Motivation
An equation can be easy to integrate without being explicitly solved for a derivative. Sometimes its two terms are the partial derivatives of one function. Along a solution that function is constant, so finding the solution means identifying a level curve rather than integrating a slope directly.
There are three separate questions. Does a potential exist on the chosen domain? Can a nonexact equation be multiplied by a nonzero factor to obtain one? Does the resulting level curve give a graph with the intended independent variable? A successful integration must answer all three. A missing domain restriction can lose a solution, while a missing graph condition can turn a vertical tangent into a nonexistent differentiable graph.
Definitions
Definition 1.6.1
(Solution curves and solution graphs)
Let the coefficients be continuously differentiable on an open set in the plane. Write the implicit equation and its curve interpretation as
A solution curve is a continuously differentiable curve on an interval satisfying (1). A regular curve also has a nonzero tangent. Its parameter need not be either coordinate. A solution graph , with as independent variable, instead satisfies
Thus a vertical curve can satisfy the implicit equation without being such a graph. The implicit equation specifies tangent directions, not a preferred speed along a curve.
Definition 1.6.2
(Exactness and a legitimate integrating factor)
The equation is exact on its domain if it has a potential whose partial derivatives are the two coefficients:
An integrating factor on that domain is a continuously differentiable, nowhere-zero function for which the multiplied equation is exact:
Nonvanishing is part of the definition here. It makes the original and multiplied curve equations equivalent at every point of the stated domain.
Exactness implies equality of the cross derivatives. We call that equality the closedness test. It is a differential condition at each point; existence of one potential on an entire domain is a global condition. On a rectangle the test is sufficient, as the next construction shows.
Theorem/Proposition
Theorem 1.6.3
(A complete exactness test on a rectangle)
Let the domain be a product of two open intervals. The equation is exact on this rectangle if and only if the cross derivatives agree everywhere. For any chosen base point in the rectangle, a potential is
Every other potential differs from this one by a constant. The original form's integral along every piecewise continuously differentiable path depends only on the endpoints, and every closed-path integral is zero.
Proof
(Integrate along two coordinate segments)
If the potential exists, its second partial derivatives are continuous and equality of mixed partials gives the test. For sufficiency, both coordinate segments in (4) stay in the rectangle. Differentiating in gives ; differentiating in and using gives
These derivatives are continuously differentiable, so the constructed function has the claimed regularity. If two potentials have the same gradient, their difference is constant along horizontal and vertical segments, and hence constant throughout the rectangle. Finally, the chain rule on each piece of a path gives
This proves path independence and the closed-path conclusion. On a general open domain, exactness still implies (5); it is the sufficiency of the pointwise cross-derivative test that requires an appropriate domain argument.
Theorem 1.6.4
(Integrating factors depending on one coordinate)
On a rectangle where never vanishes, a nowhere-zero integrating factor depending only on exists if and only if the following quotient is a function of alone:
In that case all such factors are the nonzero constant multiples of
Similarly, on a rectangle where never vanishes, a factor depending only on exists if and only if
gives it, again up to a nonzero constant multiple. The quotients are continuous under the hypotheses. These tests classify only the stated one-coordinate factors, not all possible integrating factors.
Proof
(The factor equation is a scalar linear equation)
For a factor , closedness of the multiplied equation says
The left side depends only on , proving necessity. Conversely (6) solves this scalar linear equation and never vanishes. The multiplied coefficients pass the closedness test, and the rectangle theorem supplies a global potential there. Uniqueness up to a nonzero scalar follows by taking the ratio of two solutions of the scalar factor equation. The computation for reverses which product derivative contains the extra term:
It proves (7) in the same way. The signs in the two formulas are different because the derivative of the factor enters a different side of the equality.
Proof X-ray
An integrating factor repairs a product rule
The hypothesis is used once, to divide (8) by ; it is what turns the closedness condition into an equation for alone. The key move is that a factor of one coordinate contributes exactly one extra product-rule term, , so closedness of becomes the scalar linear equation . Necessity then reads off that must be independent of , because is.
Sufficiency closes in two separate steps that should not be merged: the exponential (6) is nowhere zero, which keeps the multiplied equation equivalent to the original one, and the rectangle theorem converts pointwise closedness of into a global potential. On a domain that is not a rectangle the first step survives but the second can fail, as the punctured-plane counterexample below shows.
If a denominator vanishes somewhere, that quotient test has not been justified there. A candidate factor may nevertheless extend smoothly across the zero. Check the undivided product identity directly before making any extension claim. Conversely, failure of both one-coordinate tests does not rule out a factor depending on both coordinates.
Locally near any point where the two coefficients are not both zero, yes, though rarely by a formula.
A natural next question
Does an integrating factor always exist?
Locally near any point where the two coefficients are not both zero, yes, though rarely by a formula.
Let and near a point. Define as the value at of the solution graph of through . Differentiable dependence on initial data makes twice continuously differentiable, is constant on solution graphs, and . Constancy forces to be parallel to , so is a local nowhere-zero factor. A global factor can still fail to exist, as on the punctured plane.
Theorem 1.6.5
(Level curves, local graphs and their restrictions)
Suppose a nowhere-zero integrating factor gives a potential on an open domain. A continuously differentiable curve in that domain solves the original implicit equation exactly when the potential is constant along it. At a point where , the level through that point is locally a unique continuously differentiable graph , and it solves (2). If instead , there is a local inverse graph .
When both coefficients vanish, this regular-level conclusion makes no assertion. The level may cross itself, collapse to a point, or have other singular behavior. Constancy of the potential alone does not select one global component or one global graph.
Proof
(Constancy along curves and the implicit function theorem)
The chain rule and (3) give
Nonvanishing gives the equivalence. At a point with the potential has a nonzero partial derivative in :
The implicit function theorem gives the local graph and its uniqueness as a level graph. Every nearby solution graph has the same constant potential by the chain rule, so this also proves uniqueness among such solution graphs. Interchanging the two coordinates proves the inverse-graph statement. The gradient is zero exactly when both original coefficients are zero, because the factor never vanishes. At those points the implicit function theorem cannot be invoked.
The linear first-order integrating factor fits this same framework. For continuously differentiable coefficient functions on an interval, write
Direct differentiation gives the potential relation
Thus the familiar linear formula is a special exactness construction. The displayed product rule remains valid for merely continuous scalar coefficients, although in that case the potential is generally only continuously differentiable; the cross-derivative theorem above has its stronger stated hypotheses. If the original equation was first divided by a nonconstant coefficient, the integrating factor for that normalized equation need not be the integrating factor for the original differential form.
Worked examples
Worked example 1.6.6
(A potential and the branch selected by initial data)
Consider
Here . Integrating in leaves an unknown function ; matching determines it:
The initial condition gives , so the solution lies on the level . Solving the quadratic gives the selected branch
The other quadratic root belongs to the same level but has . On the selected branch equals the positive square root in the formula, so it never vanishes. This branch is a differentiable solution graph for every real . The global conclusion comes from the explicit branch and its nonvanishing coefficient, not from the local implicit function theorem alone.
Worked example 1.6.7
(The factor changes when the equation is normalized)
For the nonexact equation
the factor and potential are
If we first divide the original equation by , the normalized graph equation and its usual linear factor are instead
These two factors differ because they multiply two different differential forms; their final potential is the same. On the stated positive-coordinate domain, the family already includes the zero solution, with .
There is also a separately defined extension problem: the polynomial original coefficients make sense on the whole plane. For that extended implicit equation, the original factor remains nonzero and its undivided product identity still holds everywhere. Its zero level is the union of the two coordinate axes. The horizontal axis is the zero solution graph, including through ; the vertical axis is an implicit solution curve that is not such a graph. At their meeting point both original coefficients vanish, so the regular-level theorem does not apply there. A graph with diverges as and has no finite graph extension through it. These statements concern the whole-plane extension, not a point inside the original half-plane domain.
Worked example 1.6.8
(A second-coordinate factor and a vertical solution)
Take and . The test for a factor gives
On a nonzero level the graph is
The zero level is the vertical line . It is a regular implicit solution curve, but it cannot be a graph . It is naturally the graph with as independent variable. The equation did not fail; the proposed choice of independent variable did.
Common mistakes / subtle points
Counterexample mode
Closedness on a punctured domain is not global exactness
On the plane with the origin removed, consider
The cross derivatives agree, but on the counterclockwise unit circle the integral is
Hence no single-valued global potential exists there. Local rectangles still have potentials, which are local choices of angle. Even multiplication by a continuous nowhere-zero factor on the whole punctured plane cannot fix this: its values on the connected circle have one sign, so the resulting circle integral cannot be zero. The domain obstruction survives such multiplication.
The punctured plane is an obstruction coming from the domain. The next two failures come from the factor itself, which must be nowhere zero and defined on the whole domain where the original equation is studied.
Counterexample mode
A factor that vanishes can create extra curves
Start with , whose solution curves are the horizontal lines. Multiply by a continuously differentiable factor that is zero on the lower half-plane:
The multiplied form is the differential of the displayed twice continuously differentiable potential. Every curve in the lower half-plane now has constant potential and solves the multiplied equation. For example, the graph on , with values between and , does so, but it does not solve . Exactness after multiplication is therefore insufficient without the nonvanishing requirement.
Common mistake
(Dividing by a coefficient that can vanish)
The equation below is regular even at , but the factor is natural only after dividing by , which excludes that zero:
Its nonzero branches are the exponential solutions with nonzero amplitude. The original zero solution is not represented by a finite level of that logarithmic potential. It must be recovered from the original equation. Alternatively an everywhere-nonzero factor gives a global potential:
This potential includes the zero solution. Different factors can have different domains of equivalence.
Common mistake
(Swapping the signs of the two quotient tests)
The test for uses and the test for uses . Using for a factor gives the reciprocal of a correct factor, which in general fails. For the correct quotient is and ; the wrong sign gives , and then while , so the multiplied form is not closed.
At a critical level, nonuniqueness or an isolated point is possible even for an exact form. The potentials and have, at level zero, respectively two crossing lines and an isolated point. At the points of the regular level , the level is smooth but has a vertical tangent. These are distinct reasons why an implicit level description must not automatically be advertised as one differentiable graph.
Concept lensStructural
Potential, direction and clock
A potential labels integral curves by a conserved value. A nonzero factor changes the coefficients without changing the allowed tangent directions. Neither operation chooses a clock, and a level value need not distinguish different components. The independent variable and regularity conditions decide which part of that geometry is a solution graph.
Practice and solutions
Quick check
(Find a potential for the coefficients below. Can the level through the origin be a differentiable graph ?)
Match both partial derivatives, then check the graph condition at the origin instead of only finding the level constant.
Solution · Answer
The explicit coefficients, potential and initial level are
At the origin , while . Any differentiable graph contained in this level would be a solution graph by the chain rule. But (2) at the origin would read , which is impossible. Thus the level itself contains no differentiable graph with that independent variable through the origin. It is nevertheless a unique local inverse graph , because .
Quick check
(Find an integrating factor depending only on for the coefficients below on , and solve the equation.)
Compute the quotient in (6) before integrating it.
Solution · Answer
The calculation is
Differentiate the potential in each coordinate to check the multiplied equation. Its derivative is positive throughout this domain, so every level gives a differentiable graph there. No continuation across the excluded boundary is supplied by this calculation.
Quick check
(On the positive quadrant, prove that the equation below has no factor depending only on or only on , but find a factor depending on both.)
Test the coordinate quotients, then factor out the common coordinate sum.
Solution · Answer
Write the coefficients and the two quotient derivatives explicitly:
Since , the quotient is not a function of alone; since , is not a function of alone. Both one-coordinate tests fail. Nevertheless
The coordinate tests are useful classifications within their restricted ansatz, not a nonexistence theorem for general integrating factors.
Quick check
(Suppose a nowhere-zero factor has potential on a rectangle. Let be twice continuously differentiable on an open interval containing the range of , with derivative nowhere zero there. Construct another factor and potential, and compare their level curves.)
Apply the chain rule to the composition and use the nonzero derivative on an interval.
Solution · Answer
The construction is
It is continuously differentiable and nowhere zero. The continuous derivative of the outer function has one sign on its interval, so that function is strictly monotone. Its levels correspond to the old potential levels, with different labels but the same curves and components. Thus integrating factors need not differ merely by constant multiples when dependence on both coordinates is allowed. The constant-multiple uniqueness in (6) and (7) applies only within the stated one-coordinate class and its nonzero-coefficient hypotheses.