About $\operatorname{Alt}(\varphi_{i_1}\otimes\cdots\otimes\varphi_{i_k})$ ("Calculus on Manifolds" by Michael Spivak)

97 Views Asked by At

I am reading "Calculus on Manifolds" by Michael Spivak.

The author wrote as follows:

Since each $\operatorname{Alt}(\varphi_{i_1}\otimes\cdots\otimes\varphi_{i_k})$ is a constant times one of the $\varphi_{i_1}\wedge\cdots\wedge\varphi_{i_k}$, these elements span $\Lambda^k(V)$.

I wonder why this holds.

My proof of the above fact:

Let $\omega_1,\dots,\omega_n\in\Lambda^1(V)$.
By the definition of the wedge product, $$\omega_1\wedge\omega_2=\frac{(1+1)!}{1!1!}\operatorname{Alt}(\omega_1\otimes\omega_2).$$
So, $\operatorname{Alt}(\omega_1\otimes\omega_2)=\frac{1}{2}(\omega_1\wedge\omega_2)$.
By Theorem 4-4(3) in the book, $$\omega_1\wedge\omega_2\wedge\omega_3=\frac{(1+1+1)!}{1!1!1!}\operatorname{Alt}(\omega_1\otimes\omega_2\otimes\omega_3).$$
So, $\operatorname{Alt}(\omega_1\otimes\omega_2\otimes\omega_3)=\frac{1}{6}(\omega_1\wedge\omega_2\wedge\omega_3)$.
By Theorem 4-4(2), $\operatorname{Alt}(\operatorname{Alt}(\omega\otimes\eta)\otimes\theta)=\operatorname{Alt}(\omega\otimes\eta\otimes\theta)$.
$$\operatorname{Alt}(\omega_1\otimes\omega_2\otimes\omega_3\otimes\omega_4) =\operatorname{Alt}(\omega_1\otimes\omega_2\otimes(\omega_3\otimes\omega_4)) =\operatorname{Alt}(\operatorname{Alt}(\omega_1\otimes\omega_2)\otimes(\omega_3\otimes\omega_4)) \\ =\operatorname{Alt}\left(\frac{1}{2}(\omega_1\wedge\omega_2)\otimes\omega_3\otimes\omega_4\right) =\frac{1}{12}(\omega_1\wedge\omega_2\wedge\omega_3\wedge\omega_4).$$
Let $n\geq 3$.
Similarly, $\operatorname{Alt}(\omega_1\otimes\cdots\otimes\omega_n)=\frac{1}{6}\left(\frac{1}{2}\right)^{n-3}(\omega_1\wedge\cdots\wedge\omega_n)$.

So, $\operatorname{Alt}(\varphi_{i_1}\otimes\cdots\otimes\varphi_{i_k})=\frac{1}{6}\left(\frac{1}{2}\right)^{k-3}(\varphi_{i_1}\wedge\cdots\wedge\varphi_{i_k})$ if $k\geq 3$.

Is my proof ok?

If my proof is not ok, please tell me a proof.
If my proof is ok, please tell me a better standard proof.

1

There are 1 best solutions below

1
On BEST ANSWER

Your idea could be a start, but your constants are wrong.

You have showed that if each $\omega_i \in \Lambda^1(V)$, then

$$ \operatorname{Alt}(\omega_1 \otimes \omega_2) = \frac 12 (\omega_1 \wedge \omega_2) \tag{1}$$ $$ \operatorname{Alt}(\omega_1 \otimes \omega_2 \otimes \omega_3) = \frac 16 (\omega_1 \wedge \omega_2 \wedge \omega_3) \tag{2}$$

But then the first error is at

$$ \operatorname{Alt}\left(\frac 12 (\omega_1 \wedge \omega_2) \otimes \omega_3 \otimes \omega_4\right) = \frac{1}{12}(\omega_1 \wedge \omega_2 \wedge \omega_3 \wedge \omega_4) $$

You can't apply equation $(2)$ here since $\omega_1 \wedge \omega_2 \in \Lambda^2(V)$ is not an element of $\Lambda^1(V)$. Applying Spivak's Theorem 4-4(3) again, we instead have

$$ \begin{align*} \frac{(2+1+1)!}{(2!)(1!)(1!)} \operatorname{Alt}\left(\frac 12 (\omega_1 \wedge \omega_2) \otimes \omega_3 \otimes \omega_4\right) &= \frac 12(\omega_1 \wedge \omega_2 \wedge \omega_3 \wedge \omega_4) \\ \operatorname{Alt}\left(\omega_1 \otimes \omega_2 \otimes \omega_3 \otimes \omega_4\right) &= \frac{1}{24}(\omega_1 \wedge \omega_2 \wedge \omega_3 \wedge \omega_4) \tag{3} \end{align*} $$

$n$ isn't a good choice for your general formula about $\operatorname{Alt}(\omega_1 \otimes \cdots \otimes \omega_n)$ since $n$ already is used as the dimension of vector space $V$. Rather than just "Similarly" stating the general conclusion, a formal induction argument is better.

Rephrasing the essential ideas of your proof, I'd write:

If $\omega_1, \omega_2, \ldots, \omega_k \in \Lambda^1(V)$, then $$ \operatorname{Alt}(\omega_1 \otimes \cdots \otimes \omega_k) = \frac{1}{k!}(\omega_1 \wedge \cdots \wedge \omega_k) \tag{4} $$

If $k=1$, the equality is trivial.

Given $k>1$, suppose that $\operatorname{Alt}(\omega_1 \otimes \cdots \otimes \omega_{k-1}) = \frac{1}{(k-1)!}(\omega_1 \wedge \cdots \wedge \omega_{k-1})$. Then

$$ \begin{align*} \operatorname{Alt}(\omega_1 \otimes \cdots \otimes \omega_{k-1} \otimes \omega_k) &= \operatorname{Alt}(\operatorname{Alt}(\omega_1 \otimes \cdots \otimes \omega_{k-1}) \otimes \omega_k) \\ &= \frac{1}{(k-1)!} \operatorname{Alt}((\omega_1 \wedge \cdots \wedge \omega_{k-1}) \otimes \omega_k) \\ &= \frac{1}{(k-1)!} \cdot \frac{(k-1)! \, 1!}{k!} ((\omega_1 \wedge \cdots \wedge \omega_{k-1}) \wedge \omega_k) \\ \operatorname{Alt}(\omega_1 \otimes \cdots \otimes \omega_k) &= \frac{1}{k!} (\omega_1 \wedge \cdots \wedge \omega_k) \end{align*}$$ So by induction, (4) is true for every positive integer $k$.

(Or, both Spivak's 4-4(3) and this result (4) are special cases of a more general lemma not much harder to prove: If $\omega_i \in \Lambda^{j_i}$ for $i \in 1,\ldots,k$, then) $$ \operatorname{Alt}(\omega_1 \otimes \cdots \otimes \omega_k) = \frac{\prod_{i=1}^k j_i!}{\left(\sum_{i=1}^k j_i\right)!}(\omega_1 \wedge \cdots \wedge \omega_k) \tag{5} $$

However, statement (4) is not enough to finish the proof of Spivak's theorem 4-5! Note that in the basis decomposition

$$ \omega = \sum_{i_1,\ldots,i_k} a_{i_1,\ldots,i_k} \varphi_{i_1} \otimes \cdots \otimes \varphi_{i_k} $$

the indices $i_1,\ldots,i_k$ may be any of the $n^k$ sequences of indices from $1$ to $n$, in any order and possibly with repeats. But the set of wedge products $\varphi_{i_1} \wedge \cdots \wedge \varphi_{i_k}$ we claim is a basis of $\Lambda^k(V)$ only uses those ${n \choose k}$ sequences of indices satisfying $1 \leq i_1 < i_2 < \cdots < i_k \leq n$.

But a rearrangment of index orders is involved in the definition of $\operatorname{Alt}$. So we have for any $w_1, \ldots, w_k \in V$,

$$ \begin{align*} \operatorname{Alt}(\varphi_{i_{\tau(1)}} \otimes \cdots \otimes \varphi_{i_{\tau(k)}})(w_1,\ldots,w_k) &= \sum_{\sigma \in S_k} \operatorname{sgn} \sigma \cdot (\varphi_{i_{\tau(1)}} \otimes \cdots \otimes \varphi_{i_{\tau(k)}})(w_{\sigma(1)},\ldots,w_{\sigma(2)}) \\ &= \sum_{\sigma \in S_k} \operatorname{sgn} \sigma \cdot (\varphi_{i_1} \otimes \cdots \otimes \varphi_{i_k})(w_{\tau^{-1}\sigma(1)}, \ldots, w_{\tau^{-1}\sigma(k)}) \\ &= \sum_{\sigma' \in S_k} \operatorname{sgn}(\tau \sigma') \cdot (\varphi_{i_1} \otimes \cdots \otimes \varphi_{i_k})(w_{\sigma'(1)}, \ldots, w_{\sigma'(k)}) \\ &= \operatorname{sgn}\tau \cdot \operatorname{Alt}(\varphi_{i_1} \otimes \cdots \otimes \varphi_{i_k})(w_1,\ldots,w_k) \\ \operatorname{Alt}(\varphi_{i_{\tau(1)}} \otimes \cdots \otimes \varphi_{i_{\tau(k)}}) &= \operatorname{sgn}\tau \cdot \operatorname{Alt}(\varphi_{i_1} \otimes \cdots \otimes \varphi_{i_k}) \tag{6} \end{align*} $$

If any index appears more than once in the sequence $i_1,\ldots,i_k$, say $i_a=i_b$ with $a \neq b$, and if $\tau = (ab)$ is the pair swap permutation, then $i_{\tau(j)} = i_j$ for every $j$, and $\operatorname{sgn} \tau = -1$. So equation (6) gives

$$ \operatorname{Alt}(\varphi_{i_1} \otimes \cdots \otimes \varphi_{i_k}) = - \operatorname{Alt}(\varphi_{i_1} \otimes \cdots \otimes \varphi_{i_k}) $$

and therefore $\operatorname{Alt}(\varphi_{i_1} \otimes \cdots \otimes \varphi_{i_k}) = 0$. (It's technically true this is a constant times any one of the alternating wedge product basis elements, where that constant is zero. But this case could have been explained better!)

Otherwise, $i_1,\ldots,i_k$ are $k$ distinct elements of $\{1,\ldots,n\}$. Therefore they can be rearranged by some permutation $\tau$ so that $1 \leq \tau(i_1) < \cdots < \tau(i_k) \leq n$. And combining equations (4) and (6),

$$ \begin{align*} \operatorname{sgn} \tau \cdot \operatorname{Alt}(\varphi_{i_1} \otimes \cdots \otimes \varphi_{i_k}) &= \operatorname{Alt}(\varphi_{i_{\tau(1)}} \otimes \cdots \otimes \varphi_{i_{\tau(k)}}) \\ \operatorname{Alt}(\varphi_{i_1} \otimes \cdots \otimes \varphi_{i_k}) &= \frac{\operatorname{sgn} \tau}{k!} \varphi_{i_{\tau(1)}} \wedge \cdots \wedge \varphi_{i_{\tau(k)}} \end{align*} $$

This proves that the set $\{\varphi_{i_1} \wedge \cdots \wedge \varphi_{i_k} \mid 1 \leq i_1 < \cdots < i_k \leq n\}$ spans $\Lambda^k(V)$. To prove it's a basis, it remains to show these wedge products are linearly independent. For this I'd suggest the lemma:

If $1 \leq i_1 < \cdots < i_k \leq n$ and $1 \leq j_1 < \cdots < j_k \leq n$, then $$ (\varphi_{i_1} \wedge \cdots \wedge \varphi_{i_k})(v_{j_1}, \ldots, v_{j_k}) = \begin{cases} 1 & \mathrm{if\ } \forall m (i_m = j_m)\\ 0 & \mathrm{if\ } \exists m (i_m \neq j_m) \end{cases} $$