A stationary cycloid is only part of the problem
The time of descent depends on both distance traveled and speed gained from losing height. The shortest Euclidean path therefore need not be the quickest. There is also a mathematical difficulty at the starting point: a particle released from rest has zero speed there, so the time density is singular. Deriving an Euler equation where the depth is positive does not justify an endpoint substitution into a formula that is undefined at zero depth.
We will solve a precise graph problem. Its minimizer is a first cycloidal arc with an unbounded starting slope. A comparison potential proves global minimality among positive absolutely continuous graphs, and its controlled limit at the origin handles the singular endpoint. An explicit polynomial rounding then shows why demanding a continuously differentiable graph can remove attainment without changing the infimum.
Admissible graphs and the time normalization
Fix . The coordinate measures depth downward from the start. We compare
Here denotes the absolutely continuous functions on . Define the extended nonnegative functional
The value may be infinite; no integrand value at the single starting point is needed. If motion is frictionless, begins at rest and has constant downward acceleration , the energy relation gives speed and travel time . We keep the constant factor outside the variational argument.
Definition 10.1.1
(A first cycloidal arc written as a graph)
For and , set
The first coordinate is strictly increasing, although its derivative vanishes at the start. Thus the arc determines a graph. Only this first arc, with positive depth after departure and before the next return to depth zero, is used here.
The comparison class does not include overhanging parametrized routes, backward horizontal travel, or additional contacts with depth zero. Its full minimum theorem below is for this explicitly stated positive graph class.
The interior equation and its limitation
On an interval away from the start where a candidate is positive and , ordinary first variation gives
Since the density has no explicit horizontal dependence, its Beltrami quantity is
The cycloid satisfies and , so it has the expected conserved value. Direct substitution also verifies the second-order equation, including the point where . Differentiating the first integral and dividing by would not justify that equation at such a turning point. None of these interior calculations has yet proved a minimum or treated the start.
Common mistake
(A conserved Beltrami quantity does not imply the Euler equation)
Differentiating the Beltrami quantity gives , so its constancy carries no information where . A constant depth makes the Beltrami quantity equal to the constant , yet : horizontal segments are not stationary. The first integral must be used together with the Euler equation, or checked against it, never in place of it.
One arc fits each positive endpoint
Theorem 10.1.2
(Cycloidal coordinates cover the positive quadrant uniquely)
For every there are unique parameters , with the stated endpoint. They are determined by
The resulting graph belongs to and has
Proof
(Monotonicity of the endpoint ratio and finite cost)
The derivative of the ratio has positive denominator and numerator
For positivity is immediate. For , put . The numerator is , since the bracket vanishes at zero and has derivative . Also as and as . Continuity and strict monotonicity prove unique parameters. The positive derivative makes the inverse smooth throughout the positive quadrant.
To verify admissibility, use the graph derivative on truncated arcs. Its absolute integral near the start is finite, because
Together with continuity at the origin, the fundamental theorem on truncated intervals then proves absolute continuity of the full graph. Under the same parameter substitution the cost density times is the constant . This gives finite cost and the displayed travel time, even though the graph slope diverges at the start.
A smooth potential on the open quadrant
For any point with , use its unique cycloidal coordinates and define
This is smooth in the open quadrant. Differentiating in the coordinates gives
For example, the one-form has coefficient on and on , exactly the derivatives of the defining expression. This directly verifies the gradient formulas without assuming that a stationary family automatically supplies a comparison potential.
The origin needs a separate estimate. In cycloidal coordinates,
The right side is positive and continuous, with endpoint limits and as tends to zero and . It is therefore bounded. Consequently and tends to zero at the origin along every positive approach. We may define continuously, without claiming that its gradient extends there.
Concept lensGeometric
The potential as a field of equal-time curves
The theorem below shows that is the least normalized time from the origin to , so is the value function of the problem. Its identity is an eikonal equation for the local speed . The optimal slope says that is parallel to : each cycloid crosses the level curves of , the points that optimal descents from rest reach at the same time, at right angles. Equality in the Cauchy–Schwarz step is exactly this orthogonality.
Theorem 10.1.3
(The potential proves the unique global minimum in the positive graph class)
For every ,
Equality holds exactly for the cycloid graph . This comparison includes arbitrary integrable slopes; no bound at the starting point is assumed.
Proof
(Integrate away from the origin, then take its controlled limit)
Cauchy–Schwarz and the gradient identity give, for every positive point and every real slope,
Since , equality holds precisely when . A competitor of infinite cost satisfies the lower bound trivially. For a finite-cost competitor, fix . Its graph on lies in a compact part of the open quadrant. The ordinary absolutely continuous chain rule applies to there, giving
Let . The nonnegative cost integral increases to the full cost. Absolute continuity implies , and the origin estimate makes the last potential term vanish. This proves the bound. The constructed cycloid attains it by its cost formula and pointwise equality.
If a competitor attains the bound, the integral of its nonnegative excess over tends to zero as . Hence that excess is zero almost everywhere. The curve solves almost everywhere away from the origin. This slope field is smooth on the open quadrant. On any truncated interval, both it and the cycloid remain in a compact positive strip, so initial-value uniqueness applied backward from the common terminal point makes them identical. This holds for every truncation and proves global uniqueness of the graph.
Proof X-ray
The singular endpoint is handled by a limit, not by a formal substitution
The potential is smooth only where both coordinates are positive, and every competitor starts at the corner , where it is not. The proof therefore uses the chain rule only on , where the graph stays in a compact part of the open quadrant and absolute continuity of makes absolutely continuous. The missing initial segment is removed by two separate limits: monotone convergence of the nonnegative cost, and , which needs both and the uniform bound . Writing without that bound would leave a gap, and the interior Euler equation would not fill it.
The key inequality is Cauchy–Schwarz combined with the identity . It holds for every real slope, which is why no bound on competitors is needed. Equality is a separate step: zero excess forces the slope field only almost everywhere and only away from the origin, so uniqueness comes from backward initial-value uniqueness from the terminal point on each truncated interval.
Two exact endpoint geometries
Worked example 10.1.4
(The straight chord is shorter in length but slower in time)
Choose endpoints , where . The cycloid ends at , with zero terminal slope and normalized time cost . A straight graph to arbitrary positive endpoints has
For these endpoints its cost is , strictly larger than the cycloid's. Actual travel times are respectively and . Speed gained by descending changes the relevant comparison; minimizing unweighted path length would answer a different question.
Worked example 10.1.5
(The quickest graph can pass below the terminal depth and return)
Take . Its endpoint is and its terminal slope is . The curve first reaches depth at horizontal position , then moves upward to its terminal depth. Its cost is . Requiring depth to increase monotonically would exclude this proven minimizer and would define a different constrained problem. The present theorem does not solve that additional monotonicity constraint.
Worked example 10.1.6
(A nearly vertical drop recovers free fall)
Let with fixed. Since , the endpoint angle satisfies . Then
The physical time tends to , the free-fall time through depth , as it must. The straight chord has cost as well, so in this degenerate limit the two routes agree to leading order. The case itself is excluded, because a vertical segment is not a graph over .
The starting cusp and regularity of attainment
The small-angle expansions of the cycloid give
Inverting the first and differentiating through the exact slope formula yields
Theorem 10.1.7
(Attainment has a sharp Sobolev threshold, with no infimum gap)
The unique minimum belongs to exactly for . It is not a Lipschitz or graph on the closed interval. Nevertheless the infimum over positive graphs equals the attained absolutely continuous minimum.
Consequently every positive-graph class, including and , has that same infimum. It is attained exactly in the range .
Proof
(Round the initial cusp with an explicit quadratic bridge)
The derivative asymptotic makes its -th power integrable exactly when ; at three the divergence is logarithmic. The rest of the arc has bounded derivative. This proves the membership statements and excludes attainment by uniqueness.
Choose a cutoff sufficiently close to the start, before the turning point and before the terminal point. Write and . Strict concavity gives
On use the polynomial
and keep the cycloid on . Values and first derivatives agree at the join. The polynomial is positive after zero and increasing, with
Thus the joined graph is , has the correct endpoints, and its initial cost satisfies
The limit follows from the proved cusp expansion. The removed cycloidal segment also has cost tending to zero, so the joined costs converge to the global minimum. Every joined graph belongs to every class on this finite interval. Their infima are therefore no larger than this common limiting value, while the global lower bound supplies the reverse inequality. Uniqueness and the derivative asymptotic give precisely the stated attainment range. This is recovery of the infimum, not a claim that the singular graph has a finite starting slope.
Counterexample mode
A smooth graph need not have finite time cost
On , the smooth positive graph has density , whose integral is infinite. The cycloid, despite its unbounded slope, has finite cost. The singular weight depends on how quickly height leaves zero as well as on the slope. Demanding smoothness neither guarantees finite cost nor preserves attainment of this minimum.
Finite cost, attainment and the value of the infimum are therefore three different questions, and the regularity theorem answers each separately.
Concept lensAnalytic
Loss of a regular minimizer is not an infimum gap
The global comparison identifies a unique absolutely continuous optimum. The regularity test excludes it from and from at or above the critical exponent. The quadratic bridges nevertheless approach its cost within all those stricter classes. These are three separate statements: existence in the larger class, nonattainment in a smaller class, and equality of their infima. A strict Lavrentiev gap would require unequal infima, which the recovery sequence rules out here.
Practice with solutions
Quick check
(For on with , decide exactly when the graph has finite normalized time cost. Determine its membership when the starting derivative is unbounded, and compare the critical exponent with the cycloid.)
Which power graphs have finite cost? Compare the two terms under the square root near .
Solution · Guided solution
All these graphs are absolutely continuous. For , the derivative dominates the numerator and the density is asymptotic to , which is integrable. At it is . For it is asymptotic to , integrable exactly when . Thus finite cost holds precisely for .
When , the derivative is in exactly when , and is not essentially bounded. For it is bounded. At the threshold is three, matching the cycloid's leading cusp, although a power graph with that exponent is not thereby proved to be the exact optimum for its endpoints.
Quick check
(Determine when the minimizing graph is increasing in depth throughout and when it passes below the terminal depth. If both endpoint coordinates are multiplied by , find the new cycloidal parameters, graph and minimum cost, and the effect of changing the acceleration constant.)
Track the endpoint shape through the turning angle .
Solution · Guided solution
The turning angle is , and . Thus gives a strictly increasing-depth arc, equality gives zero terminal slope, and gives a descent past terminal depth followed by an ascent. These conclusions concern the minimizing graph already constructed, not an imposed monotonicity constraint.
Spatial dilation preserves the coordinate ratio and hence . It replaces by and the graph by . Its cost and physical time satisfy
For fixed geometry, replacing by multiplies physical time by and does not change the minimizing graph. An overall positive cost factor cannot change which admissible shape minimizes.
Quick check
(Fix but allow any terminal depth . Find the global least cost and terminal depth over the union of the positive graph classes, and relate the terminal slope to the natural endpoint condition without assuming that stationarity alone gives the global answer.)
Let the terminal depth vary on a vertical line and differentiate in .
Solution · Guided solution
For each terminal depth the proven minimum is . Its depth derivative is . As depth increases, the ratio and hence the angle decrease. The derivative is negative for , zero at that value, and positive above it. This proves the unique global endpoint optimum
The terminal angle is , so the slope is zero. At a regular positive endpoint, the free-height first variation requires , which gives the same condition. Here that natural condition is accompanied by the global value-function sign argument and an attaining curve; it is not used as a sufficient minimum test by itself.
Scope of the singular-endpoint theorem
The comparison and complete attainment statements concern positive absolutely continuous graphs with one initial zero-depth point and positive horizontal and depth endpoints. Additional interior zero-depth contacts, overhanging parametrized routes, friction, nonconstant force and side constraints require new admissibility and comparison arguments. The singular endpoint was treated by an actual potential limit and an explicit recovery sequence; a regular fixed-strip theorem was not silently applied at zero depth.