4

If $f(tx,ty,tz) = t^nf(x,y,z)$ then in my lecture notes it says differentiating this equation with respect to $t$ gives:

$$x\frac{df}{dx} + y\frac{df}{dy} + z\frac{df}{dz} = nt^{n-1}f(x,y,z)$$

But why isn't it:

$$x\frac{df}{d(tx)} + y\frac{df}{d(ty)} + z\frac{df}{d(tz)} = nt^{n-1}f(x,y,z)\ ?$$

AronK
  • 165
  • 1
    To whoever approved Orangutango's edit: please read carefully and examine what type of question OP want to address. The reason "nonsensical notation" Orangutango put in the edit is really nonsensical. – Shuhao Cao Jun 15 '13 at 20:07
  • @AronK I agree with your post, if I am free to interpret it as $x \frac{\partial f}{\partial x_1}+y \frac{\partial f}{\partial x_2}+ z \frac{\partial f}{\partial x_3} = nt^{n-1}f(x,y,z)$. – James S. Cook Jun 15 '13 at 20:42

3 Answers3

2

Suppose $f(tx,ty,tz)=t^nf(x,y,z)$. Consider the chain rule: $$ \frac{d}{dt} f(x_1(t),x_2(t),x_3(t)) = \frac{\partial f}{\partial x_1}(x_1(t),x_2(t),x_3(t))\frac{dx_1}{dt}+\frac{\partial f}{\partial x_2}(x_1(t),x_2(t),x_3(t))\frac{dx_2}{dt}+\frac{\partial f}{\partial x_3}(x_1(t),x_2(t),x_3(t))\frac{dx_3}{dt} $$ I am being really pedantic here, but the question raised involves this issue. The partial derivatives are properly understood to be evaluated at the inside function which happens to be $(x_1(t),x_2(t),x_3(t))$ in this context. Returning to the given identity, identify $x_1=tx$ and $x_2=ty$ and $x_3=tz$ where presumably $x,y,z$ are not functions of time (if they were we'd obtain a different identity). Note: $$ \frac{dx_1}{dt} = x, \ \ \frac{dx_2}{dt} = y, \ \ , \frac{dx_3}{dt} = z $$ Consequently, $$ \frac{d}{dt} f(x_1(t),x_2(t),x_3(t)) = x\frac{\partial f}{\partial x_1}(x_1(t),x_2(t),x_3(t))+y\frac{\partial f}{\partial x_2}(x_1(t),x_2(t),x_3(t))+z\frac{\partial f}{\partial x_3}(x_1(t),x_2(t),x_3(t)) $$ which means, $$ \frac{d}{dt} f(tx,ty,tz) = x\frac{\partial f}{\partial x_1}(tx,ty,tz)+y\frac{\partial f}{\partial x_2}(tx,ty,tz)+z\frac{\partial f}{\partial x_3}(tx,ty,tz) $$ So, I tend to think your book takes some liberty with notation.

James S. Cook
  • 16,755
  • Notice, if I was to write $\frac{\partial f}{\partial x}$ then by default that implies evaluation at $x,y,z$, so I could reasonably drop the triple and write $\frac{\partial f}{\partial x_1}$ in place of $\frac{\partial f}{\partial x_1}(tx,ty,tz)$. For me the issue is that the evaluation of the partials is not in concert with the multiplicative factors $x,y,z$ out front. – James S. Cook Jun 15 '13 at 20:39
1

Your thoughts are mostly correct, but just now written clearly.

Normally, a function in three variables is denoted $f(x, y, z)$, and the partial derivatives are $\frac{\partial f}{\partial x}$, $\frac{\partial f}{\partial y}$, $\frac{\partial f}{\partial z}$, what they mean is the partial derivative with respect to each variable.

To interpret $f(tx, ty, tz)$ nicely, we shall write $f$ as a function $f(x_1, x_2, x_3)$, and $x_1(t) = tx, x_2(t) = ty, x_3(t) = tz$, so chain rule gives

$$ \frac{d}{dt}f(x_1(t), x_2(t), x_3(t)) = \frac{dx_1}{dt}\frac{\partial f}{\partial x_1} + \frac{dx_2}{dt}\frac{\partial f}{\partial x_2} + \frac{dx_3}{dt}\frac{\partial f}{\partial x_3}. $$

So you are confused with $x, y, z$ denoting the first, second, third terms of $f$, and the fact that are used as other variables again.

1

The notes is alright but not clear enough, you are in the right direction as well, but your notation is not the best way to illustrate it.

The $\partial f/\partial x$ in your notes is rather taking derivative with respect to the first variable than taking derivative with respect to $x$.

Try to think in this way. Say you have a function with three slots to fill (notice this function satisfy the relation you gave, so it may serve as an example): $$ f(\square, \triangle,\bigcirc) = \square^n + \triangle^n + \bigcirc^n.\tag{1} $$ Now we fill the first variable by $x(t)$, the other two untouched then: $$ \frac{df}{dt} = \frac{\partial }{\partial \square}f(\square, \triangle,\bigcirc) \Bigg\vert_{\square = x(t)} \cdot \frac{d x(t)}{dt}, $$ the first term is taking derivative with respect to the first variable evaluated at $x(t)$. Taking (1) as an example: let $x(t) = tx$ $$ \frac{df}{dt} = \frac{\partial }{\partial \square}f(\square, \triangle,\bigcirc) \Bigg\vert_{\square = tx} \cdot x = n \square^{n-1}\big|_{\square = tx} \cdot x = nt^{n-1}x^n, $$ If we evaluate directly $$ f(tx, \triangle,\bigcirc) = t^n x^n + \triangle^n + \bigcirc^n, $$ you will get the same result as chain rule.

Keeping this in mind your formula can be written in a least ambiguous way: for $f(u,v,w)$

$$x\frac{\partial f}{\partial u}\Bigg\vert_{u = tx} + y\frac{\partial f}{\partial v} \Bigg\vert_{v = ty} + z\frac{\partial f}{\partial w}\Bigg\vert_{w = tz} = nt^{n-1}f(x,y,z).$$

Shuhao Cao
  • 18,935