How do you know something is alive? The standard answer involves metabolism, reproduction, homeostasis — functional descriptions that require watching over time. But Ramírez-Colón, Ni, and Carr propose something stranger: you can tell from a single chemical snapshot (2602.18490).
Their method, LUMOS, examines the distribution of HOMO-LUMO energy gaps among amino acids in a sample. The HOMO-LUMO gap measures how easily a molecule participates in chemical reactions — smaller gap, more reactive, more biochemically useful. Abiotic processes (asteroid chemistry, prebiotic synthesis) produce amino acids with uniformly distributed gaps. Biology selects for amino acids with lower gaps. The distribution itself is the signature.
This works because life doesn't just use molecules — it curates them. An abiotic sample is a random draw from thermodynamic possibility space. A biological sample is a filtered draw, biased toward molecules that do things. The filter leaves a statistical fingerprint even when you can't identify any individual molecule as “biological.”
The paper achieves >95% accuracy distinguishing biological from abiotic sources, including asteroid Bennu samples containing 15 of 20 essential amino acids. The method is composition-agnostic: it doesn't care which amino acids are present, only how their electronic properties distribute. An alien biochemistry using entirely different amino acids would still show the same bias toward reactivity.
This is elegant for a reason beyond astrobiology. The underlying principle is that selection — any selection, for any purpose — distorts the statistical properties of a population away from what chance alone would produce. Abiotic chemistry is ergodic: given enough time and energy, all accessible amino acids appear in roughly thermodynamic proportions. Biological chemistry is non-ergodic: only the useful ones accumulate.
The same logic applies to other domains. A library curated by a reader has different statistical properties than a warehouse of randomly acquired books — not because any individual book is identifiably “curated,” but because the distribution of, say, information density is shifted. A codebase maintained by a developer has different structural properties than randomly generated syntactically valid code. The selection leaves traces in aggregate statistics that are invisible at the individual level.
What makes LUMOS particularly satisfying is that it identifies life by what life does to its chemical environment, not by what life looks like. The signature is in the selection pressure, not the selected objects. This means it should generalize: any system that selects from a pool based on functional criteria will distort the pool's statistical properties in detectable ways.
The deeper question: is this what “life” means? Not metabolism, not reproduction, not homeostasis — but the presence of a selection filter that shifts a population's distribution away from thermodynamic equilibrium. If so, then “alive” is not a property of objects but a property of processes acting on populations.
And the distribution doesn't lie. You can't fake the statistical fingerprint without actually doing the selection. The HOMO-LUMO gap distribution of a biological sample isn't a representation of life — it's a consequence of life. The map is the territory.