MathIsland
← All MathIsland labs
f′(a)미분가능성 Lab

A curve is bent everywhere, so how can 'the slope at a single point' even make sense?

Differentiable means: when you zoom in, it becomes indistinguishable from a straight line. The slope of that surviving line is f′(a).

Experiment

Hands-on experiment

Curves are bent. Yet we casually speak of 'the slope at a point'. How can something bent have a single slope? The secret is magnification. Put the point (1, 1) of f(x)=x² under the microscope.

🔮 Predict first — magnify the neighborhood of (1, 1) a hundred times. What will the curve on screen look like?

📖 Read more — why it exists · insights · common mistakes · formulasExpand ▾

Why

Why does this exist?

The derivative stands on the phrase 'the slope at a single point'. But a slope originally needs two points to measure. For a slope to make sense with only one point left, the curve needs a special qualification.

That qualification is differentiability: being indistinguishable from a straight line when zoomed in. Only at points holding this qualification are the tangent, the velocity, and the rate of change defined. Skip the eligibility check and differentiate anyway, and the computation collapses at places like the corner of |x|.

Until the 19th century people believed 'if it's continuous, it's smooth almost everywhere'. In 1872 Weierstrass's monster demolished that belief, and mathematicians rebuilt differentiability as an independent concept. The order in which modern analysis stacks the derivative on top of a rigorous limit is the legacy of that event.

Insight

Insights from the video

Differentiable means indistinguishable from a straight line when zoomed in.

The tangent, the instantaneous rate of change, and the derivative are all different faces of this one sentence. If a line survives at the end of the zoom, its slope is f′(a); if none survives, there is no derivative at that point.

Continuity is connectedness; differentiability is smoothness.

The two qualifications differ. A break can't even be straightened, so differentiable always implies continuous. But being connected doesn't make it smooth. |x| is the witness, and the Weierstrass function is the extreme witness.

Misconception

Common misconceptions

If it's continuous, it's differentiable.

|x| is the counterexample. It never breaks anywhere, yet at 0 its left and right slopes split into −1 and +1. The inclusion runs the other way. Differentiable always implies continuous, but continuous does not guarantee smoothness.

If the graph looks smooth, it's differentiable.

The eye is just an impression at one zoom level. Even the Weierstrass function looks soft if you draw only a term or two. The verdict is always rendered by zooming and taking a limit. Does a line survive at every zoom? That is the only criterion.

Formula

Writing it as math

Putting what you felt in the zoom game into the language of math looks like this.

🔬 Anatomy of the definition — paired with what you touched on the zoom screen

= () /

Definition of differentiability

When this limit exists — and only then — f is differentiable at a. Whether h comes from the left or the right, it must arrive at the same value.

Why |x| is not differentiable at 0

These are the two lines you saw in stage 2. The left secant is −1 and the right secant is +1; they never meet. No limit, so no f′(0).

The zoom game in formula form (local linear approximation)

o(h) means an error that dies faster than h. It is exactly the deviation w² that shrank to a quarter each time you doubled the zoom in stage 1. It's 'zoom in and it's a line' written in symbols.

Direction of inclusion

To be smooth it must first be connected. But the converse fails. |x| is the shortest witness that the converse is false.

In Real Life

Where you meet it in real life

Linearization in engineering

Physical models like airflow around a wing or a circuit's response are mostly differentiable. So engineers swap the curve for its tangent over a narrow range to design. A guarantee of differentiability is a guarantee that you 'can compute'.

ReLU in deep learning

ReLU, the most-used activation in AI, is max(0, x) — essentially a cousin of |x|. It's not differentiable at 0, but that single point is handled by convention and used anyway. You need differentiability to read design decisions like this.

The fractal nature of stock charts

Stock charts don't get smooth when you zoom in. Beneath the daily candles there are minute wrinkles, and beneath those, second-by-second wrinkles again. That's why you should be wary of models that trust 'the slope at this instant'.

The coastline paradox

The more you increase a map's scale, the more the coastline wiggles and the longer the measured length grows. Nature's boundaries are closer to the Weierstrass monster than to a smooth curve.

Practice

Practice — conquer the types

Conquered 0 / 4
1

For f(x)=x² at the point (1, 1), when the visible interval has half-width w the right secant slope was 2+w. Pushing the zoom to the end (w→0), what value does the secant slope arrive at — that is, what is f′(1)?

2

Is f(x)=|x| differentiable at x=0?

3

Which is the correct inclusion relation between continuity and differentiability?

4

Can a function exist that has no breaks at all (continuous), yet never straightens no matter where or how much you zoom in?

Watch

Related video

Does every curve become a straight line when you zoom in? — DifferentiabilityThe video link is coming soonBrowse the YouTube channel →

Connection

Concepts connect

Previous concept

ε-δ

The ε-δ Definition

The lim inside this definition is the ε-δ game itself — the derivative only stands on a rigorous limit.

← The ε-δ Definition lab

Leads to next

F′=f

The Fundamental Theorem of Calculus

In a world where smoothness is guaranteed, differentiation and integration meet as a single theorem.

Go to the The Fundamental Theorem of Calculus lab →

Related

Labs worth exploring together

Related lab

lim

Limits

What you meet at the end of the zoom is precisely the limit.

Go to the Limits lab →

Related lab

max

Optimization

To hunt for a slope of 0, a slope must first exist.

Go to the Optimization lab →