2
$\begingroup$

I am waiting for all methods but especially define of limit of derivative.

Attemt:

Let $x$ and $y$ depend on $t$,

and we define $f(x(t),y(t))$ function that is differentiable.

$$\dfrac{df}{dt}=\dfrac{\partial f}{\partial x}\dfrac{dx}{dt}+\dfrac{\partial f}{\partial y}\dfrac{dy}{dt}$$

That is;

If we take $U(x_1(t),x_2(t),....,x_i(t))$ $$\dfrac{dU}{dt}=\dfrac{\partial U}{\partial x_1}\dfrac{dx_1}{dt}+\dfrac{\partial U}{\partial x_2}\dfrac{dx_2}{dt}+...+\dfrac{\partial U}{\partial x_i}\dfrac{dx_i}{dt}$$

But why?

To prove, I tried to apply definition limit of derivative;

Let define, $G(t)=f(x(t),y(t))$

$$\dfrac{dG}{dt}=\lim\limits_{h\to 0}\dfrac{G(t+h)-G(t)}{h}=\lim\limits_{h\to 0}\dfrac{f(x(t+h),y(t+h))-f(x(t),y(t))}{h}$$

And let's add and remove $f(x(t+h),y(t))$ and $f(x(t),y(t+h))$

$$\dfrac{dG}{dt}=\lim\limits_{h\to 0}\dfrac{2G(t+h)\pm f(x(t+h),y(t))-\pm f(x(t),y(t+h))-2G(t)}{2h}$$

$$\dfrac{dG}{dt}=A+B+C+D$$

,So they are;

$$A=\lim\limits_{h\to 0}\dfrac{f(x(t+h),y(t+h))-f(x(t+h),y(t))}{2h}$$

$$B=\lim\limits_{h\to 0}\dfrac{f(x(t+h),y(t+h))-f(x(t),y(t+h))}{2h}$$

$$C=\lim\limits_{h\to 0}\dfrac{f(x(t),y(t+h))-f(x(t),y(t))}{2h}=\dfrac12\dfrac{\partial f}{\partial y}$$

$$D=\lim\limits_{h\to 0}\dfrac{f(x(t+h),y(t))-f(x(t),y(t))}{2h}=\dfrac12\dfrac{\partial f}{\partial x}$$

$C$ and $D$ O.K. but I didnt exactly understand $A$ and $B$ and couldn't get given formula also I want to understand the physical meaning of this.

  • 0
    Just I forgot to add 1/2 C and D, now fixed it.2017-01-13
  • 0
    you should make the hypotheses on $f, x,y$ clear. Right now, the way the problem is stated, the result is false.2017-01-13

4 Answers 4

0

Suppose $(a,b) \in U,$ where $U$ is open in $\mathbb R^2.$ Assume $f:U\to \mathbb R,$ and $f$ is differentiable at $(a,b).$ This means, by definition, that there is a linear transformation $T:\mathbb R^2 \to \mathbb R$ such that

$$f(x,y) - f(a,b) = T(x-a,y-b) + ((x-a)^2+(y-b)^2)^{1/2}\,r(x,y),$$

where $\lim_{(x,y)\to (a,b)}r(x,y)=0.$

Now $T$ has the form $T(x,y) = Ax + By.$ And the constants $A,B$ are what you would expect: $A= \partial f /\partial x (a,b),B= \partial f /\partial y (a,b).$

Now assume $g(t) = (x(t),y(t))$ maps a neighborhood of $t_0\in \mathbb R$ into $U,$ with $x(t_0)=a, y(t_0)=b.$ Assume $g'(t_0) = (x'(t_0),y'(t_0))$ exists. Then

Thm: $(f\circ g)'(t_0) = \nabla f (g(t_0))\cdot g'(t_0),$ where $\cdot $ denotes the dot product.

Proof: We're all set up:

$$\frac{f\circ g(t)-f\circ g(t_0)}{t-t_0} = A\frac{x(t)-x(t_0)}{t-t_0} + B\frac{y(t)-y(t_0)}{t-t_0}$$ $$ + \frac{((x(t)-x(t_0))^2+(y(t)-y(t_0))^2)^{1/2}}{t-t_0}r(x(t),y(t)).$$

As $t\to t_0,$ the first summand on the right $\to A x'(t_0),$ and the second summand $\to By'(t_0).$ As for the third summand, verify that the fraction is bounded, and that $r(x(t),y(t))\to 0.$ This shows the third summand $\to 0$ and we have the result.

1

Let $x_i: A \subset \Bbb R \to \Bbb R$ be differentiable on $A$ and $U: B \subset \Bbb R^n \to \Bbb R$ be differentiable on $B$, where $B \supset A\times\cdots \times A$. Let $G(t) = U(x_1(t),\ldots, x_n(t))$, i.e. $G = U \circ(x_1,\ldots,x_n)$, we have by the chain rule:

$$\frac{dG}{dt} = \nabla U(x_1,\cdots, x_n) \cdot \left(\frac{dx_1}{dt}, \ldots, \frac{dx_n}{dt}\right) = \left(\frac{\partial U}{\partial x_1}, \ldots, \frac{\partial U}{\partial x_n}\right) \cdot \left(\frac{dx_1}{dt}, \ldots, \frac{dx_n}{dt}\right) \\ = \sum_{i=1}^n \frac{\partial U}{\partial x_i} \frac{dx_i}{dt}$$

Just to be clear:

  • $\frac{d(\cdot)}{dt} = (\cdot)'(t)$.

  • $\nabla U(x_1,\ldots,x_n) = \nabla U(x_1(t),\ldots,x_n(t))$.

  • $\frac{\partial U}{\partial x_i} = \frac{\partial U}{\partial x_i}(t)$

I hope this helps.

  • 1
    But he's asking for a proof of this? Particularly using the limit definition of the derivative?2017-01-13
  • 0
    @SimpleArt he's asking for a proof of what?2017-01-13
  • 0
    The chain rule I think. By the looks of the second half of the question, it looks like its particularly about 2 variables.2017-01-13
  • 0
    @SimpleArt the one I used is that if $f: \Bbb R^n \to \Bbb R$, $g:\Bbb R \to \Bbb R^n$ are differentiable at suitable points, then $(f \circ g)'(t) = \nabla f(g(t)) \cdot g'(t)$. This is well-known and has proofs all over the place.2017-01-13
  • 0
    Yes, I know of it too. The last step was a multiplication of matrices, right? But this happens to be the first sentence of the questioner's post: *I am waiting for all methods but especially define of limit of derivative.* Anyways, I'll just let the OP be the judge on what he thinks.2017-01-13
  • 0
    @SimpleArt OP doesn't seem to be a native speaker, as the first sentence doesn't make sense - it's more likely that he meant: "I'm okay with any method, but I would like to see one which uses the limit definition of the derivative". Anyway, what matrices? That's a scalar product.2017-01-13
  • 1
    Thank you a lot of both of you :)2017-01-13
  • 0
    Really, you should make clear what hypotheses you are assuming.2017-01-13
  • 1
    @zhw. I added the assumptions.2017-01-13
1

I will try to explain the formula through its geometrical meaning.

(1) $z=G(x,y)$ describes a surface in the $x,y,z$ coordinate system.

(2) $(x(t),y(t))$ describes a curve in the $x,y$ plane.

Let, for a $t_0$: $x(t_0)=x_0$ and $y(t_0)=y_0$

If $t$, the parameter of the curve changes from $t_0$ to $t_0+\Delta t$ then $G$ changes and let the total change of $G$, $\Delta G$ be defined as

$$\Delta_x G+\Delta_y G.\tag 1$$

where $$\Delta_xG=G(x(t_0+\Delta t),y_0)-G(x(t_0),y_0)$$ and $$\Delta_yG=G(x_0,y(t_0+\Delta t))-G(x(t_0),y_0).$$

Then

$$\frac{\Delta G}{\Delta t}=\frac{\Delta_xG+\Delta_xG}{\Delta t}=$$ $$=\frac{G(x(t_0+\Delta t),y_0)-G(x(t_0),y_0)}{\Delta t}+\frac{G(x_0,y(t_0+\Delta t))-G(x(t_0),y_0)}{\Delta t}.$$

Letting $\Delta t\to 0$, we get by the chain rule (if these limits exist)

$$\frac{d G}{dt}=\frac{d G(x(t),y_0)}{dt}+\frac{d G(x_0,y(t))}{dt}=$$ $$=\frac{\partial G(x,y_0)}{\partial x}\frac{d x}{d t}+\frac{\partial G(x_0,y)}{\partial y}\frac{d y}{d t}.$$

A simpler way of writing the same is

$$\frac{d G}{dt}=\frac{\partial G}{\partial x}\frac{d x}{d t}+\frac{\partial G}{\partial y}\frac{d y}{d t}.\tag 2$$

Or

$$\frac{d G}{dt}=\nabla G \cdot v$$

where $\nabla G=\left(\frac{\partial G}{\partial x},\frac{\partial G}{\partial y} \right)$ is the gradient vector field of $G$ and $v=\left(\frac{d x}{d t},\frac{d y}{d t}\right)$ is the tangent vector to the curve defined by $(x(t),y(t))$. So, $\frac{d G}{dt}$ is the directional derivative of $G$ in the direction $v$.


The definition of the "total change of $G$" by $(1)$ may seem to be arbitrary. But the result justifies that choice.

For example, if $r=(x(t),y(t))$ then $v=\dot r$ the velocity of a particle moving along the curve at stake. Then $\frac{d G}{dt}$ is the rate of change of the height (with respect to the time) measured by a traveler skating on the surface $G$ always right above $r$.

$(1)$ and $(2)$ can be generalized for higher dimensions. Then the traveler's example can be applied again.

0

This is not the way I would have done it, though the following is not at all rigorously developed. Now, if we try to apply the limit definition, (as $h\to0$)

$$\begin{align}\frac{dG}{dt}&=\frac{f(x(t+h),y(t+h))-f(x(t),y(t))}h\\&=\frac{f(x(t+h),y(t+h))\color{#3388dd}{-f(x(t+h),y(t))+f(x(t+h),y(t))}-f(x(t),y(t))}h\\&=\frac{f(x(t+h),y(t+h))-f(x(t+h),y(t))}h+\frac{f(x(t+h),y(t))-f(x(t),y(t))}h\end{align}$$

We'll call the first fraction $A$ and the second fraction $B$.

$$\begin{align}A&=\frac{f(x(t+h),y(t+h))-f(x(t+h),y(t))}h\\&=\frac{f(x(t+h),y(t+h))-f(x(t+h),y(t))}{\color{#3388dd}{y(t+h)-y(t)}}\frac{\color{#3388dd}{y(t+h)-y(t)}}h\end{align}$$

Now, if this limit exists, then,

$$A=\lim_{h_x\to0}\lim_{h\to0}\frac{f(x(t+h_x),y(t+h))-f(x(t+h_x),y(t))}{y(t+h)-y(t)}\frac{y(t+h)-y(t)}h\\=\lim_{h_x\to0}\frac{\partial f(x(t+h_x),y)}{\partial y}\frac{dy}{dt}\\=\frac{\partial f}{\partial y}\frac{dy}{dt}$$

Similarly, you will find that $B=\frac{\partial f}{\partial x}\frac{dx}{dt}$, though with no requirement of splitting the limit like we did above..

  • 0
    that's what I need2017-01-13
  • 0
    And I wonder that how we can generalize this?2017-01-13
  • 0
    @Photoneaterman One generalization is the chain rule for the differential (aka total derivative) of $f\circ g$, where $f:\mathbb R^m\to\mathbb R^n$ and $g:\mathbb R^l\to\mathbb R^m$ (or some other domains and codomains with compatible dimensions). It basically says that you can compose the functions first and find their differential, or find the two differentials first and then compose them (i.e., that differentiation and composition “commute”). The chain rule that you’re familiar with and others involving partial derivatives are essentially special cases of the one for total derivatives.2017-01-13
  • 0
    @Photoneaterman Not sure if you want to hear this, but if you apply the limit definition in the general case and follow the exact same steps (just a bit more general and messy), the result is the same as you would expect.2017-01-13
  • 0
    This proof is incorrect. There are no hypotheses stated or followed. You are dividing by things that might be $0,$ and the bit about $x(t+h)$ turning into $x(t)$ makes zero sense.2017-01-13
  • 0
    @zhw. Mhm... I see what you mean. But if $y(t+h)-y(t)=0$ for all $|h|<\epsilon$, then the chain rule fails anyways, right? And I was wondering about the $x(t+h)$ part and how that would be affected, but I couldn't figure much out of it. I suppose I assume $x(t)$ exists and all the functions are continuous, but I'm honestly bad at being rigorous XD2017-01-13
  • 0
    As @zhw pointed out, the proof presented is not complete and rigorous, but it is a nice heuristic argument. It would be good to note this somewhere in the answer.2017-01-13
  • 0
    @ZachBoyd Ok, I noted that now in my first line. :| Wondering how to do it more completely now... since it is pretty close to done-looking.2017-01-13