Eugenio Beltrami

A consistency proof, not a discovery

By 1830 the geometry was already built. Lobachevsky published in 1829 and Bolyai in 1832 a complete plane geometry in which, through a point off a line, more than one parallel passes. Neither had found a contradiction in it. Neither could show there wasn’t one. For two millennia the parallel postulate had been attacked by assuming its negation and hunting for an absurdity; Saccheri in 1733 derived page after page of hyperbolic theorems, decided the last one was “repugnant to the nature of the straight line,” and published it as a vindication of Euclid. He had proved the opposite of what he thought and could not tell, because nothing in the method distinguishes strange from false.

Beltrami’s Saggio di interpretazione della geometria non-euclidea settled it by changing the question. Take the pseudosphere — the surface of revolution generated by a tractrix — whose Gaussian curvature is the constant $K = -1/R^{2}$. Read point as point of that surface, line as geodesic, distance as arc length. Every axiom of Lobachevsky’s geometry then becomes a true statement about an ordinary object in ordinary Euclidean space.

One limit belongs here rather than in a footnote. Hilbert proved in 1901 that no complete regular analytic surface of constant negative curvature can be embedded in $\mathbb{R}^{3}$. The pseudosphere therefore models a piece of the hyperbolic plane, not the whole of it: it has an edge, and the geometry does not. The bridge is real and it is local. Klein’s and Poincaré’s later models carry the complete plane, at the cost of no longer being surfaces you can hold.

Curvature is intrinsic, and you can check it on a desk

A surface in Euclidean space carrying a non-Euclidean geometry sounds like a contradiction until Gauss’s Theorema Egregium: the Gaussian curvature $K$ is determined by the first fundamental form alone. It is a property of distances measured within the surface, not of how the surface sits in the room. An ant confined to the surface can measure $K$; it never needs to look up.

The consequence is testable without apparatus. Roll a sheet of paper into a cylinder: every intrinsic measurement is unchanged, because rolling bends without stretching. Now try to wrap that sheet onto a ball without creasing or tearing. You cannot, and the obstruction is numerical rather than practical — the plane has $K = 0$, the sphere has $K = 1/R^{2}$, and no bending can change that.

The triangles, and the dome

Gauss–Bonnet makes the same fact quantitative. For a geodesic triangle on a surface, the angle sum exceeds $\pi$ by exactly the total curvature it encloses:

$\displaystyle \alpha + \beta + \gamma - \pi \;=\; \iint_{T} K \, dA$

Excess is proportional to area. Tile a region with triangles and the excesses add, because the integral is additive — so laying triangle beside triangle and watching the surface close into a dome is not an analogy for curvature, it is the integral being evaluated one piece at a time. Keep going until the surface closes and the total is forced by topology alone: $\iint_{S} K\,dA = 2\pi\chi = 4\pi$ for any sphere, of any size and any shape. The dome is not chosen. It is what a positive curvature budget has to spend itself on.

Flip the sign and the arithmetic flips with it. On Beltrami’s surface $K = -1$, so the angle sum falls short of $\pi$ by the area, and a hyperbolic triangle’s area is therefore bounded above by $\pi$ no matter how far apart its vertices are. Triangles never close a dome there; the budget is negative and unbounded. The pseudosphere’s flaring skirt and Hilbert’s non-embedding theorem are two views of that one fact.

Why a Chladni figure bends when the plate does not stay flat

Beltrami’s other contribution to this chapter is the pair of differential parameters that generalise the Laplacian to a curved surface — what is now written $\Delta_{g}$ and called the Laplace–Beltrami operator. It is the reason the question what does a resonance look like on a curved surface? has an answer rather than an analogy.

Chladni’s figures are the nodal sets of eigenfunctions: sand collects where the surface does not move. On a stretched membrane the operator is $\Delta$; on the free metal plate Chladni actually used it is the biharmonic $\Delta^{2}$, which is why his patterns are not the ones a drumhead makes. Curve the surface and the operator becomes $\Delta_{g}$, built from the metric and nothing else.

That last clause is stronger than the intuition it replaces, and it is worth stating in the form that can be falsified. The pattern does not respond to curvature as an appearance; it responds to the intrinsic metric. Roll a flat plate into a cylinder and the spectrum and the nodal sets are unchanged, because $\Delta_{g}$ did not change. Press a flat plate into a dome and everything changes, because that cannot be done without stretching. Extrinsic bending is invisible to the resonance; only the metric is audible. Theorema Egregium, heard rather than measured.

Two consequences follow for anything carapace-shaped. Courant’s theorem bounds the $n$-th eigenfunction to at most $n$ nodal domains, so a nodal count is a real constraint and not a free parameter. And the flat disc’s famously clean patterns — the $J_{m}(kr)\cos m\theta$ families — owe their symmetry to the disc’s own symmetry group; lower the symmetry and the degenerate multiplets split. The tidy pictures are the special case, not the generic one.

The ISS is not curving

Once the metric is primitive, straight stops being a primitive. A geodesic is a curve that parallel-transports its own tangent, $\nabla_{u}u = 0$, and that is the whole definition; on a surface it is the curve of zero geodesic curvature, the one an ant walking without turning traces out. Beltrami’s dictionary works because line maps to geodesic, not because anything was bent into place.

In free fall a spacecraft follows a timelike geodesic, which is why an accelerometer aboard the ISS reads zero — the residual it does read is atmospheric drag at 400 km, which is a genuine departure from geodesic motion and the reason the station needs reboosts. What looks like a curved path is a four-dimensional worldline projected onto a three-dimensional slice, and the projection supplies most of the drama.

Drawn to scale the effect nearly vanishes. Over one orbit the worldline is a helix of radius $r \approx 6.79\times10^{6}$ m and pitch $cT \approx 1.67\times10^{12}$ m, a ratio of about $4\times10^{-6}$: on an honest $(ct, x)$ diagram it is indistinguishable from a straight line. The curvature doing the work is $GM/rc^{2} \approx 6.5\times10^{-10}$, and it closes an orbit only because it is given a very long lever in time. Every textbook picture of a “curved path through spacetime” has silently compressed the time axis by six orders of magnitude.

The carapace and the orbit are the same move made twice, and the differences between them are not decorative. A carapace is Riemannian: positive-definite metric, embedded in $\mathbb{R}^{3}$, curvature fixed in advance as a background. Spacetime is Lorentzian, embedded in nothing, and its curvature is determined by what it contains. The signature is the sharpest of the three: Riemannian geodesics locally minimise length, timelike geodesics locally maximise proper time, and the elliptic spectral theory that produces Chladni figures — discrete spectrum, nodal domains, Courant’s bound — has no Lorentzian counterpart, because the wave operator there is hyperbolic. The bridge carries the idea that the metric is primitive. It does not carry the spectra.

Place in the Series

There is a precise bridge from this chapter to the contact geometry the corpus runs on, and it is a theorem rather than a resemblance. For a Riemannian manifold $(M, g)$, the unit cotangent bundle $S^{*}M$ carries the restriction of the canonical Liouville 1-form $\lambda$; that restriction is a contact form, and its Reeb vector field generates exactly the geodesic flow of $g$. Geodesic motion is Reeb dynamics, on a manifold built from the original one. Ch Fy computes a Reeb field directly, and the Mirzakhani chapter is filed under the same heading.

The corpus already held the most famous instance of that theorem and this chapter was written without finding it. book8/ch8-0-monster.html reads the Monster’s vertex operator algebra as the chiral algebra of a conformal field theory “whose contact structure on the unit tangent bundle T¹ℍ is exactly the dm³ contact form with SL(2,ℤ)”. T¹ℍ is the unit tangent bundle of the hyperbolic plane — this chapter’s object — and T¹ℍ/SL(2,ℤ) is the modular surface, whose geodesic flow is the textbook example of the Reeb correspondence stated above. Two caveats travel with the citation. That T¹M carries a contact form whose Reeb flow is the geodesic flow is standard; that the form on T¹ℍ is exactly the dm³ contact form is the corpus’s own identification and carries no check. And the word conformal there is field-theoretic, not the angle-preserving sense used in Klein and in Escher §6 — the two meet only in dimension two, which is the subject of WP-109.

References