B Why does the Limit Process give an Exact Slope?

NoahsArk
Gold Member
Messages
258
Reaction score
24
TL;DR Summary
Why does the limit process give an exact slope of the tangent line instead of just a very close approximation?
I am taking a summer calculus class now. For years I've been stuck on the question of why the limit process gives us an exact slope of the tangent line instead of just a very close approximation. I don't need to know the reason for this class I'm taking- we are basically just learning rules of derivates and integrals and applying them. However, it bothers me that I still don't understand the reason after this long.

When you take the limit of the rise over run of any function as delta x (which I'll call h) gets closer to 0, you can never make h equal to zero because then you'd have zero in the denominator. Also, if h were zero, you'd have a point and not a line, and a point can't have a slope. Since h can never be zero, how can you ever know the exact slope?

The only explanations I can think of aren't fully satisfying to me, and I'm not sure if they are good ones. One example of an explanation is in taking the slope of ##f(x) = x^2## at the point where x = 2. Using the limit process you would get the slope function ##\frac {4h + h^2} {h}## You can't set h = to zero here but you can get an equivalent function of 4 + h. So the slope at x = 2 is 4. These two slope functions are equivalent but one has a gap and the other one doesn't, so we can say that the slope must be where the gap is since that's the only place where h is zero. This seems a bit hand wavy to me since the simplified slope function of 4 + h is not the exact same function as ##\frac {4h + h^2} {h}##.

Is the explanation more trivial then I'm thinking, and is there just some conceptual misunderstanding I'm having. Or, it it actually complicated and the subject of another class like real analysis, in which case it's ok for me not to have more than an intuitive feeling for why the limit process gives us an exact slope? Thanks
 
Physics news on Phys.org
The limit isn't a recursive process. It's a single number that has a certain, well-defined property.
 
Thank you for the response @PeroK . I know that the slope is a single number. My question is how do we know for sure what the exact number is. We can never set ##\Delta x## to zero to know an exact number for sure.
 
NoahsArk said:
Thank you for the response @PeroK . I know that the slope is a single number. My question is how do we know for sure what the exact number is. We can never set ##\Delta x## to zero to know an exact number for sure.
The slope of the curve at a point is defined as the slope of the tangent at this point, which can be visualized as the limit of the slope of secants defined through the tangent point ##p## and another one ##p+\Delta x## that is approaching it.


Tangent.png



$$
f'(p)=\left. \dfrac{d}{dx}\right|_{x=p}f(x)=\lim_{\Delta x \to 0}\dfrac{f(p+\Delta x)-f(p)}{\Delta x}
$$
 
@fresh_42 yes, my confusion is not in what a tangent line is- I'm confused about how we know the value of the slope of the tangent line from using the limit process. I am confused because the process is setting ##\Delta x ## to be zero, but you can't have a zero in the denominator. So, while we can get really really close to the slope of the tangent line by setting ##\Delta x ## as close to zero as we want, we can never set it to zero. So, how can we ever have anything more than a really close approximation? Thanks
 
NoahsArk said:
@fresh_42 yes, my confusion is not in what a tangent line is- I'm confused about how we know the value of the slope of the tangent line from using the limit process. I am confused because the process is setting ##\Delta x ## to be zero, but you can't have a zero in the denominator. So, while we can get really really close to the slope of the tangent line by setting ##\Delta x ## as close to zero as we want, we can never set it to zero. So, how can we ever have anything more than a really close approximation? Thanks
You don't get a limit by setting ##\Delta x=0##.
 
@ PeroK, I meant you you get the slope of the tangent line by taking the limit as ##\Delta x## approaches zero of the slope function ## \frac {f(x +h) - f(x)} {h}## (where h = ##\Delta x##). My confusion lies in the fact that although very small changes in x will give closer and closer approximations to the slope of a given point, you can't actually set the change in x to be zero without getting a zero in the denominator. My description of what I am getting confused on was poorly worded I think, but I hope I've clarified.
 
NoahsArk said:
@ PeroK, I meant you you get the slope of the tangent line by taking the limit as ##\Delta x## approaches zero of the slope function ## \frac {f(x +h) - f(x)} {h}## (where h = ##\Delta x##). My confusion lies in the fact that although very small changes in x will give closer and closer approximates to the slope of a given point, you can't actually set the change in x to be zero without getting a zero in the denominator. My description of what I am getting confused on was poorly worded I think, but I hope I've clarified.
That's correct. You need to look at the definition of the limit.
 
Perhaps consider a concrete example. Say ##f(x)=x^2##. What is ##\frac{f(x+\Delta x)-f(x)}{\Delta x}## and is there a ##\Delta x## in the denominator?
 
  • #10
@Ibix I know the answer to this is 2x both from doing the limit process and also from learning the power rule. I also have an intuitive visual sense of why this is the case. However, what I don't know is why it is EXACTLY 2x and not just very close to 2x. Seems like a Xeno's paradox type issue.
 
  • #11
NoahsArk said:
@Ibix I know the answer to this is 2x
That's not the answer to the question I asked. The correct answer is that ##\frac{f(x+\Delta x)-f(x)}{\Delta x}=2x+\Delta x##. So there's no ##\Delta x## in the denominator and no problem letting it tend to zero.
 
  • Like
Likes Vanadium 50
  • #12
NoahsArk said:
@fresh_42 yes, my confusion is not in what a tangent line is- I'm confused about how we know the value of the slope of the tangent line from using the limit process. I am confused because the process is setting ##\Delta x ## to be zero, but you can't have a zero in the denominator.

It is not a process, and we are not setting anything to zero. The "process" with the secants was only meant to visualize what a limit is.
$$
\lim_{\Delta x \to 0} \dfrac{f(p+\Delta x)-f(p)}{\Delta x} = s
$$
stands for a logical statement: For any given ##\varepsilon >0## there is a ##\delta(p,\varepsilon )> 0## such that ##\left|\dfrac{f(p+\Delta x)-f(p)}{\Delta x} - s\right| <\varepsilon ## whenever ##|\Delta x| < \delta(p,\varepsilon ).##

It means that we can get with our quotient - ##\dfrac{f(p+\Delta x)-f(p)}{\Delta x}## - the slope of the secants, as close as we want - ##\varepsilon ## - to the slope of the tangent - ##s## - if we are picking our second point - ##p+\Delta x ## - close enough - less than ##\delta(p,\varepsilon )## - to ##p##.

I wrote ##\delta(p,\varepsilon)## instead of the usual ##\delta## to note that the choice of ##\delta## depends on the choice of ##\varepsilon ## and the location of the tangent ##p.##

Anyway, this logical statement is neither a process nor did we anywhere replace ##\Delta x## by zero.

Let's see what my picture says:
\begin{align*}
f(x)&=\dfrac{1}{5}x^2 \wedge p=-5\\[6pt]
f'(-5)&=\lim_{\Delta x \to 0}\dfrac{f(-5+\Delta x)-f(-5)}{\Delta x}=\lim_{\Delta x \to 0}\dfrac{1}{5}\cdot \dfrac{25-10\Delta x+(\Delta x)^2 - 25}{\Delta x}=2-\dfrac{1}{5}\lim_{\Delta x \to 0} \Delta x
\end{align*}

Now, we are at a crucial point. Why is ##\lim_{A \to 0}A=0##? The logical statement says: We can get with ##A## as close as we want to ##0## if we only are sufficiently close to zero. That is certainly true, and the logical statement can easily be verified. But it is also rather trivial. We have ##\varepsilon =\delta## and it happens to be, that we get the same result if we simply identify ##A=0.## It's a matter of convenience to shorten the verification of the logical statement. And as it works, we can often make that substitution, but only in the last line of my little calculation. We did not divide anything that was zero.

NoahsArk said:
So, while we can get really really close to the slope of the tangent line by setting ##\Delta x ## as close to zero as we want, we can never set it to zero.
Correct.
NoahsArk said:
So, how can we ever have anything more than a really close approximation? Thanks
Because we ended up with ##\lim_{A \to 0}A## which equals ##0## because it is a number ##L=0##:
$$
\forall \;\varepsilon >0 \; \exists \;\delta:=\varepsilon >0\;:\; |A-L|=|A-0|=A<\varepsilon \;\forall \; |A|< \delta
$$
This is simply a true statement, so ##\lim_{A \to 0}A=0.## We have thus proven that we can compute ##L## in ##\lim_{A \to 0}A=L## by replacing ##A## by ##0.## That makes computations a lot easier. But is one thing to do it here, and another to divide by zero.
 
  • Like
Likes NoahsArk
  • #13
@Ibix Ok thanks for clarifying. While I am aware that ## \frac {f(x + \Delta x)} {\Delta x} = 2x + \Delta x ##, this explanation has still been not totally convincing for me because the two slope functions aren't really exactly the same. One of them, ## \frac {f(x + \Delta x)} {\Delta x} ## has a gap in the graph where x = 0, and the other one ## 2x + \Delta x ##, has no gap. The first of these though is the real slope function since it is in the form of rise over run. So isn't that the form what we need to use? Why can we get away with creating a new (although very similar) function?
 
  • Skeptical
Likes weirdoguy
  • #14
Is it easier to justify in your head if you take the limit in a more symmetric rfashion? (The result will be the same in the limit.)
-
$$ \frac {f(x+\Delta x)-f(x-\Delta x))} {2\Delta x}$$

Do you not believe that there exists a unique value for the slope??? How else would one find its value?
 
  • Like
Likes Ibix
  • #15
(Relative to the example ##f(x) = x^2##)
NoahsArk said:
However, what I don't know is why it is EXACTLY 2x and not just very close to 2x. Seems like a Xeno's paradox type issue.
Nothing to do with Zeno's Paradox.

This difference quotient below gives the slope of the secant line between the points ##(x, f(x))## and ##(x + h, f(x + h))##.
$$\frac{f(x + h) - f(x)}h = \frac{(x + h)^2 - x^2}h = \frac{2xh + h^2}h = 2x + h$$
For any two points x and x + h, the slope of the secant line will be 2x + h.
The derivative of f is the limit of the above difference quotient; that is,
$$f'(x) = \lim_{h \to 0} \frac{f(x + h) - f(x)}h = \lim_{h \to 0} 2x + h = 2x$$

What you seem to be having a problem with is the fact that we have h in the denominator and are taking the limit as h approaches 0. Note that the numerator is also approaching 0 as well. That makes the fraction one of several indeterminate forms -- expressions that may or may not represent a number. Every time you calculate a derivative using the definition of the derivative you're working with an indeterminate form.

The situation here is similar to the situation when we compare the two functions ##f(x) = x + 1## and ##g(x) = \frac{x^2 - 1}{x - 1}##. These two functions are almost exactly the same, differing only at exactly one point; namely, (1, 2). If we take the limit of f as x approaches 1, we trivially get 2 for the result. Although the second function, g, isn't defined at x = 1, we can take the limit of g(x) as x approaches 1, and also get a result of 2. The graphs of these two functions are nearly identical, but we say that the latter function has a "hole" at (1, 2).
 
  • Like
Likes chwala, Ibix and NoahsArk
  • #16
NoahsArk said:
@Ibix Ok thanks for clarifying. While I am aware that ## \frac {f(x + \Delta x)} {\Delta x} = 2x + \Delta x ##, this explanation has still been not totally convincing for me because the two slope functions aren't really exactly the same. One of them, ## \frac {f(x + \Delta x)} {\Delta x} ## has a gap in the graph where x = 0, and the other one ## 2x + \Delta x ##, has no gap. The first of these though is the real slope function since it is in the form of rise over run. So isn't that the form what we need to use? Why can we get away with creating a new (although very similar) function?
It's because when you take a limit, you don't care what the functions are doing at ##x## (or if they're even defined there); you only care about what they're doing in the neighborhood of that point. In that neighborhood, the two functions are exactly the same, so they approach the same limiting value as ##\Delta x \to 0##.
 
  • #17
NoahsArk said:
@Ibix I know the answer to this is 2x both from doing the limit process and also from learning the power rule. I also have an intuitive visual sense of why this is the case. However, what I don't know is why it is EXACTLY 2x and not just very close to 2x. Seems like a Xeno's paradox type issue.
It's important to dissociate the function values from the limit. Believing that the limit must be one of the function values seems to be a stumbling block for many students.

As an example, consider the set ##\{\frac 1 n\} ## for natural numbers ##n##. Clearly no member of that set is zero. But, zero is a lower bound for that set. And, in fact, it's the greatest lower bound.

We see that zero has an important relationship with that set despite not being a member of the set. This is actually the starting point for the formal study of limits. In fact, we can also say:
$$\lim_{n \to \infty} \frac 1 n = 0$$And, again, zero (precisely zero) is the limit of that sequence without being an element of the sequence.

Limits and derivatives of real-valued functions are developed from this.
 
  • Like
Likes chwala
  • #18
Definition of limit

Before formally define the notion of limit, we must take a glance at the intuitive meaning. The definition shows two concepts that need more precision.

1. The first one is the meaning of the sentence "##f(x)## is as close to the limit ##l## as you want".
We say this sentence when, for any positive quantity, the gap between ##f(x)##
and ##l## can be taken, in absolute value, less than the mentioned positive quantity.
2. The sentence "##x## enough nearby to ##a##"is the second aspect of the definition that needs to be explained.
Means that, in absolute value, the difference between ##x## and ##a## is less than a positive small quantity.

Note that, in the formal statement, the condition ##0<|x-a|## is mandatory, just to exclude from the definition the point ##a##.

I've taken the quote from a maths textbook for Spanish 17 to 18 years old students.
I've got the sensation @NoahsArk that, I am posting to you though you already know what is all about. Anyhow, best wishes!
 
  • #19
NoahsArk said:
Why does the limit process give an exact slope of the tangent line instead of just a very close approximation?
How do you define the "tangent line"?

In mathematics, notions like a "tangent line" are meaningless until we define them. Having defined them, the meaning is in the definition. There is nothing else.

Once you define the "tangent line" and the "slope" of the tangent line at a point you have, in principle, a number for the slope of the tangent line at a point.

Once you define a "first derivative" of a function as a number (if it exists) that satisfies some "limit" definition at a point you also have a number.

It is likely that you designed both definitions to yield unambiguous results. It is likely that you chose at least one of these definitions so that it would match the other.

It should then be no surprise that the two numbers, so defined, are unambiguous and matching.
 
  • Like
Likes PeroK
  • #20
Others here have mentioned this, but I'll single out the idea of the epsilon-delta proof and a key concept of calculus, that things that are "arbitrarily close" are essentially equal.
 
  • Like
Likes mcastillo356
  • #21
I don't think you're getting an exact
Ibix said:
Perhaps consider a concrete example. Say ##f(x)=x^2##. What is ##\frac{f(x+\Delta x)-f(x)}{\Delta x}## and is there a ##\Delta x## in the denominator?
There may be an exception , both in the definition, result, when we consider the linear change of a line; linear object ( n-dimensional linear object). Here, the approximation to the change is _ equal_ to the change, as , given say a line## f(x)=ax## ( ignore the constant; just a translation) the limit quotient becomes ##\frac {ax+a(\Delta x)-(ax)}{\Delta x}=\frac{a\Delta x}{\Delta x}## which equals ##a## as long as ##\Delta x \neq 0##. Basically the best local linear approximation to the change of the ( linear) function is given by the change in the linear function.
 
  • #22
WWGD said:
I don't think you're getting an exact
If ##f(x) = x^2## then the formula for the slope of a secant line: ##\frac{f(x+\Delta x) - f(x)}{\Delta x}## is exact.

It is exactly ##\frac{(x + \Delta x)^2 - x^2}{\Delta x} = \frac{x^2 + 2x\Delta x + \Delta x^2 - x^2}{\Delta x} = \frac{2x\Delta x^2 + \Delta x^2}{\Delta x} = 2x + \Delta x##

That formula is exact and is valid for ##\Delta x \ne 0##.

If we define the slope of ##f()## at ##x## to be the limit of the slope of these secant lines as ##\Delta x## approaches zero then we are asking about ##\lim_{\Delta x \to 0}\ 2x + \Delta x##.

That limit is ##2x## exactly. [Note the the limit definition carefully never involves what happens when ##\Delta x = 0##]

If we do not define the slope of ##f()## at ##x## then we have nothing to talk about.

WWGD said:
There may be an exception , both in the definition, result, when we consider the linear change of a line; linear object ( n-dimensional linear object). Here, the approximation to the change is _ equal_ to the change, as , given say a line## f(x)=ax## ( ignore the constant; just a translation) the limit quotient becomes ##\frac {ax+a(\Delta x)-(ax)}{\Delta x}=\frac{a\Delta x}{\Delta x}## which equals ##a## as long as ##\Delta x \neq 0##. Basically the best local linear approximation to the change of the ( linear) function is given by the change in the linear function.
If ##f(x) = ax## then the same process produces a formula for the slope of the secant lines: ##a##.

If we define the slope of this new ##f()## at ##x## to be limit of the slopes of these secant lines as ##\Delta x## approaches zero then we are asking about ##lim_{\Delta x \to 0}\ a##.

Unsurprisingly, that limit is ##a## exactly.
 
  • Like
Likes weirdoguy and PeroK
  • #23
jbriggs444 said:
If ##f(x) = x^2## then the formula for the slope of a secant line: ##\frac{f(x+\Delta x) - f(x)}{\Delta x}## is exact.

It is exactly ##\frac{(x + \Delta x)^2 - x^2}{\Delta x} = \frac{x^2 + 2x\Delta x + \Delta x^2 - x^2}{\Delta x} = \frac{2x\Delta x^2 + \Delta x^2}{\Delta x} = 2x + \Delta x##

That formula is exact and is valid for ##\Delta x \ne 0##.

If we define the slope of ##f()## at ##x## to be the limit of the slope of these secant lines as ##\Delta x## approaches zero then we are asking about ##\lim_{\Delta x \to 0}\ 2x + \Delta x##.

That limit is ##2x## exactly. [Note the the limit definition carefully never involves what happens when ##\Delta x = 0##]

If we do not define the slope of ##f()## at ##x## then we have nothing to talk about.


If ##f(x) = ax## then the same process produces a formula for the slope of the secant lines: ##a##.

If we define the slope of this new ##f()## at ##x## to be limit of the slopes of these secant lines as ##\Delta x## approaches zero then we are asking about ##lim_{\Delta x \to 0}\ a##.

Unsurprisingly, that limit is ##a## exactly.
But for a line, the change is constant,i.e., the limit is constant, not a function of x nor delta x, rather than a function. Unlike in other cases, the linear change in the change near a point is _ equal_ to the change of the function, and constant. In terms of differentials, for linear functions L(x), dL(x)=L(x)dx.
 
  • #24
WWGD said:
But for a line, the change is constant,i.e., the limit is constant, not a function of x nor delta x, rather than a function.
A constant function is still a function.
 
  • Like
Likes PeroK
  • #25
jbriggs444 said:
A constant function is still a function.
My point is the approximation to the change along a linear object is given _ exactly_ , by the change of the linear object, rather than by an approximation along a linear hyperplane. No limits needed. The change in the value of L(x)=ax from x to xo:=x+Deltax can be _ exactly_ determined as a(xo+Delta x)-axo== a*Delta x. It needs no use of limits, unlike the case for nonlinear functions.
 
  • #26
Do tell, @PeroK , your specific objections.
 
  • #27
Specifically, for, say, f(x)=2x, the change of f(x)=2x, from x=2, to x=2.01, is _ exactly_, independent of any limiting process, equal to 2(2.01)-2(2)= 0.02. You can't do this for , say sinx, nor x^3, etc. , where we estimate the change, rather than compute it.
 
  • #28
WWGD said:
Do tell, @PeroK , your specific objections.
I've been scrupulously avoiding reacting to your posts with scepticsm and contenting myself by liking those of the man from Maryland.
 
  • Like
Likes SammyS and jbriggs444
  • #29
WWGD said:
Specifically, for, say, f(x)=2x, the change of f(x)=2x, from x=2, to x=2.01, is _ exactly_, independent of any limiting process, equal to 2(2.01)-2(2)= 0.02. You can't do this for , say sinx, nor x^3, etc. , where we estimate the change, rather than compute it.
So your objection is that (for ##\sin x##, ##x^3##, etc) that an approximation to the slope changes depending on where you choose to measure the approximation? Yes. So?

Do you conclude that the limit approached by the set of approximations does not exist? That would be false.

Or that the limit is inexact? That would be false. If the limit exists, it is exact and unique.

Or that the notion of the "slope" of a function at a point is inherently undefined. You might argue for that position. But such an argument would be pointless. Mathematics is not about things that lack definitions. It is about things that have definitions.
 
  • Like
Likes PeroK
  • #30
jbriggs444 said:
So your objection is that (for ##\sin x##, ##x^3##, etc) that an approximation to the slope changes depending on where you choose to measure the approximation? Yes. So?

Do you conclude that the limit approached by the set of approximations does not exist? That would be false.

Or that the limit is inexact? That would be false. If the limit exists, it is exact and unique.

Or that the notion of the "slope" of a function at a point is inherently undefined. You might argue for that position. But such an argument would be pointless. Mathematics is not about things that lack definitions. It is about things that have definitions.
I never said any such thing. I said for a line , one we can find the _ actual_ change, without the use of limits. That can't be done for nonlinear functions. That's a fact.
 
  • #31
Anything, @PeroK ? Want to chime in?
 
  • #32
WWGD said:
I never said any such thing. I said for a line , one we can find the _ actual_ change, without the use of limits. That can't be done for nonlinear functions. That's a fact.
What actual change are you measuring?

Do you claim that you are actually measuring the slope of a straight line at a point? Without having first defined the slope of a function at a point.

Do you claim that our calculations for the slope of various secants are not actual or are not exact? Or are you simply pointing out that you can find secants that fail to match the slope of a function at a point. Yes. You can. We all agree about that. We can even find examples like ##f(x) = x^3## where every secant will have positive slope while (we claim that) the slope of ##f(x)## at ##x=0## is zero. Did you have a point to make beyond that?

I definitely respect the choice by @PeroK not to respond. It is difficult to argue about a viewpoint that is deeply held but which is not being coherently stated.
 
Last edited:
  • #33
I'm too busy rustling up a chicken curry.
 
  • Like
  • Haha
Likes chwala, SammyS and fresh_42
  • #34
jbriggs444 said:
What actual change are you measuring?

Do you claim that you are actually measuring the slope of a straight line at a point? Without having first defined the slope of a function at a point.

Do you claim that our calculations for the slope of various secants are not actual or are not exact? Or are you simply pointing out that you can find secants that fail to match the slope of a function at a point. Yes. You can. We all agree about that. We can even find examples like ##f(x) = x^3## where every secant will have positive slope while (we claim that) the slope of ##f(x)## at ##x=0## is zero. Did you have a point to make beyond that?

I definitely respect the choice by @PeroK not to respond. It is difficult to argue about a viewpoint that is deeply held but which is not being coherently stated.
It's your misframing my reply that is the problem here. Yes, df is the _ approximate_ change along the tangent line/plane. While ## \Delta f ## is the _ actual_ change. That is a fact, and not lacking in coherence. Maybe @fresh_42 can chime in.
 
  • #35
WWGD said:
It's your misframing my reply that is the problem here. Yes, df is the _ approximate_ change along the tangent line/plane. While ## \Delta f ## is the _ actual_ change. That is a fact, and not lacking in coherence. Maybe @fresh_42 can chime in.
I agree that the first derivative of a function will [often] give rise to a useful linear approximation to the function value in an interval. That is an immediate consequence of Taylor's theorem.

This does not mean that ##df## (or, more correctly, ##\frac{df}{dx}## or ##f'(x)##) is only approximate. It is exact.

Formally, ##df## is a piece of notation, not really a number. I assume that you are using it to refer to the first derivative of the function ##f()## at a particular point.
 
Last edited:
  • #36
It is not exact. The _actual_change in the values of ##x^2## between ##x=1## and ## x=1.01## is## \Delta f=(1.01)^2-1^2 =0.0201##. The approximate change ##df=2xdx= 2(1)(0.01)=0.02 \neq 0.0201##.
Like you said, if the change was exact, we wouldn't need the Taylor series.
But I believe we're talking at cross from each other.
 
  • #37
WWGD said:
It is not exact. The _actual_change in the values of ##x^2## between ##x=1## and ## x=1.01## is## \Delta f=(1.01)^2-1^2 =0.0201##. The approximate change ##df=2xdx= 2(1)(0.01)=0.02 \neq 0.0201##.
I agree that the change predicted by the linear approximation is 0.02 while the actual change is 0.0201.
I disagree that the notation "##df##" properly denotes the linear approximation.

In particular, I do not agree that "##df##" denotes ##f'(x) \times \Delta x## and do not accept that "##df##" is intended to be an approximation for ##f(x+\Delta x) - f(x)##.

WWGD said:
But I believe we're talking at cross from each other.
Indeed. Your usage of notation is not standard.
 
Last edited:
  • #38
The derivative of the function ##f(x) = x^2## is ##f'(x) = 2x##. This is exact. How you justify that mathematically is another question. Standard analysis uses limits - which are themselves exact.

This can also be written ##f'(x) \equiv \frac{df}{dx} = 2x##, which leads to the concept and defining property of differentials, which are not real numbers: $$df = 2xdx$$More generally, if ##y = f(x)##, then $$dy = f'(x)dx$$
 
  • Like
Likes jbriggs444
  • #39
you think this is confusing, how about this: a real number actually is a Cauchy sequence of rational numbers (modulo equivalence),

so if f is a rational polynomial like X^2, and X0 is a rational number then the slope, as a real number, of the graph y=f(X), at (X0,f(X0)), is equal to the (real number defined by the) Cauchy sequence of rational approximations:

An = [f(X0+1/n)-f(X0)]/(1/n), as n—> infinity.

I.e. the limit of the sequence {An} is actually the sequence itself!

Moral: of course, no finite part of a sequence ever gets to the limit, but we can still try to get there ourselves! (mentally).

Actually, I think the question in your title will be answered if you just decide for yourself in what sense the sequence 1, 1/2, 1/3, 1/4, ..., 1/n,... determines the exact number zero.

for more details, consult:
https://www.math.uga.edu/sites/default/files/inline-files/polynomial_tangents_without_limits.pdf

https://www.math.uga.edu/sites/default/files/inline-files/tangents_to_y.pdf
 
Last edited:
  • #40
PeroK said:
The derivative of the function ##f(x) = x^2## is ##f'(x) = 2x##. This is exact. How you justify that mathematically is another question. Standard analysis uses limits - which are themselves exact.

This can also be written ##f'(x) \equiv \frac{df}{dx} = 2x##, which leads to the concept and defining property of differentials, which are not real numbers: $$df = 2xdx$$More generally, if ##y = f(x)##, then $$dy = f'(x)dx$$
It depends on what you mean by ' exact'. Just how does this specifically counter anything Ive written here. Df is the differential, which is the change along the tangent line approximation, while ##\ Delta f:=f(x+h)-f(x)## is the actual change. The former is a limit, thus it may or may not exist. The latter is a difference of Real numbers, and will always exists. When f is a linear map, then ##\Delta f==df##. Otherwise, this is not true. The best local linear map that approximates a linear function is that linear function itself.

Just what, specifically, have I said , that is wrong or misleading?
 
  • #41
WWGD said:
It depends on what you mean by ' exact'.
There is no serious disagreement about the meaning of the word exact.

WWGD said:
Just how does this specifically counter anything Ive written here. Df is the differential, which is the change along the tangent line approximation
##df## is not the change along the tangent line approximation. That is a fiction of your own creation.
 
  • Like
Likes PeroK
  • #42
jbriggs444 said:
There is no serious disagreement about the meaning of the word exact.


##df## is not the change along the tangent line approximation. That is a fiction of your own creation.
No, it's a fiction of yours. Several sources use that, however you dislike it.
And, yes, I explained why, how I made the distinction.
 
  • #43
WWGD said:
No, it's a fiction of yours. Several sources use that, however you dislike it.
Reference, please.
 
  • #44
jbriggs444 said:
Reference, please.
Will look it up. Why don't you provide sources with an alternative definition?
 
  • #45
WWGD said:
Will look it up. Why don't you provide sources with an alternative definition?
For one, if you agree that ##df=f'(x)dx##, this agrees with the change along the tangent line at ##x=1##, with slope ##2##, giving us ##df=f'(x)dx=2dx=2(0.01)=0.02##, while the actual change, per one the prior posts, is 0.0201. The change Councidence? Since @fresh_42 may be unavailable, maybe @Mark44, your 44 colleague, can chime in.
 
  • #46
WWGD said:
For one, if you agree that ##df=f'(x)dx##
Indeed, I agree with this.
WWGD said:
this agrees with the change along the tangent line at ##x=1##, with slope ##2##, giving us ##df=f'(x)dx=2dx=2(0.01)=0.02##
0.01 is not an infinitesimal. To the extent that ##df## and ##dx## have values at all, they are infinitesimals.

There is a difference between "small" and "infinitesimal". It is not clear that 0.01 is even "small".

Let me be clear on how I understand the difference between ##\Delta f## and ##df##. It is not that the one is an actual change in the function value while the other is the change in an associated linear map. It is that the one is the actual change in the function value across a finite change in ##x## while the other is the actual change in the function value across an infinitesimal change in ##x##.

Still waiting on your references.
 
Last edited:
  • #47
jbriggs444 said:
Indeed, I agree with this.

0.01 is not an infinitesimal. To the extent that ##df## and ##dx## have values at all, they are infinitesimals.

There is a difference between "small" and "infinitesimal". It is not clear that 0.01 is even "small".

Still waiting on your references.
Will look it up. Maybe you can provide yours as well, regarding both the alternative definition and the requirement that dx be an infinitesimal. I don't remember infinitesimals being brought up when using these formulas in school. Do we also only compute Riemann sums against infinitesimals dx? Then just how is ##\int dx=x##? Or do you disagree with this too?
 
  • #48
WWGD said:
Maybe you can provide yours as well
Wikipedia has what seems to be a decent article.

WWGD said:
I don't remember infinitesimals being brought up when using these formulas in school. Do we also only compute Riemann sums against infinitesimals dx? Then just how is ##\int dx=x##?
When I went through school we stuck with the standard reals. There are no infinitesimals in the standard reals. My preferred understanding is that ##df## and ##dx## are not even values at all. They are notational placeholders. Handles on which one can hang an intuition.

But, if someone wants to treat them as having actual values, I am happy to adopt the idea of infinitesimals. It is when someone decides that they are finite that I call foul.

In integrals especially, the ##dx## is pure notation. It tells you what variable you are integrating over.

In a Riemann sum, one is taking a limit as the partition size goes to zero. That is where something akin to infinitesimals comes in -- the partition size gets small.
 
  • #49
Maybe it's time to close this thread since it turned into a discussion about notation and wording which likely won't help the OP.

I once gathered some perspectives and wordings around derivatives and I found 10 different ones, and I did not even use the word slope. We are apparently short before discussing Pfaffian forms, surreal, and hyperreal numbers.

Infinitesimals have become obsolete in mathematics by the introduction of the epsilontic as a standard in teaching calculus. Unfortunately, they are still around in the language of physics as "arbitrary small changes". The d symbol is even more ambiguous. It stands for an abbreviation of a limit - which by the way is not a process, it is a number - or a vector, or a linear operator, or a derivation, or a piece of information, or a differential form. ##\Delta x\neq dx \neq \displaystyle{\lim_{\Delta x \to 0}\Delta x =0}.## I think it cannot even be called infinitesimal. ##dx## is context sensitive. In the context of this thread, it only makes sense as a quotient
$$
\left. \dfrac{dy}{dx}\right|_{x_0} =\lim_{\Delta x \to 0} \dfrac{y(x_0+\Delta x)-y(x_0)}{\Delta x}
$$
in which case it is the slope of the function ##y(x)## at ##x_0,## a limit, and a number.
 
  • Like
Likes jbriggs444
  • #50
Differentials are often used informally in applied maths and calculus, without being formally defined. An example is an integral substitution. We start with the integral:
$$I = \int_a^b g(x) dx$$Then we let ##x = f(u)## hence ##dx = f'(u) du## and the integral becomes:
$$I = \int_{f^{-1}(a)}^{f^{-1}(b)} g(f(u))f'(u) du$$Differentials can be defined rigorously using linear maps on a manifold. But, for calculus 1, we can generally use differentials without worrying about a rigorous definition.
 
Back
Top