author | scott@6e1638ff-ae45-0410-89bd-df963105f760 |
Mon, 07 Jul 2008 03:20:11 +0000 | |
changeset 37 | 2f677e283c26 |
parent 15 | 7340ab80db25 |
child 72 | ed2594ff5870 |
permissions | -rw-r--r-- |
13 | 1 |
\nn{Here's the ``explicit'' version.} |
2 |
||
3 |
Fix a finite open cover of $X$, say $(U_l)_{l=1}^L$, along with an |
|
4 |
associated partition of unity $(r_l)$. |
|
5 |
||
15
7340ab80db25
rearranging the Hochschild section. Splitting things up into lemmas, and explaining why those lemmas are what we need.
scott@6e1638ff-ae45-0410-89bd-df963105f760
parents:
13
diff
changeset
|
6 |
We'll define the homotopy $H:I \times P \times X \to X$ via a function |
7340ab80db25
rearranging the Hochschild section. Splitting things up into lemmas, and explaining why those lemmas are what we need.
scott@6e1638ff-ae45-0410-89bd-df963105f760
parents:
13
diff
changeset
|
7 |
$u:I \times P \times X \to P$, with |
13 | 8 |
\begin{equation*} |
9 |
H(t,p,x) = F(u(t,p,x),x). |
|
10 |
\end{equation*} |
|
11 |
||
15
7340ab80db25
rearranging the Hochschild section. Splitting things up into lemmas, and explaining why those lemmas are what we need.
scott@6e1638ff-ae45-0410-89bd-df963105f760
parents:
13
diff
changeset
|
12 |
To begin, we'll define a function $u'' : I \times P \times X \to P$, and |
13 | 13 |
a corresponding homotopy $H''$. This homotopy will just be a homotopy of |
14 |
$F$ through families of maps, not through families of diffeomorphisms. On |
|
15 |
the other hand, it will be quite simple to describe, and we'll later |
|
16 |
explain how to build the desired function $u$ out of it. |
|
17 |
||
15
7340ab80db25
rearranging the Hochschild section. Splitting things up into lemmas, and explaining why those lemmas are what we need.
scott@6e1638ff-ae45-0410-89bd-df963105f760
parents:
13
diff
changeset
|
18 |
For each $l = 1, \ldots, L$, pick some $C^\infty$ function $f_l : I \to |
13 | 19 |
I$ which is identically $0$ on a neighborhood of the closed interval $[0,\frac{l-1}{L}]$ |
20 |
and identically $1$ on a neighborhood of the closed interval $[\frac{l}{L},1]$. (Monotonic? |
|
21 |
Fix a bound for the derivative?) We'll extend it to a function on |
|
15
7340ab80db25
rearranging the Hochschild section. Splitting things up into lemmas, and explaining why those lemmas are what we need.
scott@6e1638ff-ae45-0410-89bd-df963105f760
parents:
13
diff
changeset
|
22 |
$k$-tuples $f_l : I^k \to I^k$ pointwise. |
13 | 23 |
|
24 |
Define $$u''(t,p,x) = \sum_{l=1}^L r_l(x) u_l(t,p),$$ with |
|
25 |
$$u_l(t,p) = t f_l(p) + (1-t)p.$$ Notice that the $i$-th component of $u''(t,p,x)$ depends only on the $i$-th component of $p$. |
|
26 |
||
27 |
Let's now establish some properties of $u''$ and $H''$. First, |
|
28 |
\begin{align*} |
|
29 |
H''(0,p,x) & = F(u''(0,p,x),x) \\ |
|
30 |
& = F(\sum_{l=1}^L r_l(x) p, x) \\ |
|
31 |
& = F(p,x). |
|
32 |
\end{align*} |
|
33 |
Next, calculate the derivatives |
|
34 |
\begin{align*} |
|
35 |
\partial_{p_i} H''(1,p,x) & = \partial_{p_i}u''(1,p,x) \partial_1 F(u(1,p,x),x) \\ |
|
36 |
\intertext{and} |
|
37 |
\partial_{p_i}u''(1,p,x) & = \sum_{l=1}^L r_l(x) \partial_{p_i} f_l(p). |
|
38 |
\end{align*} |
|
39 |
Now $\partial_{p_i} f_l(p) = 0$ unless $\frac{l-1}{L} < p_i < \frac{l}{L}$, and $r_l(x) = 0$ unless $x \in U_l$, |
|
40 |
so we conclude that for a fixed $p$, $\partial_p H''(1,p,x) = 0$ for all $x$ outside the union of $k$ open sets from the open cover, namely |
|
41 |
$\bigcup_{i=1}^k U_{l_i}$ where for each $i$, we choose $l_i$ so $\frac{l_i -1}{L} \leq p_i \leq \frac{l_i}{L}$. It may be helpful to refer to Figure \ref{fig:supports}. |
|
42 |
||
43 |
\begin{figure}[!ht] |
|
44 |
\begin{equation*} |
|
45 |
\mathfig{0.5}{explicit/supports} |
|
46 |
\end{equation*} |
|
47 |
\caption{The supports of the derivatives {\color{green}$\partial_p f_1$}, {\color{blue}$\partial_p f_2$} and {\color{red}$\partial_p f_3$}, illustrating the case $k=2$, $L=3$. Notice that any |
|
48 |
point $p$ lies in the intersection of at most $k$ supports. The support of $\partial_p u''(1,p,x)$ is contained in the union of these supports.} |
|
49 |
\label{fig:supports} |
|
50 |
\end{figure} |
|
51 |
||
52 |
Unfortunately, $H''$ does not have the desired property that it's a homotopy through diffeomorphisms. To achieve this, we'll paste together several copies |
|
53 |
of the map $u''$. First, glue together $2^k$ copies, defining $u':I \times P \times X$ by |
|
54 |
\begin{align*} |
|
55 |
u'(t,p,x)_i & = |
|
56 |
\begin{cases} |
|
57 |
\frac{1}{2} u''(t, 2p_i, x)_i & \text{if $0 \leq p_i \leq \frac{1}{2}$} \\ |
|
58 |
1-\frac{1}{2} u''(t, 2-2p_i, x)_i & \text{if $\frac{1}{2} \leq p_i \leq 1$}. |
|
59 |
\end{cases} |
|
60 |
\end{align*} |
|
61 |
(Note that we're abusing notation somewhat, using the fact that $u''(t,p,x)_i$ depends on $p$ only through $p_i$.) |
|
62 |
To see what's going on here, it may be helpful to look at Figure \ref{fig:supports_4}, which shows the support of $\partial_p u'(1,p,x)$. |
|
63 |
\begin{figure}[!ht] |
|
64 |
\begin{equation*} |
|
65 |
\mathfig{0.4}{explicit/supports_4} \qquad \qquad \mathfig{0.4}{explicit/supports_36} |
|
66 |
\end{equation*} |
|
67 |
\caption{The supports of $\partial_p u'(1,p,x)$ and of $\partial_p u(1,p,x)$ (with $K=3$) are subsets of the indicated region.} |
|
68 |
\label{fig:supports_4} |
|
69 |
\end{figure} |
|
70 |
||
71 |
Second, pick some $K$, and define |
|
72 |
\begin{align*} |
|
73 |
u(t,p,x) & = \frac{\floor{K p}}{K} + \frac{1}{K} u'\left(t, K \left(p - \frac{\floor{K p}}{K}\right), x\right). |
|
74 |
\end{align*} |
|
75 |
||
76 |
\todo{Explain that the localisation property survives for $u'$ and $u$.} |
|
77 |
||
78 |
We now check that by making $K$ large enough, $H$ becomes a homotopy through diffeomorphisms. We start with |
|
79 |
$$\partial_x H(t,p,x) = \partial_x u(t,p,x) \partial_1 F(u(t,p,x), x) + \partial_2 F(u(t,p,x), x)$$ |
|
80 |
and observe that since $F(p, -)$ is a diffeomorphism, the second term $\partial_2 F(u(t,p,x), x)$ is bounded away from $0$. Thus if we can control the |
|
81 |
size of the first term $\partial_x u(t,p,x) \partial_1 F(u(t,p,x), x)$ we're done. The factor $\partial_1 F(u(t,p,x), x)$ is bounded, and we |
|
82 |
calculate \todo{err... this is a mess, and probably wrong.} |
|
83 |
\begin{align*} |
|
84 |
\partial_x u(t,p,x)_i & = \partial_x \frac{1}{K} u'\left(t, K\left(p - \frac{\floor{K p}}{K}\right), x\right)_i \\ |
|
85 |
& = \pm \frac{1}{2 K} \partial_x u''\left(t, (1\mp1)\pm 2K\left(p_i-\frac{\floor{K p}}{K}\right), x\right)_i \\ |
|
86 |
& = \pm \frac{1}{2 K} \sum_{l=1}^L (\partial_x r_l(x)) u_l\left(t, (1\mp1)\pm 2K\left(p_i-\frac{\floor{K p}}{K}\right)\right)_i. \\ |
|
87 |
\intertext{Since the target of $u_l$ is just the unit cube $I^k$, we can make the estimate} |
|
88 |
\norm{\partial_x u(t,p,x)_i} & \leq \frac{1}{2 K} \sum_{l=1}^L \norm{\partial_x r_l(x)}. |
|
89 |
\end{align*} |
|
90 |
The sum here is bounded, so for large enough $K$ this is small enough that $\partial_x H(t,p,x)$ is never zero. |