Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I'd appreciate help simplifying the relationship $$ \nabla\left[ \; \phi(\parallel \mathbf{x} - \mathbf{\xi}_i \parallel) \; \right] $$

for $\mathbf{x}$ and $\mathbf{\xi}_i$ in $\mathcal{R}^n$. This is how far I've come (I'm not even sure if I'm on the right track)

Setting $\mathbf{u} = \parallel \mathbf{x} - \mathbf{\xi}_i \parallel$, so that $$ \nabla[ \phi(\mathbf{u}) ] = \left( \frac{\partial \phi}{\partial u_1} , \cdots , \frac{\partial \phi}{\partial u_n} \right) $$

but

$$ \frac{\partial \phi}{\partial u_j} = \frac{\partial \phi}{\partial u_j} \frac{\partial u_j}{\partial \mathbf{x}} + \frac{\partial \phi}{\partial u_j} \frac{\partial u_j}{\partial \mathbf{\xi}_i} $$

for $j = 1 , \cdots , n$

Note: this question is related to a previous one

Edit: Your answers are correct, and I will tag them as such, but they aren't the answers I was hoping for. In my previous question, I required help proving a relationship between involving $\mathbf{x}$ and $\mathbf{\xi}$, from page 14 of these lecture slides. What I am now trying to understand is why $\phi$ is differentiated with respect to $\xi$ in the first place i.e. $\frac{\partial \phi}{\partial \xi}$. The problem I'm working on is in the area of Hermite interpolation. For example, on page 4 (column 1) of the paper Hermite variational implicit surface reconstruction it is shown that

$$ \frac{\partial}{\partial f} \mathbf{n}_i^T \nabla f(\mathbf{\xi}_i) = \mathbf{n}_i^T \nabla k(\mathbf{x} , \mathbf{\xi}_i) $$

In the past I assumed that the components of the gradient of $\nabla f(\mathbf{\xi}_i)$ and $\nabla k(\mathbf{x} , \mathbf{\xi}_i)$ were differentials of $f(\mathbf{\xi}_i)$ and $k(\mathbf{x} , \mathbf{\xi}_i)$ with respect to $x_i$. However the lecture slides and the paper suggest that the terms of the gradient are differentials with respect to $\xi_i$. What I really would like to know is why.

share|improve this question
    
What is the variable here? Are we talking about a function $f({\bf x}):=\phi(\|{\bf x}-\xi\|)$ with $\xi$ fixed, about a function $g(\xi):=\phi(\|{\bf x}-\xi\|)$ with ${\bf x}$ fixed, or about a function defined in ${\mathbb R}^n \times {\mathbb R}^n$? –  Christian Blatter Jun 9 '11 at 11:06
    
$\mathbf{\xi}$ is fixed, so that $\phi( \parallel \mathbf{x} - \mathbf{\xi} \parallel )$ is defined in $\mathcal{R}^n$ –  Olumide Jun 9 '11 at 11:17

2 Answers 2

up vote 5 down vote accepted

For the function $\rho({\bf x}):=\|{\bf x}\|$ one has ${\partial \rho\over\partial x_i}={x_i\over \|{\bf x}\|}$ $\ (1\leq i\leq n)$, or $$\nabla \rho({\bf x})={{\bf x}\over \|{\bf x}\|}\ .\qquad(1)$$ Consider now the function $$f({\bf x}):=\phi(\|{\bf x}-\xi\|)=\phi\bigl(\rho({\bf x}-\xi)\bigr)\qquad(2)$$ where $\xi$ is fixed and $\phi:\ {\mathbb R_{\geq0}}\to{\mathbb R}$ is some scalar function of a real variable $r$. Then by the one-variable chain rule and (1) one has $${\partial f\over\partial x_i}=\phi'\bigl(\rho({\bf x}-\xi)\bigr)\ {\partial \rho({\bf x}-\xi)\over\partial x_i} =\phi'(\|{\bf x}-\xi\|){x_i-\xi_i \over \|{\bf x}-\xi\|}\qquad(1\leq i\leq n)\ .$$ These $n$ scalar equations can be summarized to $$\nabla f({\bf x})=\phi'(\|{\bf x}-\xi\|){{\bf x}-\xi \over \|{\bf x}-\xi\|}\ .\qquad(3)$$ If one is sufficiently fluent with multidimensional calculus one can of course omit the use of coordinates altogether and pass directly from (2) to (3), using (1).

share|improve this answer
    
Thanks. Quick question, just to make things really clear, with respect to what variable is $\phi$ differentiated in $\phi^\prime$? –  Olumide Jun 9 '11 at 13:48
    
Got it. The differentiation is wrt $\rho$. –  Olumide Jun 9 '11 at 14:11

Since $||\mathbf{x} - \xi_i||$ is a scalar, you simply set $u=||\mathbf{x} - \xi_i||$ and then you have

$$\nabla \phi(u) = \phi'(u) \nabla u$$

and, noting that $u^2=||\mathbf{x}-\xi_i||^2$ you can deduce $\nabla u = (\mathbf{x}-\xi_i)/u$, and hence

$$\nabla \phi(||\mathbf{x}-\xi_i||) = \frac{\mathbf{x}-\xi_i}{||\mathbf{x}-\xi_i||} \phi'(||\mathbf{x}-\xi_i||)$$

or, in words, it is the vector pointing in the direction $\mathbf{x} - \xi_i$ with magnitude $\phi'(||\mathbf{x}-\xi_i||)$.

share|improve this answer
    
Thanks. Quick question, just to make things really clear, with respect to what variable is $\phi$ differentiated in $\phi^\prime$? –  Olumide Jun 9 '11 at 13:48
    
$\phi$ is only a function of one variable, which is why it's legitimate to write $\phi'(\cdot)$ in the first place. In the notation of Christian Blatter's answer above, you have $\rho:\mathbb{R}^n\to\mathbb{R}$ and $\phi:\mathbb{R}\to\mathbb{R}$, and you compose them to get a new function $f = \phi\circ\rho$, where now $f:\mathbb{R}^n\to\mathbb{R}$. –  Chris Taylor Jun 9 '11 at 13:58
    
Got it. The differentiation is wrt $u$. –  Olumide Jun 9 '11 at 14:12
    
Yes, that's right. –  Chris Taylor Jun 9 '11 at 14:19

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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