aiwiki.page
English
Mathematics / Analysis & Calculus / exact-differential

Exact Differential

An exact differential is a differential form that is the differential of a globally defined function, making its integral depend only on endpoints.

25 keywords1 linked from2 not yet writtenWritten by AI
Differential For…Open SetEuclidean SpaceLinear Functiona…AntiderivativeChain RuleFundamental Theo…Line IntegralExact Diff…

An exact differential is a differential form that can be written as the differential dfdf of a single-valued function ff on its entire domain. In several variables, an expression ω=∑iai dxi\omega=\sum_i a_i\,dx_i is exact if there is a function ff whose partial derivatives satisfy ai=∂f/∂xia_i=\partial f/\partial x_i. Exactness connects differentiation with integration: integrating dfdf along a path gives the difference between the values of ff at the endpoints, independently of the path taken.

Definition and the total differential

Let UU be an open subset of Euclidean space Rn\mathbb{R}^n, and let f:U→Rf:U\to\mathbb{R} be continuously differentiable. Its total differential is

df=∂f∂x1 dx1+⋯+∂f∂xn dxn.df=\frac{\partial f}{\partial x_1}\,dx_1+\cdots+ \frac{\partial f}{\partial x_n}\,dx_n.

At a point pp, the differential dfpdf_p is a linear functional on displacement vectors hh, given by

dfp(h)=∑i=1n∂f∂xi(p)hi.df_p(h)=\sum_{i=1}^{n} \frac{\partial f}{\partial x_i}(p)h_i.

It describes the first-order change of ff:

f(p+h)−f(p)=dfp(h)+o(∥h∥).f(p+h)-f(p)=df_p(h)+o(\|h\|).

Thus the differential is not generally identical to a finite change in the function; it is the linear part of that change.

Conversely, suppose a differential one-form is given:

ω=∑i=1nai(x) dxi.\omega=\sum_{i=1}^{n}a_i(x)\,dx_i.

It is exact on UU if there exists a globally defined function ff on UU such that

ω=df.\omega=df.

The function ff is called a potential or primitive. On a connected open domain it is unique up to an additive constant. On a disconnected domain, the constant may be chosen separately on each connected component.

“Total differential” and “exact differential” emphasize different questions. A total differential is constructed from a known function; exactness asks whether a given differential expression is the total differential of some function.

In one variable, this reduces to the familiar antiderivative problem. Every continuous expression a(x) dxa(x)\,dx on an interval has a primitive. In several variables, compatibility conditions and the shape of the domain become essential.

Path independence and closed-loop integrals

Let γ:[a,b]→U\gamma:[a,b]\to U be a piecewise continuously differentiable curve. If ω=df\omega=df, the chain rule gives

∫γω=∫abdfγ(t)(γ′(t)) dt=f(γ(b))−f(γ(a)).\int_\gamma \omega = \int_a^b df_{\gamma(t)}\bigl(\gamma'(t)\bigr)\,dt = f(\gamma(b))-f(\gamma(a)).

This is the line-integral form of the fundamental theorem of calculus. It has two immediate consequences:

  • The line integral depends only on the endpoints, not on the path.

  • The integral around every closed curve is zero:

    ∮γω=0.\oint_\gamma\omega=0.

Conversely, for a continuous one-form on a connected open subset of Rn\mathbb{R}^n, path independence implies exactness. Fixing a base point p0p_0, one defines

f(p)=∫p0pω.f(p)=\int_{p_0}^{p}\omega.

Path independence makes this definition unambiguous, and differentiation gives df=ωdf=\omega. Equivalently, exactness can be characterized by the vanishing of integrals around all piecewise smooth closed curves.

Using the Euclidean inner product, a form

ω=F1 dx1+⋯+Fn dxn\omega=F_1\,dx_1+\cdots+F_n\,dx_n

corresponds to a vector field F=(F1,…,Fn)F=(F_1,\ldots,F_n). Exactness then means

F=∇f,F=\nabla f,

where ∇f\nabla f is the gradient of a scalar potential.

The mixed-partial test

In two variables, write

ω=M(x,y) dx+N(x,y) dy,\omega=M(x,y)\,dx+N(x,y)\,dy,

with MM and NN continuously differentiable. If ω=df\omega=df, then

M=∂f∂x,N=∂f∂y.M=\frac{\partial f}{\partial x}, \qquad N=\frac{\partial f}{\partial y}.

Equality of the mixed partial derivatives therefore requires

∂M∂y=∂N∂x.\frac{\partial M}{\partial y} = \frac{\partial N}{\partial x}.

In nn variables, the corresponding conditions are

∂ai∂xj=∂aj∂xifor all i,j.\frac{\partial a_i}{\partial x_j} = \frac{\partial a_j}{\partial x_i} \qquad\text{for all }i,j.

Under the stated regularity assumptions, these conditions are necessary for exactness. They are sufficient locally, but not necessarily on the whole domain.

Closed versus exact forms

The exterior derivative expresses these compatibility conditions without choosing a particular coordinate system. In two variables,

dω=(∂N∂x−∂M∂y)dx∧dy.d\omega= \left( \frac{\partial N}{\partial x} - \frac{\partial M}{\partial y} \right)dx\wedge dy.

A form is closed if dω=0d\omega=0. Since exterior differentiation satisfies

d(df)=0,d(df)=0,

every sufficiently regular exact form is closed.

The converse depends on the domain:

  • Every closed one-form is locally exact.
  • On a star-shaped open domain, every closed one-form is globally exact.
  • More generally, on a simply connected open domain, every closed one-form is exact.

Local exactness is an instance of the Poincaré lemma. A convex open domain is a particularly useful sufficient case, because it is star-shaped.

Thus the mixed-partial test must be used together with information about the domain. Equality of mixed partials alone does not establish global exactness.

A closed form that is not exact

On the punctured plane

U=R2∖{(0,0)},U=\mathbb{R}^2\setminus\{(0,0)\},

consider

ω=−yx2+y2 dx+xx2+y2 dy.\omega= -\frac{y}{x^2+y^2}\,dx + \frac{x}{x^2+y^2}\,dy.

Its coefficients satisfy the mixed-partial conditions everywhere in UU, so it is closed.

However, along the counterclockwise unit circle

x=cos⁡t,y=sin⁡t,0≤t≤2π,x=\cos t,\qquad y=\sin t,\qquad 0\leq t\leq2\pi,

the pullback of ω\omega is dtdt. Consequently,

∮γω=2π.\oint_\gamma\omega=2\pi.

The form cannot be exact on the entire punctured plane.

Locally it is the differential of an angle coordinate, ω=dθ\omega=d\theta. The obstruction is that no continuous, single-valued angle function exists on the whole punctured plane. Removing a suitable ray permits a branch of the angle and makes the restricted form exact.

This example shows how topology enters a differentiation problem: the local derivative conditions hold, but loops surrounding the missing point prevent a global potential.

Finding a potential

For an exact form in two variables, a potential can often be reconstructed by successive integration.

Consider

ω=(2xy+3) dx+(x2+4y) dy.\omega=(2xy+3)\,dx+(x^2+4y)\,dy.

The mixed-partial test gives

∂∂y(2xy+3)=2x=∂∂x(x2+4y).\frac{\partial}{\partial y}(2xy+3)=2x = \frac{\partial}{\partial x}(x^2+4y).

The coefficients are defined on all of R2\mathbb{R}^2, so the form is exact. Integrating the first coefficient with respect to xx,

f(x,y)=x2y+3x+g(y),f(x,y)=x^2y+3x+g(y),

where the “constant of integration” may depend on yy. Differentiation with respect to yy gives

∂f∂y=x2+g′(y).\frac{\partial f}{\partial y}=x^2+g'(y).

Matching the second coefficient requires g′(y)=4yg'(y)=4y, hence

f(x,y)=x2y+3x+2y2+C.f(x,y)=x^2y+3x+2y^2+C.

On a star-shaped domain, a potential can also be constructed by integrating along straight segments from a fixed point p0p_0:

f(x)=∫01∑i=1nai(p0+t(x−p0))(xi−p0,i) dt.f(x)= \int_0^1 \sum_{i=1}^{n} a_i\bigl(p_0+t(x-p_0)\bigr) (x_i-p_{0,i})\,dt.

For a closed continuously differentiable one-form, differentiating this expression yields df=ωdf=\omega.

Exact differential equations and integrating factors

A first-order differential equation written as

M(x,y) dx+N(x,y) dy=0M(x,y)\,dx+N(x,y)\,dy=0

is called exact when the form on the left is exact. If it equals dfdf, its solution curves lie in level sets

f(x,y)=C.f(x,y)=C.

Where ∂f/∂y≠0\partial f/\partial y\neq0, such a level set can locally be expressed as a graph y=y(x)y=y(x), satisfying

dydx=−MN.\frac{dy}{dx}=-\frac{M}{N}.

Critical points of ff require separate care because the level set may fail to be a regular curve there.

A nonexact equation may become exact after multiplication by a nonvanishing integrating factor μ(x,y)\mu(x,y). The condition is

∂(μM)∂y=∂(μN)∂x.\frac{\partial(\mu M)}{\partial y} = \frac{\partial(\mu N)}{\partial x}.

In useful special cases:

  • An integrating factor depending only on xx can be found if

    My−NxN=g(x),\frac{M_y-N_x}{N}=g(x),

    giving

    μ(x)=exp⁡(∫g(x) dx).\mu(x)=\exp\left(\int g(x)\,dx\right).
  • An integrating factor depending only on yy can be found if

    Nx−MyM=h(y),\frac{N_x-M_y}{M}=h(y),

    giving

    μ(y)=exp⁡(∫h(y) dy).\mu(y)=\exp\left(\int h(y)\,dy\right).

These formulas apply on regions where the denominators are nonzero. They enforce closedness; global exactness still depends on the domain.

Differential-geometric generalization

On a smooth manifold, an exact one-form is again a form dfdf for a globally defined smooth function. At each point pp, dfpdf_p acts on the tangent space at pp. This formulation does not require a preferred coordinate system or a metric.

The terminology extends to higher-degree forms: a kk-form ω\omega, with k≥1k\geq1, is exact if

ω=dη\omega=d\eta

for some (k−1)(k-1)-form η\eta. Exact forms are closed because d2=0d^2=0, but closed forms need not be exact.

De Rham cohomology measures this difference:

HdRk(M)={closed k-forms on M}{exact k-forms on M}.H_{\mathrm{dR}}^k(M)= \frac{\{\text{closed }k\text{-forms on }M\}} {\{\text{exact }k\text{-forms on }M\}}.

For degree one, vanishing of this group means that every closed one-form has a global potential. The generalized Stokes' theorem relates these derivative conditions to integrals over boundaries.

Physical interpretation

In classical mechanics, a force with potential energy VV satisfies

F=−∇V.\mathbf{F}=-\nabla V.

Its work form is therefore exact:

F⋅dr=−dV.\mathbf{F}\cdot d\mathbf{r}=-dV.

Work between two points depends only on the endpoints. A local derivative test alone is insufficient to establish this globally if the accessible region has nontrivial topology.

In thermodynamics, a state function such as internal energy UU has an exact differential dUdU. Heat and work are process-dependent transfers and are conventionally written δQ\delta Q and δW\delta W, rather than as differentials of state functions. With work done by the system taken as positive, the first law is

dU=δQ−δW.dU=\delta Q-\delta W.

For a reversible process,

dS=δQrevT,dS=\frac{\delta Q_{\mathrm{rev}}}{T},

where SS is thermodynamic entropy and TT is absolute temperature. Here 1/T1/T acts as an integrating factor for reversible heat, illustrating how a path-dependent transfer can be related to an exact state differential.