Extensions of the Euler–Lagrange Equation
CSIR NET calculus of variations: Euler–Poisson, systems, Euler–Ostrogradsky for PDEs, isoperimetric problems, Lagrange multipliers and transversality.
- 20 min
On this page · 15 sections
- Prerequisites
- The Hook: Dido’s Rope
- The Pattern Behind Every Extension
- Functionals with Higher Derivatives: Euler–Poisson
- Several Unknown Functions
- Several Independent Variables: Euler–Ostrogradsky
- Isoperimetric Problems
- Variable End Points and Transversality
- Memory Tricks
- Don’t Confuse
- Common Mistakes
- Practice
- Quick Revision Table
- Key Takeaways
- What’s Next
Prerequisites
- Functionals, First Variation and the Euler–Lagrange Equation: every result here is the same three steps (vary, integrate by parts, apply the fundamental lemma).
- Metric Spaces and Several Variables: Lagrange multipliers for constrained extrema in \( \mathbb{R}^n \), which the isoperimetric rule copies.
- Green’s theorem (divergence theorem in the plane), used for functionals of several variables.
The Hook: Dido’s Rope
The Pattern Behind Every Extension
graph LR
A["Write J of y plus epsilon eta"] --> B["Differentiate at epsilon = 0"]
B --> C["Integrate by parts until only eta appears"]
C --> D["Fundamental lemma: bracket = 0"]
C --> E["Boundary terms: natural or transversality conditions"]Each extra derivative in \( F \) means one more integration by parts, so one more sign change and one more boundary term. That single observation reproduces every formula below.
Functionals with Higher Derivatives: Euler–Poisson
For \( J[y] = \int_a^b F(x, y, y', y'')\, dx \) with \( y \) and \( y' \) prescribed at both ends, the first variation is \( \int (F_y\eta + F_{y'}\eta' + F_{y''}\eta'')\, dx \). Integrating the second term by parts once and the third twice:
$$\boxed{F_y - \frac{d}{dx}F_{y'} + \frac{d^2}{dx^2}F_{y''} = 0.}$$In general, for \( F(x, y, y', \dots, y^{(n)}) \):
$$F_y - \frac{d}{dx}F_{y'} + \frac{d^2}{dx^2}F_{y''} - \cdots + (-1)^n\frac{d^n}{dx^n}F_{y^{(n)}} = 0.$$| Feature | Value |
|---|---|
| Order of the Euler–Poisson equation | \( 2n \) (when \( F_{y^{(n)}y^{(n)}} \ne 0 \)) |
| Boundary conditions needed | \( 2n \): \( y, y', \dots, y^{(n-1)} \) at both ends |
| Sign pattern | alternating: \( +, -, +, \dots \) with the \( k \)-th total derivative of \( F_{y^{(k)}} \) |
| Natural conditions (e.g. \( n = 2 \), free \( y(b), y'(b) \)) | \( F_{y''} = 0 \) and \( F_{y'} - \frac{d}{dx}F_{y''} = 0 \) at \( b \) |
Several Unknown Functions
For \( J[y_1, \dots, y_n] = \int_a^b F(x, y_1, \dots, y_n, y_1', \dots, y_n')\, dx \), vary one function at a time. You get a system, one equation per function:
$$F_{y_i} - \frac{d}{dx}F_{y_i'} = 0, \qquad i = 1, \dots, n.$$A system of \( n \) second-order equations has order \( 2n \), so the general extremal involves \( 2n \) constants, matched by \( 2n \) end values.
Hamilton’s principle
For a mechanical system with generalised coordinates \( q_1, \dots, q_n \) and Lagrangian \( L = T - V \), Hamilton’s principle says the actual motion makes the action \( \int_{t_0}^{t_1} L\, dt \) stationary. The E–L system is then Lagrange’s equations
$$\frac{d}{dt}\frac{\partial L}{\partial \dot q_i} - \frac{\partial L}{\partial q_i} = 0.$$The first integrals of the previous page reappear as conservation laws: if \( L \) has no \( q_i \), the momentum \( \frac{\partial L}{\partial \dot q_i} \) is conserved; if \( L \) has no explicit \( t \), the Beltrami quantity \( \sum \dot q_i\frac{\partial L}{\partial \dot q_i} - L \) (the energy, for natural systems) is conserved. The mechanics side is developed in Generalized Coordinates and Lagrange’s Equations.
Several Independent Variables: Euler–Ostrogradsky
For \( u(x, y) \) on a region \( D \) with \( u \) prescribed on the boundary \( \partial D \), and
$$J[u] = \iint_D F(x, y, u, u_x, u_y)\, dx\, dy,$$the first variation is \( \iint (F_u\eta + F_{u_x}\eta_x + F_{u_y}\eta_y) \). Green’s theorem moves the derivatives off \( \eta \) (the boundary integral vanishes because \( \eta = 0 \) on \( \partial D \)), giving the Euler–Ostrogradsky equation
$$\boxed{F_u - \frac{\partial}{\partial x}F_{u_x} - \frac{\partial}{\partial y}F_{u_y} = 0,}$$a partial differential equation. Here \( \frac{\partial}{\partial x} \) is again a total derivative in \( x \) (with \( y \) fixed): it acts on \( u \), \( u_x \) and \( u_y \) as well. In three variables add \( -\frac{\partial}{\partial z}F_{u_z} \).
| \( F \) | Euler–Ostrogradsky equation | Name |
|---|---|---|
| \( u_x^2 + u_y^2 \) | \( u_{xx} + u_{yy} = 0 \) | Laplace (Dirichlet integral) |
| \( u_x^2 + u_y^2 + 2f(x, y)u \) | \( \Delta u = f \) | Poisson |
| \( u_x^2 + u_y^2 - k^2u^2 \) | \( \Delta u + k^2u = 0 \) | Helmholtz |
| \( \frac{1}{2}(u_t^2 - c^2u_x^2) \) | \( u_{tt} = c^2u_{xx} \) | wave equation (action of a string) |
| \( u_xu_y \) | \( u_{xy} = 0 \) | |
| \( \sqrt{1 + u_x^2 + u_y^2} \) | \( (1 + u_y^2)u_{xx} - 2u_xu_yu_{xy} + (1 + u_x^2)u_{yy} = 0 \) | minimal surface equation |
These are the functionals used, in reverse, to solve PDE boundary value problems on Variational Methods for Boundary Value Problems.
Isoperimetric Problems
Problem. Extremise \( J[y] = \int_a^b F(x, y, y')\, dx \) subject to fixed ends and the integral constraint
$$K[y] = \int_a^b G(x, y, y')\, dx = \ell.$$This is the function-space version of Lagrange multipliers. The rule is symmetric: the extremals of “\( J \) with \( K \) fixed” and of “\( K \) with \( J \) fixed” are the same family (the reciprocity principle).
Dido’s problem solved
Maximise the area \( \int y\, dx \) with the length \( \int \sqrt{1 + y'^2}\, dx = \ell \) fixed. \( H = y + \lambda\sqrt{1 + y'^2} \) has no \( x \), so by Beltrami
$$y + \lambda\sqrt{1 + y'^2} - \frac{\lambda y'^2}{\sqrt{1 + y'^2}} = y + \frac{\lambda}{\sqrt{1 + y'^2}} = c.$$Thus \( (y - c)^2(1 + y'^2) = \lambda^2 \), whose solutions are the circles \( (x - x_0)^2 + (y - c)^2 = \lambda^2 \). Isoperimetric extremals for area are circular arcs; with the ends at \( (\pm 1, 0) \) and \( \ell = \pi \) the arc is the unit semicircle, enclosing area \( \frac{\pi}{2} \).
Eigenvalue problems are isoperimetric problems
Extremise \( \int_a^b (py'^2 + qy^2)\, dx \) subject to \( \int_a^b ry^2\, dx = 1 \), \( y(a) = y(b) = 0 \). With \( H = py'^2 + qy^2 - \lambda ry^2 \) the E–L equation is
$$-(py')' + qy = \lambda ry,$$the Sturm–Liouville eigenvalue problem (see Sturm–Liouville Problems and Green’s Function). The multiplier is the eigenvalue, and the minimum of the functional is the smallest eigenvalue. This is the basis of the Rayleigh–Ritz method.
Pointwise (holonomic) constraints
For a constraint \( g(x, y, z) = 0 \) that must hold at every point (a curve on a surface, as in geodesic problems), the multiplier becomes a function \( \lambda(x) \): use \( H = F + \lambda(x)g \) and treat \( \lambda \) as one more unknown function.
Variable End Points and Transversality
Suppose the left end is fixed and the right end \( (x_1, y(x_1)) \) may slide along a given curve \( y = \varphi(x) \). Varying both the curve and the end point, \( \delta J = 0 \) gives the E–L equation and, at the free end, the transversality condition
$$\boxed{\Big[F + (\varphi' - y')F_{y'}\Big]_{x = x_1} = 0.}$$| End-point freedom | Condition at the free end |
|---|---|
| on the curve \( y = \varphi(x) \) | \( F + (\varphi' - y')F_{y'} = 0 \) |
| on a vertical line \( x = x_1 \) (\( y \) free) | \( F_{y'} = 0 \) (the natural condition) |
| on a horizontal line \( y = y_1 \) (\( x_1 \) free) | \( F - y'F_{y'} = 0 \) |
| \( F = f(x, y)\sqrt{1 + y'^2} \), \( f \ne 0 \), end on \( y = \varphi(x) \) | \( y'\varphi' = -1 \): the extremal meets the curve orthogonally |
For the last row: \( F + (\varphi' - y')F_{y'} = f\,\frac{(1 + y'^2) + (\varphi' - y')y'}{\sqrt{1 + y'^2}} = f\,\frac{1 + \varphi'y'}{\sqrt{1 + y'^2}} \).
Memory Tricks
Don’t Confuse
| This | vs | That |
|---|---|---|
| Euler–Poisson (one variable, higher derivatives, ODE of order \( 2n \)) | Euler–Ostrogradsky (several variables, PDE) | |
| isoperimetric constraint \( \int G = \ell \): \( \lambda \) constant | holonomic constraint \( g = 0 \): \( \lambda(x) \) a function | |
| natural condition \( F_{y'} = 0 \) (end on a vertical line) | transversality \( F + (\varphi' - y')F_{y'} = 0 \) (end on a curve) | |
| \( \frac{\partial}{\partial x}F_{u_x} \) in Euler–Ostrogradsky (total in \( x \), \( y \) fixed) | \( F_{xu_x} \) (partial with \( u, u_x, u_y \) frozen) | |
| the system for \( (y, z) \): order 4, four constants | each equation alone: order 2 |
Common Mistakes
Practice
All questions below are original practice questions written for this page. They are not from any CSIR NET paper. In Level 3, as in Part C, one or more options may be correct.
Level 1: Foundation
- What is the order of the Euler–Poisson equation for \( F(x, y, y', y'', y''') \), and how many boundary conditions does the fixed-end problem need?
- Write the Euler–Ostrogradsky equation of \( \iint (u_x^2 + u_y^2)\, dx\, dy \).
- In an isoperimetric problem, is the multiplier \( \lambda \) a constant or a function of \( x \)?
Answers
- Order 6 (twice the highest derivative, 3); six conditions: \( y, y', y'' \) at both ends.
- \( -2u_{xx} - 2u_{yy} = 0 \), i.e. Laplace’s equation \( u_{xx} + u_{yy} = 0 \).
- A constant. (A pointwise constraint \( g(x, y, z) = 0 \) needs a function \( \lambda(x) \).)
Level 2: Part B Style
Practice Q1. The extremal of \( J[y] = \int_0^1 (y''^2 - 48y)\, dx \) with \( y(0) = 0 \), \( y'(0) = 0 \), \( y(1) = 1 \), \( y'(1) = 4 \) is:
(a) \( x^4 \) (b) \( x^3 \) (c) \( 3x^2 - 2x^3 \) (d) \( 2x^4 - x^2 \)
Answer: (a)
Euler–Poisson: \( F_y + \frac{d^2}{dx^2}F_{y''} = -48 + 2y'''' = 0 \), so \( y'''' = 24 \) and \( y = x^4 + c_3x^3 + c_2x^2 + c_1x + c_0 \). The four conditions force \( c_0 = c_1 = c_2 = c_3 = 0 \). (b) fails \( y'''' = 24 \) and has \( y'(1) = 3 \); (c) has \( y'(1) = 0 \); (d) has \( y(1) = 1 \) but \( y'(1) = 6 \).
Practice Q2. The Euler–Ostrogradsky equation of \( J[u] = \iint_D (u_x^2 + u_y^2 + 2f(x, y)u)\, dx\, dy \) is:
(a) \( \Delta u = 0 \) (b) \( \Delta u = f \) (c) \( \Delta u = -f \) (d) \( \Delta u = 2f \)
Answer: (b)
\( F_u - \partial_xF_{u_x} - \partial_yF_{u_y} = 2f - 2u_{xx} - 2u_{yy} = 0 \).
Practice Q3. Among \( y \in C^2[0, \pi] \) with \( y(0) = y(\pi) = 0 \) and \( \int_0^\pi y^2\, dx = 1 \), the minimum of \( \int_0^\pi y'^2\, dx \) is:
(a) 1 (b) \( \pi \) (c) \( \frac{2}{\pi} \) (d) \( \pi^2 \)
Answer: (a)
\( H = y'^2 - \lambda y^2 \) gives \( y'' + \lambda y = 0 \) with \( y(0) = y(\pi) = 0 \): \( \lambda = n^2 \), \( y = \sqrt{2/\pi}\,\sin nx \). On such \( y \), \( \int y'^2 = \lambda\int y^2 = n^2 \) (multiply the equation by \( y \) and integrate by parts). The smallest is \( n = 1 \): minimum 1. (This is Wirtinger’s inequality \( \int_0^\pi y'^2 \ge \int_0^\pi y^2 \).)
Practice Q4. The shortest distance from the origin to the line \( x + 2y = 5 \), found as an extremal of \( \int\sqrt{1 + y'^2}\, dx \) with the transversality condition, is:
(a) \( \sqrt{5} \) (b) 5 (c) \( \frac{5}{\sqrt{3}} \) (d) 2
Answer: (a)
Extremals through the origin: \( y = mx \). The line has slope \( \varphi' = -\frac{1}{2} \); orthogonality \( m\varphi' = -1 \) gives \( m = 2 \). Intersection: \( x + 4x = 5 \), so \( (1, 2) \), at distance \( \sqrt{1 + 4} = \sqrt{5} \).
Level 3: Part C Style
Practice Q5. Consider \( J[y, z] = \int_0^{\pi/2} (y'^2 + z'^2 + 2yz)\, dx \) with \( y(0) = 0 \), \( y(\frac{\pi}{2}) = 1 \), \( z(0) = 0 \), \( z(\frac{\pi}{2}) = -1 \). Which of the following are true?
(a) Every extremal satisfies \( y'''' = y \). (b) \( y = \sin x \), \( z = -\sin x \) is an extremal satisfying the boundary conditions. (c) \( y = \frac{\sinh x}{\sinh(\pi/2)} \), \( z = -y \) is an extremal satisfying the boundary conditions. (d) The general extremal contains eight arbitrary constants.
Answer: (a) and (b)
- (a) True. The system is \( y'' = z \), \( z'' = y \); so \( y'''' = z'' = y \).
- (b) True. \( y'' = -\sin x = z \) and \( z'' = \sin x = y \); the four boundary values hold.
- (c) False. With \( z = -y \), \( z'' = -y'' = -y \ne y \); the equation \( z'' = y \) fails.
- (d) False. Two second-order equations: order 4, so four constants (\( y \) carries four, and \( z = y'' \) adds none).
Practice Q6. Which of the following Euler–Ostrogradsky equations are correct?
(a) \( \iint \sqrt{1 + u_x^2 + u_y^2}\, dx\, dy \): \( (1 + u_y^2)u_{xx} - 2u_xu_yu_{xy} + (1 + u_x^2)u_{yy} = 0 \). (b) \( \iint \frac{1}{2}(u_t^2 - c^2u_x^2)\, dx\, dt \): \( u_{tt} = c^2u_{xx} \). (c) \( \iint (u_x^2 + u_y^2 + u^2)\, dx\, dy \): \( \Delta u = 0 \). (d) \( \iint u_xu_y\, dx\, dy \): \( u_{xy} = 0 \).
Answer: (a), (b) and (d)
- (a) True. With \( W = \sqrt{1 + u_x^2 + u_y^2} \), the equation \( \partial_x\frac{u_x}{W} + \partial_y\frac{u_y}{W} = 0 \), multiplied by \( W^3 \), is the stated minimal surface equation.
- (b) True. \( -\partial_t(u_t) - \partial_x(-c^2u_x) = 0 \).
- (c) False. \( F_u = 2u \) gives \( 2u - 2\Delta u = 0 \), i.e. \( \Delta u = u \).
- (d) True. \( F_{u_x} = u_y \), \( F_{u_y} = u_x \): \( -\partial_xu_y - \partial_yu_x = -2u_{xy} = 0 \).
Practice Q7. Among curves of length \( \pi \) joining \( (-1, 0) \) to \( (1, 0) \) and lying in \( y \ge 0 \), we maximise the area \( \int_{-1}^1 y\, dx \). Which of the following are true?
(a) The extremals of the associated Euler–Lagrange equation are circular arcs. (b) The maximising curve is \( x^2 + y^2 = 1 \), \( y \ge 0 \). (c) The maximum area is \( \frac{\pi}{2} \). (d) The Lagrange multiplier must be taken as a function \( \lambda(x) \).
Answer: (a), (b) and (c)
- (a) True. \( H = y + \lambda\sqrt{1 + y'^2} \); Beltrami gives \( (y - c)^2(1 + y'^2) = \lambda^2 \), circles of radius \( \lvert\lambda\rvert \).
- (b) True. A circular arc of length \( \pi \) through \( (\pm 1, 0) \): the unit semicircle has length \( \pi \) and fits.
- (c) True. Half the unit disc: \( \frac{\pi}{2} \).
- (d) False. The constraint is an integral (isoperimetric), so \( \lambda \) is a constant.
Quick Revision Table
| Problem | Equation or condition |
|---|---|
| \( F(x, y, y', y'') \) | \( F_y - (F_{y'})' + (F_{y''})'' = 0 \); order 4; \( y, y' \) at both ends |
| \( F(x, y_1, \dots, y_n, y_1', \dots, y_n') \) | \( F_{y_i} - (F_{y_i'})' = 0 \) for each \( i \); \( 2n \) constants |
| \( F(x, y, u, u_x, u_y) \) | \( F_u - \partial_xF_{u_x} - \partial_yF_{u_y} = 0 \) |
| Dirichlet integral | Laplace; with \( 2fu \): Poisson \( \Delta u = f \) |
| Isoperimetric \( \int G = \ell \) | E–L of \( F + \lambda G \), \( \lambda \) constant, found from \( \int G = \ell \) |
| Holonomic \( g = 0 \) | \( F + \lambda(x)g \) |
| Sturm–Liouville | min of \( \int(py'^2 + qy^2) \) with \( \int ry^2 = 1 \) is \( \lambda_1 \) |
| End on \( y = \varphi(x) \) | \( F + (\varphi' - y')F_{y'} = 0 \) |
| \( F = f\sqrt{1 + y'^2} \) | transversality = orthogonality |
| Hamilton’s principle | \( \delta\int L\, dt = 0 \) gives Lagrange’s equations |
Key Takeaways
What’s Next
- Conditions for Extrema: which extremals actually minimise.
- Variational Methods for Boundary Value Problems: Ritz, Galerkin and Rayleigh quotients built on the functionals of this page.
- Back to Functionals, First Variation and the Euler–Lagrange Equation for the basic case.
- Calculus of Variations: Quick Revision.