Manifold · Visual Differential Geometry

Fields of slabs
you pierce

Forget arrows for a moment. Needham draws a 1-form not as a vector but as a field of sheets — a stack of parallel surfaces filling space. To evaluate the form on a vector, you don't take a dot product; you just count how many sheets the vector punches through. Out of that one picture falls gradient, curl, flux, and a single theorem — Stokes — that eats Green's, Gauss's, and even Gauss–Bonnet whole.

Scroll to begin

00 A 1-form is a stack of sheets

Counting, not projecting.

Here is a 1-form, drawn as Needham draws it: a family of evenly spaced sheets (here, lines). The form is the field of sheets, not any one of them. Now grab the arrowhead and drag. The number the form spits out is simply how many sheets your arrow crosses — positive if you go with the grain, negative against it. A long arrow that runs along the sheets crosses none and evaluates to zero; the same arrow turned across them rings up its full count.

That's the entire definition of "a 1-form eats a vector and returns a number." No magnitude, no cosine — just a tally of pierced sheets. And turn on the tone: a dense stack of sheets is a fine grating, so dragging across it ticks faster and rings higher, exactly like a card in bicycle spokes. A 1-form is something you can hear.

ω(arrow) = 0.0 sheets tilt

Drag the arrowhead. tilt is a constant form (evenly spaced parallel sheets). bowl and saddle come from functions — their sheets bunch up where the function changes fast. Sheet density = how strong the form is there.

01 Forms from functions

The sheets are just contour lines.

The cleanest way to get a 1-form is to take a function f and form its differential df. Its sheets are nothing exotic — they're the contour lines of f, the curves where f holds a constant value, drawn at equal steps. Where f climbs steeply the contours crowd together (dense sheets, strong form); where f is flat they spread apart.

And evaluating df on an arrow from A to B is then obvious: the number of contour lines you cross is just the change in height,

df(v) = f(B) − f(A) = sheets crossed the directional derivative, read off as a count. The "bowl" above is f = x²+y² (circular contours); the "saddle" is f = xy (hyperbolic contours).

This is why the gradient and the contour map carry the same information drawn two ways: the gradient is the arrow pointing straight across the densest sheets, and the 1-form is the sheets. Needham's quiet point is that the sheets are the more honest object — they're what you actually integrate, and they generalize to curved spaces where "the gradient vector" needs a metric and the sheets don't.

★ The boundary law

What happens inside shows up on the edge.

Now the theorem the whole subject is built to deliver. Walk a closed loop and add up the form as you go — the net sheets you pierce all the way around, ∮ω. For a form that came from a function (ω = df), you end where you started, on the same contour, so the net is zero: what you gain on one side you give back on the other. The loop is conservative.

But some forms don't come from any function — they swirl. For those, going around nets a nonzero total, and that total equals the swirl enclosed by the loop. That enclosed swirl is the exterior derivative dω — itself a 2-form, a honeycomb of tubes — and the equality is Stokes' theorem:

∮_∂S ω = ∬_S dω the form summed around the boundary = its exterior derivative summed over the interior. Drag the loop and watch the two numbers stay equal.

∮ω boundary 0.00 ∬dω interior 0.00

smallbig

Drag the loop's center; drag the slider to resize it. In gradient mode both readouts sit at zero — no swirl anywhere. Switch to swirl and the boundary circulation exactly matches the curl enclosed, and both grow with the area. That equality is Stokes.

03 One grand law

Stokes eats everything.

The single line ∮_∂S ω = ∬_S dω is the whole of vector calculus wearing one coat. Let the dimensions and the form change and it becomes every integral theorem you were made to memorize separately:

∫_∂S ω = ∫_S dω ⟹ gradient · Green's · Stokes' curl · divergence f(b)−f(a)=∫f′; circulation=∬curl; flux out=∭div. One law, four costumes — pick the degree of the form.

A 0-form is a function; its d is the gradient (the fundamental theorem of calculus). A 1-form's d is the curl (Green's and Stokes'). A 2-form's d is the divergence (Gauss's theorem). The wedge product ∧ stacks the sheets into honeycombs of tubes, and the operator d always means the same thing: take the boundary. "The integral of a derivative over a region equals the original over the boundary" — said once, it covers them all.

And it reaches back into this very wing. Curvature, it turns out, is a 2-form; Gauss–Bonnet is Stokes' theorem applied to it, which is why summing curvature over a closed surface gave a boundary-less answer fixed by topology. The pictures connect: holonomy is the boundary integral, curvature is its d. Forms are the language in which all of it is one sentence.

One act remains: geometry rebuilt entirely from forms — Cartan's moving frames, the connection and curvature as forms, and the final payoff where parallel transport through curved spacetime becomes gravity, with Maxwell's equations collapsing to dF = 0.

After Tristan Needham, Visual Differential Geometry and Forms (Princeton, 2021). A lesson in the Manifold wing of Reverbs.