Morse Theory Flashcards
Discuss brief history of Morse theory
Started in 1930’s by Morse. In 1980’s entirely new way of thinking of field emerged –> Floer theory
What is Morse theory about?
P = Reasonable space
f: P –> R a reasonably smooth function
Goal: Relate critical points of f to the topology of P (homology, homotopy etc)
can work both ways topology –> critical points and critical points –> topology
In particular prove Morse Inequalities
|Crit_k(f)| >= dim H_k(P)
Inductive move from min to max. See how topology of sublevels change as we go through critical points
pg 9, 19
Discuss setting of Morse theory for f.d. manifolds
P = M = Closed manifold.
f at least C^2 normally just C^inf
Crit(f) = {p | df(p) = 0}
Main question: Produce meaningful lower bound for Crit(f)
pg 10
What is very weak lower bound on number of critical points of closed manifold?
2 - max min
pg 10
What happens if f:M –> R has exactly 2 critical points? M closed manifold
M homeo to S^2, not nec. diffeo
Use Morse theory
pg 10, pg 63
Discuss critical points of height function for closed surface. Lower bound? Draw examples
p in Crit(f) <=> T_pP horizontal <– prove
Almost true that >= 2 + 2g
It is possible to construct a function on surface of g >0 with exactly 3 critical points
But is true for broad class of functions –> Morse functions
pg 11-12
Discuss Lie derivatives
See Lee book
Define: Hessian and discuss/prove properties
quadratic form/bilinear pairing
14-16
Can we define Hessian when p is not critical point?
Example?
No - not well-defined
Can find coordinates so f(x) = x_1 <– Hessian = 0
or more generally f(x) = x_1 + any quadratic <– Hessian = whatever you want
pg 15
Consider P^n < R^n+1 and f projection onto some coordinate - x_n+1. Discuss how P is locally graph(f)
pg 17
Define: p non-degenerate
Morse index
Morse function
Crit_k(f)
pg 18
What happens to {f<c} sublevel sets as we move through an interval where c is regular?
Proof?
Nothing happens – diffeomorphism type does not change
Pf. Use anti-gradient flow, key tool of Morse theory. Fix Riemannian metric on P. Now have well-defined gradient that looks the same as classical gradient as long as we choose nice coordinates
Scale gradient to have speed -1 everywhere. Call this X.
phi^t = flow of X
By construction, f decreases along the flow lines of X with unit speed => f(phi^t(x)) = f(x) - t
phi^(b-a) gives diffeo
pg 20 -21
What is Morse Lemma? Importance? Examples?
Answers question of what happens at critical points
Prop. p in Crit(f) is non-degen => exists a coordinate system s.t. f(x) = f(p) + 1/2 (-x1^2 - x2^2 - … - xk^2 + xk+1^2 + … + xn^2) = f(p) + 1/2 d_p^2f
k = ind_p (f)
By local change of coordinates at p, eliminate higher order terms in Taylor expansion
going from d^2_pf to (-x1^2 - x2^2 - … - xk^2 + xk+1^2 + … + xn^2) just linear algebra - diagonalize
Index is only local invariant of non-degen critical point. Two critical points with same index are locally diffeo
Notice in examples critical points f = 0 lie on a variety with singularities – not smooth. All other level sets give smooth manifolds
pg 22-23
What is Hadamard Lemma? Why is result non-obvious? Proof?
Corollary?
Lemma. f: R^n –> R, f(0) = 0. Then there exist g_i: R^n –> R s.t. g_i(0) = df/dx_i(0) and f(x) = sum g_i(x) x_i near 0.
Can replace R^n by any star-shaped neighborhood.
Pf. FTC used cleverly
Corollary. Assume f(x) = f(0) + h_ij(x)x_ix_j
Note: In general, Hadamard lemma takes over role of converging Taylor expansion
pg 24-27
Prove Morse Lemma
- The question is local - work in R^n
- f(x) = Q(x) + R(x). Want a diffeo phi so f o phi = Q
- Use Moser Homotopy Method
- Set f_t = Q + tR <– homotopy between Q and f. Look for family of diffeos phi_t s.t. f_t o phi_t = Q
- Differentiate and look for a generating time dependent vector field v_t
- Get linear system and solve
pg 28 - 32
Discuss normal forms and singularities of functions
Consider examples
1. Matrices - Jordan normal form
- Symmetric matrices - diagonal with +1, -1, 0 (or eigenvalues if we only allow orthogonal transformations)
- Functions near 0 (germs) - ?
Now (3) in more detail:
(a) df_p != 0, then can write f(x) = x1 + c
(b) df_p = 0. Morse Lemma provides normal form as long as Hessian non-degenerate - only invariant is index
(c) Can go down hierarchy to more and more degenerate functions - classification of singularities very popular in 1970s - difficult
pg 33-35
Discuss attaching cells to topological space. Define CW complex. Discuss cell structures on surfaces and projective spaces over R and C
Just form disjoint union and quotient by relation. Build CW inductively, attaching cells in increasing dimension
pg 36-38
Discuss 2 criteria for homotopy equivalence from Hatcher
See chapter 0 of Hatcher
Prove: Attaching a cell to a CW complex arbitrarily gives a space homotopic to a CW complex
If we are attaching a k+1 cell, we need to push the attaching map, which might hit cells of dim > k, down into k-skeleton.
Idea: If S^k hits interior of ball, we can homotope S^k to boundary of ball by radially projection from point of ball that is not in image of S^k. Then continue pushing inductively until we have pushed into k-skeleton
Problem: Space filling curves.
Solution: Approximate attaching map by map which is smooth on part mapped into ball (Weierstrass approximation theorem). Then Sard’s thm shows curve not space filling.
pg 39-41
Discuss cellular homology intuitively - chain complex, differential
C_k = free module over Z generated by the k-cells
differential = sum degree of attaching maps * e_k-1
pg 42-43
Discuss cellular homology with differential coming from l.e.s. of pair. Prove we get a chain complex…
See Chris/Hatcher
Attempt to do from purely geometric description as well…he failed so don’t spend too much time
pg 44-46
Show cellular homology = singular homology whenever both are defined
So cellular homology is a homotopy invariant.
See Chris/Hatcher
pg 47-49
Use CW structure to calculate homology groups of surfaces and projective spaces over R and C
See Chris/Hatcher
pg 49
What can we say about topology of a sublevel containing a critical point? Corollaries? Proof of corollaries? Don’t prove main.
Thm. If only critical value in [a,b] is c, then Pb obtained from Pa by attaching a k-cell for every critical point x on the critical level f = c, where k = index(x).
Cor. P has homotopy type of CW complex with one cell for each crit pt with dim = index
In essense exists a complex (C_k, d) with C_k generated by Crit_k(f) over F
Cor. Morse Inequalities. |Crit_k(f)| >= dim H_k(P, F) = b_k or rk H_k((P, Z).
pg 50-51