Let $f_i : \mathbb{R}^n \setminus \{0\} \to \mathbb{R}$ be a homogeneous function of degree $p \neq -1$ for every $1 \le i \le n$. Namely, $f_i(tx) = t^pf_i(x)$ for every $ x \in\mathbb{R}^n \setminus \{0\}$ and $t >0$.

We define $\omega = \sum_{i=1}^n f_i(x)dx_i$ and assume it is closed.

I want to prove that $\omega = dF$ when $F = \frac{1}{p+1} \sum_{i=1}^n x_if_i(x)$

Basically want we want to show is that $\frac{\partial F}{\partial x_i} = f_i(x)$.

By straightforward deriving I get: $$ \frac{\partial F}{\partial x_i} = \frac{1}{p+1} (\sum_{j=1}^n x_j\frac{\partial f_j(x)}{\partial x_i} + f_i(x))$$

Here I got stuck. I don't know how to continue. I also don't see how the information that $\omega$ is closed (namely, $\frac{\partial f_i(x)}{\partial x_j} = \frac{\partial f_j(x)}{\partial x_i}$ for $i \neq j$) helps me here. I also don't know how to use the fact thar $f_i$ are homogeneous because it is not given that $x_i > 0$.

Help would be appreciated.

  • $\begingroup$ If I understand correctly, you should sum over, say, $j$ in your final formula: for instance $\partial F/\partial x_i=(p+1)^{-1}\left(\sum_j x_j \partial f_j/\partial x_i+f_i\right)$ $\endgroup$ – Brightsun Jul 2 '19 at 11:50
  • $\begingroup$ Thanks, I corrected that. $\endgroup$ – Gabi G Jul 2 '19 at 11:51

Hint: ${d\over{dt}}_{t=1}f_i(tx)={d\over{dt}}_{t=1}t^pf_i(x)=pf_i(x)$.

You can also compute it as follows

${d\over{dt}}_{t=1}f_i(tx)=(\sum_{j=1}^{j=n}x_j{{\partial f_i}\over{\partial x_j}})_{t=1}$

$=\sum_{j=1}^{j=n}x_j{{\partial f_i}\over{\partial x_j}}$.

If you replace the second formula by the first in your computations, you obtain the result.

  • $\begingroup$ Could you explain the first line in your answer? Why $t=1$ becomes $t=0$ and where does the $pf_i(x)$ come from? Moreover, I am not sure to which computations in my question you are referring, so if you can clear it up I would be glad. $\endgroup$ – Gabi G Jul 2 '19 at 13:17
  • 1
    $\begingroup$ @Gabi G For any positive $t$, we have $f_i(tx)=t^pf_i(x)$. Now if you take the ordinary derivative w.r. to $t$ on both sides you get $d/dt\, f_i(tx)=p t^{p-1}f_i(x)$, which is what Tsemo wrote in the first line substituting $t=1$. Then, you can also calculate the derivative using the chain rule: $d/dt\, f_i(tx)=\sum_j \partial f_i(tx)/\partial x_j\, d(tx_j)/dt=\sum_j x_j \partial f_i/\partial x_j\,$; substituting again $t=1$ yields the last formula by Tsemo. $\endgroup$ – Brightsun Jul 2 '19 at 13:41
  • $\begingroup$ @Tsemo Aristide, I edited out a $t$ from your next-to-last line, guessing it was a typo. $\endgroup$ – Brightsun Jul 2 '19 at 13:43
  • 1
    $\begingroup$ @GabiG After you establish $p f_i = \sum_j x_j \partial f_i/\partial x_j$ you can use the closedness hypothesis to conclude that $\sum_j x_j \partial f_j/\partial x_i=p f_i $ which is precisely what you need in order to conclude your argument. $\endgroup$ – Brightsun Jul 2 '19 at 13:45
  • $\begingroup$ Thanks to both of you, I get it now $\endgroup$ – Gabi G Jul 2 '19 at 13:52

Assume $x_1$, $x_2$, $\ldots$, $x_n$ are all positive. Then, for fixed $i$, we can write $$ f_i(x_1,\ldots, x_i, \ldots, x_n) = x_i^{p} f_i\left(\tfrac{x_1}{x_i},\ldots,\tfrac{x_{i-1}}{x_i}, 1,\tfrac{x_{i+1}}{x_i}, \ldots,\tfrac{x_n}{x_i}\right)\,. $$ Now, $$ F(x)=\sum_{i=1}^n x_i f_i(x)=\sum_{i=1}^n x_i^{p+1}f_i\left(\tfrac{x_1}{x_i},\ldots,\tfrac{x_{i-1}}{x_i}, 1,\tfrac{x_{i+1}}{x_i}, \ldots,\tfrac{x_n}{x_i}\right)\,. $$ Consider, for fixed $j$, $$ \frac{\partial}{\partial x_j}F= (p+1)x_j^{p+1}f_j\left(\tfrac{x_1}{x_j},\ldots,\tfrac{x_{j-1}}{x_j}, 1,\tfrac{x_{j+1}}{x_j}, \ldots,\tfrac{x_n}{x_j}\right)\\ +\sum_{i=1}^n x_i^{p+1}\frac{\partial}{\partial x_j}f_i\left(\tfrac{x_1}{x_i},\ldots,\tfrac{x_{i-1}}{x_i}, 1,\tfrac{x_{i+1}}{x_i}, \ldots,\tfrac{x_n}{x_i}\right)\,. $$ We recognize the first term on the right-hand side as $(p+1)f_j(x)$. The second term, appropriately separating the terms in the sum, gives $$ \sum_{i\neq j} x_i^{p+1}\frac{\partial }{\partial x_j} f_i\left(\tfrac{x_1}{x_i},\ldots,\tfrac{x_{i-1}}{x_i}, 1,\tfrac{x_{i+1}}{x_i}, \ldots,\tfrac{x_n}{x_i}\right)\\ + x_j^{p+1} \frac{\partial }{\partial x_j}f_j\left(\tfrac{x_1}{x_j},\ldots,\tfrac{x_{j-1}}{x_j}, 1,\tfrac{x_{j+1}}{x_j}, \ldots,\tfrac{x_n}{x_j}\right)\\ = \sum_{i\neq j} \frac{x_i^{p+1}}{x_i}\frac{\partial f_i}{\partial x_j}\left(\tfrac{x_1}{x_i},\ldots,\tfrac{x_{i-1}}{x_i}, 1,\tfrac{x_{i+1}}{x_i}, \ldots,\tfrac{x_n}{x_i}\right)\\ +\sum_{k\neq j} \frac{-x_j^{p+1}x_k}{x_j^2} \frac{\partial f_j}{\partial x_k}\left(\tfrac{x_1}{x_j},\ldots,\tfrac{x_{j-1}}{x_j}, 1,\tfrac{x_{j+1}}{x_j}, \ldots,\tfrac{x_n}{x_j}\right)\\ =\sum_{k\neq j} x_k\left(\frac{\partial }{\partial x_j}f_k-\frac{\partial}{\partial x_k}f_j\right)=0\,. $$ In the last step we used the closedness property; note that, for $k\neq j$, $$ \frac{\partial}{\partial x_k}f_j(x)=x_j^{p-1} \frac{\partial f_j}{\partial x_k}\left(\tfrac{x_1}{x_j},\ldots,\tfrac{x_{j-1}}{x_j}, 1,\tfrac{x_{j+1}}{x_j}, \ldots,\tfrac{x_n}{x_j}\right)\,. $$ When some $x_i$ is allowed to be negative, the argument can be repeated by replacing the $1$ with $-1$ appropriately.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.