Motivation
A stationary curve can fail to minimize because neighboring extremals focus back onto it. Jacobi fields describe the first-order separation of nearby extremals. When such a field vanishes at both endpoints, an infinitesimal change of the initial slope becomes invisible at the final position. This is the meaning of a conjugate endpoint in the model developed here.
There are two distinctions to keep in view. An endpoint may be nonconjugate even though an earlier conjugate point lies inside the interval. Also, a conjugate endpoint detects a zero direction in the quadratic approximation; the original functional may still have a strict minimum because of a positive fourth-order term. Neither a successful shooting calculation nor a vanishing determinant can replace the full minimum argument.
We study an explicit scalar family in which every statement can be proved directly. The results extend the existing second-variation discussion but do not assert a general conjugate-point theorem for arbitrary integrands, vector curves or free endpoints.
Definitions
Definition 5.1.1
(The reference extremal and its second variation)
Fix and real parameters . On
consider the smooth functional
The zero curve is stationary. For endpoint-zero , its second variation is
The exact objective difference is . The fourth-order term does not enter .
Euler–Lagrange for a classical extremal reads . Linearizing at zero gives . If a family of classical extremals is differentiable in its parameter with sufficient mixed derivatives to differentiate this equation, its parameter derivative at the zero member solves that linear equation. For the quadratic case , we will exhibit such a family explicitly; no assumption about a global nonlinear family is needed for the calculations below.
Definition 5.1.2
(Jacobi field and conjugacy in this model)
A Jacobi field along the zero extremal is a solution of . A number is conjugate to the left endpoint zero if a nonzero Jacobi field satisfies .
Normalize the left-endpoint field by and . For , the extremals have initial slope , so is exactly their separation derivative . At a conjugate endpoint the derivative of the shooting map vanishes.
The length here is a location on the independent-variable axis. It is not a value of the curve. Conjugacy is defined relative to the reference extremal and its linearized equation. We use zero endpoint values for the minimum classification; the shooting interpretation can also help solve a quadratic problem with a prescribed nonzero terminal height.
Theorem/Proposition
Theorem 5.1.3
(The complete Jacobi field and conjugate times)
The normalized field is
Every field vanishing at zero is a multiple of . If , its positive conjugate times are , . If , there are none.
These formulas follow by solving the constant-coefficient equation and imposing the two initial conditions. In the quadratic problem with terminal height , shooting requires . If , this determines one initial slope. If it is zero, nonzero is impossible, whereas leaves every slope available. This describes boundary-value solvability in this model, not a general minimum criterion.
Theorem 5.1.4
(Second-variation threshold and its endpoint kernel)
Put . For every ,
with equality exactly for multiples of . Consequently:
- If , then for a positive constant, where for and for .
- If , then and its kernel consists of those sine multiples.
- If , the first sine gives a negative direction.
For positive , the three cases mean that the first conjugate point is beyond , exactly at , or inside , respectively.
Concept lensAnalytic
Conjugate times as Dirichlet eigenvalues
A number is conjugate to zero exactly when is a Dirichlet eigenvalue of on : a nonzero Jacobi field vanishing at and is precisely such an eigenfunction. Those eigenvalues are . In sine coordinates , Parseval's identity gives
So the number of negative sine modes equals the number of conjugate times inside , and a conjugate endpoint is a zero mode.
Theorem 5.1.5
(What the quadratic test proves for the nonlinear functional)
For the explicit family above:
- If , zero is a strict strong local minimum for every real . If , it is also the unique global minimum.
- If and , zero is the unique global minimum despite the Jacobi kernel. If , every first-sine multiple is a global minimum. If , zero is not a local minimum.
- If , zero is not a local minimum for any .
For , the functional is unbounded below on the full admissible class, including in the subcritical case where zero is a strict local minimum. Here “strong local” compares all competitors with sufficiently small , without requiring their slopes to be small.
Proof sketch or proof idea
Proof
(A square identity with the singular endpoints handled)
Take , which is positive on and has simple zeros at both endpoints. On a truncated interval away from its zeros, direct differentiation using gives
We do not divide by at an endpoint. Instead integrate on and let . Since and has zero endpoints, near zero and near . The factor has only a simple pole there. Thus both boundary products tend to zero. The integral on the left converges to , so the nonnegative square integral has a finite improper limit and
Equality makes throughout the interior, hence is a multiple of ; such multiples do give equality. This proves the sharp inequality and its equality case. For , substitute it into to obtain . For , use directly, since multiplying an inequality by a negative number would reverse its direction.
Finally, for ,
This proves negativity above the threshold. It also shows why endpoint conjugacy supplies a zero quadratic direction rather than an automatically negative one.
Proof
(A quantitative minimum estimate and the degenerate cases)
Below the threshold, write . The sharp inequality gives
The exact objective therefore satisfies
If , choose ; then for nonzero . If , no smallness is needed. This proves the subcritical conclusions. Strong locality follows here because the exact remainder depends only on height; a general second-variation theorem need not control large slopes in this way.
At the threshold, . When , every nonzero continuous has , so the objective is strictly positive. When , the equality case of the sharp inequality gives exactly the family of minimizers. To handle the other cases, use the exact sine formula
At the threshold with negative it is negative for every nonzero . Above the threshold its negative quadratic term dominates for small , whatever is. These perturbations approach zero even in . Finally, if , scaling any fixed nonzero admissible function to arbitrarily large amplitude makes its negative fourth-order term dominate, proving unboundedness below.
Proof X-ray
What a Jacobi zero actually decides
The square identity needs a solution of the Jacobi equation that is positive on the whole open interval . The field qualifies precisely because the first conjugate time of is . Dividing by is the one delicate point, and the proof never does it at an endpoint: it truncates to and then uses two facts together, that has simple zeros and that is with . Neither alone suffices. If , the boundary product behaves like and does not disappear.
Once the boundary terms vanish, the quadratic form is a weighted square, which settles the sign and the kernel at the same time. Above the threshold every nonzero solution of has a zero inside , since its zeros are spaced apart, so no such identity can exist; this matches the negative direction. The identity sees only the quadratic part. On its kernel the decision passes to the quartic term, which is why the second proof evaluates exactly.
Worked examples
Worked example 5.1.6
(Unique shooting solution with an interior conjugate point)
Take , , and . Since , the zero-endpoint boundary-value problem has only the extremal . The endpoint itself is nonconjugate, but the first conjugate time lies inside the interval.
For and , both endpoint-zero,
The objective changes by half these quantities times the square of the perturbation amplitude. Arbitrarily small perturbations can therefore lower or raise it, so zero is a saddle. Uniqueness of a classical stationary boundary-value solution did not imply that it minimized the functional.
The explorer below moves and , showing the zeros of the Jacobi field , the exact threshold , the sign of the second variation on sine modes, and the minimum verdict, starting from the case , of this example.
Worked example 5.1.7
(A variable-coefficient problem reducible to the same model)
Let and . With zero endpoints, consider
Put . Then . The cross term integrates to , giving exactly
Multiplication by is an invertible transformation of the endpoint-zero class. The Jacobi equation is , with normalized field . Its conjugate times are .
Thus zero is the unique global minimum when , is one of the family of minima at equality, and has a negative direction when . The positive leading coefficient never changes sign; it cannot by itself prevent instability. The reduction works because we verified the exact transformation and its boundary term, not because every variable-coefficient problem has the same conjugate times.
Worked example 5.1.8
(A strict minimum with a conjugate endpoint)
On , let and . The field vanishes at both endpoints, and . Nevertheless
for every nonzero admissible , so zero is the unique global minimum. Along the null direction,
There is no positive quadratic growth constant along these perturbations. Replacing by leaves the Jacobi equation and second variation unchanged but reverses this fourth-order sign, destroying the minimum. The conjugate endpoint identifies degeneracy, while the higher-order term resolves it.
Common mistakes
Counterexample mode
Endpoint invertibility is not sufficient for minimization
The false rule is: if the normalized Jacobi field is nonzero at the terminal point, a stationary solution is a strict minimum. The first worked example disproves it: at the field is nonzero, yet an earlier zero has already occurred and the quadratic form has a negative direction. The correct model criterion for positive definite second variation is that the interval end before the first conjugate time, rather than merely avoiding the discrete set of conjugate endpoints.
Two further errors concern what the quadratic test can and cannot see.
Common mistake
(A Jacobi field is not a nearby minimizer)
A Jacobi field is an infinitesimal separation of extremals, not necessarily a second feasible finite-amplitude curve in a nonlinear problem. At the threshold with , the field spans the kernel of , yet for every , and does not even solve . The kernel of the Hessian does not describe the exact minimizers until the remaining terms have been examined.
Common mistake
(A strict minimum need not grow quadratically)
Do not call a strict minimum quadratically coercive merely because every nonzero competitor has larger objective. In the critical positive-quartic example the increase along the kernel direction is of fourth order in the amplitude, so no estimate with holds near zero; the exercise below on the critical value proves this.
Summary
The explicit Jacobi equation computes conjugate times and connects them to a sharp second-variation threshold. A conjugate point at the endpoint is a degenerate case, while the first interior conjugate point permits descent in this model. Coercivity controls perturbations below the threshold; at equality, higher-order terms decide strictness. The weighted example transfers the conclusions only after an exact change of variables and endpoint calculation.
Exercises
Quick check
(Let , and . Locate the first conjugate point, give a quantitative strong-local minimum estimate at zero, and decide whether the functional is bounded below globally.)
Compute and use the height-only remainder estimate.
Solution
Here , so the first conjugate point is . Also and . The estimate above gives whenever , proving a strict strong local minimum. But along a fixed nonzero sine with amplitude tending to infinity, the negative fourth-order term dominates, so . Local minimality is not a global lower-bound statement.
Quick check
(In the weighted quadratic example, set , and . Find the normalized Jacobi field and all minimizers. What changes after adding ?)
Use and keep the zero endpoints.
Solution
The normalized field is , whose derivative at zero is one. The interval ends at its first positive zero. The transformation gives , with equality exactly for . Hence all minimizers are . Adding makes the energy strictly positive for every nonzero curve. The Jacobi field remains unchanged, but the only minimizer is now zero.
Quick check
(At with , prove that zero is a strict global minimum but that no gives throughout a neighborhood of zero.)
Test the ratio on as .
Solution
The sharp inequality gives , and continuity makes whenever is nonzero. Thus the positive-quartic functional has a unique global minimum at zero. For the first sine,
These curves tend to zero in both uniform and norms. Therefore no positive uniform quadratic-growth constant works even in a weak neighborhood. The conclusion does not contradict strictness, which asks only for a positive objective gap for each nonzero competitor.