Fractal Mechanics discovers a revolutionary nanostructure

Strong as steel, light as foam — five times stronger than titanium at the same density. This is the result achieved by Peter Serles and his team from the University of Toronto and Caltech last year.

How did they achieve such a feat? In a completely intuitive and empirical way, with the help of Machine Learning.

Fractal Mechanics has made it possible to get out of this “black box” mode by offering a rigorous mathematical framework, paving the way for a new era for nanomaterials.

The problem of nanonetworks

Traditionally, improving a material’s strength involves increasing its density, and therefore its mass. But nanonetworks change the game, because they are composed primarily of… empty space.

A nanonetwork is a microscopic architecture composed of uprights and nodes, like a tiny carbon scaffold. The challenge: how to connect thousands of these uprights without creating weak points ?

Traditional truss designs use geometric shapes: cubes, octahedrons, hexagons. They are elegant. But they share a major flaw: wherever two struts meet, the acute angle creates what engineers call a stress concentration — a point where forces multiply locally and break the structure.

One of the first nanostructures with fractal characteristics, Julia Greer (2015)

Many forms have been attempted since the birth of this field (around 2010), including fractal structures, as explained by Julia Greer , a professor at Caltech and a pioneer in the development of nanometric structures:

“These structures can contain nearly 99% air while being as strong as steel. Their fractal design allows us to integrate a hierarchical structure into the materials architecture, which promises other advantageous properties.” — Prof. Julia Greer , Caltech (2014)

But the promised revolution is taking its time. Why? Because without an explanatory theoretical framework, we don’t really know what we’re looking for among an infinite number of possibilities. That’s why Serles’ team had the idea of ​​using machine learning to test thousands of candidate forms.

The team therefore let the algorithm perform its own simulations using brute force, starting from a given topology (see image below). What they sought to optimize was only the shape of the cross-section of each strut (from among 400 possibilities), but not the basic structure.

What the algorithm discovered: wider spacers at their junctions and narrower ones at the edges. No sharp angles. Smooth, curved profiles. The result: the material’s specific compressive strength doubled.

“By applying the Bayesian optimization algorithm, we can efficiently reshape the nanolattice’s struts to double their overall strength without adding any material — a remarkable result I couldn’t believe at first!” — Peter Serles

The nanonetwork created by Peter Serles’ team. In the lower right corner, the larger diameter of the junctions towards the center is noticeable.

What their AI ​​found — and what it missed

The AI ​​operated within a parameterization limited to a few Bézier control points — a design space perceived as efficient, but with the drawback of limiting the possibilities to explore. Thus, as the team explains (who have no way of knowing if optimizations are possible, or to what extent):

“We will continue to explore new designs that allow us to further reduce material density while maintaining high strength and rigidity.”

Without an explanatory theoretical framework, it’s impossible to move forward except by trying all the possibilities one by one, and without knowing whether we’re approaching the optimal result or still far from it. In this case, the AI ​​tested structures with regular patterns, with an optimal junction diameter ratio of approximately 1.44.

That’s where Fractal Mechanics comes in.

Recently, our theoretical model has solved the turbulence regularity problem (Navier-Stokes , one of the 7 Millennium Prize Problems). In fractal mechanics, the Fibonacci cascade (based on the Golden Ratio φ) guarantees that energy, or stress, cannot catastrophically concentrate at a single scale.

Thus, by applying the Fibonacci Cascade Theorem — the same one that guarantees the regularity of the Navier-Stokes equations — to structural mechanics, we have shown that the stress limit imposes a precise profile: A(x) = A_midpoint × φ^(2|x/L − 0.5|). This profile predicts exactly D_junction/D_midpoint = φ = 1.618 (vs. 1.44 for Serles). Not a coincidence; a mathematical consequence.

But that was only the first step.

The problem of individual cells

It’s not the profile of the cross braces that’s most revealing. It’s what happens when you build a larger structure.

Imagine a single CFCC cell subjected to a compressive load. The struts (the edges of the cubes) are extremely thin. Under these conditions, the dominant failure mode is “buckling”: a strut subjected to compression, beyond a critical load, suddenly bends sideways and collapses.

Strength tests on a triangular structure (Professor Julia Greer’s team)

A lattice with a single scale (regular pattern), like the one proposed by the Caltech team, has a critical buckling wavelength, a kind of resonance frequency that gives it a certain fragility, like a crystal glass. Any load that resonates with this wavelength can cause all the cells to buckle simultaneously — what engineers call a “catastrophic” failure: the structure fractures suddenly and completely, rather than gradually.

The revolutionary idea

Fractal Mechanics introduces a new element: a fractal hierarchy of cells, each scaled by φ.

• Level 0 (blue squares): 20 μm cells, 300 nm spacer diameter (Caltech cell scale)
• Level 1 (red squares): 32.4 μm cells, 485 nm spacer diameter — exactly φ× larger.
• Level 2 (yellow square): 52.4 μm cells, 785 nm spacer diameter — larger φ².

No new manufacturing technology is required — the same two-photon polymerization process used by Caltech applies at all levels. Note that the scale is different (the Caltech cell is the same size that the blue FM cells).

This is not simply any fractal structure, like those attempted by Julia Greer: the key lies in the unique self-similarity properties of the φ cascade.

Here’s what this means: each level resists buckling independently at its own wavelength. If the smallest cells reach their critical load, the structure above them has not yet reached its own. Thus, failure is sequential, not simultaneous.

Like electrical fuses in a circuit: the weakest fuse blows first, protecting the others. The circuit degrades gradually instead of collapsing instantly.

This phenomenon is directly analogous to the Fibonacci cascade in turbulent fluids: energy cannot pass directly from level 0 to level 2 — it can only flow towards adjacent scales (which we proved in Theorem 1 of our Navier-Stokes article), thanks to the unique properties of φ. This same principle prevents stress from propagating catastrophically through a material.

Turbulence has a fractal structure, with clearly distinct scales (artist’s impression)

This innovative principle gives our FM φ structure a remarkable capability: at low impact speeds, only the smallest scale (20 μm) deforms. At higher speeds, the energy penetrates to the next level (32 μm), and so on. Thus, the total cascading absorption capacity of our nanomaterial increases with speed — exactly the opposite of a conventional material that saturates. FM naturally predicts this behavior; a single-scale structure like the one from Caltech does not benefit from it.

Result: at the same material density, our solution offers a specific strength 18% higher. And by allowing a higher density, mechanical strength performance is increased tenfold.

Technological Applications

The applications range from the very small to the very large in architectural terms.

Currently: the Caltech team has produced a 6.3 × 6.3 × 3.8 mm³ sample in less than two days. This represents approximately 18 million network cells in a space smaller than a fingertip. An extraordinary achievement, but on a very small scale.

Near future (aerospace): This is precisely where space structures are most important, as weight is paramount. A satellite panel using FM φ fractal nanostructures offering the same structural performance could be five times lighter. NASA and ESA are discreetly monitoring this field.

In the medium term (protective structures): The unique absorption capabilities of our nanostructure have obvious applications in ballistic protection, collision structures, nuclear containment…

In the longer term (civil infrastructure): A bridge? The strength-to-weight ratio isn’t the limiting factor for most bridges — it’s manufacturing costs and corrosion. But for very long spans where dead weight dominates (the next generation of suspension bridges, or structures in zero-gravity environments), FM nanostructures could be a game-changer.

And there is a particular advantage for bridges: the FM hierarchy naturally resists multiple resonant frequencies simultaneously. The catastrophic collapse of the Tacoma Narrows Bridge in 1940 was caused by wind resonating at a specific frequency. A fractal hierarchy φ would resist resonance across an entire frequency spectrum — for exactly the same reason it works for turbulence.

A deeper connection

The Caltech team writes in their article: “We will continue to explore new designs that reduce material density while maintaining high rigidity and strength.”

We think we know what these designs will look like.

The same mathematics that explain why turbulent fluids do not explode in a finite time — the Fibonacci cascade, spacing according to the golden ratio, local-scale energy transfer — explain why fractal φ nanostructures do not suffer catastrophic failure under stress.

AI empirically explores what the mathematics of Fractal Mechanics describes analytically.

φ is not a coincidence. It is a convergence.

Papers:

• 📄 Phi-Fractal Hierarchical Nanolattice — From the Navier-Stokes Fibonacci Cascade to Structural Mechanics: Zenodo (juin 2026)
• 📄 Fractal Mechanics v2: Zenodo (mai 2026)
• 📄 Global Regularity of Navier-Stokes via Fibonacci Cascade Decomposition: A Physically-Constrained Approach: Zenodo (mai 2026)