Closed and exact forms; the Poincaré lemma; the angle 1-form

Intuition Beginner

Some quantities you measure along a path depend only on where you start and where you stop, not on the route. The change in your altitude as you hike is like this: loop back to camp and the net climb is zero, no matter how winding the trail. A measurement with a "height" behind it that it merely reports the change of is called exact. The height itself is the potential.

There is a weaker, more local condition. A measurement can be free of any local swirl — pinch off a tiny loop anywhere and it reads zero around that loop — and still fail to come from a single global height. Such a measurement is called closed. Every exact measurement is closed, because reporting changes in a height never produces local swirl.

The interesting question is the reverse. When does closed force exact? On a region with no holes, always. On a region with a hole, a closed measurement can circle the hole and refuse to be the change of any single height.

Visual Beginner

Picture two regions side by side. On the left, a solid disk with no holes; on the right, the same disk with a single point punched out at the center, so a small puncture remains.

On the solid disk, draw any small loop and the swirl through it cancels: closed and exact agree. On the punctured disk, a measurement can wind once around the missing point, adding up to the same nonzero amount on every loop that encircles the hole. That winding is what no global height can match, because a height would have to increase by a fixed step each lap and never reset.

Worked example Beginner

Take the measurement that adds up $2x$ times each step east plus $2y$ times each step north, written $2xdx+2ydy$. Ask whether it comes from a height.

Try the height $h=x^2+y^2$. Moving east changes $h$ at the rate $2x$; moving north changes it at the rate $2y$. These match the two pieces of the measurement exactly. So the measurement is the change of $h$, and it is exact.

Now total it along any closed loop. Start and end at the same point, so $h$ returns to its starting value, and the net total is $0$. Check a square loop from $(0,0)$ to $(1,0)$ to $(1,1)$ to $(0,1)$ and back: the four leg totals are $1$, then $2$, then $-1$, then $-2$, summing to $0$.

What this tells us: when a measurement comes from a height, every closed loop totals zero, and finding the height is the whole game.

Check your understanding Beginner

Formal definition Intermediate+

Let $U\subseteq {\mathbb{R}}^n$ be open and let ${\Omega }^k(U)$ denote the smooth $k$-forms on $U$. The exterior derivative is the sequence of linear maps $d:{\Omega }^k(U)\to {\Omega }^{k+1}(U)$ with $d\circ d=0$.

A form $\omega \in {\Omega }^k(U)$ is closed if $d\omega =0$. It is exact if there exists $\eta \in {\Omega }^{k-1}(U)$ with $\omega =d\eta$; such an $\eta$ is a primitive (or potential) of $\omega$. By the relation $d\circ d=0$, every exact form is closed, so the de Rham group $$ H^k_{\mathrm{dR}}(U) = \frac{\ker\big(d:\Omega^k(U)\to\Omega^{k+1}(U)\big)}{\operatorname{im}\big(d:\Omega^{k-1}(U)\to\Omega^k(U)\big)} $$ measures the failure of closed to imply exact ^{[Shifrin Ch. 8 §8.5]}. A class in $H_{\mathrm{dR}}^k(U)$ is a closed $k$-form modulo exact forms.

An open set $U\subseteq {\mathbb{R}}^n$ is star-shaped with respect to a point $p\in U$ if for every $x\in U$ the whole segment $\{p+t(x-p):t\in [0,1]\}$ lies in $U$. Convex sets are star-shaped with respect to each of their points; ${\mathbb{R}}^n$ itself is star-shaped. The star-shaped hypothesis is the geometric input to the Poincaré lemma below.

The canonical closed-but-not-exact form lives on the punctured plane $U={\mathbb{R}}^2\setminus \{0\}$. The angle 1-form (or winding form) is $$ \theta = \frac{x,dy - y,dx}{x^2 + y^2}. $$ Its name records that on the upper half-plane, where one may write $\varphi =\arctan (y/x)$ as a single-valued function, $\theta =d\varphi$; but no such $\varphi$ extends to a single-valued smooth function on all of $U$, because the angle increases by $2\pi$ each time one circles the origin. A common slip is to conclude from the local formula $\theta =d\varphi$ that $\theta$ is exact on $U$; it is only locally exact, and the obstruction to a global primitive is precisely the puncture.

Key theorem with proof Intermediate+

Theorem (Poincaré lemma on a star-shaped set). Let $U\subseteq {\mathbb{R}}^n$ be open and star-shaped with respect to the origin. Then every closed $k$-form on $U$ with $k\ge 1$ is exact. Equivalently, $H_{\mathrm{dR}}^k(U)=0$ for all $k\ge 1$.

Proof. The argument is constructive: a single linear operator $K:{\Omega }^k(U)\to {\Omega }^{k-1}(U)$, the homotopy operator, produces the primitive. For a $k$-form written in coordinates as $$ \omega = \sum_{i_1 < \cdots < i_k} a_{i_1\cdots i_k}(x), dx^{i_1}\wedge\cdots\wedge dx^{i_k}, $$ define $$ (K\omega)(x) = \sum_{i_1 < \cdots < i_k} \sum_{r=1}^{k} (-1)^{r-1}\left(\int_0^1 t^{k-1} a_{i_1\cdots i_k}(tx), dt\right) x^{i_r}; dx^{i_1}\wedge\cdots\wedge\widehat{dx^{i_r}}\wedge\cdots\wedge dx^{i_k}, $$ where the hat means the factor is omitted. The integral over $t\in [0,1]$ runs along the radial segment from the origin to $x$; star-shapedness with respect to the origin guarantees $tx\in U$ for $t\in [0,1]$, so $a_{i_1\cdots i_k}(tx)$ is defined and the integral converges. The operator $K$ is the integration of the pullback of $\omega$ along the straight-line homotopy $H(x,t)=tx$ from the constant map to the identity, contracted against the radial vector field ${\partial }_t$.

The defining identity is $$ d(K\omega) + K(d\omega) = \omega \qquad \text{for } k \ge 1. $$ Granting it, the theorem follows at once: if $\omega$ is closed then $d\omega =0$, so $K(d\omega )=0$ and the identity reduces to $d(K\omega )=\omega$. Thus $\eta =K\omega$ is a primitive, and $\omega$ is exact.

It remains to verify the identity. It suffices to check it on a single term, since $d$, $K$, and the right-hand side are all linear. Take $\omega =a(x)d{x}^I$ with $I=(i_1<\cdots <i_k)$. Differentiating $K\omega$ and using the product rule on each $x^{i_r}$ factor produces two kinds of terms: those where $d$ lands on the explicit $x^{i_r}$, giving $d{x}^{i_r}$, and those where $d$ lands inside the integral on $a(tx)$, pulling down a factor $t$ from the chain rule. The first kind, summed over $r$, reassembles $$ k\left(\int_0^1 t^{k-1} a(tx),dt\right) dx^I, $$ after the alternating signs cancel the reordering signs. Computing $K(d\omega )$ produces an integral of ${\sum }_j{\partial }_{j}a(tx)x^j$ against the same wedge, whose terms split into one piece matching the second kind from $d(K\omega )$ with the opposite sign — these cancel — and one piece equal to $$ \left(\int_0^1 \frac{d}{dt}\big(t^{k} a(tx)\big),dt - k\int_0^1 t^{k-1} a(tx),dt\right) dx^I, $$ using $\frac{d}{dt}(t^{k}a(tx))=k{t}^{k-1}a(tx)+t^k{\sum }_j{\partial }_{j}a(tx)x^j$. Adding $d(K\omega )$ and $K(d\omega )$, the two $k{\int }_0^1{t}^{k-1}a(tx)dt$ contributions cancel against this, leaving $$ \left(\int_0^1 \frac{d}{dt}\big(t^{k} a(tx)\big),dt\right) dx^I = \big(t^k a(tx)\big)\Big|_{t=0}^{t=1}; dx^I = a(x), dx^I = \omega, $$ since $k\ge 1$ kills the boundary term at $t=0$. The identity holds, and the theorem is proved. $\square$

Bridge. The homotopy operator $K$ built here is the prototype that builds toward 03.04.06 (de Rham cohomology), where the same $dK+Kd=\mathrm{id}-(\text{pullback by a homotopy endpoint})$ template proves homotopy invariance of $H^{\ast }$, and it appears again in 03.04.05 (Stokes' theorem), whose period pairing turns the closed-not-exact angle form into a nonzero integral. The local vanishing it establishes is exactly the input that 03.04.20 (surface integrals and flux) needs for its sphere computation, and the obstruction it isolates is the seed of 03.06.06 (Chern-Weil theory), where global non-exactness of locally exact curvature data becomes a characteristic class.

Exercises Intermediate+

Lean formalization Intermediate+

lean_status: none — Mathlib carries the alternating-form de Rham complex and the nilpotence $d\circ d=0$, but it does not package the closed-versus-exact dichotomy with the explicit star-shaped homotopy operator $K$ and the cochain identity $d(K\omega )+K(d\omega )=\omega$ that proves the Poincaré lemma constructively.

A formalization route would need: the radial homotopy $H(x,t)=tx$ on a star-shaped domain, the integral operator $K$ with the contraction against ${\partial }_t$, the chain-homotopy identity above, and the closed-not-exact witness — the angle form $\theta$ on ${\mathbb{R}}^2\setminus \{0\}$ with $d\theta =0$ and ${\int }_{S^1}\theta =2\pi$ — together with the period pairing certifying a nonzero class in $H^1({\mathbb{R}}^2\setminus \{0\})$.

Advanced results Master

The Poincaré lemma is the local triviality of the de Rham complex, and three facts organise what survives globally.

First, the lemma is homotopy invariance in disguise. The operator $K$ of the Key theorem is the contraction of the homotopy $H(x,t)=tx$ between the constant map $c_0$ and the identity. For any smooth homotopy $H:U\times [0,1]\to V$ between $f_0$ and $f_1$, the same construction yields $K:{\Omega }^k(V)\to {\Omega }^{k-1}(U)$ with $f_1^{\ast }-f_0^{\ast }=dK+Kd$ on forms ^{[Spivak Ch. 4]}. A star-shaped $U$ is contractible — the identity is homotopic to a constant — so ${\mathrm{id}}^{\ast }-c^{\ast }=dK+Kd$, and on closed forms of positive degree the right side is $d(K\omega )$ while $c^{\ast }\omega =0$, recovering $\omega =d(K\omega )$. Thus $H_{\mathrm{dR}}^k$ depends only on homotopy type, and contractible sets have $H^k=0$ for $k\ge 1$.

Second, the failure of the lemma on the punctured plane is measured by periods, and the period map is an isomorphism. For a closed 1-form $\omega$ on ${\mathbb{R}}^2\setminus \{0\}$, the period is $P(\omega )={\int }_{S^1}\omega$. Stokes' theorem makes $P$ vanish on exact forms, so it descends to a linear map $P:H^1({\mathbb{R}}^2\setminus \{0\})\to \mathbb{R}$. Surjectivity holds because $P(\theta )=2\pi$. Injectivity holds because a closed 1-form with zero period has a single-valued primitive: fixing a basepoint, the path integral $f(x)={\int }_{x_0}^x\omega$ is independent of path, since any two paths differ by a loop whose integral is a multiple of the period. Hence $H^1({\mathbb{R}}^2\setminus \{0\})\cong \mathbb{R}$, generated by the class $[\theta ]$, and the integer $\frac{1}{2\pi }{\int }_{\gamma }\theta$ is the winding number of a loop $\gamma$ about the origin.

Third, the lemma converts the vector calculus of ${\mathbb{R}}^3$ into one complex. Under the dictionary sending functions to $0$-forms, fields $F$ to the 1-form ${\alpha }_F=F^{1}dx+F^{2}dy+F^{3}dz$, fields to the 2-form ${\omega }_F={\iota }_F(dV)$, and functions to $3$-forms via $gdV$, the exterior derivative becomes the gradient, curl, and divergence in succession: $$ \Omega^0 \xrightarrow{;d = \mathrm{grad};} \Omega^1 \xrightarrow{;d = \mathrm{curl};} \Omega^2 \xrightarrow{;d = \mathrm{div};} \Omega^3. $$ The identities $\mathrm{curl}\mathrm{grad}=0$ and $\mathrm{div}\mathrm{curl}=0$ are both instances of $d^2=0$. On a star-shaped domain the Poincaré lemma reads: an irrotational field ($\mathrm{curl}F=0$) is a gradient, and a solenoidal field ($\mathrm{div}F=0$) is a curl. The conservative-vector-field criterion — that $\mathrm{curl}F=0$ on a simply connected domain implies $F=\nabla f$ — is exactly $H^1=0$ for such domains, and the angle form supplies the standard counterexample on the non-simply-connected punctured plane.

Synthesis. Closed and exact are the kernel and image of $d$, and the Poincaré lemma is the statement that on a star-shaped — hence contractible — open set the two coincide above degree zero, proved constructively by the radial homotopy operator $K$ satisfying $dK+Kd=\mathrm{id}$ on positive-degree forms. The operator is not an artifact of the proof: it is the contraction of the straight-line homotopy, and the same chain-homotopy identity, written for an arbitrary smooth homotopy, is precisely what makes de Rham cohomology a homotopy invariant. The punctured plane is the minimal place where the lemma fails, and the angle form $\theta =(xdy-ydx)/(x^2+y^2)$ is its witness: closed because $d\theta =0$ off the origin, not exact because its period $2\pi$ around the generating loop is an obstruction no single-valued primitive can absorb. That period is an isomorphism $H^1({\mathbb{R}}^2\setminus \{0\})\cong \mathbb{R}$, and the integer it produces on a loop is the winding number. Read through the grad-curl-div dictionary, the whole apparatus is the assertion that on simply connected regions irrotational means conservative and solenoidal means a curl, with the angle form marking the exact point where topology, not analysis, controls the answer.

Full proof set Master

Proposition (exact implies closed). If $\omega \in {\Omega }^k(U)$ is exact, then $\omega$ is closed.

Proof. Write $\omega =d\eta$ for some $\eta \in {\Omega }^{k-1}(U)$. Then $d\omega =d(d\eta )=(d\circ d)\eta =0$ by the nilpotence of the exterior derivative. Hence $\omega$ is closed, and $\mathrm{im}(d)\subseteq \ker (d)$, so the de Rham quotient is defined. $\square$

Proposition (the angle form is closed but not exact). The 1-form $\theta =(xdy-ydx)/(x^2+y^2)$ on $U={\mathbb{R}}^2\setminus \{0\}$ satisfies $d\theta =0$, yet there is no $f\in C^{\infty }(U)$ with $\theta =df$.

Proof. Write $\theta =adx+bdy$ with $a=-y/(x^2+y^2)$ and $b=x/(x^2+y^2)$. Then $$ \frac{\partial b}{\partial x} = \frac{(x^2+y^2) - x(2x)}{(x^2+y^2)^2} = \frac{y^2 - x^2}{(x^2+y^2)^2}, \qquad \frac{\partial a}{\partial y} = \frac{-(x^2+y^2) + y(2y)}{(x^2+y^2)^2} = \frac{y^2 - x^2}{(x^2+y^2)^2}, $$ so $d\theta =({\partial }_{x}b-{\partial }_{y}a)dx\wedge dy=0$ on $U$. For non-exactness, parametrise the unit circle by $\gamma (s)=(\cos s,\sin s)$, $s\in [0,2\pi ]$. There $xdy-ydx$ pulls back to $\cos s(\cos sds)-\sin s(-\sin sds)=ds$ and $x^2+y^2=1$, so ${\gamma }^{\ast }\theta =ds$ and $$ \int_{S^1}\theta = \int_0^{2\pi} ds = 2\pi. $$ Were $\theta =df$ for a global $f\in C^{\infty }(U)$, Stokes' theorem on the closed loop $S^1=\partial (\text{nothing within }U)$ — more directly, the fundamental theorem of line integrals — would give ${\int }_{S^1}\theta ={\int }_{S^1}df=0$, since $\gamma$ is a loop. This contradicts $2\pi \ne 0$. Hence $\theta$ is closed but not exact, and $[\theta ]\ne 0$ in $H^1(U)$. $\square$

Proposition (period isomorphism on the punctured plane). The period map $P:H^1({\mathbb{R}}^2\setminus \{0\})\to \mathbb{R}$, $P([\omega ])={\int }_{S^1}\omega$, is a linear isomorphism; in particular $H^1({\mathbb{R}}^2\setminus \{0\})\cong \mathbb{R}$.

Proof. $P$ is well-defined on classes: if ${\omega }^{\prime }=\omega +dg$, then ${\int }_{S^1}{\omega }^{\prime }={\int }_{S^1}\omega +{\int }_{S^1}dg={\int }_{S^1}\omega$ because $S^1$ is a loop. Surjectivity: $P([\theta ])=2\pi$, so every real value is attained by a scalar multiple of $[\theta ]$. Injectivity: suppose $P([\omega ])=0$ for a closed $\omega$. Fix the basepoint $x_0=(1,0)$ and set $f(x)={\int }_c\omega$ along any smooth path $c$ in $U$ from $x_0$ to $x$. Two such paths $c_1,c_2$ form a loop $c_1-c_2$ homotopic in $U$ to an integer multiple $m$ of $S^1$; since $\omega$ is closed, its integral over the loop equals $m\cdot P([\omega ])=0$, so $f$ is well-defined. By construction $df=\omega$, so $\omega$ is exact and $[\omega ]=0$. Thus $\ker P=0$, and $P$ is an isomorphism. $\square$

Proposition (homotopy operator identity, general homotopy). Let $H:U\times [0,1]\to V$ be smooth with $H_0=f_0$, $H_1=f_1$. There is a linear $K:{\Omega }^k(V)\to {\Omega }^{k-1}(U)$ with $f_1^{\ast }\omega -f_0^{\ast }\omega =d(K\omega )+K(d\omega )$ for all $\omega \in {\Omega }^k(V)$.

Proof. Let ${\iota }_t:U\to U\times [0,1]$, ${\iota }_t(x)=(x,t)$. Every form on $U\times [0,1]$ decomposes uniquely as $\alpha +dt\wedge \beta$ with $\alpha ,\beta$ having no $dt$ factor. Define $K\omega ={\int }_0^1{\iota }_t^{\ast }({\iota }_{{\partial }_t}(H^{\ast }\omega ))dt$, the fiber integration of the ${\partial }_t$-contraction of the pullback. A direct computation of $d$ on $U\times [0,1]$, splitting into the $dt$-free and $dt$-carrying parts and applying the fundamental theorem of calculus in $t$, gives ${\iota }_1^{\ast }(H^{\ast }\omega )-{\iota }_0^{\ast }(H^{\ast }\omega )=d(K\omega )+K(d\omega )$. Since ${\iota }_1^{\ast }{H}^{\ast }=f_1^{\ast }$ and ${\iota }_0^{\ast }{H}^{\ast }=f_0^{\ast }$, the identity follows. The Key theorem's $K$ is this construction for $H(x,t)=tx$, $f_0=$ const, $f_1=\mathrm{id}$. $\square$

Connections Master

De Rham cohomology 03.04.06 is the global home of this unit: closed modulo exact is its definition, the Poincaré lemma is the computation $H^k(\text{contractible})=0$ that anchors every Mayer-Vietoris induction, and the homotopy operator $K$ is the engine of homotopy invariance.

The exterior derivative 03.04.04 supplies $d^2=0$, which makes exact imply closed; the grad-curl-div complex of ${\mathbb{R}}^3$ is this single operator read through the form dictionary, and $\mathrm{curl}\mathrm{grad}=0$, $\mathrm{div}\mathrm{curl}=0$ are its nilpotence.

Stokes' theorem 03.04.05 converts the period ${\int }_{S^1}\theta =2\pi$ into the obstruction to exactness, since an exact form integrates to zero over a closed loop; the same pairing certifies that $[\theta ]$ is nonzero.

Surface integrals and flux 03.04.20 reuse the closed-not-exact mechanism one degree up: the area form of $S^2$ is closed but not exact, generating $H^2(S^2)$, exactly as $\theta$ generates $H^1$ of the punctured plane.

Chern-Weil theory 03.06.06 globalises the obstruction: locally exact curvature data fails to admit a global primitive, and the resulting class is a characteristic class, the higher-degree descendant of the winding number.

Historical & philosophical context Master

The integrability conditions distinguishing locally exact from globally exact differential expressions were studied by Volterra in 1889, who isolated the closedness equations ${\partial }_i{a}_j={\partial }_j{a}_i$ as the local obstruction and recognised that multiply connected domains carry additional global obstructions ^{[Volterra 1889]}. Poincaré's Analysis Situs of 1895 organised the global theory: he introduced the language of periods of a closed differential expression over cycles and observed that a closed expression with all periods zero is a total differential, the statement that becomes the period isomorphism for $H^1$ ^{[Poincaré 1895]}. The lemma that bears his name — closed implies exact on a contractible domain — appears in this circle of ideas, and the punctured plane with its angle form is the canonical example separating the two notions.

The cohomological reading was completed by de Rham, whose 1931 thesis proved that the periods pair closed forms against cycles to give an isomorphism between the analytic invariant (closed modulo exact) and the topological one (real homology), so that the dimension of $H_{\mathrm{dR}}^k$ equals the $k$-th Betti number ^{[de Rham 1931]}. Spivak's 1965 Calculus on Manifolds gave the homotopy-operator proof its now-standard undergraduate form, and Shifrin's Multivariable Mathematics threads the conservative-field and angle-form computations to the form-theoretic statement at the level of a rigorous undergraduate course.

Bibliography Master

@book{shifrin2005multivariable,
  author    = {Shifrin, Theodore},
  title     = {Multivariable Mathematics: Linear Algebra, Multivariable Calculus, and Manifolds},
  publisher = {John Wiley \& Sons},
  year      = {2005}
}

@book{spivak1965calculus,
  author    = {Spivak, Michael},
  title     = {Calculus on Manifolds},
  publisher = {W. A. Benjamin},
  year      = {1965}
}

@article{poincare1895analysis,
  author    = {Poincar\'e, Henri},
  title     = {Analysis Situs},
  journal   = {Journal de l'\'Ecole Polytechnique},
  volume    = {1},
  pages     = {1--121},
  year      = {1895}
}

@article{derham1931analysis,
  author    = {de Rham, Georges},
  title     = {Sur l'analysis situs des vari\'et\'es \`a $n$ dimensions},
  journal   = {Journal de Math\'ematiques Pures et Appliqu\'ees},
  volume    = {10},
  pages     = {115--200},
  year      = {1931}
}

@article{volterra1889variabili,
  author    = {Volterra, Vito},
  title     = {Delle variabili complesse negli iperspazi},
  journal   = {Rendiconti della Reale Accademia dei Lincei},
  volume    = {5},
  year      = {1889}
}

@book{botttu1982differential,
  author    = {Bott, Raoul and Tu, Loring W.},
  title     = {Differential Forms in Algebraic Topology},
  publisher = {Springer},
  year      = {1982}
}