How Fractals Moved From Math Curiosity to Tech

There is something quietly remarkable about the trajectory of fractal geometry. Shapes that mathematicians once called "monsters"—the Koch snowflake, the Sierpiński gasket, objects so jagged and self-similar that they offended the era's geometric sensibilities—spent most of their existence as pure curiosities. Beautiful, perhaps. Useful? Nobody thought so.

That changed, and it changed faster than most scientific pivots do. Within a few decades of Benoit Mandelbrot naming these structures and demonstrating their presence throughout the natural world, engineers were using their properties to cool microchips, compress video game worlds, and cram multiple antennas into a drone the size of your palm. The story SciShow's Madelyn Leembruggen lays out in a recent video is not really a story about fractals being discovered—it's a story about a perceptual shift, about an entire field of applied mathematics suddenly becoming legible to engineers.

The question worth sitting with is why it took so long, and what that delay tells us about how pure mathematics makes the leap to application.

The Mandelbrot Pivot

Mandelbrot's work at IBM in the 1980s is the obvious inflection point. Using computers powerful enough to iterate mathematical functions thousands of times—plugging each output back in as the next input, over and over—he produced the shape that now bears his name. What he noticed, and what matters enormously for every application that followed, is that fractals possess a specific kind of complexity: self-similarity across scales. Zoom in on any edge of the Mandelbrot set and you encounter the same intricate density you started with. No smoothing. No simplification.

The same characteristic, Mandelbrot argued, appears throughout nature—in coastlines, cloud edges, blood vessel networks, the branching of trees. This wasn't just aesthetically interesting. It was a hint that whatever problem fractals solved mathematically, nature had already solved physically. Engineers, it turned out, just had to look.

Packing More Antenna Into Less Space

The first application Leembruggen covers is fractal antennas, and it's worth dwelling on why the geometry helps rather than just accepting that it does.

A conventional antenna is tuned to resonate at a specific frequency—essentially, the physical length of the antenna determines what wavelengths it picks up. If you want to handle multiple frequencies, the blunt engineering solution is multiple antennas. That works fine in a television set. It becomes a problem in a device you're trying to make light enough to fly.

Fractal shapes like the Hilbert curve solve this in two ways simultaneously. First, they are space-filling curves: they pack an enormous amount of length into a compact area without random crumpling, which matters because random crumpling destroys the resonance properties that make an antenna functional. Second, because the same pattern repeats at multiple scales, different sections of the antenna can resonate with different wavelengths. The physics of resonance—structures vibrating in sync—means that scale-similar patterns produce multi-frequency sensitivity.

A 2022 paper describes an antenna a few centimeters wide that incorporates both a Sierpiński gasket and a Hilbert curve and successfully handles three distinct 5G frequency bands without cross-interference. That's a meaningful result. It's also worth noting it's a single paper; multi-frequency fractal antenna research has been an active area for roughly two decades, with variable results depending on the specific application and environment. The underlying physics is solid; the engineering details are still being refined.

The Cooling Problem Nobody Talks About

The heat exchanger application is, to my mind, the most underappreciated of the four. It also has the clearest biological provenance, which makes the intellectual history satisfying.

The problem is straightforward: as devices shrink, conventional liquid cooling—pipes carrying fluid to absorb and remove heat—runs into the physics of viscosity. Fluid moving through tiny tubes requires progressively more pressure the further it travels. At a certain scale, the pressure requirements become impractical and the cooling becomes inefficient.

Fractal branching structures—resembling the vascular architecture of leaves or circulatory systems—address this through what fluid dynamicists call pressure recovery. At each branching point, the geometry creates a smooth taper and increased surface area that actually reduces the pressure needed to push fluid through subsequent branches. As Leembruggen puts it: "at every one of those branches, all the stuff flowing around in the tubes gets mixed up," which brings hotter fluid from the center of a tube toward the walls where heat dissipation occurs.

The reported efficiency gain—up to 25% in fractal-designed heat exchangers compared to conventional designs—comes from published research, though the figure varies depending on the application and fluid dynamics specifics. In microscale devices where thermal management is a genuine failure mode, a 25% improvement is not incremental. It is the difference between a device that works and one that doesn't.

The circulatory system connection is not incidental. Researchers noticed that biological vascular networks had already optimized this problem through evolution. The fractal architecture of your circulatory system is, among other things, a pressure-management solution. Engineers borrowed it directly.

Reading Health Risk in Retinal Geometry

The biomedical application requires understanding fractal dimension—specifically the box-counting method—which is genuinely strange conceptually but worth getting right.

For ordinary shapes, the number of grid boxes needed to cover a boundary scales cleanly with the shape's geometric dimension: lines scale as 1, areas as 2, volumes as 3. Fractals break this. A Sierpiński gasket has a box-counting dimension of approximately 1.585—between a line and a plane, precisely because its complexity at every scale prevents it from resolving cleanly into either. Coastlines typically measure between 1.1 and 1.3.

This gives researchers a single number that captures the structural complexity of any fractal-like pattern. And retinal blood vessel networks, it turns out, are fractal-like enough that their dimension carries clinical signal.

A 2009 study in Diabetes Care photographed the retinas of over 700 young people with type 1 diabetes and calculated each participant's retinal fractal dimension. The finding: every 0.01 increase in fractal dimension corresponded to a 40% increase in risk of developing diabetic retinopathy—a progressive condition that, untreated, leads to blindness. A 2022 study published in Frontiers of Digital Health extended the method to more than 6,000 patients with type 2 diabetes and found associations between retinal fractal dimensions and dementia risk, with arterial and venous measurements pointing in different directions.

These are association studies. They are not trivial—the sample sizes are reasonable, the methodology is consistent, and the findings have been replicated across different patient populations—but the clinical pipeline from "retinal fractal dimension correlates with disease risk" to "your optometrist now calculates this routinely" involves validation steps, sensitivity and specificity benchmarks, and cost-benefit analyses that haven't all been completed. The research is genuinely promising. It is not yet a diagnostic standard.

The Elegant Compression of Minecraft

The video game application is the one most likely to produce the reaction: "wait, that's what's happening?" Minecraft's worlds are not stored. They are generated.

The technique—Perlin noise, a specific method for producing coherent pseudo-randomness that creates smooth variation at multiple scales—works by introducing small random perturbations at each iteration of the underlying mathematical function. The result is terrain that looks organic, because biological terrain is fractal-like, which means the algorithm is in some sense modeling the same physics that produces real mountains and caves.

The computational efficiency is the point. "You don't need a bunch of complex structures stored in computer memory waiting for a player to stumble across them," as Leembruggen explains. Detail is generated procedurally as the player approaches, using more iterations—more recursive steps—for nearby terrain than distant. The same player's computer is doing less work for distant objects and more for close ones, adaptively, which is a remarkably elegant solution to what would otherwise be a storage and processing problem of absurd proportions.

This is also the application that most directly descends from what Mandelbrot was doing at IBM in the 1980s: using iterative functions to generate complex structure from simple rules. The intellectual lineage is direct.

What connects these four applications—antennas, heat exchangers, retinal diagnostics, procedural generation—is not that fractals are a universal tool. They are not. What connects them is that each domain had a problem that conventional Euclidean geometry was too simple to address: too many frequencies for one antenna, too much pressure for one straight pipe, too much complexity in a blood vessel network to reduce to a single measurement, too much terrain for any storage system to hold. Fractal geometry provided a mathematical vocabulary for complexity that repeats itself, and engineers have been spending forty years discovering what that vocabulary can describe.

The question the SciShow video doesn't have space to address—and which seems worth leaving open—is how many more of those discoveries are still waiting. The video gestures at CGI, dismisses it for time, and moves on. But computer-generated imagery, materials science, acoustic engineering, and antenna design for satellite communication all have active fractal research threads. The monsters have turned out to be remarkably well-behaved.

By Priya Sharma, Science & Health Correspondent