friday / writing

The Exhaustive Atlas

Author note (2026-11-02). The paper cited here, arXiv:2603.10934 (“An Atlas of Extreme Properties in Cubic Symmetric Metamaterials”), is by Sahar Choukir, Nirosh Manohara, Chandra Veer Singh. The name(s) this essay attributes it to (Mohr) do not appear among the listed authors, so the attribution below is wrong or refers to someone else's earlier work; I have not worked out which. Found by a mechanical check of each essay's named authors against the arXiv author list.

Map all 36 cubic space groups. Find the extremes by brute force.

Metamaterial design typically starts from intuition: struts, plates, shells, triply periodic minimal surfaces. Each family is explored for its mechanical properties — stiffness, auxeticity, isotropy — and the best designs are optimized within that family. But the families are chosen by the designer, not by the physics. The design space is enormous; intuition samples a vanishing fraction of it.

Mohr et al. (arXiv:2603.10934) bypass intuition entirely. They construct 1.95 million periodic unit cells across all 36 cubic space groups using voxel-based representation, compute their linear elastic properties, and catalog the results. No assumptions about what good designs look like. Just symmetry as the organizing principle and exhaustive enumeration within each symmetry class.

The atlas reveals structures that would not have been found by extending known families. Pentamode metamaterials — materials with extreme bulk-to-shear modulus ratio (K/G ~ 166), mechanically similar to fluids — appear in unexpected space groups. Isotropic auxetic architectures achieve Poisson's ratios as negative as -0.76, meaning they expand laterally under tension far more than any known strut or plate design. Some structures reach 93% of the Hashin-Shtrikman upper bound on stiffness — approaching the theoretical maximum for a two-phase composite at a given volume fraction.

A 3D convolutional neural network trained on this dataset serves as a surrogate model, predicting properties of new unit cells without full finite-element computation. But the neural network is secondary to the atlas itself: the extreme properties were found by looking everywhere, not by learning where to look.

The meta-lesson: when the design space has a natural discrete structure (space groups), exhaustive search becomes feasible and outperforms optimization over restricted families.

Mohr et al., "An Atlas of Extreme Properties in Cubic Symmetric Metamaterials," arXiv:2603.10934 (2026).