Tags: #projects/my-talks #projects/reading-groups
See Floer Reading Group Fall 2020
Summary/Outline
Outline
What we’re trying to prove:
- 8.1.5: (dF)u is a Fredholm operator of index μ(x)−μ(y).
What we have so far:
- Define L:W1,p(R×S1;R2n)⟶Lp(R×S1;R2n)Y⟼∂Y∂s+J0∂Y∂t+S(s,t)Y where S:R×S1→Mat(2n;R)S(s,t)s→±∞→S±(t).
Outline
-
Took R±:I→Sp(2n;R): symplectic paths associated to S±
-
These paths defined μ(x),μ(y)
-
Section 8.7: R±∈S:={R(t) | R(0)=id, det(R(1)−id)≠0}⟹L is Fredholm.
-
WTS 8.8.1: Ind(L)Thm?=μ(R−(t))−μ(R+(t))=μ(x)−μ(y).
From Yesterday
-
Han proved 8.8.2 and 8.8.4.
- So we know Ind(L)=Ind(L1)
-
Today: 8.8.5 and 8.8.3:
- Computing Ind(L1) by computing kernels.
8.8.5: dimkerF,F∗
Recall
L:W1,p(R×S1;R2n)⟶Lp(R×S1;R2n)Y⟼∂Y∂s+J0∂Y∂t+S(s,t)YL1:W1,p(R×S1;R2n)⟶Lp(R×S1;R2n)Y⟼∂Y∂s+J0∂Y∂t+S(s)YL⋆1:W1,q(R×S1;R2n)⟶Lq(R×S1;R2n)Z⟼−∂Z∂s+J0∂Z∂t+S(s)tZ
Here 1p+1q=1 are conjugate exponents.
Reductions
L∗1=−∂∂s+J0∂∂t+S(s)t.
-
Since cokerL1≅kerL∗1, it suffices to compute kerL∗1
-
We have
J10:=[0−110]⟹J0=[J10J10⋱J10]∈n⨁i=1Mat(2;R).
- This allows us to reduce to the n=1 case.
Setup
L1 used a path of diagonal matrices constant near ∞: S(s):=(a1(s)00a2(s)), with ai(s):={a−iif s≤−s0a+iif s≥s0.
\begin{center} \includegraphics[width = \textwidth]{figures/image_2020-05-27-20-10-07.png} \end{center}
Statement of Later Lemma (8.8.5)
Let p>2 and define F:W1,p(R×S1;R2)⟶Lp(R×S1;R2)Y↦∂Y∂s+J0∂Y∂t+S(s)Y.
Note: F is L1 for n=1: L1:W1,p(R×S1;R2n)⟶Lp(R×S1;R2n)Y⟼∂Y∂s+J0∂Y∂t+S(s)Y.
Statement of Lemma
F:W1,p(R×S1;R2)⟶Lp(R×S1;R2)Y↦∂Y∂s+J0∂Y∂t+S(s)Y.
Suppose a±i∉2πZ.
- Suppose a1(s)=a2(s) and set a±:=a±1=a±2. Then
dimKerF=2⋅♯{ℓ∈Z | 2πℓ∈(a−,a+)⊂R}dimKerF∗=2⋅♯{ℓ∈Z | 2πℓ∈(a+,a−)⊂R}.
- Suppose sups∈R‖S(s)‖<1, then
dimKerF=♯{i∈{1,2} | a−i<0 and a+i>0}dimKerF∗=♯{i∈{1,2} | a+i<0 and a−i>0}.
Statement of Lemma
In words:
-
If S(s) is a scalar matrix, set a±=a±1=a±2 to the limiting scalars and count the integer multiples of 2π between a− and a+.
-
Otherwise, if S is uniformly bounded by 1, count the number of entries the flip from positive to negative as s goes from −∞→∞.
\begin{center} \includegraphics[width = \textwidth]{figures/image_2020-05-27-20-10-07.png} \end{center}
Proof of Assertion 1
- Suppose a1(s)=a2(s) and set a±:=a±1=a±2. Then
dimKerF=2⋅♯{ℓ∈Z | 2πℓ∈(a−,a+)⊂R}dimKerF∗=2⋅♯{ℓ∈Z | 2πℓ∈(a+,a−)⊂R}.
Step 1: Transform to Cauchy-Riemann Equations
- Write a(s):=a1(s)=a2(s).
- Start with equation on R2, Y(s,t)=[Y1(s,t),Y2(s,t)].
- Replace with equation on C: Y(s,t)=Y1(s,t)+iY2(s,t).
Assertion 1, Step 1: Reduce to CR
-
Expand definition of the PDE F(Y)=0⇝¯∂Y+SY=0∂∂sY+(0−110)∂∂tY+(a(s)00a(s))Y=0.
-
Change of variables: want to reduce to ¯∂˜Y=0
-
Choose B∈GL(1,C) such that ¯∂B+SB=0
-
Set Y=B˜Y, which (?) reduces the previous equation to ¯∂˜Y=0.
Assertion 1, Step 1: Reduce to CR
Can choose (and then solve) B=[b(s)00b(s)]where∂b∂s=−a(s)b(s)⟹b(s)=exp∫s0−a(σ) dσ:=exp−A(s).
Remarks:
- For some constants Ci, we have
A(s)={C1+a−s,s≤−σ0C2+a+s,s≥σ0.
-
The new ˜Y satisfies CR, is continuous and L1loc, so elliptic regularity ⟹C∞.
-
The real/imaginary parts of ˜Y are C∞ and harmonic.
Assertion 1, Step 2: Solve CR
-
Identify s+it∈R×S1 with u=e2πz
-
Apply Laurent’s theorem to ˜Y(u) on C∖{0} to obtain an expansion of ˜Y in z.
-
Deduce that the solutions of the system are given by ˜Y(u)=∑ℓ∈Zcℓuℓ⟹˜Y(s+it)=∑ℓ∈Zcℓe(s+it)2πℓ. where {cℓ}ℓ∈Z⊂C converges for all s,t.
Assertion 1, Step 2: Solve CR
Use es+it=es(cos(t)+isin(t)) to write in real coordinates: ˜Y(s,t)=∑ℓ∈Ze2πsℓ[cos(2πℓt)−sin(2πℓt)sin(2πℓt)cos(2πℓt)][αℓβℓ].
Use Y=B˜Y=[e−A(s)00e−A(s)]˜Y
to write Y(s,t)=∑ℓ∈Ze2πsℓ[e−A(s)00e−A(s)][cos(2πℓt)−sin(2πℓt)sin(2πℓt)cos(2πℓt)][αℓβℓ].
For s≤s0 this yields for some constants K,K′: Y(s,t)=∑ℓ∈Ze2πℓ−a−[eK(αℓcos(2πℓt)−βℓsin(2πℓt))eK′(αℓsin(2πℓt)+βℓcos(2πℓt))].
Condition on Lp Solutions
For s≤s0 we had Y(s,t)=∑ℓ∈Ze(2πℓ−a−)s[eK(αℓcos(2πℓt)−βℓsin(2πℓt))eK′(αℓsin(2πℓt)+βℓcos(2πℓt))]
and similarly for s≥s0, for some constants C,C′ we have: Y(s,t)=∑ℓ∈Ze(2πℓ−a+)s[eC(αℓcos(2πℓt)−βℓsin(2πℓt))eC′(αℓsin(2πℓt)+βℓcos(2πℓt))].
Then Y∈Lp⟺exponential termsℓ→∞→0.
Condition on Lp Solutions: Small Tails
Y(s,t)=∑ℓ∈Ze(2πℓ−a−)s[eK(αℓcos(2πℓt)−βℓsin(2πℓt))eK′(αℓsin(2πℓt)+βℓcos(2πℓt))]
- ℓ≠0: Need αℓ=βℓ=0 or 2πℓ<a+
-
ℓ=0: Need both
- α0=0 or a+>0 and
- β0=0 or a+>0.
Counting Solutions
{αℓ=βℓ=0 or 2πℓ∈(a−,a+)ℓ≠0(α0=0or0∈(a−,a+))and(β0=0or0∈(a−,a+))ℓ=0.
- Finitely many such ℓ that satisfy these conditions
- Sufficient conditions for Y(s,t)∈W1,p.
Compute dimension of space of solutions: ` \begin{align*} \operatorname{dim} \operatorname{Ker} F &=2 \cdot {\sharp}\left{{\ell \in \mathbb{Z}^{*} {~\mathrel{\Big\vert}~} 2\pi\ell \in (a^-, a^+) }\right}
- 2\cdot \indic{0 \in (a^-, a^+)} \ &=2 \cdot {\sharp}\left{\ell \in \mathbb{Z} {~\mathrel{\Big\vert}~}2\pi\ell \in (a^-, a^+) \right} .\end{align*} `{=html}
Note: not sure what Z∗ is: most likely Z∖{0}.
Counting Solutions
Use this to deduce dimkerF∗:
-
Y∈kerF∗⟺Z(s,t):=Y(−s,t) is in the kernel of the operator \begin{align*} \tilde F: W^{1, q}\left(\mathbb{R} \times S^{1} ; \mathbb{R}^{2}\right) &\longrightarrow L^{p}\left(\mathbb{R} \times S^{1} ; \mathbb{R}^{2}\right) \\ Z &\mapsto \frac{\partial Z}{\partial s}+J_{0} \frac{\partial Z}{\partial t}+S({\color{red}-s}) Y .\end{align*}
-
Obtain \ker F^* \cong \ker \tilde F.
-
Formula for \dim \ker \tilde F almost identical to previous formula, just swapping a^- and a^+.
Assertion 2
Assertion 2: Suppose \sup_{s\in {\mathbb{R}}} {\left\lVert {S(s)} \right\rVert} < 1, then \begin{align*} \operatorname{dim} \operatorname{Ker} F &= {\sharp}\left\{i \in\{1,2\} {~\mathrel{\Big\vert}~}~a_{i}^{-}<0 < a_{i}^{+}\right\}\\ \operatorname{dim} \operatorname{Ker} F^{*} &={\sharp}\left\{i \in\{1,2\} {~\mathrel{\Big\vert}~}~ a_{i}^{+}<0 < a_{i}^{-} \right\} .\end{align*}
We use the following:
-
Lemma 8.8.7: \begin{align*} \sup_{s\in {\mathbb{R}}} {\left\lVert { S(s) } \right\rVert} < 1 \implies \text{the elements in }\ker F,~ \ker F^* \text{ are independent of }t .\end{align*}
-
Proof: in subsection 10.4.a.
Proof of Assertion 2
\begin{align*} F: W^{1, p}\left(\mathbb{R} \times S^{1} ; \mathbb{R}^{2}\right) &\longrightarrow L^{p}\left(\mathbb{R} \times S^{1} ; \mathbb{R}^{2}\right) \\ Y &\mapsto \frac{\partial Y}{\partial s}+J_{0} \frac{\partial Y}{\partial t}+S(s) Y .\end{align*}
-
Given as a fact: ` \begin{align*} \mathbf{Y} \in \ker F \implies {\frac{\partial }{\partial s},}\mathbf{Y} = \mathbf{a}(s)\mathbf{Y} ~~~
-a_1(s) & 0 \\ 0 & -a_2(s) \end{bmatrix} \mathbf{Y} .\end{align*} <span>`{=html}
- Therefore we can solve to obtain \begin{align*} \mathbf{Y}(s) = \mathbf{c}_0 \exp{-\mathbf{A}(s)}{\quad \operatorname{where} \quad} \mathbf{A}(s) = \int_0^s -\mathbf{a}(\sigma) ~d\sigma .\end{align*}
Proof of Assertion 2
-
Explicitly in components: \begin{align*} \begin{cases} {\frac{\partial Y_1}{\partial s}\,} &= -a_1(s) Y_1 \\ {\frac{\partial Y_s}{\partial s}\,} &= -a_2(s) Y_2 \\ \end{cases} \quad \implies \quad Y_i(s) = c_i e^{-A_i(s)}, \quad A_i(s) &= \int_0^s -a_i(\sigma) ~d\sigma .\end{align*}
-
As before, for some constants C_{j, i}, \begin{align*} A_i(s) = \begin{cases} C_{1, i} + a_i^-\cdot s & s \leq -\sigma_0 \\ C_{2, i} + a_i^+\cdot s & s \geq \sigma_0 \\ \end{cases} .\end{align*}
-
Thus \begin{align*} Y_i \in W^{1, p} \iff 0 \in (a_i^-, a_i^+) ,\end{align*}
establishing
\begin{align*} \dim \ker F = {\sharp}\left\{i \in\{1,2\} {~\mathrel{\Big\vert}~}0 \in (a_i^-, a_i^+) \right\} .\end{align*}
8.8.3: \operatorname{Ind}(L_1) = k^- - k^+
Statement and Outline
Statement: let k^\pm \coloneqq\operatorname{Ind}(R^\pm); then \operatorname{Ind}(L_1) = k^- - k^+.
- Consider four cases, depending on parity of k^\pm - n
- Show all 4 lead to \operatorname{Ind}(L_1) = k^- - k^+
- k^- \equiv k^+ \equiv n \operatorname{mod}2.
- k^- \equiv n, k^+ \equiv n-1 \operatorname{mod}2
- k^- \equiv n-1, k^+ \equiv n \operatorname{mod}2.
- k^- \equiv k^+ \equiv n-1 \operatorname{mod}2
\begin{center} \includegraphics[width = 0.3\textwidth]{figures/image_2020-05-27-22-54-44.png} \end{center}
Case 1: k^+ \equiv k^- \equiv n \operatorname{mod}2
\begin{align*} S_{k^-} & = \begin{bmatrix} -\pi & & & & & & & \\ & -\pi & & & & & & \\ & & \ddots & & & & & \\ & & & & -\pi & & & \\ & & & & & -\pi & & \\ & & & & & & (n-1-k^-)\pi & \\ & & & & & & & (n-1-k^-)\pi \\ \end{bmatrix} \\ S_{k^+} & = \begin{bmatrix} -\pi & & & & & & & \\ & -\pi & & & & & & \\ & & \ddots & & & & & \\ & & & & -\pi & & & \\ & & & & & -\pi & & \\ & & & & & & (n-1-{\color{blue}k^+})\pi & \\ & & & & & & & (n-1-{\color{blue}k^+})\pi \\ \end{bmatrix} .\end{align*}
Case 1: k^- \equiv k^+ \equiv n \operatorname{mod}2
- Take a_1(s) = a_2(s) so a_1^\pm = a^\pm
- Apply the proved lemma to obtain
\begin{align*} \dim \ker L_1 &= 2\cdot {\sharp}\left\{{\ell \in {\mathbb{Z}}{~\mathrel{\Big\vert}~}2\ell \in (n-1-k^-, n-1-k^+)}\right\} \\ &= \begin{cases} k^- - k^+ & k^- > k^+ \\ 0 & \text{else} \end{cases} \\ \\ \dim \ker L_1^* &= 2\cdot {\sharp}\left\{{ \ell \in {\mathbb{Z}}{~\mathrel{\Big\vert}~}2\ell \in (k^- - n + 1, k^+ - n + 1)}\right\} \\ &= \begin{cases} k^+ - k^- & k^+ > k^- \\ 0 & \text{otherwise} \end{cases} \\ \\ \implies \operatorname{Ind}(L_1) &= \qty{k^- - k^+ \over 2} - \qty{k^+ - k^- \over 2} = k^- - k^+ .\end{align*}
Case 2: k^+ \not\equiv k^- \equiv n \operatorname{mod}2
\begin{align*} S_{k-} & = \begin{bmatrix} -\pi & & & & & & & \\ & -\pi & & & & & & \\ & & \ddots & & & & & \\ & & & & -{\color{red}{\varepsilon}}\pi & & & \\ & & & & & -{\color{red}{\varepsilon}}\pi & & \\ & & & & & & (n-1-{\color{red}k^-})\pi & \\ & & & & & & & (n-1-{\color{red}k^-})\pi \\ \end{bmatrix} \\ S_{k^+} & = \begin{bmatrix} -\pi & & & & & & & \\ & -\pi & & & & & & \\ & & \ddots & & & & & \\ & & & & {\color{red}{\varepsilon}} & & & \\ & & & & & -{\color{red}{\varepsilon}} & & \\ & & & & & & (n-{\color{red}2}-k^+)\pi & \\ & & & & & & & (n-{\color{red}2}-k^+)\pi \\ \end{bmatrix} .\end{align*}
Case 2: k^+ \not\equiv k^- \equiv n \operatorname{mod}2
- Take a_1(s) = a_2(s) everywhere except the n-1st block, where we can assume \sup_{s\in {\mathbb{R}}} {\left\lVert {S(s)} \right\rVert} < 1.
- Assertion 2 applies and we get
\begin{align*} \dim \ker L_1 &= 2\cdot {\sharp}\left\{{\ell \in {\mathbb{Z}}{~\mathrel{\Big\vert}~}2\ell \in (n-1-k^-, n-2-k^+)}\right\} + 1 \\ &= \begin{cases} \qty{k^- - k^+ - 1} + 1 & k^- > k^+ \\ 1 & \text{otherwise} \end{cases} \\ \\ \dim \ker L_1^* &= 2\cdot {\sharp}\left\{{\ell \in {\mathbb{Z}}{~\mathrel{\Big\vert}~}2\ell \in (k^- - n + 1, k^+ - n + 2)}\right\} \\ &= \begin{cases} k^+ - k^- + 1, & k^+ > k^- \\ 0 & \text{otherwise} \end{cases} \\ \implies \operatorname{Ind}(L_1) &= \qty{ {k^- - k^+ -1 \over 2} + 1} - \qty{k^+ - k^- + 1 \over 2} = k^- - k^+ .\end{align*}
The other 2 cases involve different matrices S_{k^\pm}, but proceed similarly.