =n∑
k=0
(−1)kβk =n∑
k=0
(−1)kMk .
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 25 / 1024
Torus Example
The Weak Morse Inequalities give good estimates on the the Bettinumbers. For example, we have for T2
β0 ≤ M0 = 1
β1 ≤ M1 = 2
β2 ≤ M2 = 1
χ(T2) = M0 −M1 + M2 = 1− 2 + 1 = 0,
using the height function from before.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 26 / 1024
Witten’s Proof
We sketch the idea of Edward Witten’s remarkable proof:By a result from Hodge Theory,
βk = dim ker ∆: Ωk → Ωk .
Let f be Morse. Then we define the ‘twisted exterior derivative’
dt = e−tf detf
from which we can construct the ‘Witten Laplacian’
∆t = dtδt + δtdt : Ωk → Ωk ,
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 27 / 1024
There is an induced co-chain complex
0→ · · · dt−→ Ωk(M)dt−→ Ωk+1(M)
d−→ · · · → 0
which is isomorphic to the deRham complex, so that
βk = dim ker ∆k = dim ker ∆kt .
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 28 / 1024
But this is a remarkable improvement, leading to the conclusion that ast →∞ the elements of the kernel of ∆t will concentrate around thecritical points of f . Computations can then be approximated in localcoordinates, leading to the Morse Inequalities.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 29 / 1024
Strong Morse Inequalities
We can make the inequalities sharper.
Theorem (Strong Morse Inequalities)
Let M be compact, βi be the Betti numbers of M, f : M → R be Morse,and Mk be the Morse numbers of f . Then for any 0 ≤ k ≤ n,
βk − βk−1 + · · · ± β0 ≤ Mk −Mk−1 + · · · ±M0
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 30 / 1024
Polynomial Morse Inequalities
Define the Poincare Polynomial Pt =∑n
i=0 βi ti and the Morse Polynomial
Mt =∑n
i=0Mi ti .
Theorem (Polynomial Morse Inequalities)
Assumptions as before. For t ∈ R there exist some non-negative integersQi such that
Mt − Pt = (1 + t)n−1∑i=0
Qi ti
Lemma (Banyaga)
The Strong Morse Inequalities and the Polynomial Morse Inequalities areequivalent.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 31 / 1024
Raoul Bott writes (Morse Theory Indomitable):“The (1 + t) term on the right gives this inequality much more power thanit would have without it. The (1 + t) term feeds back information fromthe critical points of f to the topology of M.”
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 32 / 1024
Existence of Morse Functions
So, given a manifold M and a Morse function f we have nice results, butcan we actually find Morse functions?
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 33 / 1024
Yes. In fact, there is an ‘easy’ construction:
Theorem (Milnor)
Let M be a compact smooth manifold, and ι : M → RN be an embeddingof M into RN . For p ∈ RN , define Lp : M → R by
Lp(q) = ‖p − ι(q)‖2
where ‖ · ‖ is the standard Euclidean norm on RN . Then Lp is Morse foralmost every p ∈ RN .
Corollary
On any compact smooth manifold M there exists a Morse function, forwhich each Ma is compact.
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 34 / 1024
Theorem (Milnor)
Let M be a smooth manifold, K ⊂ M compact, and k ≥ 0 an integer. Anybounded smooth function f : M → R can be uniformly approximated by aMorse function g . Furthermore, for 1 ≤ i ≤ k it is possible to choose gsuch that the i-th derivatives of g on K uniformly approximate thecorresponding derivatives of f .
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 35 / 1024
Applications
There are a number of important applications, including
Classification of compact 2-manifolds
h-cobordism Theorem
Lefschetz Hyperplane Theorem
Yang-Mills Theory
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 36 / 1024
Openings
There is much active research deriving from Morse Theory:
Index Theory
Witten Helffer-Sjostrand Theory
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 37 / 1024
Further Reading
John Milnor, Morse Theory
Raoul Bott, Morse Theory Indomitable
Edward Witten, Supersymmetry and Morse Theory
Augustin Banyaga, Lectures on Morse Homology
Gianmarco Molino (UConn Sigma Seminar) An Introduction to Morse Theory 27 July, 2017 1024 / 1024