Isoperimetric Problems

An isoperimetric problem asks for the function that makes one integral functional stationary while a second integral functional is pinned at a prescribed value. The name comes from the oldest optimization question on record — Dido's problem: among all closed plane curves of a given perimeter, which encloses the greatest area? The answer is the circle, and every constrained variational problem built on the same template inherits the label isoperimetric (equal-perimeter). The physics is far wider than geometry: the shape of a hanging chain, the ground-state wavefunction of a quantum particle, the equilibrium of a soap film, and the profile of a self-supporting arch are all the same calculation wearing different clothes. The method is the exact function-space analogue of the finite-dimensional Lagrange-multiplier rule. To extremize $J[y]$ subject to $K[y]=\ell$ , introduce a single constant $\lambda$ and seek free stationary points of the auxiliary functional $J-\lambda K$ . The Euler–Lagrange equation for the combined integrand $h=f-\lambda g$ , solved together with the constraint $K[y]=\ell$ , determines both the extremal curve $y(x)$ and the number $\lambda$ . The remarkable fact — Euler's — is that a single constant suffices, because the constraint is a single scalar condition. That constant is never a nuisance: it is a price. In Dido's problem it is the radius of the optimal circle; in the catenary it is set by the horizontal tension in the chain; in quantum mechanics it is the energy eigenvalue itself. This article takes the multiplier machinery of its prerequisite on constrained variation and turns it loose on integral constraints. It sits beside the unconstrained Euler–Lagrange theory and points forward to the Beltrami first integral, to geodesics, and to Hamilton's principle — all of which reuse exactly this apparatus.

Key takeaways

An isoperimetric problem extremizes $J[y]=\int_a^b f(x,y,y')\,\mathrm{d}x$ subject to a fixed value of a second functional $K[y]=\int_a^b g(x,y,y')\,\mathrm{d}x=\ell$ .
The constraint is absorbed by a Lagrange multiplier: stationarity of $J-\lambda K$ gives the Euler–Lagrange equation for the auxiliary integrand $h=f-\lambda g$ .
The multiplier $\lambda$ is a single constant — one number, because the constraint is one scalar condition — fixed afterwards by the constraint itself plus the boundary data.
The archetype, maximizing area at fixed perimeter, yields the circle and saturates the isoperimetric inequality $4\pi A\le L^2$ ; a square of equal perimeter captures only 78.5 percent of the circle's area, an equilateral triangle only 60.5 percent.
The hanging chain of fixed length is the same machinery run through the Beltrami first integral, giving the catenary $y=C\cosh((x-x_0)/C)+\mathrm{const}$ .
Fixing $\int|\psi|^2\,\mathrm{d}x=1$ while minimizing energy turns the multiplier into the energy eigenvalue and the Euler–Lagrange equation into the time-independent Schrödinger equation.
Key idea: To make $J[y]$ stationary subject to $K[y]=\ell$ , extremize $J-\lambda K$ freely — the Euler–Lagrange equation for $h=f-\lambda g$ together with the constraint $K[y]=\ell$ determine both $y(x)$ and the single constant $\lambda$ .

The core insight

To make $J[y]$ stationary subject to $K[y]=\ell$ , extremize $J-\lambda K$ freely: the Euler–Lagrange equation for $h=f-\lambda g$ plus the constraint $K[y]=\ell$ determine both the curve $y(x)$ and the single constant multiplier $\lambda$ .

Where this lives today

The isoperimetric template is quietly everywhere modern physics meets optimization. In materials science and biology it governs the equilibrium of interfaces: a soap film or lipid vesicle minimizes surface area at fixed enclosed volume, and the multiplier is the pressure difference across the membrane — the Young–Laplace law is an isoperimetric result. In architecture and engineering the inverted catenary is the ideal compression-only arch; Gaudí used hanging-chain models to design the Sagrada Família, and the Gateway Arch in St. Louis is a weighted catenary. In spectral geometry the Faber–Krahn inequality — the disk minimizes the fundamental drum frequency at fixed area — is an isoperimetric statement about eigenvalues, and its cousins constrain everything from acoustic design to quantum-dot spectra. Most pervasively, the normalization-constrained energy minimization of Worked Example 3 is the engine of computational quantum mechanics: every density-functional-theory calculation, every Hartree–Fock self-consistent field, and every variational quantum eigensolver run on today's quantum hardware is solving an isoperimetric problem, with Lagrange multipliers enforcing normalization and orthogonality and appearing physically as orbital energies. The circle's ancient monopoly on efficiency turns out to be a special case of a principle that now sizes molecules and designs spacecraft trusses.

The physical picture

Picture the admissible curves as points in an infinite-dimensional space. The objective $J$ and the constraint $K$ each define a gradient — their first variations $\delta J$ and $\delta K$ . Without the constraint you would slide downhill on $J$ until its gradient vanished. With the constraint you are confined to the level surface $K=\ell$ , and you can keep improving $J$ only while some allowed motion along that surface still changes $J$ . You stop exactly when the objective's gradient has no component along the constraint surface — that is, when $\delta J$ is parallel to $\delta K$ . Parallelism is the single equation $\delta J=\lambda\,\delta K$ , and the proportionality constant is the multiplier. The first figure is the finite-dimensional cartoon of this statement, the one every reader already met in ordinary calculus.

Figure

At a constrained extremum the objective's gradient is parallel to the constraint's — no motion along the level curve K = ℓ can change J. This tangency is exactly δJ = λ δK, and the single number λ is the Lagrange multiplier

The infinite-dimensional version is the whole subject. When the objective is enclosed area and the constraint is perimeter, the tangency condition becomes a differential equation whose solution has constant curvature, and the only closed curve of constant curvature is a circle. The multiplier turns out to be the radius. The second figure is the punchline made quantitative: three shapes of identical perimeter, and the price you pay in enclosed area for departing from the circle. The dimensionless ratio $Q=4\pi A/L^2$ is $1$ for the circle and strictly less than $1$ for everything else — that inequality is the isoperimetric theorem.

Figure

Three shapes with identical perimeter L: the circle encloses the most area (Q = 4πA/L² = 1.000), a square of the same perimeter only 78.5 percent of it (Q = 0.785), an equilateral triangle only 60.5 percent (Q = 0.605). The circle is the unique maximizer

The core equations

An isoperimetric problem is defined by a pair of functionals: one to extremize and one to hold fixed. Both are integrals of local integrands over the same interval with the same fixed endpoints.

J[y]=\int_a^b f(x,y,y')\,\mathrm{d}x \quad\text{stationary, subject to}\quad K[y]=\int_a^b g(x,y,y')\,\mathrm{d}x=\ell \qquad (1)

where: $J$ — objective functional to be extremized; $K$ — constraint functional held at a fixed value; $\ell$ — prescribed constant; $f,g$ — smooth integrands in $x,y,y'$ ; $y(x)$ — unknown curve on $[a,b]$ with fixed endpoints. Reading: you are choosing a whole function, not a few numbers, and the constraint removes one "direction" of freedom in that infinite-dimensional choice. The constraint is enforced by demanding that the first variation of the objective be proportional to the first variation of the constraint. Equivalently, the auxiliary functional $J-\lambda K$ is made stationary with no constraint at all.

\delta J-\lambda\,\delta K=0 \quad\Longleftrightarrow\quad \delta\!\left(J-\lambda K\right)=0 \qquad (2)

where: $\delta J,\delta K$ — first variations of objective and constraint; $\lambda$ — the Lagrange multiplier, a single real constant. Reading: this is the function-space image of "gradients parallel". The single scalar $\lambda$ is the exchange rate between the objective and the constraint. Carrying out the free variation of $J-\lambda K$ gives an ordinary Euler–Lagrange equation, but for the combined integrand $h=f-\lambda g$ rather than for $f$ alone.

\frac{\partial h}{\partial y}-\frac{\mathrm{d}}{\mathrm{d}x}\frac{\partial h}{\partial y'}=0,\qquad h\equiv f-\lambda g \qquad (3)

where: $h$ — auxiliary integrand $f-\lambda g$ ; $\lambda$ — constant, still unknown at this stage, pinned afterwards by the constraint. Reading: solve this like any Euler–Lagrange equation, treating $\lambda$ as a parameter; the family of solutions carries $\lambda$ and the usual integration constants. If the auxiliary integrand has no explicit dependence on $x$ , the Euler–Lagrange equation admits a first integral — the Beltrami identity — which lowers the order by one and is the workhorse for the catenary.

h-y'\,\frac{\partial h}{\partial y'}=C \qquad(\text{valid when } \partial h/\partial x=0) \qquad (4)

where: $C$ — constant of integration; $y'=\mathrm{d}y/\mathrm{d}x$ . Reading: an autonomous integrand behaves like an energy-conserving system, and $h-y'\,\partial h/\partial y'$ is the conserved "energy" of the variational problem. For the geometric archetype the two functionals are enclosed area (in Green/shoelace form) and perimeter, and the answer is captured by a sharp inequality.

A=\tfrac12\oint\!\left(x\,\mathrm{d}y-y\,\mathrm{d}x\right),\qquad L=\oint\!\sqrt{\mathrm{d}x^2+\mathrm{d}y^2},\qquad 4\pi A\le L^2 \qquad (5)

where: $A$ — enclosed area; $L$ — perimeter, the fixed constraint; equality holds if and only if the curve is a circle. Reading: the closer a shape is to circular, the closer $4\pi A/L^2$ climbs toward $1$ ; the circle alone reaches it. Finally, the same template with the physicist's favourite constraint — fixed normalization — turns a minimum-energy problem into an eigenvalue problem, with the multiplier appearing as the eigenvalue.

\delta\!\left[\int\!\Big(\tfrac{\hbar^2}{2m}|\psi'|^2+V|\psi|^2\Big)\mathrm{d}x-E\!\int|\psi|^2\,\mathrm{d}x\right]=0 \;\Rightarrow\; -\frac{\hbar^2}{2m}\psi''+V\psi=E\psi \qquad (6)

where: $E$ — the multiplier enforcing $\int|\psi|^2\mathrm{d}x=1$ , identified as the energy eigenvalue; $V$ — potential. Reading: the ground state is literally the solution of an isoperimetric problem, and its energy is the price of the normalization constraint.

Derivation AWhy a single constant multiplier suffices

The goal is to prove the central structural fact of the whole subject: although $y(x)$ is an entire function with infinitely many degrees of freedom, one integral constraint is removed by exactly one constant multiplier — not a function $\lambda(x)$ , just a number. We do this with a two-parameter variation that enforces the constraint exactly, then show that the ratio of variations that emerges is independent of which test functions we chose, so it deserves the name of a single constant.

Step 1. Perturb the candidate extremal with two independent admissible variations, $y_\varepsilon=y+\varepsilon_1\eta_1+\varepsilon_2\eta_2$ , where $\eta_1,\eta_2$ both vanish at the endpoints. Why: one parameter would leave no freedom to stay on the constraint surface while probing the objective; two is the minimum that lets us hold $K$ fixed and still vary $J$ .

Step 2. Expand the constraint to first order, $K[y_\varepsilon]=\ell+\varepsilon_1\,\delta K[\eta_1]+\varepsilon_2\,\delta K[\eta_2]$ , and demand it stay equal to $\ell$ . Provided $\delta K[\eta_2]\ne0$ this fixes $\varepsilon_2$ as a function of $\varepsilon_1$ : $\varepsilon_2=-\varepsilon_1\,\delta K[\eta_1]/\delta K[\eta_2]$ . Why: the constraint is non-degenerate exactly when some admissible variation changes $K$ ; that is the constraint-qualification condition, and it is what lets us solve for $\varepsilon_2$ .

Step 3. Impose stationarity of the objective along the constrained family, $\mathrm{d}J/\mathrm{d}\varepsilon_1=0$ at the origin, and substitute the relation from Step 2 to get $\delta J[\eta_1]-\dfrac{\delta K[\eta_1]}{\delta K[\eta_2]}\,\delta J[\eta_2]=0$ for every admissible $\eta_1$ . Why: this is the true stationarity statement — the objective is stationary only among perturbations that respect the constraint, not among all perturbations.

Step 4. Rearrange into $\delta J[\eta_1]=\lambda\,\delta K[\eta_1]$ with $\lambda\equiv\dfrac{\delta J[\eta_2]}{\delta K[\eta_2]}$ . The left side depends on the arbitrary $\eta_1$ , but $\lambda$ is built only from the fixed reference direction $\eta_2$ , so $\lambda$ is one and the same number for every $\eta_1$ . Why: this is the crux — the ratio is forced to be constant because it must reproduce the same proportionality for every test function, which is impossible unless it is a genuine constant.

Step 5. Since $\delta J[\eta]=\lambda\,\delta K[\eta]$ holds for all admissible $\eta$ , the fundamental lemma of the calculus of variations converts it into the pointwise Euler–Lagrange equation for $h=f-\lambda g$ . Why: an integral that vanishes against every test function forces its integrand to vanish, delivering equation (3) and closing the argument.

Go deeper · Derivation B · 1 minDido's problem — the circle encloses the most area

Derivation BDido's problem — the circle encloses the most area

The goal is to show, from the multiplier rule alone, that among all closed plane curves of a fixed perimeter the one enclosing maximum area is the circle, and to identify the multiplier as its radius $L/2\pi$ . We parameterize the curve by arc length so the length constraint becomes algebraic, treat the two coordinate functions as the unknowns, write the auxiliary integrand, and let the Euler–Lagrange equations force constant curvature.

Step 1. Parameterize the closed curve by arc length $s\in[0,L]$ , so that $x'^2+y'^2=1$ with $'=\mathrm{d}/\mathrm{d}s$ , and write the two functionals as $A=\tfrac12\int_0^L(x y'-y x')\,\mathrm{d}s$ and $L=\int_0^L\sqrt{x'^2+y'^2}\,\mathrm{d}s$ . Why: arc-length parameterization makes the perimeter integrand simply $1$ along the true curve and treats $x(s)$ and $y(s)$ symmetrically, which the geometry demands.

Step 2. Form the auxiliary integrand $h=\tfrac12(x y'-y x')-\lambda\sqrt{x'^2+y'^2}$ and write one Euler–Lagrange equation for each of $x$ and $y$ . Using $\sqrt{x'^2+y'^2}=1$ on the extremal, these reduce to $y'+\lambda x''=0$ and $-x'+\lambda y''=0$ . Why: with two unknown functions the single scalar constraint still needs only one multiplier, and it enters both equations identically — the hallmark of an integral, not pointwise, constraint.

Step 3. Integrate each once to get $y+\lambda x'=c_1$ and $-x+\lambda y'=c_2$ , then differentiate and substitute to decouple, yielding $x''=(x-c_2)/\lambda^2$ and $y''=(y-c_1)/\lambda^2$ . Combining, $(x-c_2)^2+(y-c_1)^2=\lambda^2$ . Why: constant curvature $1/\lambda$ pops out with no further assumptions, and constant curvature is the geometric signature of a circle of radius $|\lambda|$ .

Step 4. Fix the multiplier with the constraint: a circle of radius $|\lambda|$ has perimeter $L=2\pi|\lambda|$ , so $\lambda=L/2\pi$ , and its area is $A=\pi\lambda^2=L^2/4\pi$ , which saturates $4\pi A\le L^2$ . Why: the constraint is the extra equation that turns the family of extremals into a definite answer, and the multiplier acquires a clean physical reading — the radius, equivalently the reciprocal curvature.

Go deeper · Derivation C · 1 minThe hanging chain and the Beltrami first integral

Derivation CThe hanging chain and the Beltrami first integral

The goal is to derive the shape of a uniform flexible chain of fixed length hung between two fixed points by minimizing its gravitational potential energy, and to show that because the auxiliary integrand is autonomous, the Beltrami identity reduces the problem to a first-order equation whose solution is the catenary. The multiplier here is a physical constant tied to the chain's tension, and it is fixed jointly with the integration constants by the endpoints and the length.

Step 1. Write the energy and length functionals, $U=\rho g\int y\sqrt{1+y'^2}\,\mathrm{d}x$ and $L=\int\sqrt{1+y'^2}\,\mathrm{d}x$ , and form the auxiliary integrand $h=(\rho g\,y-\lambda)\sqrt{1+y'^2}$ . Dividing through by $\rho g$ , set $\mu=\lambda/\rho g$ so that $h\propto(y-\mu)\sqrt{1+y'^2}$ . Why: the potential energy of a heavy chain is its weight times the height of its centre of mass, which is exactly $\int y\,\mathrm{d}s$ ; subtracting $\lambda$ times the length enforces the fixed-length constraint.

Step 2. Note that $h$ has no explicit $x$ , so the Beltrami first integral applies: $h-y'\,\partial h/\partial y'=C$ . Computing the derivative and simplifying gives $(y-\mu)/\sqrt{1+y'^2}=C$ , that is $y-\mu=C\sqrt{1+y'^2}$ . Why: an $x$ -independent integrand conserves the Beltrami quantity, collapsing a second-order Euler–Lagrange equation to a first-order one and saving all the labour of differentiating a square root twice.

Step 3. Separate variables, $\mathrm{d}x=\dfrac{\mathrm{d}y}{\sqrt{(y-\mu)^2/C^2-1}}$ , and integrate to obtain $y-\mu=C\cosh\!\dfrac{x-x_0}{C}$ — the catenary. Why: the integral of $1/\sqrt{u^2-1}$ is $\operatorname{arcosh}$ , so inverting gives the hyperbolic cosine; the shape is not a parabola, a distinction Galileo guessed wrong and Leibniz, Huygens, and Johann Bernoulli settled in 1691.

Step 4. Determine the three constants $x_0$ , $C$ , and $\mu$ from the two endpoint heights and the length constraint $L=\int\sqrt{1+y'^2}\,\mathrm{d}x=2C\sinh(a/C)$ for a symmetric span of half-width $a$ . The parameter $C=T_H/(\rho g)$ is the horizontal tension divided by weight per length, and $\mu$ sets the baseline. Why: the multiplier is not left dangling — it and the integration constants are fixed together by the boundary data and the length, which is the general pattern for every isoperimetric problem.

Limiting cases

Special structures collapse the general machinery to simpler, familiar problems:

Case	Result	Meaning
No constraint ( $K$ trivial)	$\lambda=0$	The auxiliary integrand is just $f$ ; recovers the ordinary Euler–Lagrange problem.
Autonomous integrand	$h-y'\,\dfrac{\partial h}{\partial y'}=\mathrm{const}$	If $h=f-\lambda g$ has no explicit $x$ , the Beltrami first integral reduces the order by one.
Multiple constraints	$\delta\!\left(J-\sum_i\lambda_i K_i\right)=0$	Each integral constraint $K_i=\ell_i$ contributes its own constant multiplier $\lambda_i$ .
Integral vs pointwise	$\lambda=\mathrm{const}$ vs $\lambda(x)$	One integral constraint gives a constant; a holonomic constraint $g(x,y)=0$ at every $x$ needs a function $\lambda(x)$ .
Degenerate constraint	$\mu_0\,\delta J=\lambda\,\delta K$	If $\delta K\equiv0$ at the candidate, the normal rule fails; an "abnormal" multiplier $\mu_0$ (possibly zero) is required.

Common trapThe multiplier is a constant, and the constraint is one of your equations

Two traps snare almost everyone. First: in an isoperimetric (integral-constraint) problem the multiplier $\lambda$ is a single constant, not a function of $x$ . A function $\lambda(x)$ belongs to a pointwise (holonomic) constraint $g(x,y)=0$ that must hold at every point — that is infinitely many conditions and needs infinitely many multipliers. Confusing the two is the most common conceptual error in the subject. Second: do not treat $\lambda$ as a free parameter you may leave undetermined. The constraint equation $K[y]=\ell$ is not optional scenery — it is one of the equations you must solve, and it is precisely what fixes $\lambda$ together with the integration constants. An answer with an unpinned $\lambda$ is an unfinished answer. Finally, remember the method assumes the constraint is non-degenerate ( $\delta K\not\equiv0$ at the extremal); at points where the candidate is itself an extremal of $K$ the simple rule can break down.

Worked examples

1 · How much area you lose by not being a circle (confidence builder)

A closed loop of string has perimeter $L=1\ \mathrm{m}$ . Compute the area it encloses as a circle, as a square, and as an equilateral triangle, and form the isoperimetric ratio $Q=4\pi A/L^2$ for each.

Step 1. Circle. Perimeter $L=2\pi r$ gives $r=L/2\pi=1/(2\pi)=0.15915\ \mathrm{m}$ , and $A=\pi r^2=L^2/4\pi=1/(4\pi)=\mathbf{0.07958\ m^2}$ . Its ratio is $Q=4\pi A/L^2=4\pi\cdot(1/4\pi)=\mathbf{1.000}$ — the circle saturates the inequality by construction.

Step 2. Square. Side $=L/4=0.25\ \mathrm{m}$ , so $A=(0.25)^2=\mathbf{0.06250\ m^2}$ and $Q=4\pi(0.0625)=\mathbf{0.785}$ (exactly $\pi/4$ ). The square encloses only $0.0625/0.07958=78.5\%$ of the circle's area at the same perimeter.

Step 3. Equilateral triangle. Side $=L/3=0.3333\ \mathrm{m}$ , and $A=\tfrac{\sqrt3}{4}(L/3)^2=\sqrt3/36=\mathbf{0.04811\ m^2}$ , giving $Q=4\pi(0.04811)=\mathbf{0.605}$ (exactly $\pi/3\sqrt3$ ). The triangle captures only $60.5\%$ of the circle's area. Circle $Q=1.000$ , square $Q=0.785$ , triangle $Q=0.605$ : rounder is always bigger, and only the circle reaches $Q=1$ .

2 · The parameter and sag of a hanging chain

A uniform chain of length $L=2.3504\ \mathrm{m}$ hangs between two supports at equal height, a horizontal distance $2a=2\ \mathrm{m}$ apart (supports at $x=\pm1\ \mathrm{m}$ ). Find the catenary parameter $C$ and the sag (how far the lowest point falls below the supports).

Step 1. From Derivation C the shape is $y=C\cosh(x/C)$ (choosing the low point at $x=0$ ), and the length of a symmetric span is $L=2C\sinh(a/C)$ . Insert the data: $2.3504=2C\sinh(1/C)$ , i.e. $C\sinh(1/C)=1.1752$ . This is transcendental — it fixes $C$ implicitly.

Step 2. Solve numerically. Writing $u=1/C$ , the equation is $\sinh(u)=1.1752\,u$ . Since $\sinh(1)=1.17520$ , the root is $u=1.000$ , so $\mathbf{C=1.000\ m}$ . The multiplier physics: $C=T_H/(\rho g)$ , so the horizontal tension is $T_H=\rho g\,C=\rho g\times1\ \mathrm{m}$ — a taut chain has large $C$ , a slack one small $C$ .

Step 3. The lowest point sits at $y(0)=C=1.000\ \mathrm{m}$ ; the supports at $x=\pm1$ sit at $y(1)=C\cosh(1/C)=\cosh(1)=1.5431\ \mathrm{m}$ . The sag is $y(1)-y(0)=1.5431-1.000=\mathbf{0.543\ m}$ . A chain 2.35 m long across a 2 m gap forms a catenary of parameter $C=1$ m and sags 0.543 m — not a parabola, though over a shallow span the two are nearly indistinguishable.

3 · Normalization as a constraint gives the eigenvalue equation

Show that minimizing the energy functional of a particle subject to fixed normalization reproduces the time-independent Schrödinger equation, and identify the multiplier.

Step 1. Take $J[\psi]=\int\big(\tfrac{\hbar^2}{2m}|\psi'|^2+V|\psi|^2\big)\mathrm{d}x$ as the objective and $K[\psi]=\int|\psi|^2\,\mathrm{d}x=1$ as the constraint. Form $h=\tfrac{\hbar^2}{2m}|\psi'|^2+V|\psi|^2-E|\psi|^2$ , calling the multiplier $E$ .

Step 2. The Euler–Lagrange equation for $h$ (treating $\psi$ and $\psi^\ast$ independently) is $-\tfrac{\hbar^2}{2m}\psi''+V\psi=E\psi$ — the time-independent Schrödinger equation. The multiplier is not a bookkeeping device: $E$ is the energy eigenvalue, and the constraint $\int|\psi|^2=1$ selects which eigenfunction and eigenvalue you land on.

Step 3. This is why the variational principle underlies practical quantum chemistry: minimizing $\langle\psi|H|\psi\rangle$ over trial functions with the normalization held fixed always returns an upper bound on the ground-state energy, and $E$ emerges as the multiplier. The ground state of any quantum system is the solution of an isoperimetric problem, with the energy playing the role of $\lambda$ .

Physics of the portrait

Isoperimetric problems are where the calculus of variations was born: Euler wrote the first systematic treatise on them, and the multiplier trick that tames them is his.

Origins · 1 minLeonhard Euler and how this was discovered

Leonhard Euler

1707–1783 · calculus of variations and the isoperimetric multiplier rule

Leonhard Euler, the most prolific mathematician in history, gave the calculus of variations both its name and its founding treatise. His 1744 book Methodus inveniendi lineas curvas maximi minimive proprietate gaudentes ("A method for finding curves enjoying a maximum or minimum property") collected and systematized the scattered optimization problems of the previous half-century — Newton's solid of least resistance, the Bernoullis' brachistochrone, the shape of the hanging chain — into a single theory built on what we now call the Euler–Lagrange equation. Crucially, an entire appendix is devoted to isoperimetric problems, where a curve must be optimal while a second integral is held fixed, and it is there that Euler works out how a single auxiliary constant absorbs an integral constraint. When the nineteen-year-old Joseph-Louis Lagrange wrote to him in 1755 with a slicker, purely analytic $\delta$ -algorithm that mechanized the whole business, Euler generously set aside his own geometric methods, renamed the field the "calculus of variations" after Lagrange's variation symbol, and delayed publishing so the young man could take the credit. The isoperimetric multiplier is the direct ancestor of the Lagrange multipliers of constrained optimization, of the Lagrangian of mechanics, and of the constrained functionals that pervade modern physics from density-functional theory to general relativity.

History

1691

— Leibniz, Huygens, and Johann Bernoulli show the hanging chain is a catenary, not a parabola

1744

— Euler's Methodus inveniendi founds the calculus of variations and treats isoperimetric problems

1755

— the young Lagrange sends Euler his δ-algorithm, systematizing constrained variation

1838

— Jakob Steiner gives synthetic proofs that the circle maximizes area at fixed perimeter

1902

— Hurwitz proves the isoperimetric inequality 4πA ≤ L² cleanly using Fourier series

Going further

Texts. Stone and Goldbart, Mathematics for Physics, Ch. 1 (Calculus of Variations) — the cleanest physics-oriented treatment of the multiplier method · Gelfand and Fomin, Calculus of Variations, §12 (the isoperimetric problem and the reciprocity theorem) · Arfken and Weber, Mathematical Methods for Physicists, the calculus-of-variations chapter · Riley, Hobson and Bence, Mathematical Methods for Physics and Engineering, Ch. 22 · Courant and Hilbert, Methods of Mathematical Physics, Vol. I, Ch. IV · Hassani, Mathematical Physics, and Byron and Fuller, Mathematics of Classical and Quantum Physics, both for the finite-to-infinite-dimensional analogy. The key anchor results are: the Dido derivation gives constant curvature, radius $L/2\pi$ , and area $L^2/4\pi$ ; the Beltrami integration yields the catenary $C\cosh((x-x_0)/C)+\mathrm{const}$ , with the numeric case $L=2.3504$ m across $2a=2$ m giving $C=1$ m and sag $0.543$ m; the isoperimetric ratios are $Q=1,\;0.785=\pi/4,\;0.605=\pi/3\sqrt3$ for the circle, square, and equilateral triangle at equal perimeter; and the normalization-constrained energy minimization reproduces $-\tfrac{\hbar^2}{2m}\psi''+V\psi=E\psi$ .

Builds on. Euler–Lagrange Equation · First Variation and Stationarity · Constrained Variation and Lagrange Multipliers · Functionals and Function Spaces. Leads to. Beltrami Identity and First Integrals · Geodesics as Variational Problems · Hamilton's Principle of Stationary Action.

Ask the tutor: Hand me any "make this stationary while that stays fixed" problem — a shape, a chain, a wavefunction — and I will build $J-\lambda K$ , run the Euler–Lagrange or Beltrami step, and pin down the multiplier with you.

What exactly makes a problem "isoperimetric" rather than an ordinary variational problem?

The presence of an integral constraint. An ordinary variational problem asks only that $J[y]=\int f\,\mathrm{d}x$ be stationary among curves with fixed endpoints. An isoperimetric problem adds the requirement that a second integral, $K[y]=\int g\,\mathrm{d}x$ , equal a prescribed value $\ell$ . The classical case fixes the perimeter (hence "iso-perimetric", equal-perimeter) while maximizing area, but the name now covers any integral constraint — fixed length, fixed charge, fixed normalization, fixed volume. The defining structure is the pair of integrals, one to extremize and one to hold fixed, tied together by a single constant multiplier.

Why can one constant handle a constraint on an entire function?

Because the constraint is a single number, not a condition at every point. $K[y]=\ell$ is one scalar equation, so it removes exactly one direction of freedom from the infinite-dimensional space of curves. The two-parameter variation argument makes this precise: enforcing $K=\ell$ ties one perturbation parameter to another, and eliminating it leaves the objective's variation proportional to the constraint's variation with a proportionality constant — the multiplier — that comes out the same no matter which test functions you use. That "same for every test function" is exactly what it means to be a constant. A pointwise constraint $g(x,y)=0$ , by contrast, is a separate condition at each $x$ , so it needs a whole function $\lambda(x)$ .

How do I actually find the value of λ in a concrete problem?

Never from the Euler–Lagrange equation alone — that equation contains $\lambda$ as an unknown parameter and gives you a family of solutions, one for each $\lambda$ . You pin $\lambda$ down by imposing the constraint. Solve the Euler–Lagrange equation for $h=f-\lambda g$ carrying $\lambda$ and the integration constants along, then substitute into $K[y]=\ell$ and the boundary conditions. Those extra equations fix everything at once. In Dido's problem the perimeter constraint gives $\lambda=L/2\pi$ ; in the catenary the length plus the two support heights fix the multiplier together with the integration constants. Treat the constraint as one of your equations, not as an afterthought.

Is the hanging chain really a catenary and not a parabola?

It is a catenary, $y=C\cosh(x/C)$ , and the distinction is real. Galileo guessed a parabola in 1638; Leibniz, Huygens, and Johann Bernoulli proved it was the hyperbolic cosine in 1691. The parabola is the shape of a suspension-bridge cable, which carries a load uniform in the horizontal coordinate (the roadway), so it minimizes energy against a different weight distribution. A free chain carries a load uniform along its own arc length, and that changes the integrand from $\int y\,\mathrm{d}x$ to $\int y\,\mathrm{d}s=\int y\sqrt{1+y'^2}\,\mathrm{d}x$ — the extra $\sqrt{1+y'^2}$ is exactly what turns the parabola into a catenary. Over a shallow, nearly flat span the two curves are almost identical, which is why the confusion persisted.

What is the physical meaning of the multiplier λ?

It is a price, or an exchange rate, between the objective and the constraint — mathematically, the sensitivity of the extremal value of $J$ to the constraint level $\ell$ , so $\lambda=\mathrm{d}J^\ast/\mathrm{d}\ell$ . Its concrete identity changes with the problem. In Dido's problem $\lambda$ is the radius of the optimal circle (equivalently the reciprocal of the constant curvature). In the catenary the associated constant $C=T_H/(\rho g)$ is fixed by the horizontal tension, so $\lambda$ measures the cost of buying more length. For a soap film at fixed volume the multiplier is the pressure difference across the surface. In quantum normalization it is the energy eigenvalue. In every case $\lambda$ is a genuine physical quantity, not just algebraic scaffolding.

Does the method extend to several constraints or higher dimensions?

Yes, transparently. For several integral constraints $K_1=\ell_1,\dots,K_m=\ell_m$ you introduce one constant multiplier per constraint and make $J-\sum_i\lambda_i K_i$ stationary; the $m$ constraint equations then fix the $m$ multipliers. For a curve in the plane, like Dido's, you simply write one Euler–Lagrange equation per coordinate function while still using a single multiplier for the single scalar constraint. Surfaces and fields work the same way, with the ordinary derivative promoted to a functional (variational) derivative and the multiplier still constant for each integral constraint. The one caveat is degeneracy: if the constraints are not independent, or if the candidate is itself a stationary point of some $K_i$ , the naive rule can fail and an "abnormal" multiplier is needed.