Equivalent definitions for the differentiability in general

466 Views Asked by At

I'm looking for equivalent definitions for the differentiability in general . I'm really confused about the differentiation of multivariable functions . Also I want the proofs for equivalency of the definitions .

Here is the definition in the Stewart calculus but it doesn't make sense to me : enter image description here

2

There are 2 best solutions below

1
On

There are plenty of definitions for differentiability. For example, within the context of a $\mathbb{R}^n$, we can suggest the definition to be:

A function $f: \mathbb{R}^n \rightarrow \mathbb{R}^m$ is differentiable at $x \in \mathbb{R}^n$ provided there exists a linear mapping $A$ such that the limit $$\lim_{\epsilon \rightarrow 0, \ \ h \in B_{\epsilon}(x)} \frac{|| f(x + h) - (f(x) + h A(x)) ||}{||h||} = 0 $$

Clearly, in the context you are looking for, $n = 2, m = 1$.

It becomes a little complicated to prove equivalence however, simply by Analysis and it's ability to fit within a variety of different contexts and applications. An immediate question that is often discussed after the definition I wrote above would be how it relates to the "differential", which seems to be the approach to a derivative the excerpt you included implies. So, firstly, we would consider the differential of this function to simply be: $$df_x(h) = A(x) \cdot h$$ If we express $x = \pmatrix{a \\ b}$, $A(x) = \pmatrix{f_x(a,b) \\ f_y(a,b)}$ and $h = \pmatrix{\Delta x \\ \Delta y}$, then we see that: $$df_x(h) = f_x(a,b) \Delta x + f_y(a,b) \Delta y$$ Now giving ourselves a room of epsilon, we see that: $$\Delta z = \Delta f_x(h) = f_x(a,b) \Delta x + f_y(a,b) \Delta y + \Delta x \cdot \epsilon_1 + \Delta y \cdot \epsilon_2$$ where we ensure $\sqrt{\epsilon_1^2 + \epsilon_2^2} < \epsilon$ in the limit statement. So you see this statement is the same as the one you provided. It is rather specific, which is indivative of the source Stewart Calculus.

0
On

Stewart's definition does seem a little confusing. I assume the notations $\Delta z,$ $\Delta x,$ and $\Delta y$ were defined earlier, since the definition is meaningless otherwise.

The essence of the derivative of a single-variable function $g$ is that in the neighborhood of an input value $x=a,$ the function is very well approximated by a first-degree polynomial function of $x,$ which is a function describing a line. That is, $$ g(x) \approx g(a) + m (x-a).\tag1$$ The derivative of $g$ at $x=a$ is then $g'(a) = m.$ The rest of the mechanics of the derivative are there to ensure that the linear function in $(1)$ actually is a good approximation, and if we need it to be better, we can just restrict ourselves to look at a smaller neighborhood of $x=a.$

The idea of the multivariate derivative is the same, but the input now is the pair of coordinates of a point $(x,y).$ The derivative of $f$ at $(x,y)=(a,b)$ should allow us to approximate $f$ near $(x,y)=(a,b)$ by first-degree polynomial in $x$ and $y$, which describes a plane. That is, $$ f(x,y)\approx f(a,b) + m(x-a) + n(y-b).$$ The parameters $m$ and $n$ in the approximate equation are the partial derivatives of $f$ at $(x,y)=(a,b)$, which are often written $f_x(a,b)$ and $f_y(a,b)$.

Stewart introduces the factors $\epsilon_1$ and $\epsilon_2$ to account for the fact that the approximation often is not exact and that it gets worse as you get farther from $(x,y)=(a,b).$ This is a little awkward, since (unlike in the single-variable case) you cannot determine both of these values at an arbitrary point $(x,y)$ just by looking at the values and derivative of $f$ at $(a,b)$ and the arbitrary $(x,y).$ You would have two unknowns but only one equation. Still, we could "guess" a relationship between arbitrary coordinates and the epsilons at those coordinates that would make everything in the definition true, provided that $f$ actually is well approximated by a plane at $(x,y)=(a,b).$

That part of the definition seems a bit messy. You have another answer that defines the derivative by looking at coordinates within a disk around $(x,y)=(a,b)$ and shrinking the radius of that disk to zero in the limit. It also lets us use just one "epsilon" instead of two. That seems to me to be a better way to define the limit than just writing $(\Delta x,\Delta y)\to(0,0).$ But I suspect that if you read the material before (or perhaps after) this definition carefully, the idea is the same.