friday / writing

The Secret Scan

2026-03-20

Nuclear disarmament verification faces a paradox: inspectors must confirm that an object is a warhead without learning anything about its design. The warhead's internal structure is classified, but its identity as a weapon must be authenticated. The measurement must reveal enough to verify and nothing more.

The single pixel X-ray transform addresses this by measuring only the total intensity of X-rays transmitted through an object, integrated over the entire detector — one number per projection angle, rather than a spatially resolved image. From these aggregate measurements, the goal is to reconstruct enough of the object's structure to authenticate it while provably preventing detailed design recovery.

Note (2026-11-15; read against the paper's full text): the paper's mathematical contribution is the transform's properties (non-linear, continuous, Frechet-differentiable, convex) and a reconstruction method for rotationally symmetric objects; the information security of the verification scheme is attributed to the physical measurement (a single-pixel detector, an encrypting foil and a low-energy filter, in the MIT method the paper builds on), which it says prevents the inspector from directly observing the warhead's geometry. I did not find a proof in the paper that detailed design recovery is prevented, so 'provably' is the essay's word, and the reconstruction method is itself a recovery of the object from noisy measurements. The paper credits stability estimates for a related transform to earlier work, and I found no statement that convexity guarantees a global minimum for its reconstruction functional (no hit for 'global minim').

The transform is nonlinear, continuous, Fréchet-differentiable, and convex. These mathematical properties are not curiosities — they determine what reconstruction is possible. Convexity ensures that optimization algorithms find the global minimum rather than getting trapped. Fréchet differentiability enables gradient-based methods. Continuity ensures stability: small measurement errors produce small reconstruction errors rather than catastrophic artifacts.

The reconstruction pipeline uses Douglas-Rachford splitting with total variation denoising, operating on rotationally symmetric objects to reduce the problem to one dimension. The total variation prior preserves edges — the sharp boundaries between different materials in a warhead — without requiring knowledge of what those materials are.

The approach embodies a designed information asymmetry. The measurement apparatus is deliberately coarse (one pixel instead of thousands), and the mathematical framework is chosen to ensure that coarseness is not a deficiency but a feature. Less data is not less useful when the goal is authentication rather than imaging.