Why the first variation need not vanish
Imagine a stretched curve whose endpoints are fixed and which must stay above a prescribed obstacle. Away from the obstacle it can move up or down. At a contact point, a downward displacement may be forbidden. A minimum therefore need not have zero first variation in every endpoint-zero direction. It must instead have a nonnegative first variation in every feasible direction into the admissible set.
This distinction changes the equation as well as the proof. The ordinary Euler equation survives on the open region where there is clearance. On contact, a nonnegative reaction can balance the applied source. The contact region is part of the unknown: imposing it in advance can give a curve with the wrong joining slopes, or a curve that passes through the obstacle.
We study a real, scalar, quadratic problem on an interval. This setting permits a complete existence proof, an exact characterization of the minimum and a nontrivial explicit contact region. It also lets us see precisely why checking a differential equation almost everywhere can miss a concentrated force at a corner.
The admissible set
Definition 9.1.1
(A curve above an obstacle)
Let be the real vector space of absolutely continuous functions on with and . Its inner product is and its norm is .
Assume with , and let be real. Define
We minimize over . The factor is part of our energy convention throughout.
These functions have specified continuous representatives, so the pointwise condition is meaningful. The derivative map identifies isometrically with the closed subspace ; its inverse is . Thus is complete. Cauchy–Schwarz gives the useful, deliberately nonoptimal bounds
The set is nonempty: is Lipschitz, has zero endpoints and lies above . It is convex because convex combinations preserve the pointwise inequality. It is closed in because convergence in implies uniform convergence by the first bound. These are substantive properties of this obstacle set; arbitrary closed constraint sets need not be convex.
Existence and the variational inequality
Theorem 9.1.2
(The unique minimum is characterized by an inequality)
Under the preceding assumptions, there is a unique minimizer . A curve is that minimizer if and only if
Moreover, for every the minimizer satisfies the quantitative gap
Proof
(Project onto a closed convex set, then take a one-sided derivative)
The functional is bounded on , since . Riesz representation supplies with for every . Completing the square gives
For completeness, the nearest point to in exists without assuming norm compactness. Put and choose whose distances tend to . Their midpoints belong to . The parallelogram identity implies
Completeness and closedness give a limit in , attaining the distance. The same midpoint identity shows uniqueness. Hence the completed square has a unique minimum.
If minimizes and , the segment is feasible for . Its energy has a minimum at the left endpoint , so its right derivative is nonnegative. This is exactly the stated variational inequality, not an equality.
Conversely, direct expansion gives for any
The inequality makes the final two terms nonnegative for feasible . This proves the gap and gives uniqueness again: zero derivative norm and zero endpoint values force .
Proof X-ray
Why convexity changes stationarity into an inequality
Convexity gives a feasible segment from the candidate toward every competitor. It gives only the right derivative at the initial point; the segment need not extend backwards. The exact quadratic expansion then turns this necessary inequality into a sufficient global comparison. Setting every first variation to zero would discard the contact force rather than derive it.
Free regions and reaction forces
Definition 9.1.3
(Free region, contact set and free boundary)
For the minimizer , the free region is and the contact set is . Continuity makes open. Its boundary points inside are free-boundary points: their locations are not prescribed in the problem.
For any set
If is the minimizer and , then for every , so . If instead is smooth and compactly supported in , the gap has a positive minimum on its support. Both signs of a sufficiently small multiple of are feasible, so . In particular
This is the weak equation on the free region. On the whole interval, means the functional on smooth compactly supported tests; it is nonnegative on nonnegative tests. We have not asserted that it is an ordinary function or that an arbitrary minimizer has a classical second derivative.
What nonnegative tests actually prove
The passage from an integral inequality to a pointwise sign is a one-sided version of the fundamental lemma. First suppose is continuous on an open interval and for every nonnegative . If , continuity makes on a small interval around . A nonzero, nonnegative smooth bump supported there then gives a strictly negative integral, a contradiction. Thus everywhere. With zero integrals for every test, apply the sign argument to both and to get .
For a merely locally integrable , the corresponding conclusion is almost everywhere, not at every chosen representative's value. To see why, fix a compact subinterval and convolve locally with a nonnegative smooth unit-mass mollifier . For small enough , the test is supported in for every . Therefore on . The standard approximation property of mollifiers gives in ; it follows from continuity of translations in and averaging those translations against . Since , the negative part satisfies
An exhaustion by compact intervals proves the claim on . If all test integrals vanish, apply the same argument to and . This is the test-function step used below when a density is known. Changing an function at a single point changes no test integral, so an everywhere sign conclusion would be too strong.
Worked example 9.1.4
(A positive reaction need not have a density)
For this example only, relax the earlier obstacle assumption to the Lipschitz obstacle on . Set and . The feasible set is nonempty (it contains ), convex and closed by the same uniform-convergence argument. The tent belongs to . Integrating on either side of zero gives, for ,
Every feasible satisfies , hence ; the quadratic gap above proves is the unique minimizer. This is a positive point reaction at the corner. It cannot equal for any and all smooth compactly supported : tests supported away from zero would force almost everywhere on each side, leaving a zero integral even for a test with . Thus positivity of the reaction functional alone does not supply an ordinary density.
Quick check
(Replace the tent by and keep . Find the reaction and decide whether minimizes.)
Integrate by parts separately on each side of the corner.
Solution · Answer
The same piecewise integration gives . For feasible , , so the variational inequality holds; the nonnegative quadratic gap proves the unique minimum. The reaction is concentrated at the corner and has no density by the same test argument.
Theorem 9.1.5
(An ordinary contact-force density certifies the minimum)
Suppose a candidate admits a function such that
Then is the unique minimizer. All the integrals are well defined because members of are bounded. The conditions say that the reaction is nonnegative and acts only where there is contact; the last condition is called complementarity.
Proof
(The density makes every first variation nonnegative)
For , complementarity replaces by inside the integral, and with gives
The previous theorem therefore shows that is the unique minimizer.
This is a sufficient certificate with an explicitly stated density assumption, not a claim that every reaction must have such a density. If a minimizer does have this density representation, positivity on tests gives almost everywhere, and the free-region equation gives almost everywhere on . Thus complementarity also follows in that case. These conclusions use the usual test-function characterization of a nonnegative or zero locally integrable function.
There is a convenient way to verify the representation for explicit curves. If and is up to the ends of finitely many subintervals, integration by parts on those pieces gives almost everywhere. Internal boundary terms cancel because the first derivative is continuous. Endpoint terms vanish because . Without the derivative matching, extra point evaluations remain; they cannot be hidden in an almost-everywhere formula.
Concept lensGeometric
Contact restricts directions, not just values
In a free region a small localized direction and its negative are allowed, so a linear first variation must be zero. On contact, raising the curve is allowed while lowering it may fail. A nonnegative reaction is exactly compatible with those one-sided directions. Contact alone does not mean a strictly positive reaction: a curve can touch an obstacle without needing a force there.
Worked examples
Worked example 9.1.6
(A minimum with a nonzero first variation)
Take and . For every feasible ,
Thus is the unique minimum, and its density is . Yet is positive for every nonzero nonnegative test . The first variation does not vanish in all directions. The unconstrained Euler solution has and zero endpoints, but lies below the obstacle. Solving that equation would have solved a different problem.
Worked example 9.1.7
(A parabolic obstacle with an unknown contact interval)
Let and with . Seek a symmetric curve that equals the obstacle on and is linear from each contact point to the corresponding endpoint. At the right contact point, matching the slope requires the outer line to be . Matching its value gives
The other algebraic root lies outside the interval. Define
This curve has zero endpoints and is . On the outer pieces, , so it really stays above the obstacle and has no additional contact there. Its contact set is exactly .
Piecewise integration by parts gives . This is nonnegative and supported on contact, so the certificate proves the unique global minimum. The joining equations were not merely a guess about the shape: they removed the concentrated reactions at the joins, and the certificate checked every admissible competitor.
With the convention , its value is
For and , , the outer pieces are , and the minimum is . When increases from zero toward , the formula makes increase from zero toward one. This describes this explicit family; it is not a regularity or free-boundary theorem for arbitrary obstacles.
Concept lensStructural
The minimizer is an energy-norm projection of the free solution
Let be the unconstrained solution, for all . The completed square says that the obstacle minimizer is the point of nearest to in the energy norm , and the variational inequality is the obtuse-angle characterization of that nearest point: for every . The clipped curve is nearest to pointwise, a different metric, which is why it generally fails, as the next mistake computes.
Common mistake
(Clipping the unconstrained solution at the obstacle)
It is tempting to solve without the constraint and then take . The result is feasible but in general not the minimizer, because its joins are not tangential. With and , the unconstrained solution is , and the clipped curve equals on with and outside. Piecewise integration by parts gives
so the joins carry negative point reactions, which no certificate allows. Its energy is , larger than the minimum found above. The contact interval comes from slope matching, not from the set where .
A corner can defeat an almost-everywhere check
Practice
Quick check
(Let the obstacle be with , and let the source be the constant . Find the minimizer for every real and its reaction density. Can every point be in contact while the reaction is zero?)
Separate the unconstrained parabola from the case in which the obstacle itself provides a certificate.
Solution
The unconstrained solution is . If , it is feasible and has zero reaction, so it is the minimum. If , use the obstacle itself, whose density is . Thus
Here . At the curve equals the obstacle everywhere and . Complementarity says positive reaction requires contact; it does not say contact requires positive reaction.
Quick check
(Keep one obstacle fixed, and let and be the minima for sources . Prove that .)
Use each minimum as a competitor in the other variational inequality. The constant two comes from the interval estimate already proved, not an optimal Poincaré constant.
Solution
Write . Test the inequality for with and the one for with . Adding and rearranging gives
Divide if ; otherwise the desired bound is immediate. Both curves satisfy the same obstacle, so these competitors are legitimate. With different obstacles this particular argument would not be justified.
Quick check
(For zero source and obstacle , find a contact interval and linear outer pieces. Verify feasibility and a contact-force certificate.)
The contact interval is no longer symmetric. Match each tangent line to its own zero endpoint.
Solution
Write , so . A left tangent reaches when ; a right tangent reaches when . Solving gives
Set on , on , and on . Values and first derivatives match. Since the obstacle has second derivative , the outer gaps are and . Thus the curve is feasible and its contact interval is exactly . The integral representation holds with , which proves the unique minimum.