Which Curvature, and What Is Curved

1. One word, two invariants

At every point of a smooth surface there are two principal curvatures, κ₁ and κ₂: the extreme bending rates of the normal sections through that point. From them come the two quantities that do the physical work.

H = (κ₁ + κ₂) / 2 — mean curvature, extrinsic: how the sheet sits in the surrounding space.

K = κ₁ · κ₂ — Gaussian curvature, intrinsic: detectable from distances measured inside the sheet, with no reference to any embedding.

The distinction is not a refinement. It is the hinge. A developable wrinkle can carry a very large H with K ≈ 0; a dome has K > 0; a saddle has K < 0. The topology of the sheet can be identical in all three cases — still a disk, still genus zero, still no cut and no glue. What changed is the embedding, and the two invariants of the embedding are answerable to different physics.

2. What mean curvature does

Flexoelectric polarisation tracks the curvature of the embedding, not the genus. In the nanowrinkle regime the working approximation is

P ≈ f (κ₁ + κ₂) ≈ f / R

when one principal radius is tiny and the other nearly flat. That form has a consequence worth stating on its own: sharpness dominates height. Polarisation scales with 1/R, not with wrinkle amplitude, which is why sub-nanometre radii put the reported polarisation orders of magnitude above mesoscale flexoelectricity, and why a tall gentle fold does comparatively little. The sheet is still a disk. The π cloud is simply no longer symmetric across the local normal.

Mean curvature and strain gradient are cousins rather than twins. For a thin sheet of effective thickness t, bending puts opposite in-plane strains on the two faces, of order ± t / 2R, and therefore a through-thickness gradient ∂ε/∂n ~ 1/R. In graphene t is not an ionic thickness at all but the spatial extent of the π orbitals, so the two “faces” are electronic. That is the differential-geometric content of the phrase quantum orbital flexoelectricity: an extrinsic curvature couples to an out-of-plane shift of charge.

3. What Gaussian curvature does

Gaussian curvature has a different job, and Gauss’s theorema egregium is what assigns it. K cannot be removed by isometric flattening. If a lattice with nonzero K insists on remaining a crystal, it must stretch: the metric is forced. That in-plane strain renormalises electron hopping, generates a pseudomagnetic field for Dirac electrons, and can shift or open van Hove singularities.

So a cylinder (K = 0, H ≠ 0) and a bump (K ≠ 0) change the electronics by different routes: the first chiefly through orbital asymmetry and flexoelectric dipoles, the second additionally through metric strain. A paper that calls both “topological” has collapsed two mechanisms and one non-mechanism into a single adjective.

The bending energy of a membrane already carries the split in its two terms:

E_bend ∝ ∫ κ H² dA + ∫ κ_G K dA

The first term cares how the sheet is embedded. The second is topological in the global Gauss–Bonnet sense — ∫ K dA plus boundary terms equals 2πχ — and yet, locally, K(x) is still acting as a geometric source of strain. Function follows the local densities H(x) and K(x). It does not follow the Euler characteristic of the sample.

4. The drill

For any shape you meet, ask whether the new property survives an isometric flattening.

• If it dies when H → 0 at fixed metric, it was mean-curvature physics.
• If it survives that and dies only when K → 0, it was Gaussian, which is to say metric, physics.
• If it survives both, the deformation was not the cause and something else is carrying the effect — composition, a boundary, an invariant.

The nanowrinkle battery dies with H. That single sentence places it: geometric activation of a field on an invariant form, not a topological change, and the label follows the test rather than the press release.

5. Name the carrier before crossing scales

The temptation the previous essay opened is to run the same word from a graphene crease to the curvature of the universe to a folded protein to entanglement, and call the through-line a discovery. It is not a discovery until the carrier is named, because these are different objects and only the vocabulary is shared.

• A 2D crystal — an embedded surface in ordinary 3-space. Local markers: H, K, strain ε_ab. Topology can tell you the sheet is still a disk. It cannot tell you why a dipole appeared.
• A biological membrane — also a 2-surface, but fluid, in water, built of dipolar lipids and worked on by proteins. The same schematic P = f(c₁ + c₂) appears and is discussed in hearing and ion-channel contexts. Same differential-geometric slot; different physics inside the slot, because biology adds active remodelling and the curvature is often maintained rather than merely suffered.
• Spacetime — a 4-manifold with a Lorentzian metric, curvature sourced by stress-energy. Local marker: the Riemann and Ricci tensors. Measured spatial curvature in the observable universe is consistent with flatness within present limits, which is a statement about triangles and the Friedmann parameter, not about whether the whole cosmos is a 3-sphere or a 3-torus.
• A quantum state — not a bent surface at all. Entanglement is a correlation structure on a Hilbert space. Its natural measures are entropies and Bell-type quantities, not H and K.

A large H on a sheet implies nothing about the Riemann tensor of the universe, and a flat cosmos forbids no wrinkled graphene. Scale does not unify the curvature type; it changes which terms dominate. At angstrom radii 1/R is enormous and quantum orbital effects rule. At cellular radii thermal and viscous terms rule. At cosmic scale the measured spatial curvature is tiny and gravity is the curvature of a different manifold entirely.

6. Local and nonlocal is not micro and macro

Two axes get crossed here often enough to deserve separating. Local curvature is a tensor at a point or in a small neighbourhood: Riemann in spacetime, H and K on a membrane. Nonlocal, in physics, usually means a correlation or an integral that cannot be reduced to one point — ∫ K dA = 2πχ is nonlocal in exactly that sense, and so is the entanglement entropy of a region. Neither has anything to do with being large.

On present evidence, entanglement does not let you change local membrane curvature, or local spacetime curvature, at a distant laboratory faster than causal contact allows. ER = EPR and its relatives treat entanglement as a kind of spacetime connectivity; that is a live research programme and interesting, and it is not a measured replacement for P ≈ f(κ₁ + κ₂) on a carbon sheet. Putting the two on one shelf is the same error the companion essay catalogued under conflation, moved up a few orders of magnitude.

7. Four questions

The discipline compresses to four questions, asked of every claim that curvature explains a function:

• What manifold, and what is actually curved?
• Which curvature invariant — H, K, Riemann, or a band-structure invariant wearing the word?
• Is the effect local, or an integral constraint?
• Is entanglement being used as correlation, or as a stand-in for geometry?

Those four keep macroscopic and microscopic, living and non-living, from being forced onto one curve. They also satisfy this archive’s standing rule about borrowed shapes: each question can come back with an answer that kills the analogy, which is the only property that makes an analogy worth using.

8. Status and falsifiers

Established. The definitions of H and K from principal curvatures; Gauss’s theorema egregium and the intrinsic character of K; the Gauss–Bonnet relation; the Helfrich form of bending energy; strain-induced pseudomagnetic fields in graphene; curvature–polarisation coupling in lipid bilayers as a modelled and measured effect; spatial flatness of the observable universe within current observational limits.

Reported, unconfirmed here. The graphene nanowrinkle flexoelectric result, single group, at scanning-probe resolution, with polarisation figures that mix measurement with model. The f/R scaling is carried as the reported working approximation, not as a verified law across the whole wrinkle range.

Interpretive. The isometric-flattening drill as a classification test; the carrier list as a way of stopping cross-scale transfer; the claim that most cross-scale “curvature of everything” language is vocabulary sharing rather than physics sharing.

Contested. Whether entanglement geometry (ER = EPR and successors) will eventually license treating correlation structure as curvature of an emergent manifold. Held open, not denied, and kept off the same shelf as measured membrane physics until a measurement does the moving.

Falsifiers. (1) If a reported flexoelectric response is shown to persist after the mean curvature is removed at fixed metric, the effect was not mean-curvature physics and the classification in section 4 misassigned it. (2) If a curvature-driven function in a 2D crystal is shown to scale with amplitude rather than with 1/R, the sharpness-over-height claim is wrong for that system. (3) If an experiment demonstrates a change in local curvature or a local mechanical field brought about by entanglement alone, without causal contact, then section 6 is wrong and the shelf separation collapses. (4) If cosmological measurement resolves nonzero spatial curvature at a scale that changes local predictions, the “flat within limits” framing here requires restatement. (5) Turned on this essay: if the H/K/carrier split is ever applied to a case where every branch returns the same answer and no prediction differs, the split did no work in that case and should not be invoked there. An abstraction that cannot fail earned nothing.

← All essays