The Epistemology of Geometry

For most of Western thought, geometry was the very model of certain knowledge: necessary, exact, and known a priori. Euclid's theorems seemed to describe the structure of physical space with a certainty no empirical science could match, and this certainty demanded a philosophical explanation. Kant supplied the most influential one — geometry as synthetic a priori knowledge grounded in the form of intuition — and then, within a century, two developments dismantled it: the discovery that non-Euclidean geometries are perfectly consistent, and the discovery, in general relativity, that physical spacetime is not Euclidean at all. The epistemology of geometry is the story of how the a priori certainty of geometry gave way, and of what took its place: geometric empiricism, and the conventionalist challenge to it.

This page traces that arc; it presupposes the space and geometry overview and connects forward to conventionalism and to the physics of curved spacetime.

Kant's synthetic a priori

Kant faced a puzzle. Geometrical truths — "the shortest path between two points is a straight line," "the interior angles of a triangle sum to two right angles" — seemed to be:

• A priori — known independently of experience, and with necessity and universality no amount of measurement could confer; we do not check whether triangles might have angle-sums of 179°.
• Synthetic — not mere logical or definitional truths (their denials are not self-contradictory in the way "a bachelor is married" is); the predicate is not contained in the subject. Geometry tells us something substantive about space.

How can knowledge be both substantive and prior to experience? Kant's answer: space is not a feature of things-in-themselves but the a priori form of outer intuition — the framework the mind imposes on all sensory experience of outer objects. Geometry describes the structure of that framework. Because every possible experience must conform to it, geometry is necessary (nothing could appear to us non-Euclideanly) yet substantive (it constrains the content of experience). Kant took the geometry in question to be specifically Euclidean: Euclidean structure is built into the form of our intuition. This elegantly explained geometry's certainty and its unfailing applicability to the world.

The non-Euclidean earthquake

Kant's account had a fatal presupposition: that Euclidean geometry is the only coherent geometry, forced by the very form of intuition. The nineteenth century refuted this.

By denying Euclid's fifth (parallel) postulate — that through a point not on a line there is exactly one parallel — and replacing it (many parallels, or none), Gauss, Bolyai, and Lobachevsky (hyperbolic, constant negative curvature) and Riemann (elliptic, positive curvature) developed geometries that are perfectly consistent. Their consistency was later secured by relative-consistency proofs: Beltrami and Klein constructed models of non-Euclidean geometry within Euclidean geometry (e.g. the hyperbolic plane as the interior of a disc), showing that if Euclidean geometry is consistent, so are the others. Non-Euclidean geometry is therefore not a confused fantasy but genuine mathematics.

This shattered the necessity limb of Kant's account. If alternative geometries are conceivable and consistent, then Euclidean geometry is not forced by reason or by the form of intuition; we can perfectly well represent non-Euclidean spaces. Riemann went further, developing the general theory of manifolds of variable curvature (see topology and manifolds), and explicitly raised the possibility that the geometry of physical space is an empirical question, to be settled by measurement — perhaps varying from place to place.

The empirical turn — and its vindication in relativity

If geometry is not a priori, what is it? The natural successor is geometric empiricism: the geometry of physical space (or spacetime) is a contingent, empirical matter, discovered by measurement like any other physical quantity. Gauss is said to have tried to test it directly, measuring the angles of a triangle formed by three mountain peaks to see whether they summed to 180° (they did, within error — space is very nearly flat on that scale).

Empiricism was decisively vindicated by general relativity. There the geometry of spacetime is encoded in a metric field that is dynamical — determined by the distribution of matter and energy through the Einstein field equations — and generally non-Euclidean (curved). Spacetime geometry is not a fixed a priori backdrop but a physical variable, and its curvature has been measured: the bending of starlight near the Sun (Eddington, 1919), the precession of Mercury's orbit, gravitational lensing, and the propagation of gravitational waves all test the local geometry (see the experimental status). Physical geometry turned out to be exactly what Kant said it could not be — contingent, variable, and known a posteriori. The mathematics that makes this precise is the theory of curved manifolds.

Salvaging something from Kant

Kant's specific thesis — that physical space is necessarily Euclidean — is dead. But two weaker descendants survive:

• A priority of some structure. Perhaps not the full metric, but weaker features — that space is a continuous manifold, or is three-dimensional, or has some topological structure — might be a priori conditions of experience even if the metric is empirical. Reichenbach distinguished the "relativised a priori" (constitutive framework principles that are a priori relative to a theory but revisable across theories) precisely to preserve a Kantian insight without the false necessity.
• The transcendental point about representation. That we must represent objects as in space at all may remain a condition of outer experience, even though which geometry that space has is left open. This is a common modern gloss: Kant was right that spatiality is a form of our experience, wrong that its metric is fixed a priori.

The unfinished business: conventionalism

Geometric empiricism is not the final word, because of a problem that Kant's fall exposed rather than solved. We never measure geometry nakedly: every geometrical measurement uses physical objects — rods, light rays, clocks — that obey physical laws, so what confronts experience is always a package of geometry-plus-physics. A discordant measurement can always be blamed on the physics (a "force" distorting the rods) instead of the geometry. This underdetermination is the engine of conventionalism (Poincaré, Reichenbach, Grünbaum): the claim that the choice of geometry is, in part, a stipulation rather than a discovery. Whether the empirical geometry of general relativity is a genuine fact or a convenient convention is the subject of the next page.

Where this sits

The epistemology of geometry is the historical core of the geometric question: it is the arc from Kant's synthetic a priori, through the non-Euclidean revolution, to the empirical geometry vindicated by general relativity. It presupposes that the continuum itself is coherent, and it hands on to conventionalism the unresolved question of whether physical geometry is discovered or partly chosen. It also bears on the substantivalism debate: if geometry is a dynamical field, as relativity says, then geometry is something physical — which is grist for the substantivalist's mill. The next page presses the conventionalist challenge.