Reasons for Non-Differentiability

Take f(x) = |x| and try to compute its derivative at x = 0. From the right side, the difference quotient is

From the left side,

The left-hand limit is -1, the right-hand limit is +1. They disagree, so the two-sided limit does not exist. The derivative at x = 0 does not exist.

But |x| is perfectly continuous at x = 0 — no gap, no jump, no missing point. It just has a sharp corner. Continuity promises no breaks in the graph; differentiability promises the graph has a well-defined tangent line at every point. A corner breaks the second promise without breaking the first.

This article catalogues all five ways the derivative can fail to exist. Each one is a different geometric pathology, and recognising them by sight is a skill that pays off everywhere in calculus.

Way 1: Corner points

A corner (sometimes called a kink) is a point where the graph makes an abrupt change of direction. The curve is continuous — the two arms meet — but the slope jumps from one value to a different value.

The prototype is f(x) = |x| at x = 0. The left arm has slope -1, the right arm has slope +1. At the join, there is no single tangent line. Any line through the corner that isn't steeper than +1 or shallower than -1 could reasonably "touch" the curve — but a tangent is supposed to be unique. Since no single slope describes the graph at x = 0, the derivative does not exist there.

Formally, the left-hand derivative is

and the right-hand derivative is

Since f'_-(0) \neq f'_+(0), the two-sided limit does not exist, and f'(0) is undefined.

Corner points appear naturally in piecewise-defined functions. The function

is continuous at x = 1 (both pieces give the value 3), but the slope jumps from 2 to -1. That is a corner, and g'(1) does not exist.

The test for a corner is simple: compute the left-hand and right-hand derivatives. If both exist but are different finite numbers, the point is a corner.

Way 2: Cusps

A cusp is like a corner, but more extreme. Instead of the slopes jumping between two finite values, the slopes blow up to +\infty on one side and -\infty on the other side (or vice versa). The graph comes to a sharp point, but the two arms are nearly vertical at the tip.

The prototype is f(x) = x^{2/3} at x = 0.

Compute the derivative from the definition at x = 0:

As h \to 0^+, this is \frac{1}{h^{1/3}} \to +\infty. As h \to 0^-, this is \frac{1}{h^{1/3}} \to -\infty. The two one-sided limits are infinite and have opposite signs.

The difference between a corner and a cusp: at a corner, both one-sided derivatives are finite but unequal. At a cusp, at least one of them is infinite, and typically they have opposite signs. Geometrically, a corner looks like two straight roads meeting at an angle; a cusp looks like two walls closing to a point.

Another cusp you will encounter is f(x) = |x|^{1/2} at x = 0. The difference quotient is |h|^{1/2}/h. For h > 0, this is 1/\sqrt{h} \to +\infty. For h < 0, writing h = -k with k > 0 gives \sqrt{k}/(-k) = -1/\sqrt{k} \to -\infty. Opposite infinities again — a cusp. The general pattern: |x|^p at x = 0 has a cusp when 0 < p < 1, a corner when p = 1 (that is just |x|), and is differentiable when p > 1.

Way 3: Vertical tangents

A vertical tangent is the case where the tangent line exists and is unique, but it is vertical — its slope is undefined (or, if you prefer, infinite).

The prototype is f(x) = x^{1/3} (the cube root function) at x = 0.

As h \to 0 from either side, h^{2/3} is always positive (since the exponent 2/3 gives a positive result for both positive and negative h). So \frac{1}{h^{2/3}} \to +\infty from both sides.

Notice the key difference from a cusp: at a vertical tangent, both one-sided limits of the difference quotient are +\infty (or both are -\infty). They agree on the direction — the tangent line is genuinely there, unique and well-defined — but its slope is not a finite number. Since the derivative is defined as a real number, a vertical tangent means the derivative does not exist, even though the tangent line does.

This is a subtler failure than a corner or a cusp. The function is not doing anything wrong geometrically — the graph is perfectly smooth. It is just too steep for the derivative to capture.

Way 4: Jump discontinuities

All three failures above happened at points where the function was continuous. This one does not even give you that.

A jump discontinuity is a point where the function has a sudden gap — it jumps from one value to a different value. The left-hand limit and the right-hand limit both exist, but they are not equal.

Take the function

At x = 2, the function jumps from 1 to 3. The difference quotient from the left is

As h \to 0^-, this blows up: \frac{-2}{h} \to +\infty. The quotient does not approach any finite number.

The fundamental reason is straightforward: differentiability requires continuity. If a function is not continuous at a point, the derivative cannot exist there. The proof is a one-line consequence of the limit definition: if f'(a) exists, then

which means \lim_{h \to 0} f(a+h) = f(a), i.e., f is continuous at a. Contrapositive: if f is not continuous at a, then f'(a) does not exist.

So jump discontinuities automatically kill differentiability — and so do removable discontinuities, infinite discontinuities, and every other flavour of discontinuity. Continuity is the minimum entry requirement for differentiability.

Way 5: Oscillatory behaviour

This is the strangest case. There are functions that are continuous at a point, have no corner, no cusp, no vertical tangent — and yet the derivative still does not exist because the function oscillates so wildly near the point that the difference quotient never settles down.

The classic example is

This function is continuous at x = 0 (since |x \sin(1/x)| \leq |x| \to 0). But the difference quotient at x = 0 is

As h \to 0, the argument 1/h blows up, so \sin(1/h) oscillates between -1 and +1 infinitely often without ever settling down. The limit does not exist.

Contrast this with g(x) = x^2 \sin(1/x), which is differentiable at x = 0. The difference quotient becomes h\sin(1/h), and since |h\sin(1/h)| \leq |h| \to 0, the limit is 0. The extra factor of x tames the oscillation just enough for the derivative to exist. The difference between x\sin(1/x) and x^2\sin(1/x) is the difference between oscillation that is too wild for a derivative and oscillation that is just barely controlled.

Here is a concrete way to see the difference. For f(x) = x\sin(1/x), the secant slopes from the origin are \sin(1/h) — they oscillate between -1 and +1 forever, never settling. For g(x) = x^2\sin(1/x), the secant slopes from the origin are h\sin(1/h) — they still oscillate, but the oscillations are being squeezed to zero by the factor h. The squeeze theorem finishes the job: -|h| \leq h\sin(1/h) \leq |h|, and both bounds go to 0, so the limit is 0.

This is the only failure mode in this article that you cannot diagnose by looking at the graph. The graph of x\sin(1/x) near the origin looks like a mess of oscillations at any zoom level — there is no corner, no cusp, no vertical tangent, and yet the derivative fails. The failure is not a geometric feature you can point to; it is the absence of convergence in the limit.

Analysing piecewise functions

Many exam problems hand you a piecewise-defined function and ask where it is differentiable. Here is the systematic approach.

Step 1. Check continuity at every join point. If the function is not continuous, stop — the derivative does not exist.

Step 2. If continuous, compute the left-hand derivative f'_-(a) and the right-hand derivative f'_+(a) at the join.

Step 3. If f'_-(a) = f'_+(a) and both are finite, the function is differentiable at a, and f'(a) is that common value. Otherwise, it is not.

Common confusions

• "If a function is continuous, it must be differentiable." This is the single most common misconception in introductory calculus. It is false. Continuity is necessary for differentiability, but it is not sufficient. Every example in this article except the jump discontinuity is a continuous function that fails to be differentiable. The implication goes only one way: differentiable \Rightarrow continuous, never the reverse.

• "A cusp and a corner are the same thing." They are not. At a corner, both one-sided derivatives are finite but unequal (f'_- \neq f'_+, both in \mathbb{R}). At a cusp, at least one one-sided derivative is infinite, and typically they have opposite signs. The geometry is different too: a corner looks like a bent wire; a cusp looks like a knife edge.

• "A vertical tangent is the same as no tangent." No — at a vertical tangent, there is a unique tangent line. It just happens to be vertical, meaning its slope is not a finite number. At a corner, there is genuinely no unique tangent. These are different kinds of failure.

• "If the formula for f'(x) is undefined at x = a, the function is not differentiable there." Be careful. The formula for the derivative might have a removable issue at a. Always go back to the limit definition or the one-sided derivative test. The formula is a shortcut; the limit is the authority.

• "Piecewise functions are always non-differentiable at the join." False — Example 1 above is a piecewise function that is differentiable at the join. A piecewise function is differentiable at the break point whenever both pieces arrive with the same value and the same slope.

Going deeper

A function that is continuous everywhere and differentiable nowhere

All the examples above fail to be differentiable at one or a few isolated points. There exist functions that are continuous everywhere but differentiable nowhere — not at a single point on the entire real line. The first known example was constructed by Weierstrass in the 19th century:

where 0 < a < 1, b is an odd positive integer, and ab > 1 + \frac{3\pi}{2}. Each term a^n \cos(b^n \pi x) is a smooth cosine wave, but as n grows, the frequency b^n increases far faster than the amplitude a^n decreases. The result is a function whose graph is a continuous curve that is so jagged at every scale that no tangent line exists at any point — like a coastline that never smooths out, no matter how close you zoom in.

This is not a curiosity. It shows that differentiability is a much stronger condition than continuity. Continuous functions can be arbitrarily wild. The derivative is a filter that lets through only the well-behaved ones.

Differentiability as a hierarchy

The five failures in this article can be arranged by how badly the derivative fails:

1. Jump discontinuity — the function itself is broken. The derivative never had a chance.
2. Corner — the function is continuous but changes direction abruptly. Both one-sided derivatives exist (finite) but disagree.
3. Cusp — the function is continuous but comes to a sharp point. One or both one-sided derivatives blow up with opposite signs.
4. Vertical tangent — the function is continuous and smooth, but too steep. Both one-sided derivatives agree but are infinite.
5. Oscillation — the function is continuous and bounded, but oscillates too wildly for any limit to form.

Failures 1 through 4 can each be diagnosed from a graph by eye. Failure 5 cannot — the oscillation is invisible at any finite zoom level. That is what makes it the most subtle.

A systematic decision tree

When you need to check differentiability at a point x = a, follow this decision tree:

1. Is f defined at a? If not, the derivative cannot exist.
2. Is f continuous at a? Compute \lim_{x \to a} f(x) and check if it equals f(a). If f is not continuous, the derivative does not exist. Stop here — you have a discontinuity (jump, removable, or infinite).
3. Do the one-sided derivatives f'_-(a) and f'_+(a) both exist? Compute \lim_{h \to 0^-} \frac{f(a+h) - f(a)}{h} and \lim_{h \to 0^+} \frac{f(a+h) - f(a)}{h}.
4. Are both finite? If one or both are \pm\infty:
   • If they have opposite signs (+\infty and -\infty), you have a cusp.
   • If they have the same sign (both +\infty, or both -\infty), you have a vertical tangent.
5. Are both finite and equal? If yes, f is differentiable at a and f'(a) is that common value. If they are finite but unequal, you have a corner.
6. Does neither one-sided limit exist? (Neither settles on a finite value nor heads to \pm\infty.) Then you have oscillatory non-differentiability.

This decision tree covers every case. The vast majority of exam problems land at step 2 or step 5. The oscillatory case (step 6) appears occasionally in competitive exams, but it is rare in standard board-level work.

Where this leads next

You now know every way a derivative can fail to exist. The next step is to learn how to compute derivatives efficiently when they do exist — which, for the functions you will meet in practice, is almost always.

• Derivatives of Basic Functions — the power rule, the exponential, and the logarithm, all derived from first principles.
• Rules of Differentiation — the sum, product, and quotient rules that let you differentiate any combination of basic functions.
• Differentiability — the formal definition, the relationship with continuity proved rigorously, and what left and right derivatives really mean.
• Continuity — the prerequisite: what it means for a function to have no jumps, and why that alone is not enough.