A constraint at every point changes the test directions
Requiring the integral of a curve's squared norm to have a fixed value leaves considerable freedom at individual points. Requiring the norm itself to equal one at every point does not. A displacement must now be tangent to a sphere at each position, and those tangent planes change along the curve. Varying one Cartesian coordinate independently will usually leave the admissible set.
The unit sphere gives a concrete setting in which we can derive the constrained equations without assuming a general multiplier theorem. We will build actual sphere-valued variations, identify all stationary curves and then prove which have the least energy. The distinction matters: a long great-circle arc satisfies the same differential equation as the short arc, but on the two-dimensional sphere it has an explicit energy-decreasing direction.
The fixed time interval also matters. The energy depends on how fast a path is traversed. Its relation to geometric length supplies both a global lower bound and the correct constant-speed parameterization. A time-dependent weight will later change the optimal timing without changing the short geometric arc.
Curves, tangent directions and exact feasibility
Definition 13.1.1
(The sphere-constrained energy)
Work on . Fix and consider curves with and . Their energy and length are
A tangent variation field is a vector function with and at every . A feasible variation is a family of admissible curves whose dependence on its parameter is differentiable at zero as a map into . Its derivative is therefore a tangent field; it need not be . A curve is stationary when the energy derivative is zero along every such feasible variation.
Differentiating the sphere constraint in any feasible family gives the tangent condition. For constructing exact families within our admissible class, we use tangent fields. Every such field is the derivative of
Its denominator is , so it never vanishes. It retains both endpoints, lies on the sphere and satisfies . Thus tangency is not merely a formal linearized condition here: we have an explicit feasible realization for every test field we use.
The straight perturbation is generally off the sphere by a second-order amount. Normalization removes that error without changing the first-order direction, and it records the second-order normal correction that the constrained second variation will need.
The constrained Euler equation
Theorem 13.1.2
(Stationary curves are constant-speed great-circle arcs)
A curve in the stated class is stationary if and only if
Its speed is a constant . If , it is constant. If , writing gives
These are great-circle arcs with constant speed, including arcs that wind around the circle more than once.
Proof
(Test tangent fields and recover the normal acceleration)
For a tangent field and its normalized family, differentiation under the integral and integration by parts give
For any smooth compactly supported vector test , use . It is an admissible tangent field. The componentwise fundamental lemma therefore gives
Twice differentiating gives and , proving the equation. It also makes zero for every endpoint-zero tangent field, including the derivative of any stated feasible variation. This proves the converse.
The equation gives . Thus , whose initial-value solution is the displayed trigonometric formula. Its initial constraints make and orthonormal, so its image really lies on a great circle.
With the sign convention , the normal multiplier is . The defining constraint has nonzero gradient on the sphere. We obtained its normal direction directly, rather than importing the constant multipliers from finitely many integral equalities. The multiplier happens to be constant for this particular energy; a later example shows why pointwise constraints do not generally imply constant multipliers.
The global energy minimum
Theorem 13.1.3
(The short arc attains the global lower bound)
Let . Every admissible curve satisfies
For , the unique minimizer in the stated class is
For , the unique minimum is the constant curve. For , the minimum is and all minimizing curves are with any unit vector perpendicular to .
Proof
(Bound angular displacement without differentiating at a pole)
Put . Since ,
Direct differentiation of could be invalid when . Instead, for put and . Here is an angular regularizer, separate from the variation parameter. The regularized angle is differentiable throughout the interval and
The fundamental theorem of calculus now gives
As , the endpoint values satisfy and . Thus . Cauchy–Schwarz on the unit time interval gives .
The stated short arc has speed , so it attains the bound. Any other minimizer is stationary by feasible first variations and hence has constant speed . Its energy forces . When , its endpoint equation uniquely determines , proving uniqueness. For the energy is zero and the curve is constant. For the stationary formula permits precisely the stated perpendicular unit vectors, all of which attain the bound.
Proof X-ray
Where the pole and the clock enter
The proof splits the lower bound into a geometric step, , and a timing step, . The geometric step uses only one fact about the sphere, , which bounds by , the speed of the angle measured from . The non-obvious move is the regularizer . The function is not differentiable at , the curve starts at and may pass through , so the angle itself cannot be differentiated along the whole curve. Shrinking the argument by keeps it inside , and the derivative bound survives because .
The timing step is Cauchy–Schwarz on an interval of length one, with equality exactly at constant speed. Uniqueness closes through the first variation: a minimizer is stationary, hence a constant-speed great circle, and only the endpoint equation remains. At that equation no longer determines , which is exactly where uniqueness is lost.
Common mistake
(A shortest arc need not minimize energy)
Energy and length are different functionals. Equality in requires constant speed. A nonuniform traversal of the same short geometric arc, such as the angle , still has length but energy .
The antipodal failure of uniqueness is geometric: many different half-circles have the same least length and the same constant speed.
A constrained second variation detects the long arcs
Theorem 13.1.4
(The normal correction contributes a negative curvature term)
Let be a stationary great-circle curve of speed , and let be a tangent field. For its normalized feasible family,
If , there is a negative direction, so the curve is not a local minimum even among -nearby admissible curves on .
Proof
(Differentiate the normalized family twice)
Normalization gives . Therefore
The derivative of contributes no term because . Choose a constant unit vector perpendicular to the great-circle plane and set . It is tangent to the sphere along the whole curve and vanishes at the endpoints. Direct integration gives
For this is negative. The first variation is zero, so Taylor expansion along this explicit feasible family decreases energy for small nonzero parameters. Such a perpendicular vector exists because the constraint is . The same negative direction is unavailable for curves confined to a circle; this distinction will be tested in practice.
Proof X-ray
One test direction rules out every long stationary arc
Stationarity is used twice. It removes the first-order term, so the sign of decides the energy change for small parameters, and it makes a great circle of constant speed , so in the last integral. The cross term in vanishes for a different reason: holds for every curve on the unit sphere. What remains is the normal correction , the sole source of . The constant normal is tangent along the whole arc only because the arc stays in one plane. The threshold is the sharp constant of the fixed-endpoint Poincaré inequality on , with extremal . One direction suffices: for the endpoint condition gives , so every stationary curve other than the short arc has .
Common mistake
(Perturbing straight off the sphere)
Treating as if it were admissible gives for every , which would wrongly suggest that every stationary arc, long or short, is a local minimum. The term comes entirely from the second-order normal correction that keeps the family on the sphere.
At the displayed direction has zero second variation. That alone would not prove minimality, but the global bound already does. The antipodal worked example below shows that this zero direction is tangent to an exact family of distinct minimizing half-circles with the same endpoints, so a zero quadratic direction can reflect nonuniqueness at a genuine global minimum.
Worked examples
Worked example 13.1.5
(Two stationary arcs between the coordinate axes)
Take and . Then and the unique minimum is
The opposite traversal has the same endpoints and is stationary, but has energy . For its second variation is . This proves more than a comparison with a distant better curve: there are arbitrarily nearby sphere curves of smaller energy.
Worked example 13.1.6
(Antipodal endpoints: a zero direction at a global minimum)
Take and , so . For the unit vector , the curve
equals , has speed , joins the poles and has energy , the global minimum. At the curve lies in the plane with unit normal , and gives .
This null direction is no accident: . The zero of the second variation is the derivative of an exact family of minimizers, so it signals nonuniqueness rather than a failure of minimality.
Worked example 13.1.7
(A variable weight makes the multiplier vary with time)
Fix endpoints with . Replace the energy by with . We will show attainment of its lower bound. The tangent-field argument now gives
Weighted Cauchy–Schwarz yields . Hence the lower bound is . It is attained by the short arc with angle
Here is constant, so substituting verifies the weighted Euler equation. Its normal multiplier, in the displayed convention, is
which is not constant when . The curve moves faster when the energy weight is lower. A pointwise constraint produces a multiplier field; constancy in the unweighted problem was a consequence of its particular dynamics.
Concept lensGeometric
Separate the geometric arc from its clock
The angle determines where the curve lies on its great circle. The derivative of the angle determines its speed and energy. Ordinary energy chooses constant angular speed; weighted energy chooses angular speed proportional to the reciprocal weight. Both traverse the same short arc, while their multiplier functions and energy values differ.
A bad defining function can destroy a multiplier formula
The multiplier formulas above depend on how the constraint is written, not only on the constraint set.
Counterexample mode
The same constraint set need not have a regular gradient
Replace by . The feasible set is still exactly the sphere, but vanishes everywhere on it. For a nonconstant short minimizing arc, . No finite multiplier function times can balance that acceleration in a normal Euler equation with objective coefficient one.
The minimum has not disappeared. Its normal multiplier representation failed because this defining function lost rank. An abnormal relation with zero coefficient on the objective can still hold, since every constraint-gradient term is zero; it supplies no acceleration equation. The explicit normalized-variation proof remains valid and explains the correct normal direction. One cannot apply a regular multiplier rule merely because the zero set is a familiar smooth surface.
Practice
Quick check
(Restrict curves to the unit circle in instead of the sphere in , and let a stationary arc have angle . Prove that it is a strict energy minimum among fixed-endpoint circle curves at uniform distance less than , even when .)
Use a continuous relative angle in with zero endpoint values. Explain why this does not contradict the sphere's negative direction.
Solution
Every circle curve in this uniform neighborhood has a unique relative angle , with , and can be written . Identify the plane with the complex numbers. The relative unit point avoids the negative real axis because the uniform distance is less than two, so its continuous principal argument provides this angle. For curves it is . Consequently
Equality forces . Thus each such arc is a strict local minimum within the circle, even when it is not the global shortest arc. The sphere perturbation moved perpendicular to the great-circle plane; it leaves the circle and is not an allowed direction in this question.
Quick check
(Replace the unit sphere by the sphere of radius . For endpoints on that sphere, derive the stationary equation and the minimum energy in terms of their central angle.)
The multiplier's coefficient and the energy scale differently with the radius. State the exceptional endpoint cases.
Solution
Put , and . Tangent fields still satisfy . Now and , so the normal equation is
Scaling a unit-sphere curve by scales length by and energy by . The minimum is therefore , attained by the scaled constant-speed short arc. It is unique for distinct nonantipodal endpoints; equal endpoints give the unique constant curve, and antipodal endpoints give the family of scaled half-circles.
Quick check
(Let be a positive function on and let the endpoint angle satisfy . Find an exactly minimizing curve for the weighted energy and compute its multiplier.)
Write . Check the lower bound and the weighted Euler equation, rather than only proposing a reparameterization.
Solution
Define and . The assumed range makes the unit vector from the short-arc theorem well defined. Set
It is an admissible curve, has angular speed , and has energy . Every competitor satisfies , so this value is the global minimum. Since is constant, differentiation also gives the weighted Euler equation. Its multiplier is , which need not be constant.