Closed, Exact & the Poincaré Lemma

Two questions you can ask of any form

From the previous guide you carry one fact in your pocket: the exterior derivative d turns a k-form into a (k+1)-form, and applying it twice always gives zero, d of d of anything is 0. With d in hand, two natural yes-or-no questions present themselves about a form omega. First: is d omega = 0? Second: is omega itself equal to d of some other form? These are not the same question, and the entire subject of this guide lives in the gap between them.

Give the two properties names. A form omega is closed if d omega = 0 — the derivative d kills it, leaving nothing. A form omega is exact if there exists some form alpha with omega = d alpha — omega is itself the output of d, born as a derivative. So closed means d sends omega to zero on the way up; exact means omega arrived as the d of something one degree lower. One looks forward through d, the other looks backward through d. The drama is in how, and whether, these two notions match.

You have met both ideas before in disguise, back in the vector-calculus rung. A conservative field is a gradient field F = nabla f — that is exactly the 1-form omega = F . dr being exact, with the potential f playing the role of alpha. And the test you used, 'curl F = 0' or 'the mixed partials match', is precisely the statement that omega is closed. So the closed-versus-exact question is the grown-up, all-dimensions version of a problem you already half-solved by hand. Forms just let you ask it cleanly for every degree at once.

Exact implies closed — for free

One direction is automatic and costs nothing. Suppose omega is exact, so omega = d alpha for some alpha. Then d omega = d(d alpha) = 0, because applying d twice always vanishes. So every exact form is closed, full stop, in every dimension and on every space. There is no hypothesis to check, no smoothness to worry about beyond what you already need to write d. The single fact d-squared = 0 does all the work: it is the engine that makes 'exact' a special case of 'closed'.

This is genuinely useful as a negative test. If you ever compute d omega and get something nonzero, you may immediately conclude omega is not exact — there is no point hunting for a potential alpha, because none can exist. In the vector-calculus translation this is the everyday move: if curl F is not the zero vector, F cannot be a gradient, so stop looking for a potential. 'Exact implies closed' run in reverse is 'not closed implies not exact', and that contrapositive saves a lot of wasted effort.

The Poincaré lemma: locally, closed is exact

Now the converse, under a condition. The Poincaré lemma says: on a region that is contractible — one you can continuously shrink to a single point without leaving it, like a ball, a cube, or all of n-dimensional space — every closed form is exact. So locally there is no gap at all: if d omega = 0 there, then omega = d alpha for some alpha you can actually build. The word 'locally' is the safety valve: any nice region is contractible in a small enough neighborhood of each point, so the lemma always applies in the small.

The proof is not a mystery — it hands you an actual recipe for the potential. For a 1-form on a star-shaped region, you simply integrate omega along the straight ray from the center to each point; the resulting function is the alpha you wanted, and differentiating it gives omega back. This is the exact higher-dimensional echo of the fundamental theorem of calculus, where you recover a function as the integral of its derivative. The 'homotopy operator' that does this for k-forms is just that ray-integration done systematically, degree by degree.

The famous hole: the angle form on the punctured plane

Here is the one example everyone remembers, and it is worth carrying for life. On the plane with the origin removed, consider the 1-form omega = (-y dx + x dy)/(x^2 + y^2). A short computation shows d omega = 0 everywhere it is defined, so omega is closed. It even looks exact: away from a cut, omega = d(theta), where theta is the polar angle. The trouble is that theta is not a single honest function on the whole punctured plane — walk once around the origin and it jumps by 2 pi. There is no global alpha. So omega is closed but not exact, and the obstruction is precisely the missing point, the hole.

omega = (-y dx + x dy) / (x^2 + y^2)     on the plane minus the origin

   d omega = 0  everywhere defined   ->  CLOSED

   Integrate omega around a loop:
     loop NOT enclosing the origin  ->  integral = 0      (looks exact here)
     loop enclosing the origin once ->  integral = 2 pi   (cannot be exact!)

   An exact form would integrate to 0 around EVERY closed loop.
   The nonzero 2 pi is the hole, detected.

The loop integral is the detector. Around a loop missing the origin it gives 0 (consistent with exactness); around a loop circling the origin it gives 2 pi. Since an exact form must integrate to zero over every closed loop, that 2 pi proves omega is closed yet not exact — and counts how the loop wraps the hole.

Why does the loop integral decide it? Because if omega were exact, omega = d alpha, then its integral around any closed loop would be alpha-at-the-end minus alpha-at-the-start — and on a closed loop those are the same point, so the answer would have to be zero. A nonzero loop integral is therefore a certificate of non-exactness. This is again the conservative-field story you know: a path-independent, zero-around-loops field is exactly a gradient. The punctured plane is where that story first breaks, and forms tell you precisely why.

The gap is the shape: a first look at cohomology

Step back and admire what just happened. Closed forms are easy to find; exact forms are a strict subset of them; and on a contractible region the two sets coincide exactly. So the only way they can differ is if the region is not contractible — if it has holes. The size of the discrepancy, 'closed forms modulo exact forms', is therefore a number you can compute by calculus that reports a fact of pure shape. That quotient is called de Rham cohomology, and it is one of the most beautiful bridges in mathematics: differentiation, which feels local and analytic, ends up counting the global holes of a space.

The punctured plane illustrates the dictionary perfectly. It has one essential hole, and correspondingly there is essentially one closed-but-not-exact 1-form — our angle form omega — with every other such form being a multiple of it plus an exact piece. 'How many independent closed-but-not-exact k-forms exist' literally counts the k-dimensional holes: one-dimensional holes are loops you cannot fill, two-dimensional holes are voids like the inside of a sphere. The calculus you run is identical everywhere; what changes is the answer, and the answer is the topology.

Compute d omega. If it is nonzero, omega is not closed and certainly not exact — you are done, no potential exists.
If d omega = 0, omega is closed. Now ask about the domain: is it contractible (a ball, a box, all of space, no holes)?
If the domain is contractible, the Poincaré lemma guarantees omega is exact — build alpha by integrating omega along rays from a chosen center.
If the domain has holes, test exactness directly: integrate omega around loops circling each hole. A nonzero answer means closed but not exact — and it is measuring that hole.