From measuring on a surface to watching it bend
The previous guide handed you the first fundamental form, ds^2 = E du^2 + 2F du dv + G dv^2 — the tool that lets a flatlander living inside a regular surface measure lengths, angles, and areas without ever leaving home. That form is purely intrinsic: it knows nothing about which way the surface curls in the surrounding space. A flat sheet of paper and the same sheet rolled into a cylinder have identical first fundamental forms, because rolling does not stretch or tear anything. Yet your eye plainly sees the cylinder bending and the paper not. So the first fundamental form, for all its power, is deaf to bending — and bending is exactly what this guide is about.
To hear the bending we must step outside the surface and look at how it sits in three-dimensional space — this is extrinsic geometry. The key new character is a single arrow: at each point P of the surface there is a tangent plane, the flat plane that best hugs the surface there, and standing straight up from it is the unit normal vector N, perpendicular to every tangent direction. As you slide P around the surface, N pivots and tilts. The whole secret of curvature is hidden in how fast and in which way N turns as you move — a rolling sheet of paper keeps N pointing the same way along the roll's axis but swings it around the curl, and that swing is the bending your eye detects.
Slice the surface and read each slice's curvature
Here is the concrete experiment that turns bending into a number. Stand at a point P with its normal N. Pick a direction to walk in — a unit tangent vector v in the tangent plane. Now take the plane that contains both N and v (a vertical plane, slicing straight down through the surface) and look at the curve where it cuts the surface. That sliced curve, a normal section, is just an ordinary plane curve, and you already know from the curves rung how to measure its bending: its curvature kappa, the rate at which its unit tangent turns. Give that bending a sign — positive if the slice curves toward N, negative if it curves away — and you have the normal curvature k_n(v) in the direction v.
The decisive point is that this number depends on the direction v. Imagine standing in the middle of a Pringle-style saddle. Walk along the ridge that rises in front of you and the slice curves upward, toward N: positive normal curvature. Turn ninety degrees and walk along the valley that falls away to your sides and the slice curves downward, away from N: negative normal curvature. Same point, same surface, opposite bending — purely because you chose a different walking direction. A surface does not have one curvature at a point; it has a whole spread of normal curvatures, one for every direction you might face.
The two that rule them all: principal curvatures
So we have a normal curvature k_n(v) for every direction v — an entire function on the compass of directions. Does it wander wildly, or is there hidden order? Here is the beautiful fact, due to Euler: as v sweeps once around all directions, k_n(v) varies smoothly and reaches a clean maximum and a clean minimum. Those two extreme values are the principal curvatures, written k_1 and k_2 (with k_1 the larger). Even better, the maximum and minimum are always achieved along directions that are perpendicular to each other — the principal directions. Two numbers and two perpendicular directions: that is the entire curvature data of the surface at the point.
Euler went further and gave the exact formula for every in-between direction. If you walk at an angle theta measured from the first principal direction, the normal curvature is k_n(theta) = k_1 cos^2 theta + k_2 sin^2 theta. Check the corners: at theta = 0 this is k_1 (cos is 1, sin is 0); at theta = 90 degrees it is k_2; and in between it slides smoothly between the two extremes. So the maximum and minimum genuinely control all the others — every direction's bending is a weighted blend of just k_1 and k_2. This is why two numbers suffice where you might have feared needing infinitely many.
Euler's formula for normal curvature k_n(theta) = k_1 cos^2 theta + k_2 sin^2 theta theta = 0 (first principal direction) -> k_n = k_1 (max) theta = 90 (second principal direction) -> k_n = k_2 (min) the two principal directions are perpendicular, and are the lines of curvature through the point.
Where the principal directions come from: the Gauss map
Why should there be exactly two extremes, neatly perpendicular? The honest engine behind Euler's theorem is the Gauss map. Recall that at every point the unit normal N is itself a vector of length 1, so it names a point on the unit sphere. The Gauss map sends each surface point to the tip of its normal on that sphere, recording which way the surface faces there. As you move across the surface, the image point slides across the sphere — and the rate at which N changes as you step in direction v is what bending really is. Differentiating the Gauss map produces a linear operator on the tangent plane, the shape operator, that eats a direction v and returns how N tilts when you move that way.
Now the linear algebra you carried up the ladder pays off all at once. The shape operator is a symmetric linear operator (a fact that the second fundamental form encodes — it is precisely the shape operator measured against the first fundamental form). And a symmetric operator, by the spectral theorem, always has real eigenvalues and perpendicular eigenvectors. Those eigenvalues are the principal curvatures k_1 and k_2; those eigenvectors are the principal directions. The perpendicularity you were promised is not a happy accident — it is the spectral theorem wearing a geometric coat. The curves you get by always following a principal direction are the lines of curvature, the natural grid the surface draws on itself.
Two numbers, two famous averages: Gaussian and mean
Once you hold k_1 and k_2, two combinations of them carry almost all the fame in surface theory. Their product is the Gaussian curvature K = k_1 * k_2, and its sign tells you the local shape at a glance. If K > 0 both principal curvatures bend the same way — a dome or a bowl, curving toward N in every direction, like a sphere. If K < 0 they bend opposite ways — a saddle, the Pringle, hill in one direction and valley in the perpendicular one. If K = 0 at least one principal curvature is zero — the surface is flat in some direction, like a cylinder or a flat plane. One number, three verdicts.
Their average is the mean curvature H = (k_1 + k_2) / 2, and it governs an entirely different question: area. Surfaces with H = 0 everywhere are the minimal surfaces — the shapes a soap film spontaneously settles into when it spans a wire loop, because zero mean curvature is exactly the condition for locally least area. So K answers 'what does it look like here?' while H answers 'how would a soap film behave here?' — two faces of the same pair of numbers. A point where the two principal curvatures happen to be equal, k_1 = k_2, is an umbilic point: there the surface bends the same in every direction, with no preferred principal direction at all, exactly as happens at every single point of a sphere.
A quick sphere of radius r makes all of this concrete. By symmetry every direction bends identically, so k_1 = k_2 = 1/r (a bigger sphere is gentler, so smaller curvature — the 1/r is honest). Then K = 1/r^2, a positive constant matching its bowl-everywhere look, and H = 1/r. Every point is umbilic, as promised. Try the same on a cylinder of radius r: along the straight axis the slice is a straight line with k = 0, while around the circular waist k = 1/r, so K = 0 * (1/r) = 0. The cylinder has zero Gaussian curvature — the precise statement of the fact we opened with, that you can unroll it flat onto paper without distortion.
What you can now do, and the bombshell ahead
- At a point P, find the unit normal N and the tangent plane; this fixes the extrinsic frame in which bending is measured.
- Slice in each direction v to read the normal curvature k_n(v) — the signed curvature of the normal section through N and v.
- Take the max and min over all directions to get the principal curvatures k_1, k_2, achieved along perpendicular principal directions (the eigenvalues and eigenvectors of the shape operator).
- Combine them: product gives Gaussian curvature K = k_1 k_2 (the local shape), average gives mean curvature H = (k_1 + k_2)/2 (the soap-film question).
Be honest about the limits of what we have shown. We have defined these curvatures and computed a couple of clean examples, but the Euler formula and the spectral structure rest on the second fundamental form and a careful differentiation of the Gauss map — machinery we have sketched in spirit rather than proved in full. That is appropriate for an introductory rung; the ideas here are faithful, but the airtight proofs live a level deeper. What matters is that you now have a true mental picture: a surface bends by a different amount in each direction, two perpendicular directions catch the extremes, and two numbers summarise it all.
And here is the bombshell the next guide detonates. Everything in this guide was avowedly extrinsic — we had to leave the surface, look at N in the surrounding space, and watch it tilt. The Gaussian curvature K = k_1 k_2 is built from two of those extrinsic numbers, so surely K too is extrinsic, invisible to the flatlander. Gauss proved otherwise. His Theorema Egregium — the 'remarkable theorem' — shows that K can be computed entirely from the first fundamental form, from intrinsic measurements alone. A flatlander who can only measure distances and angles, never seeing the third dimension, can nevertheless feel the Gaussian curvature of their world. That is why the cylinder (K = 0) can be unrolled onto paper but the sphere (K = 1/r^2) can never be flattened without tearing — and it is the heart of the next guide.