2
$\begingroup$

I am asking myself wether the following fact is true:

Let $f(x_1,\ldots,x_k) \colon U \subseteq \mathbb{R}^n \to \mathbb{R}$ be a multivariate function.

Do the multivariate Taylor series $$MT_{f}(x_1,\ldots,x_k) = \sum_{\alpha\in\mathbb{N}^k} \frac{\partial^\alpha f(0,\ldots,0)}{\alpha!~\partial x^\alpha} x^\alpha \qquad \text{where $\alpha$ is used as a multi-index}$$ and iterating univariate Taylor series expansion (i.e. treating the other variables as constants) $$ \begin{align*} T_{f}(x_1,\ldots, x_k) &= \sum_{\alpha\in\mathbb{N}} \frac{\partial^\alpha f(0,x_2,\ldots,x_k)}{\alpha!~\partial{x_1^\alpha}} x_1^\alpha\\ T_{T_{f}}(x_1,\ldots,x_k) &= \sum_{\alpha\in\mathbb{N}} \frac{\partial^\alpha T_{f}(x_1,0,\ldots,x_k)}{\alpha!~\partial{x_2^\alpha}} x_2^\alpha\\ &\;\vdots\\ IT_{f}(x_1, \ldots, x_k)&= \sum_{\alpha\in\mathbb{N}} \frac{\partial^\alpha T_{T_{\ddots_{T_{f}}}}(x_1,x_2,\ldots,0)}{\alpha!~\partial{x_k^\alpha}} x_k^\alpha \end{align*} $$ coincide? Specifically:

  • if the multivariate Taylor series $MT$ does not exist, the iterated one $IT$ also does not exist
  • if they both exist, they are equal and the iteration method is invariant under variable permutation, e.g., the sequences $x_1\to x_2\to x_3\to \ldots \to x_k$ and $x_2 \to x_1 \to x_k \to \ldots$ should result in the same Taylor series.
$\endgroup$
8
  • $\begingroup$ Factorials are missing here. $\endgroup$ Commented May 10 at 12:23
  • $\begingroup$ What is $U$ here? What do you mean by "series [...] does not exist"? Exist in what sense? $\endgroup$ Commented May 10 at 12:42
  • $\begingroup$ @Iosif For example for $f(x,y) = \frac{xy^3}{x^2+y^2}$ the first iteration yields $T_f(x,y) = 0 + yx - \frac{6}{y} x^3 + \frac{120}{y^3} x^5 ...$. Then the second iteration $T_{T_f}(x,y)$ does not exist as the first derivative of $\frac{-6}{y} x^3$ with respect to $y$ at $y=0$ does not exist hence we cannot construct the second iteration as described in the above scheme. $\endgroup$
    – WaveL
    Commented May 10 at 12:58
  • 1
    $\begingroup$ We would surely want to assume at the very least that $f\in C^{\infty}$ near zero, and then the two methods obviously produce the same formal object. Convergence is a separate issue of course. $\endgroup$ Commented May 10 at 13:38
  • 1
    $\begingroup$ @WaveL: That doesn't matter here since (for differentiable functions, which your example is not) $\partial f(x,y)/\partial x$ at $(x,y)=(0,0)$ is the same whether you first take the derivative and then set $x=y=0$ or first fix $y=0$, differentiate the resulting function of $x$ and finally take $x=0$. $\endgroup$ Commented May 10 at 18:46

0

You must log in to answer this question.

Browse other questions tagged .