4

Let $f(0)=0, f(1)=t $ and $$ \int_0^1 f(x) dx = 1 $$ How find the minimum length of $f(x)$ in $[0, 1]$ ?

For example when t=2, the minimum length of f(x) is the segment of OA

(O is the origin and A=(1, 2))

But how find the minimum length when t=3 ?

c-2785
  • 361

2 Answers2

1

Following Gil Strang's calculus of variations notes online, you want to use a Lagrangian of the form $$ \mathscr{L}(y,y',\lambda) = \int_0^1 \sqrt{1+(y')^2}\,dx + \lambda \left(\int_0^1 y\,dx - 1\right) = \int_0^1\left(\sqrt{1+y'^2}+\lambda y\right)\,dx -\lambda $$ Let the term in parentheses be $F(y,y',\lambda)$. If $\dfrac{\delta \mathscr{L}}{\delta y} =0$, the Euler-Lagrange equations are $$ 0 = \frac{\partial F}{\partial y} -\frac{d}{dx}\frac{\partial F}{\partial y'} = \lambda - \frac{d}{dx}\frac{y'}{\sqrt{1+(y')^2}} $$ So we can integrate once to get: $$ \frac{y'}{\sqrt{1+y'^2}} = \lambda x + c $$ for some constant $c$. Solving, $$ y' = \frac{\lambda x + c}{\sqrt{1-(\lambda x + c)^2}} $$ Let $u = \lambda x + c$. Then $$ y = \int \frac{\lambda x + c}{\sqrt{1-(\lambda x + c)^2}}\,dx = \frac{1}{\lambda}\int \frac{u}{\sqrt{1-u^2}}\,du = -\frac{1}{\lambda} \sqrt{1-u^2} + d $$ Or, \begin{align*} (\lambda y - \lambda d)^2 &= 1-(\lambda x + c)^2\\ \implies \left(x + \frac{c}{\lambda}\right)^2 + (y-d)^2 &= \frac{1}{\lambda^2} \end{align*} The curve is a circular arc. We can find $c$, $d$, and $\lambda$ by placing $(0,0)$ and $(1,t)$ on the curve, and solving $\int_0^1 y\,dx = 1$.

0

When you did $t=2$ the shortest distance was a straight line. If we increase $t$ then the shortest distance is likely to be an arc of a circle.

This is better justified with we rotate the graph so that $(1,t)$ is now on the x-axis at $(\sqrt{1+t^2},0)$.

enter image description here

The area can be found by considering the triangle take away the sector. The radius can be adjusted until the area enclosed is equal to $1$.

Ian Miller
  • 11,844
  • It is a nice insight about circular arc, but the wording "is likely to be" is the weakest part, which is not much convincing. It is possible though to make a rigorous proof of the claim without using calculus of variations. To minimize the length with given area is related to minimizing/maximizing area with given length. The latter is the classical isoperimetric problem: Dido's problem. – A.Γ. Oct 28 '16 at 00:06
  • I deliberately uses weak wording as I knew it wasn't a rigorous proof. I'm glad to see my intuition was right even though I couldn't justify it. – Ian Miller Oct 28 '16 at 01:14
  • Thank you Ian, I read your wording just this time. But how about a question that the red arc can meet the right black slope at another point , then it is not function ? – c-2785 Nov 02 '16 at 01:29
  • @c-2785 I started doing the math behind the calculators - i.e. find the area of the segment of the circle and set it equal to half the area of the triangle and the radius will always be such that it doesn't intersect twice. I didn't add that to my answer as I saw the other answer which I felt was neater and more precise (and it seemed that you were happy with it). If you are still interested in a geometric approach rather than a calculus one I can update my answer. Let me know. – Ian Miller Nov 02 '16 at 04:53