The Minimum Structure

Some functions don’t degrade gracefully. Below a structural minimum, they don’t exist at all.

A molecular amplifier built from dimers — complexes of two monomers — cannot amplify a signal at thermodynamic equilibrium. Not weakly. Not with noise. It is mathematically impossible. The proof is general: no network restricted to pairwise interactions can achieve equilibrium amplification regardless of the network’s size or connectivity. Add one monomer — make trimers — and amplification becomes possible. The maximum amplification then scales linearly with interaction free energy. The dimer-trimer boundary is not a quantitative threshold where performance improves. It is a qualitative boundary where a function switches from impossible to possible.

This pattern appears across domains that share no obvious connection.

In granular physics, packings of purely repulsive particles obey marginal stability — the system sits at the minimum coordination needed for mechanical rigidity. Add cohesion (attractive interactions between particles), and marginal stability breaks. The shear modulus develops hysteresis under compression and decompression. Pressure alone can no longer describe the mechanical state. One interaction type (attraction) added to another (repulsion) doesn’t gradually modify the physics. It creates a qualitatively different material.

In atmospheric chemistry, carbon monoxide exposed to UV radiation produces a narrow range of organic hazes — particles between 10 and 80 nanometers with limited chemical diversity. Replace CO with methane — both are single-carbon molecules, both are common in planetary atmospheres — and the haze yield jumps dramatically. The particles become chemically complex, dense, and diverse enough to support prebiotic chemistry. The switch from an oxidizing to a reducing carbon source doesn’t improve haze formation. It unlocks an entirely different category of molecular complexity.

In mechanism design, a seller investigating a buyer before setting a price faces a type space that could be arbitrarily complex — thousands of buyer types with different valuations and constraints. The optimal investigation, regardless of the type space’s complexity, requires at most three signal outcomes. The bound comes from the problem’s effective policy dimension: two independent decisions (whether to allocate and what to charge) require three outcomes. No additional resolution helps, no matter how many types exist. The minimum information structure is set by the problem’s intrinsic dimensionality.

In spectral graph theory, the eigenvalues of a symmetric matrix can uniquely identify the graph it represents — you can “hear the shape” of an undirected network. Make the matrix asymmetric (directed edges), and spectral uniqueness is destroyed. Almost all directed graphs have spectral twins that are structurally different but mathematically indistinguishable from eigenvalues alone. The symmetry requirement isn’t a convenience. It is a structural prerequisite for spectral fingerprinting. Remove it, and the function vanishes.

What these examples share is not the familiar story of phase transitions, where a continuous parameter crosses a threshold and the system reorganizes. These are discrete structural prerequisites. You cannot have 2.5-mers. You cannot have half-cohesion in the relevant sense. You cannot have 0.7 symmetry in a matrix. The function either has its structural minimum or it doesn’t.

The coarse screening result makes this sharpest. The seller doesn’t need complex signals because the problem isn’t complex in the relevant sense — it has two decisions, so it needs three outcomes. The type space’s apparent complexity is irrelevant to the information structure required. The minimum is set by the problem’s dimensionality, not by the space it’s embedded in.

This distinction matters because it resists the intuition that more resolution, more components, or more sophistication always helps. In each case above, the system below the minimum isn’t merely underperforming. It is categorically incapable. And the system above the minimum doesn’t need to be far above it. Trimers suffice; you don’t need tetramers. Three outcomes suffice; you don’t need thirty. The minimum structure is a floor, not a target.

The practical implication is diagnostic. When a system fails to exhibit an expected function, the question isn’t always whether the parameters are tuned correctly. Sometimes the question is whether the structure has the minimum dimensionality the function requires. If a dimer network can’t amplify, no amount of optimization within the dimer architecture will produce amplification. The architecture must change.

The minimum structure is the smallest thing that can do the job — not because smaller is better, but because below it, there is no job at all.