The equation and the minimum are two separate tasks
The hanging-chain problem combines three basic ideas: a scalar integral constraint, an autonomous Euler equation and a sufficient comparison valid for every competitor. Finding a hyperbolic cosine solves only the candidate equation. It does not explain which stationary curve hangs down, why that curve has the prescribed length, or why an asymmetric competitor cannot have lower energy.
We will prove each step for equal-height supports. The global comparison uses arc length and a pointwise Cauchy–Schwarz inequality. Symmetry will emerge from equality; it is not imposed on the comparison class.
The model and its feasible lengths
Fix . A curve is a graph on with . Height is measured upwards, and mass per unit arc length and gravitational acceleration are normalized to one. Define
The length is fixed at , and the objective is gravitational potential. The density is uniform along the chain, which is why the measure is arc length rather than horizontal distance.
Definition 12.1.1
(The length-constrained comparison class)
For let
The main case is . Stationarity of means that the first derivative of vanishes along every feasible family with , defined for and differentiable at zero as a map into . A local extremum is stationary by the two-sided scalar derivative test. We derive the stationary equation for members of this comparison class and prove the final minimum against every member. Corners, overhanging curves, elasticity and variable material density are not part of the model.
Since , no curve has length below . Equality forces everywhere, hence . For we will explicitly construct a feasible curve. The objective is bounded: from either support a point's vertical displacement is at most the intervening arc length, so and . This estimate does not yet identify a minimum.
Concept lensAnalytic
The measure in the model determines the curve
The objective integrates height per unit actual chain length. Replacing by would assign mass per horizontal distance instead, and the Euler equation would change with it. The stationary equations and the admissible length constraint must be derived from the same model. The parabolic approximation in the exercises has a stated small-slope limit; it is not an exact replacement for the catenary at arbitrary slack.
Why the normal multiplier rule applies in the slack case
At a admissible the length derivative is
If it vanished for every test, integration by parts and the fundamental lemma would make constant. This is a strictly increasing function of , so would be constant. Equal endpoint heights would then give and length . Thus is a nonzero functional whenever .
For completeness choose a endpoint-zero with . For any endpoint-zero direction , the scalar implicit function theorem corrects to have the same length, with
At a constrained stationary curve, differentiating along this family gives
Hence a real constant makes stationary. The sign convention here is plus ; the fixed term has no curve derivative. The multiplier is constant because only one scalar integral is fixed.
The hypothesis is not a formality, as the boundary length shows.
Counterexample mode
The taut chain is a minimum with no normal multiplier
At the only feasible curve is , so it is the unique minimum. Yet for every direction, whereas is not always zero; for it equals . No finite makes vanish for every endpoint-zero .
There is no contradiction with the slack-chain multiplier argument: its explicit nonzero-derivative hypothesis has failed. The global straight-chain conclusion is immediate from feasibility, not from a normal augmented Euler equation.
Find every classical stationary graph
For the augmented density
Euler–Lagrange reduces to
In particular never vanishes along a classical stationary graph. The Euler equation gives
Thus Beltrami, used as a consequence of Euler–Lagrange, gives a nonzero constant
Combining the two equations yields
No division by occurs; a horizontal tangent is allowed. Integration gives . Substituting back into Beltrami fixes the additive height constant, since . Hence
Equal endpoint heights force . Indeed, equality of the two endpoint coshes makes their real arguments equal or opposite; equality would require , whereas opposite arguments give . Write . Both possible graphs therefore have the form
The endpoint value fixes . The plus choice lies below the supports and has ; the minus choice lies above them and has the opposite multiplier. For either sign, , so its length is .
Theorem 12.1.2
(Exactly two classical stationary graphs have a slack prescribed length)
For every there is exactly one satisfying
The two graphs above, with this , are the only constrained stationary members of . Both are smooth; the theorem has not yet selected the minimum.
Proof
(Solving the length equation and checking both graphs)
Put . Differentiation gives
The sign follows because vanishes at zero and has derivative for . The length tends to infinity as and to as . Continuity and strict monotonicity prove existence and uniqueness of .
The preceding derivation puts every classical stationary graph into one of the two forms , and its prescribed length fixes this unique . Conversely, substitution verifies the length, endpoint and augmented Euler equations for both forms. Their augmented first variation is zero, so their objective derivative vanishes along every feasible constant-length family.
Common mistake
(Solving the Euler equation is not finding the minimum)
Both and satisfy the same augmented equation ; only the multiplier differs. Choosing the sagging branch "because a chain hangs down" uses physical intuition in place of an argument. In this problem the upward branch is in fact the global maximum, so the selection has to come from a comparison with all competitors.
A sufficient certificate against every equal-length graph
Fix the unique and set
The second expression follows from and .
Theorem 12.1.3
(The downward catenary is the unique global minimum)
For the minimum of over the whole class is
uniquely attained by
The sufficient conditions are the declared graph model, fixed supports, fixed arc length and the unique positive solution of the length equation. Competitor symmetry and stationarity are not assumptions of the comparison.
Proof
(Arc-length reparametrization and a pointwise Cauchy–Schwarz bound)
For any competitor define centered arc length
It is and strictly increasing, with positive derivative, so its inverse defines a curve on . It has endpoints and , and
The last equality uses integration by parts and the zero endpoint heights. Also . At every , Cauchy–Schwarz applied to and the unit vector gives
Integrating yields the universal lower bound
For the evaluation of the integral, use
and .
Equality in the pointwise inequality requires
If the integrated bound is an equality, the continuous nonnegative pointwise gap vanishes everywhere. Integrating these two components and using the endpoint positions gives
Inverting gives and hence exactly . Conversely the arc-length derivatives of have these values, so it attains the bound. This proves existence, global sufficiency and uniqueness without assuming a minimizer existed before the calculation.
Proof X-ray
Where each hypothesis enters the certificate
The fixed length enters only once: it makes the arc-length parameter run over the same interval for every competitor, so all competitors are integrated against the same weight . The fixed supports enter twice: they remove the boundary term in , and they make the same for every competitor. That is why the term can be added inside the inequality at no cost; it plays the role of the multiplier.
The single non-obvious move is the choice of the vector . The constant is not free: the equality curve started at the left support reaches , and this equals exactly when . With any other constant the equality curve misses the support, the bound is strict for every competitor and identifies no minimizer. The argument closes by equality: zero integrated gap forces a prescribed unit tangent at every arc-length value, which leaves one curve. Neither the multiplier rule nor the Euler equation is used here; they only suggested which curve to test.
The certificate is easiest to absorb on explicit numbers, then on a scaling check and a near-competitor.
Worked example 12.1.4
(A fully specified chain with elementary parameter values)
Take and . Then solves the length equation, since , and . The minimizing curve, sag and minimum energy are
Its slope is , so its length integral is . The support values vanish because . Every equal-length graph has energy at least the displayed , with equality only for this curve. Checking the length and the lower bound are separate from verifying the differential equation.
Worked example 12.1.5
(Scaling the supports and the length together)
Replace and by and with . If solves , then solves , and by uniqueness it is the parameter of the scaled problem. Hence becomes and
For the previous chain with , so and , the minimizer is and . The energy scales by because both height and arc length scale by ; the shape is the same curve magnified. This is a quick test for any proposed closed formula for .
Worked example 12.1.6
(A circular arc of the same length loses, but barely)
Keep and . A circular arc of radius through both supports, with half-angle at its center , is a smooth graph. Its half length and half span give
Bisection gives , . Parametrize by with height and :
This exceeds by about , as the theorem requires. The margin is small because the two curves are close; the comparison is decided by the certificate, not by how different the pictures look.
Practice with solutions
Quick check
(Show that the reflected curve is the unique global maximum of on and find its multiplier in the convention . Why can the augmented Euler equation alone not select the minimum?)
Reflection preserves the comparison class.
Solution · Answer
Reflection preserves endpoints and length, and . The unique global lower bound therefore gives the unique upper bound , attained by . For this curve the multiplier is : both and reverse sign, so their product still equals .
Thus the upward and downward graphs satisfy the same type of stationary equation with different multipliers. The lower bound and its equality case, not the candidate equation, distinguish the minimizing graph. A positive gravitational potential for the upward graph does not violate any sign convention: heights were measured upwards.
Exercise 12.1.7
(Keep fixed and let decrease to zero. Prove the expansions below, with the second remainder in , then find the limits of and , where is the sag, and describe the multiplier.)
Synthesis
Solution(4)
- 1
Send the parameter to infinity and expand the length
As we must have . With ,
Because: The length is strictly decreasing in with limit as .
- 2
Expand the curve and its slope uniformly
Expanding both coshes in gives the second formula. The derivative remainder is uniformly for , proving the assertion.
- 3
Expand the sag and the minimum value
Key step
The sag is , and gives
Because: with .
- 4
Divide by the square root of the slack
Since ,
The normal multiplier tends to : the limiting feasible graph is straight, but no finite normal multiplier survives there, consistently with the taut-chain counterexample. The parabola describes this controlled limiting shape, not the exact finite-slack solution.
Exercise 12.1.8
(For fixed and , prove that is differentiable with , using the explicit parameter equations rather than a general envelope theorem. Interpret the sign.)
Core
Solution(3)
- 1
Make a differentiable function of
The implicit function theorem gives a differentiable inverse on .
Because: .
- 2
Differentiate the value with respect to
Key step
Because: and .
- 3
Divide by the length derivative and read the sign
By the chain rule . This matches the convention : increasing the constraint value contributes , and on the minimizing branch. Since , extra chain length lowers the least potential energy, and as . The result concerns this solved family only; it is not a claim that every constrained variational value is differentiable.
Exact scope
The argument solves the fixed-length, uniform-density, equal-height-support graph problem in , classifies its stationary graphs and handles the degenerate straight case separately. Unequal supports, nonuniform density, elastic extension, obstacles, overhangs, dynamics and numerical convergence are not included. The sufficient comparison is tied to the declared potential and arc-length constraint.