Motivation
Penalizing a slope discourages steepness. Penalizing a change of slope is a different choice: it depends on the second derivative. A straight line then has no bending cost even if it is tilted, while a curve with rapidly changing slope can be expensive. This distinction changes both the differential equation and the information needed at the endpoints.
When an integral depends on , integration by parts must be performed twice. The boundary expression contains both the variation of the endpoint height and the variation of its slope. Fixing a height does not fix a slope, and freeing a slope does not free the height. A fourth-order equation needs four independent conditions in a regular boundary-value problem, but those conditions must be derived from the actual admissible variations rather than guessed by counting alone.
We develop a classical scalar theory with fixed endpoint abscissas. Beam energies provide explicit examples of why the distinction matters.
Definitions
Definition 6.1.1
(A functional involving the second derivative)
Let and consider
Write the last three arguments of as . Assume and the partial derivatives are jointly continuous on an open neighborhood of the relevant quadruples. Only curves staying in that domain are compared. Endpoint heights or slopes may be prescribed independently; a direction has at each prescribed height and at each prescribed slope.
A weak local extremum here uses the norm . Openness and compactness ensure that sufficiently small such variations remain in the integrand's domain.
For a fixed direction, differentiating under the integral gives
This formula needs only the stated continuous partial derivatives in the curve variables. To integrate by parts into a classical differential equation, we need additional regularity along the candidate: put and , and assume , . The convenient stronger assumptions and ensure this. A admissible class does not itself prove that every extremal is .
Definition 6.1.2
(The two endpoint coefficients)
Under this along-curve regularity, define
At the right endpoint the first-variation boundary expression is ; at the left endpoint it has the opposite sign. Thus acts on the endpoint slope variation, while acts on the height variation. The endpoint abscissas remain fixed throughout this unit.
The labels identify coefficients in a variational formula. For the beam functional below, is the bending-moment coefficient and with our chosen sign convention. External forces and moments are introduced through their actual endpoint work, which fixes the signs without relying on a drawing.
Theorem/Proposition
Theorem 6.1.3
(First variation and the higher-derivative equation)
Assume the integrand and along-curve regularity stated above. Then
At a weak local extremum with independently fixed or free endpoint heights and slopes, compactly supported directions give the interior equation
This is a necessary stationarity equation; minimization still needs a separate argument.
Theorem 6.1.4
(Natural conditions depend on which endpoint datum is free)
For the integral functional alone, at either endpoint:
| Height | Slope | Condition supplied by the boundary variation |
|---|---|---|
| Fixed | Fixed | Neither nor is forced to vanish |
| Fixed | Free | |
| Free | Fixed | |
| Free | Free |
If an endpoint cost is added at the right end, a free height instead requires , and a free slope requires . A left-end cost gives and for its respective free variables. Assume these costs are .
“Free slope” means that slope is not prescribed kinematically. A spring or an applied endpoint moment can still contribute a nonzero derivative of an endpoint cost. Such a slope need not satisfy once that extra cost is present.
Theorem 6.1.5
(A global minimum certificate for quadratic bending energy)
Let with , let , and take
with linear endpoint work and any fixed selection of endpoint heights and slopes. If a candidate satisfies and all the corresponding boundary conditions, then every admissible competitor satisfies
It is a global minimum. It is unique precisely when the homogeneous endpoint restrictions admit no nonzero affine function .
The theorem is conditional on having a classical candidate. It does not prove existence for every selection of loads and endpoint freedoms. The beam integrand with merely continuous fits the earlier continuous-partial formulation: , , and has the required regularity along a candidate.
The affine-kernel criterion is for the stated linear endpoint work. A nonnegative quadratic endpoint spring adds its own quadratic gap; an affine direction can then preserve equality only if it also produces zero spring variation. Such a spring can remove a freedom that the bending term alone does not control.
Concept lensStructural
Bending energy as a squared norm on admissible directions
Let be the space of admissible directions: functions satisfying the homogeneous versions of the prescribed endpoint data. On the form is symmetric and bilinear, and it is positive definite exactly when contains no nonzero affine function. In this language the boundary-value problem is the weak equation for all , where is the distributed and endpoint work done on , and the energy gap is . Uniqueness is definiteness of the form.
Proof sketch or proof idea
Proof
(Two integrations by parts and four independent endpoint variations)
The first variation vanishes at a local extremum because the scalar function has an extremum at zero for every allowed direction. The differentiated integrand is uniformly bounded on a compact neighborhood of the chosen perturbed quadruples, which justifies differentiating the integral.
For the second-derivative term,
Also . Adding these expressions gives the formula. Compactly supported tests eliminate both boundary terms. If the continuous interior coefficient were positive on a small interval , the test there and zero outside would give a strictly positive integral, a contradiction. This test is at its joins. Applying the same argument to a negative coefficient proves the fundamental-lemma conclusion that the coefficient is zero.
Once the interior equation holds, the only remaining first variation is the boundary expression. Endpoint heights and slopes can indeed be chosen independently. With , a cubic Hermite polynomial is
It has , , , and . Set the entries corresponding to fixed data to zero and vary the others. Each free coefficient must vanish, proving the table. Adding the differential of each endpoint cost gives the signed modifications stated above.
Proof X-ray
Why four boundary coefficients do not mean four zero conditions
The proof runs in a fixed order, and the order matters. First the interior equation is extracted using compactly supported tests, which see none of the boundary terms; only then does the first variation reduce to the four numbers multiplied by . The key step is the Hermite cubic: it shows that these four endpoint values can be chosen independently of each other, so a free variation can be switched on alone and its coefficient isolated. Prescribed data suppress the corresponding variations, and only the remaining free variations produce natural conditions. Thus a clamped beam has prescribed height and slope, not an additional zero-moment condition. Interior order suggests how much endpoint information a regular problem needs, but the first-variation formula determines which information belongs to the problem.
The certificate for the quadratic bending energy uses the same formula, now applied to an exact expansion rather than to a derivative at zero.
Proof
(Exact expansion of the bending energy)
For an admissible competitor , the energy is quadratic in . The part linear in is plus the linear endpoint work. The interior equation and boundary conditions make it zero by the first-variation formula. The remaining term is . Equality and force , hence is affine. This proves both the certificate and the precise uniqueness condition. Fixing a height and a slope at one endpoint kills every affine direction; fixing heights at both distinct endpoints does so as well.
Worked examples
Worked example 6.1.6
(A uniformly loaded beam clamped at both ends)
Let , and be constants, and minimize
The equation is . The quartic
satisfies it and all four prescribed data. No natural zero condition is added. In fact , which is nonzero for nonzero load. These coefficients balance reactions at the prescribed slopes; they are not required to vanish.
Every admissible perturbation has zero height and slope at both ends, so the gap is , strictly positive unless . Hence this is the unique global minimum. Integration gives . The negative total value reflects the load-work term; the bending part alone is nonnegative.
Worked example 6.1.7
(Heights held, slopes allowed to rotate)
Keep the same load and energy but prescribe only . Both endpoint slopes are free, so the natural conditions are . Solving with these four conditions gives
Its endpoint slopes are and ; setting them to zero would solve a different problem. Its values are and . These need not vanish because endpoint heights are fixed. The exact gap again proves a unique global minimum: an affine perturbation vanishing at two distinct endpoints is zero.
Here , lower than the clamped minimum when . This is consistent with admissible-set inclusion: freeing slopes enlarges the comparison class. The equation in the interior stayed the same, while the boundary conditions and minimizing curve changed.
Worked example 6.1.8
(A clamped left end and endpoint force and moment)
Take no distributed load, clamp , and let both right-end data vary. Include the endpoint work
The interior equation is . At the right end the natural conditions are
They follow from the cost derivatives and . Integrating with the left clamp gives
Its exact gap is positive for every nonzero perturbation satisfying the left clamp. The minimum is therefore unique and global, with value . When , the tip displacement is and the tip slope is . A free endpoint can carry prescribed external work; freedom describes allowed variations, not the absence of loads.
Common mistakes
Counterexample mode
Positive bending stiffness does not guarantee a free-beam minimum
Remove all endpoint restrictions and keep a nonzero uniform load . The admissible constant curves have zero bending energy and total objective , which is unbounded below. No minimum exists. The formal free-end conditions would require at both ends, yet integrating would give , a contradiction. The interior equation alone misses this load imbalance.
The load is not the only obstruction to a well-posed free problem; the kernel of the bending term also matters.
Common mistake
(Expecting positive stiffness to fix the curve)
With no load and no endpoint restrictions, every affine function has zero bending energy and is a global minimum. Strictly positive controls second derivatives, not the two-dimensional affine kernel. Uniqueness needs endpoint data that exclude nonzero affine directions.
Common mistake
(Imposing a natural condition on prescribed data)
In the clamped loaded example, imposing as well as the fixed slope adds conditions the variational problem never asked for and wrongly excludes the actual minimizer, whose moment coefficient is . Fixed and free data must be distinguished before applying a natural boundary rule.
Summary
A second-derivative functional produces a fourth-order stationarity equation when its along-curve derivatives have the stated regularity. Twice integrating by parts gives separate height and slope coefficients. Fixed data remove variations; free data produce natural equations, modified by endpoint costs. For quadratic bending energy, exact expansion proves global minimization and identifies affine motions as the entire possible nonuniqueness.
Exercises
Quick check
(For the uniformly loaded constant-stiffness beam, clamp the left end, hold , and leave free. Derive the boundary conditions, solve for the minimizing curve, and justify uniqueness.)
The endpoint abscissas are ; do not prescribe the free right slope.
Solution
This is one clamped end and one rotating support. The conditions are and , with . Their solution is
It satisfies each condition directly. The exact gap is ; the left clamp excludes every nonzero affine perturbation, proving unique global minimality. The objective value is . Adding would change the comparison class.
Quick check
(On , minimize with and free endpoint slopes. Derive the interior and natural equations and prove the minimum.)
Both and contribute to the integrated first variation.
Solution
The first-derivative term must remain. Here , , so the equation is . Free slopes give . The zero curve satisfies these and the fixed heights. Since the functional itself is , it is nonnegative, and equality implies and then by the endpoints. Thus zero is the unique global minimum. Omitting would incorrectly replace the interior equation by .
Quick check
(Let and . Add the rotational spring cost to the unloaded beam energy, clamp the left end, and hold . Find the natural slope condition and the unique minimizing curve.)
The right slope is free to vary, but its endpoint cost is not constant.
Solution
Here a free slope is balanced by a spring. The right condition is . With and the three prescribed endpoint data, write . Substitution gives
For an admissible perturbation, the stationary linear terms cancel and the exact gap is . It is strictly positive for nonzero because of the left clamp. At the minimizer is zero. As , the right slope tends to ; a stiff spring approaches a prescribed-slope condition without making a finite-stiffness free slope identical to fixed data.