The key question is global: does the work done by a field depend only on the endpoints, or also on the route?
Definition 5.4.1
(Potential and path independence)
Let be open and continuous. A potential for on is a function with ; a field that has a potential is called conservative. The field is path independent on if takes the same value for all piecewise paths in with the same initial point and the same terminal point.
From a gradient to endpoint work
Let be piecewise . The line integral is
summed over the smooth pieces.
Theorem 5.4.2
(Fundamental theorem for line integrals)
If for a scalar function on , then for every piecewise path
Proof
(Chain rule on each smooth piece)
Let be a subdivision such that is on each . On such a piece the chain rule gives , a continuous function. The one-variable fundamental theorem therefore gives . Summing over telescopes, because is continuous at each , to .
Therefore the integral is independent of the path between fixed endpoints and every closed path has integral zero. Reversing a path reverses its integral; concatenating paths adds their integrals. These properties turn global statements about routes into algebraic statements about closed curves.
Path independence also constructs a potential
Theorem 5.4.3
(A potential from path independence)
Let be open and path connected, and let be continuous and path independent on . Fix and put . Then is well defined, , and satisfies . Any two potentials of on differ by a constant.
Proof
(Appending a segment to a route)
Every point can be reached from by a piecewise path, and the value of the integral does not depend on the chosen path, so is meaningful. To find its th derivative at , choose small enough that the segment from to stays in . Append that segment to a path from to and use path independence:
Continuity of justifies the limit, for positive and negative . Thus , and is because its partial derivatives are continuous. The same conclusion follows if every closed piecewise smooth path has zero integral: concatenate one path with the reverse of another to see that two paths with the same endpoints give equal integrals. On an open path-connected set, polygonal paths exist between any two points, so the definition is available throughout .
If two potentials exist on connected , their difference has zero gradient and is constant along any piecewise smooth path, hence constant. A constant is therefore the only freedom on each connected component. On a disconnected domain, different components may carry different constants.
Proof X-ray
Where path independence is used in the construction
Path independence is used twice, and the two uses are different. First, it makes a single number, rather than a value depending on the route chosen from . Second, it allows us to choose the route: to compute we may go to first and then along the short segment, so the difference quotient becomes exactly an average of over , with no leftover from the earlier part of the route. Openness of is needed so that this segment exists for small of both signs, and continuity of is what turns the average into . Connectedness is what lets be defined at every point from a single base point. If path independence fails, as for the field on the punctured plane in the counterexample below, the integral from to depends on the route, and no single-valued is produced.
Mixed partials: a necessary local check
For of class on an open set, a potential would satisfy , . Its second partials are continuous, so equality of mixed partials requires
In three dimensions the corresponding necessary condition is . Failure at even one point rules out a potential on any neighborhood containing that point. Passing the test is not automatically sufficient on every domain: it gives a local differential condition, whereas a loop may travel around a hole.
Quick check
(Does have a potential on any neighborhood of the origin?)
Compare and at the origin.
Solution · Answer
Here and , so at we get . A potential on a neighborhood of the origin would be and would force equality of these mixed partials at every point of the neighborhood, but the equality holds only on the line , which contains no open set. Hence there is no potential.
On a star-shaped open domain, the local condition is sufficient.
Proposition 5.4.4
(Potentials on star-shaped domains)
Let be open and star-shaped about , and let satisfy for all . Then
is a potential for on .
Proof
(Differentiating along the rays)
To check the formula, differentiate under the finite integral. Writing , the th derivative of the integrand is
The cross-partial equality justifies the middle step. Integration over gives , which is continuous, so is . Star-shapedness is not cosmetic: it keeps the entire segment inside the domain, where and its derivatives are defined. The differentiation under the integral sign is justified in the section below on differentiating the segment formula.
Integrating and recovering the missing function
Worked example 5.4.5
(Potential and endpoint work in the plane)
Let on . Determine a potential and the work from to .
Here . Since the domain is all of , the star-shaped criterion applies. Alternatively construct a potential directly: integrating in gives
Differentiate in and match : , so . Thus . We have and , up to the same constant, so the work is . Check the sign by keeping destination minus starting point. The arbitrary function is essential: integrating a partial derivative only determines the potential up to a function of the remaining variable.
Common mistake
(A constant of integration is a function of the other variables)
Writing for the field above omits the term , since this has . The omission can go unnoticed when two endpoints happen to have equal , as for and , but it shows at other endpoints. From to the wrong potential gives work , whereas along the axis and the work is , which the correct potential reproduces: .
Quick check
(Find a potential for and the work from to .)
Check and first, then integrate with respect to .
Solution · Answer
on , which is star-shaped. Integrating in gives , and matching gives , so is constant and . The work is .
The hole matters
Counterexample mode
Equal cross partials but no potential
One might claim that a field with on an open set always has a potential on that set. On let
The field is on , and direct differentiation gives
Nevertheless there is no global potential on . For the unit circle , , we have . Therefore
A gradient would have zero integral around this closed curve, contradiction. The loop surrounds the missing origin; the star-shaped theorem cannot be applied to the punctured plane. On the right half-plane, by contrast, is a valid potential, as direct differentiation verifies. The repaired statement is that the condition suffices on a star-shaped domain, such as a disk or the right half-plane, and a local formula need not extend to a single-valued potential on the full punctured domain.
Common mistake
(Extending across the punctured plane)
The formula is a potential only where (and it is not even defined on the axis). Using it as a potential on all of fails visibly: it takes the value at both and , so the endpoint formula would give work between them. But along the upper unit semicircle , , the work is , and along the lower semicircle, run with from to , it is . The two routes with the same endpoints give different values, so the work is path dependent on .
Quick check
(For the same field on the punctured plane, what is the integral around the circle , ?)
Compute directly.
Solution · Answer
and , so and the integral is , the same as for the unit circle. Any circle around the origin gives .
Using a route to detect failure
Worked example 5.4.6
(Two routes for a non-conservative field)
For compare the path from to first along the -axis then vertically with the reversed order of those two segments. On the first route, on the horizontal segment and on the vertical segment, so the integral is 0. On the second, the vertical segment contributes 0 and the horizontal segment has , giving integral 1. Thus work depends on path. The local test reaches the same conclusion: but .
Choosing a method
First identify the actual open domain and smoothness. A failed mixed-partial test or nonzero closed-loop integral disproves conservativeness. A successful local test proves a global potential only with an appropriate domain theorem, such as the star-shaped construction, or after a potential is explicitly constructed and differentiated. Finally evaluate destination minus origin; do not assume that a familiar formula remains defined across a deleted point or branch cut.
Simply connected domains suffice in space, but a missing line still blocks potentials.
A path to generalization
Which condition on the domain replaces star-shaped?
Simply connected domains suffice in space, but a missing line still blocks potentials.
In remove the -axis and let . Only the third component of its curl could be nonzero, and it is ; yet the unit circle in the plane has circulation , so there is no potential. Removing only a point creates no such obstruction: a standard theorem says that a curl-free field on a simply connected open set in , where every closed loop can be shrunk to a point inside the set, has a potential. Its proof uses Stokes' theorem and lies beyond this unit.
Why differentiating the segment formula is legitimate
Fix in the star-shaped domain. The segment from to is compact and lies in open , so a positive-radius tube around it also lies in : finitely many open balls covering the segment provide a uniform buffer. For all nearby , the segments from to lie inside that tube. On a compact closed tube within , the field and its first derivatives are continuous and uniformly continuous. Difference quotients of the finite-interval integrand therefore converge uniformly in , allowing the derivative to pass under the integral. This supplies the local justification behind the earlier computation rather than assuming that an integral with a variable argument is automatically differentiable.
Changing a piecewise parametrization by an increasing bijection leaves the oriented work unchanged by the substitution rule, under the usual piecewise regularity assumptions. A decreasing reparametrization reverses direction and changes its sign. By contrast, a scalar line integral uses and does not change sign on reversal. Reparametrization changes a description of the same oriented path; path independence is the stronger statement that different paths with the same endpoints give the same work.
A potential in three variables
Worked example 5.4.7
(Three-variable potential by successive matching)
Find a potential for
on . Integrating the first component with respect to gives ; the undetermined term may depend on both remaining variables. Matching the second component gives , hence . Matching the third gives , so . Thus
Differentiating the final expression in all three variables verifies the construction, including the dependence of the integration constants. For example, the work from to is . The explicit potential proves path independence; a separate curl test is unnecessary once this verification is complete.