I am new to calculus of variations, till now I know how to get the extremal functions for a given functional using Euler-Lagrange equation.

What if I have a functional but I am not looking for minimizing/maximizing it, but instead solving equations involving functionals, say: $$I = \int_{x_{1}}^{x_2}{F(x,y,y') \,\mathrm{d}x}=\alpha\quad \,,\text{for }\alpha\in \mathbb{R}$$ How to solve for $y(x)$ that satisfy this equation? can I transform it to a classical problem then solve it using Euler-Lagrange equation?

EDIT: for example say, we have the following problem : $$I = \int_{0}^{1}{\left(f(x)+2f'(x)\right) \, \mathrm{d}x}= 1/2$$

I appreciate any ideas,

Thank you

  • 2
    $\begingroup$ It's possible this could be "transform[ed].... to a classical problem", but not enough information about $F$ is given to suggest ways that might succeed. It appears that you are trying to determine a function $y$ (on what interval?) from a single scalar equation, which generically is not enough information to make a full determination of $y$. $\endgroup$
    – hardmath
    Jun 23, 2016 at 0:11
  • $\begingroup$ y(x) for $x \in \, [x_1,\, x_2]$ $\endgroup$
    – Freshman42
    Jun 23, 2016 at 0:17
  • $\begingroup$ I hope you take my point, that a single equation is not enough to determine the infinite degrees of freedom that define $y(x)$ on $[x_1,x_2]$. There are variational formulations (see Galerkin methods) that use a family of integral equations to determine a function $y(x)$ (and for which the variational formulation may be more tractable than the classical formulation of the problem). $\endgroup$
    – hardmath
    Jun 23, 2016 at 0:30
  • $\begingroup$ @Hardmath I really do not understand what kind of additional info to solve the problem, can you give me an example please ( for me it seems logical because I replaced the min/max requirement by $\alpha$)? $\endgroup$
    – Freshman42
    Jun 23, 2016 at 0:36
  • $\begingroup$ @Freshman42 To your example: choose some arbitrary function $g$ which is finite with finite derivative on $[0,1]$. Now, compute $I_g=\int_0^1g(t)+2g'(t)\ dt$. Finally, set $f\equiv \frac{1}{2I_g}g$ (this assumes that $I_g$ isn't zero, of course, but it will only be zero for a small number of functions). Now this $f$ - which is essentially arbitrary - satisfies your equation. $\endgroup$ Jun 23, 2016 at 0:49

2 Answers 2


You can rig this to have unique solutions for very particular $\alpha$ (e.g. $F = (y-f)^2$, $\alpha = 0$ has unique solution $y=f$) but in general you should expect a large family of solutions.

Intuitively this is because the space of functions is much bigger than the space of possible values of the functional $J(y)=\int F[y]$, so $J$ can't be anything close to injective. Making this rigorous isn't quite as simple as a cardinality argument (since e.g. $C^1([0,1])$ has the same cardinality as $\mathbb R$), but reasonable requirements on $F$ will make $J$ a differentiable map when restricted to a finite-dimensional space of functions. Sard's theorem then tells you that $J(y)=\alpha$ has multiple solutions (or no solution) for almost every $\alpha$, even amongst a 2-parameter family of functions $y$.

This shouldn't be too surprising - prescribing the value of the functional (a single real number) is much less information than prescribing the derivative of the functional (an element of some infinite-dimensional function space).

Another way of putting it: You're prescribing some kind of average of $y$, but not the local behaviour.


Suppose we have a point particle of mass $m$ moving along the $y$ axis. The Lagrangian is the difference between the kinetic ($K$) and potential ($U$) energies

$$\mathcal{L} (y, \dot y) := K - U = \frac{1}{2} m \dot y^2 - m g y$$

and the action is the functional

$$S (y) := \int_0^T \mathcal{L} (y, \dot y) \, \mathrm{d}t$$

The Euler-Lagrange equation gives us the 2nd order differential equation $\ddot y = -g$. However, suppose we have the equality constraint

$$\int_0^T \mathcal{L} (y, \dot y) \, \mathrm{d}t = S_0$$

which can be rewritten in the form

$$\frac{S_0}{T} = \left(\frac{1}{T} \int_0^T \frac{1}{2} m \dot y^2 \, \mathrm{d}t \right)- \left(\frac{1}{T} \int_0^T m g y \, \mathrm{d}t \right) = \langle K \rangle - \langle U \rangle$$

where $\langle K \rangle$ and $\langle U \rangle$ are the average kinetic and potential energies, respectively. Thus, the equality constraint merely imposes a constraint on the average kinetic and potential energies.

  • 1
    $\begingroup$ thank you but $K = \dfrac{1}{2}m\dot y^2$, can you apply this for the given example above? $\endgroup$
    – Freshman42
    Jun 23, 2016 at 0:53
  • $\begingroup$ @Freshman42 Thanks. $\endgroup$ Jun 23, 2016 at 0:55
  • 1
    $\begingroup$ @Freshman42 $$\int_0^1 f (x) \, \mathrm{d}x + 2 f (1) - 2 f (0) = \frac{1}{2}$$ $\endgroup$ Jun 23, 2016 at 1:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.