nucleation

LIT-872: Hyperbolic geometry of the olfactory space

3 min read

Properties4
StatusActive — read 2026-10-09 (NOTE-672); worth reading as the first application of clique topology to stimulus statistics rather than neural data, and as the Sharpee lab's first claim of a hyperbolic geometry in biology. For four natural odour sources (strawberry, tomato, blueberry, mouse urine; 45–78 compounds over 50–101 samples), the rank order of absolute concentration correlations gives Betti curves that match points sampled near the surface of a 3D hyperbolic ball (R_max = 7, R_min = 0.9 R_max, the same for all four) and reject uniform Euclidean cubes of any dimension tried. Human descriptor profiles for 127 odorants (Dravnieks) match a full 3D hyperbolic ball and no Euclidean cube. Linear axes for pleasantness, boiling point and acidity are found in the fitted fruit-odour space. The tree that motivates the hyperbolic model is argued, not recovered: compounds do not cluster by chemical class, and the only null geometry is a uniform Euclidean cube.
Tagsneuroscience, natural-sciences, mathematics, cognition
Cited as “paper” byNOTE-672 — Hyperbolic geometry of the olfactory space
Cited as “source” byTHEORY-191 — In four natural odour sources (strawberry, tomato, blueberry and mouse-urine volatiles), the rank order of correlations between compound concentrations has the Betti-curve signature of points near the boundary of a three-dimensional hyperbolic ball, not of any uniform Euclidean cube tried, and human odour-descriptor profiles for 127 odorants have that of a full three-dimensional hyperbolic ball, THEORY-618 — Neural circuits that must represent hierarchically organised states in few dimensions use a hyperbolic geometry, three-dimensional where it has been measured, because hyperbolic space holds exponentially many well-separated states within a small radius; the argument from Zipf's law to this geometry does not hold, and the evidence is two Betti-curve studies tested only against uniform Euclidean nulls

Graph View

Yuansheng Zhou, Brian H. Smith and Tatyana O. Sharpee (2018), Science Advances 4(8):eaaq1458 — DOI-10.1126/sciadv.aaq1458, open access at PubMed Central (PMC6114987)

Key takeaways

  • The test is rank-order topology. Each odour pair gets the distance −|corr| of their concentrations across samples. Clique topology (Giusti et al. 2015) counts 1-, 2- and 3-cycles as the edge threshold sweeps, so the Betti curves depend only on the rank order of the correlations. Candidate geometries are scored by where the data’s integrated Betti values (and, as a check, L1 distances between curves) fall among 300 samples of as many points from each geometry, with multiplicative noise of 4–5% on the model distances.
  • Natural odours: a thin hyperbolic shell. For strawberry (78 compounds, 54 samples), tomato (66, 79), blueberry (45, 101) and mouse urine (69, 50), a 3D hyperbolic ball of curvature −1 sampled only in the shell 6.3 ≤ r ≤ 7 matches all three Betti curves (P > 0.19 in every case), with one radius for all four sources. The best Euclidean cube (dimension 8 or 10) is rejected (P < 0.03 for blueberry, < 0.003 for the others), as are shuffled concentrations. Log-transforming the concentrations changes nothing. Higher-dimensional hyperbolic spaces are not ruled out; 3D fits best.
  • Perception: a full hyperbolic ball. Dravnieks’ 146 descriptor ratings of 127 odorants, with Euclidean distances between descriptor profiles, match a full 3D hyperbolic ball (R_max = 1.6) and no Euclidean dimension; hyperbolic dimension 9 and above is rejected. The biphasic Betti curves are accounted for by non-uniform sampling: the embedded odorants fill only half of the space.
  • Linear axes in the fitted space. After a hyperbolic nonmetric MDS (angles only, radius fixed), directions for pleasantness, boiling point and acidity are found in the strawberry–tomato embedding and validated on held-out compounds: pleasantness R = 0.66 for single compounds, and R = 0.34 when a strawberry-trained axis predicts tomato liking. Because the shell is essentially 2D, the three axes are not independent.
  • No chemical clustering. Compounds do not group by functional class in the embedding, which the authors attribute to co-production in shared biochemical pathways.
  • The hierarchy is the motivation, not a finding. The paper’s case for a hyperbolic model is that a set system (Venn diagram) maps to points of a half-space by (centre, radius), with containment as height, and that biochemical networks are hierarchical. No tree is inferred from the odour data.

Standing in the record

Filed at the owner’s request on 2026-10-09, in the second part of the batch on hierarchy and hyperbolic geometry that followed QUESTION-025. The first part held Cagnetta et al.’s random hierarchy model (LIT-877), Krioukov et al. 2010 (LIT-865), Sala et al. 2018 (LIT-874), Lin et al. 2023 (LIT-870), Zhang et al.’s hippocampal paper (LIT-876), Yang et al. 2023 (LIT-871) and Ganea et al.’s entailment cones (LIT-866).

Read on 2026-10-09 (NOTE-672). The reading is the source of THEORY-191. Its nearest sibling is LIT-876: the same lab, the same Betti-curve test and the same uniform-cube null, applied four years later to hippocampal spike correlations, where it gives THEORY-192. This paper is the earlier one and applies the test to stimulus statistics and perceptual ratings, not neural activity. It cites Krioukov et al. (LIT-865) for the claim that hyperbolic space approximates hierarchical networks.

It is not a reading for the anthology: it involves no machine-learning practice, and its subject is a measurement of natural odours and human perception.

Backlinks