Geodesics: Straightest Paths on a Surface

What can replace a straight line?

On a flat plane a straight line wears two hats at once. It is the shortest route between two points, and it is the straightest — the path you walk when you never turn the steering wheel. On the plane these two descriptions name the very same lines, so we rarely notice they are different ideas. But move onto a curved surface, a sphere or a saddle, and there are no honest straight lines left living inside it: every path bends. The question of this guide is what, if anything, deserves to inherit the role of 'the line' on a surface — and the surprise is that the two hats can be picked up separately.

Take the most familiar curved surface, the sphere. Fly from one city to another along the shortest possible track and you trace an arc of a great circle — a circle whose plane cuts through the centre of the sphere, like the equator or any line of longitude. That is why long-haul flight paths look like curves bowing toward the pole on a flat map: the genuinely shortest route on the globe only looks bent because the map lied about distances. Great circles are the sphere's answer to 'line', and noticing why is the first step.

Splitting the bend in two

To make 'straightest' precise we lean on the curvature of curves you already met low on this ladder. Walk along a curve drawn on the surface at unit speed, and at each instant your acceleration vector measures how the path is bending. Here is the key move: that acceleration lives in space, so split it into two perpendicular pieces — the part pointing along the surface's normal, straight out of the surface, and the part lying *flat in the tangent plane*, the part that would turn you left or right as a flatlander walking on the surface.

The normal piece is the normal curvature from the earlier guides — it measures how much the surface itself forces the path to bend simply by being curved, and you cannot escape it. Even the straightest possible walk on a sphere still curves in space, because the sphere is round under your feet; that bending is the surface's doing, not yours. The in-the-tangent-plane piece is different. It is the geodesic curvature, written kappa_g, and that is the part you control with the steering wheel. It records how hard you are turning relative to the surface, the genuinely sideways effort a flatlander would feel.

for a unit-speed curve on a surface, the
in-space bending splits perpendicularly:

  (total curvature)^2  =  kappa_n^2  +  kappa_g^2

  kappa_n  =  normal curvature  (the surface's doing,
              points along the surface normal)

  kappa_g  =  geodesic curvature  (your doing,
              lies in the tangent plane)

  GEODESIC  <=>  kappa_g = 0  everywhere
             (no sideways steering at all)

The definition, and why a great circle obeys it

Now the definition writes itself. A geodesic is a curve whose geodesic curvature is zero everywhere: kappa_g = 0. In plain words, its acceleration has no component in the tangent plane — every bit of its bending is the unavoidable normal kind the surface imposes, and none of it is steering. That is the honest meaning of 'straightest path': not unbent in space (impossible on a curved surface), but unbent as far as the surface allows. The little locked-wheel car traces exactly such a curve, because a locked wheel applies no sideways turn.

Let us test this against the great circle and watch it pass. Walk the equator of a unit sphere at unit speed. Your acceleration at every point aims straight at the centre of the sphere — and on a sphere the line to the centre is the surface normal at your feet. So all the acceleration is normal; there is nothing left over in the tangent plane, kappa_g = 0, and the equator is a geodesic. The same argument works for every great circle, by symmetry. Meanwhile a circle of latitude away from the equator — say a tight ring near the north pole — visibly curves toward the pole; that pull is sideways within the tangent plane, so its kappa_g is not zero and it is not a geodesic, however 'round' it looks.

Carrying an arrow without twisting it: parallel transport

There is a second, deeper way to see 'straightest', and it unlocks something genuinely new. On a flat plane you can slide an arrow around keeping it pointing the same way — translating a vector without rotating it. On a curved surface there is no global 'same direction' to appeal to, because the tangent plane tilts as you move. The repair is parallel transport: to carry a tangent vector along a path while turning it as little as the surface permits, allowing only the unavoidable tilt that keeps it flush with the surface, and adding no extra spin of your own.

With that idea in hand, geodesics get a strikingly clean restatement: a geodesic is a curve that parallel-transports its own velocity vector. As you walk it, your direction of travel is being carried along by the surface's own no-extra-spin rule — you are not turning, you are letting the surface carry your heading 'straight ahead' for you. This connects to the machinery the manifolds rung formalizes as the Levi-Civita connection: parallel transport, geodesics, and 'no sideways acceleration' are three faces of one structure, the surface's built-in notion of what it means to keep going the same way.

Geodesics are intrinsic — and that is the whole point

Why does any of this matter beyond bookkeeping? Because — and this is the payoff that ties the whole rung together — geodesics are intrinsic. The condition kappa_g = 0 can be written entirely in terms of the first fundamental form, the metric ds^2 = E du^2 + 2F du dv + G dv^2 that the surface carries inside it. The recipe for a geodesic is encoded by the so-called Christoffel symbols, which are built purely from E, F, G and their derivatives — no peek at how the surface sits in space is needed anywhere.

This is the Theorema Egregium's spirit again, made practical. Recall that guide's punchline: an ant confined to the surface, knowing only distances and angles measured within it, can still discover the surface's Gaussian curvature without ever leaving home. Geodesics are how such an ant gets around — they are the routes it finds by walking as straight as it can, using nothing but its own internal sense of direction. Bend a sheet of paper into a cylinder without stretching it and its straight lines become helices and circles, yet they are still geodesics: the intrinsic metric never changed, so the straightest paths did not either.

One honest caveat keeps the two hats from being glued back together carelessly. A geodesic is locally shortest — over a short enough stretch it really is the minimizing path — but it need not be globally shortest. On a sphere, the short equatorial arc from a city to its neighbour and the long way round the other side are both geodesics; only one is the shortest journey. So 'straightest' (kappa_g = 0) is the clean, local, always-true definition; 'shortest' is a global prize a geodesic may or may not win. Honoring that gap is exactly the kind of caveat this subject rewards rather than glossing over.

What you can now do, and where it leads

1. Given a curve on a surface, split its acceleration into a normal part and a tangential part; read off the normal curvature kappa_n and the geodesic curvature kappa_g.
2. Declare the curve a geodesic exactly when kappa_g = 0 everywhere — no sideways steering, all bending forced by the surface.
3. Recognize a geodesic as the curve that parallel-transports its own velocity, and on a sphere recognize the geodesics as the great circles.
4. Trust that geodesics are intrinsic — fixed by the first fundamental form alone — so bending a surface without stretching it leaves its geodesics in place.

This guide quietly assembled the two ingredients the rung has been building toward. From the curvature guides we have curvature that lives in the surface; from this guide we have geodesics and parallel transport, the surface's own sense of 'straight ahead'. The next and final guide of the rung, on the Gauss-Bonnet theorem, marries them in one astonishing equation, tying the total Gaussian curvature of a region to the turning of its boundary and even to the Euler characteristic chi = V - E + F you met in topology. Geodesics are the threads that stitch local curvature to global shape; you have just learned to follow them.