The Weierstrass Function: Continuous Everywhere, Differentiable Nowhere

Imagine a coastline. Zoom in, and the coastline doesn't get smoother — it gets jaggier, revealing more inlets and headlands at every scale, seemingly forever. Now imagine a mathematical curve that does something even stranger: it is perfectly continuous, with no jumps or breaks anywhere, and yet at every single point, it is infinitely jagged. You could never draw a tangent line to it, not even for an instant, because it has no slope anywhere at all. For over a century, mathematicians didn't believe such a thing could exist. Then, in 1872, Karl Weierstrass built one — and broke calculus in the process.

The Concept

To understand why this function was such a shock, you have to understand what 19th-century mathematicians thought they knew about continuous functions. A function is "continuous" if you can draw its graph without lifting your pen off the page — no sudden jumps, no holes. A function is "differentiable" at a point if it has a well-defined slope there, meaning if you zoom in close enough, the curve starts to look like a straight line.

For decades, most mathematicians — including giants like André-Marie Ampère — believed, and tried to prove, that any continuous function had to be differentiable almost everywhere, or at least at most points. It seemed obvious: if a curve never breaks, surely it has to smooth out somewhere when you zoom in close enough.

Karl Weierstrass, a German mathematician working at the University of Berlin, proved that intuition catastrophically wrong. He constructed a function — now called the Weierstrass function — that is continuous at every point but differentiable at no point. Not "differentiable almost nowhere." Not "differentiable except at a few weird spots." Nowhere. Anywhere you zoom in, no matter how far, the curve keeps jittering up and down with no settled direction, forever.

The trick to building it is deceptively simple: add together infinitely many wave functions (cosine curves), each one higher-frequency and lower-amplitude than the last. Formally:

W(x) = Σ aⁿ cos(bⁿ π x), summed over n = 0 to infinity, where 0 < a < 1, b is an odd integer, and the product ab is greater than 1 + 3π/2 (roughly 5.71).

Each term in the sum is a smooth, gentle wave on its own. But because each successive wave oscillates faster (governed by b) while shrinking in height only a little (governed by a), the higher-frequency wiggles never get smoothed away — they just keep piling detail on top of detail, at every scale, forever. Zoom into any piece of the curve and you find a smaller, roughened copy of the same chaotic wiggling. It's a fractal, though that word wouldn't be coined for another century.

Why It Matters

Weierstrass first presented his monster to the Royal Prussian Academy of Sciences in Berlin on July 18, 1872, though he had reportedly been discussing versions of the idea in his lectures as early as 1861. It wasn't formally published until 1875, when fellow mathematician Paul du Bois-Reymond wrote it up and credited Weierstrass — deliberately putting the "monster" in front of the entire mathematical community.

The reaction was not polite curiosity. It was closer to outrage. Henri Poincaré, one of the most celebrated mathematicians of the era, called functions like Weierstrass's "an outrage against common sense." Charles Hermite, another towering figure of French mathematics, wrote to his colleague Thomas Stieltjes in 1893 that he turned "with fright and horror from this lamentable evil of functions which do not have derivatives." These weren't cranks — they were among the best mathematicians alive, and they found the object genuinely offensive to their sense of how mathematics should behave.

Why the visceral reaction? Because calculus, the entire toolkit of derivatives and rates of change that had powered Newtonian physics and engineering for two centuries, quietly assumed that "reasonable" functions were differentiable almost everywhere. Weierstrass's function showed that continuity — the seemingly weaker, more permissive property — didn't guarantee differentiability at all. The two ideas, which mathematicians had treated as close cousins, turned out to be almost unrelated. It forced the field to rebuild its foundations on much more careful, rigorous footing — a project Weierstrass himself helped lead, which is part of why he's remembered as "the father of modern analysis."

Interestingly, Weierstrass wasn't even first — he was just first to publish and force the issue. The Czech mathematician and priest Bernard Bolzano had sketched a similar pathological function around 1830, more than 40 years earlier. But Bolzano's manuscript sat unread among his papers and wasn't discovered until 1921 (published in 1922), long after Weierstrass's version had already reshaped the field. Being right first, in an unread notebook, changes nothing; being right loudly, in front of the Berlin Academy, changes everything.

The Details

Picture building the curve one layer at a time. Start with a single, gentle cosine wave — a smooth rolling hill. Now add a second cosine wave that oscillates several times faster but is shorter — it rides on top of the first, adding small bumps to the big hill. Add a third wave, faster and shorter still, riding on top of the bumps. Keep going, forever.

If the higher-frequency waves shrank fast enough, this process would converge to something smooth — the bumps would eventually become invisible, like sanding a plank of wood smoother and smoother. That's what everyone expected. But Weierstrass tuned the two parameters, a and b, so that the frequency of each new wave (controlled by b) grows faster than its height shrinks (controlled by a) can compensate for. The condition ab > 1 + 3π/2 is precisely the threshold at which the roughness never gets sanded away — every zoom level reveals fresh, undiminished jaggedness.

The resulting graph has a genuinely fractal character: a small piece of the curve, magnified, looks statistically like the whole curve. This is exactly the self-similarity that Benoit Mandelbrot would later put at the center of fractal geometry in the 1970s and 80s, and Mandelbrot explicitly built on Weierstrass's construction — the "Weierstrass–Mandelbrot function" is a generalized, randomized version used to model rough, self-similar processes in the real world.

That's where this "useless monster" turns out to be surprisingly practical. Randomized cousins of the Weierstrass function are used to model financial market volatility — researchers have used generalized Weierstrass-Mandelbrot functions to reproduce the fractal dimension and long-memory statistical behavior (measured by something called the Hurst exponent) seen in real stock market indices, since prices jitter unpredictably at every timescale from seconds to years, much like the function itself. Similar constructions show up in modeling atmospheric turbulence, where physical quantities like temperature and velocity fluctuate roughly across a huge range of scales in a way that smooth, classical functions can't capture. And the function is mathematically related to fractional Brownian motion, the family of random processes used to model everything from stock prices to the erratic path of a pollen grain jostled by water molecules.

More broadly, Weierstrass's function is the ancestor of an entire family of "pathological" objects — the Koch snowflake, the Cantor set, Mandelbrot's coastline paradox — that once seemed like mathematical curiosities but turned out to describe real, rough phenomena that smooth Euclidean geometry never could: coastlines, mountain ranges, blood vessel networks, and the branching patterns of lungs and trees. The fractal-geometry tradition his function kicked off eventually fed into practical engineering too, including the compact, self-similar antenna designs that let modern cell phones handle multiple frequency bands in a tiny footprint.

Takeaways

Karl Weierstrass presented a function in 1872 that is continuous everywhere but differentiable nowhere — a curve with no breaks that nonetheless has no well-defined slope at any point, ever.
It's built by stacking infinitely many cosine waves of increasing frequency and decreasing (but not fast-enough-decreasing) amplitude, so the jaggedness never smooths out no matter how far you zoom in.
The mathematical establishment initially hated it — Poincaré called it "an outrage against common sense," and Hermite recoiled "with fright and horror" — because it broke the assumed link between continuity and differentiability.
The Czech mathematician Bernard Bolzano had devised something similar decades earlier, around 1830, but it lay undiscovered in his papers until 1921 — a reminder that priority in math depends on publication, not just discovery.
Far from a useless curiosity, randomized versions of Weierstrass's function (the Weierstrass–Mandelbrot function) now help model rough, self-similar real-world processes: financial market volatility, atmospheric turbulence, and fractional Brownian motion.

Resources: - The Jagged, Monstrous Function That Broke Calculus — Quanta Magazine - Weierstrass function — Wikipedia - Math's Beautiful Monsters — Nautilus