2

$$ \frac{\partial \phi}{\partial \xi} ( \parallel \mathbf{x} - \xi_i \parallel) = - \frac{\partial \phi}{\partial x} ( \parallel \mathbf{x} - \xi_i \parallel) $$

from page 14 of these lecture slides, where $\phi$ is (a radial basis) function the variables $\mathbf{x}$ and $\xi_i$.

Willie Wong
  • 73,139
Olumide
  • 1,241

1 Answers1

2

In fact we have $\frac{\partial}{\partial x}f(x-\xi) = - \frac{\partial}{\partial \xi}f(x-\xi)$ for any function $f$, of which your formula is a special case.

This can be seen by applying the chain rule for differentiation:

$\frac{\partial}{\partial x}f(x-\xi) = f^\prime(x-\xi)\cdot\frac{\partial}{\partial x}(x-\xi) = f^\prime(x-\xi)$ whereas $\frac{\partial}{\partial \xi}f(x-\xi) = f^\prime(x-\xi)\cdot\frac{\partial}{\partial \xi}(x-\xi) = -f^\prime(x-\xi)$.

gfes
  • 736
  • Thanks. Just one small question before I check the answer button, what is the prime in the first and second expressions? A dummy variable which is equal to $x - \xi$? – Olumide Jun 05 '11 at 23:23
  • It's just shorthand for the derivative of f. – gfes Jun 05 '11 at 23:26
  • Yes, but with respect to what? Surely, it's not with respect to $x$ or $\xi$. – Olumide Jun 06 '11 at 09:47
  • No. It's a normal derivative, not a partial one. It's $\frac{df(t)}{dt}$ evaluated at $x-\xi$. – gfes Jun 06 '11 at 10:12
  • But if $f^\prime$ is the derivative of $f$ wrt to $t$, where $t = x - \xi$ then the final result contains $\frac{\partial f}{\partial t}$ and not $\frac{\partial f}{\partial x}$. Or is $\frac{df(t)}{dt}$ the total derivative? – Olumide Jun 06 '11 at 11:19
  • It's the total derivative indeed. – gfes Jun 06 '11 at 16:16
  • Now I get it. If $u = x - \xi$. You only have to compare or equate both expressions so that $\frac{\partial f}{\partial x} = \frac{\partial f}{\partial u} = -\frac{\partial f}{\partial \xi}$ – Olumide Jun 09 '11 at 02:11