How to derive the associated Euler-Lagrange equation?

  • Thread starter Thread starter Math100
  • Start date Start date
  • Tags Tags
    Derivation
Math100
Messages
813
Reaction score
229
Homework Statement
For each of the following functionals, find the Gateaux differential and use it to derive the associated Euler-Lagrange equation. In each case, be sure to specify all boundary conditions that the stationary path must satisfy.
a) ## S[y]=\int_{1}^{2}(3y'^2-2y^2)dx, y(1)=1, y(2)=3 ##.
b) ## S[y]=y(0)+\frac{1}{2}\int_{0}^{1}(2y'^2+x^2y^2)dx, y(1)=2 ##.
Relevant Equations
Gateaux differential: ## \bigtriangleup S[y, h]=\displaystyle \lim_{\epsilon \to 0}\frac{d}{d\epsilon}S[y+\epsilon h] ## is defined on admissible paths ## y+\epsilon h ## for all ## \epsilon ## in the neighborhood of zero.

A path ## y(x) ## is a stationary path of a functional if the Gateaux differential ## \bigtriangleup S[y, h] ## is zero for all admissible paths ## y+\epsilon h ##.

For the functional ## S[y]=\int_{a}^{b}F(x, y, y')dx, y(a)=A, y(b)=B ##, the Euler-Lagrange equation is ## \frac{d}{dx}(\frac{\partial F}{\partial y'})-\frac{\partial F}{\partial y}=0, y(a)=A, y(b)=B ##.
a) We have ## S[y+\epsilon h]=\int_{1}^{2}[3(y'+\epsilon h')^2-2(y+\epsilon h)^2]dx ##.
Note that ## \frac{d}{d\epsilon}S[y+\epsilon h]=\int_{1}^{2}[6(y'+\epsilon h')h'-4(y+\epsilon h)h]dx=2\int_{1}^{2}[3(y'+\epsilon h')h'-2(y+\epsilon h)]dx ##.
Then ## \bigtriangleup S[y, h]=\frac{d}{d\epsilon}S[y+\epsilon h]\rvert_{\epsilon=0}=2\int_{1}^{2}(3y'h'-2yh)dx ##.
Thus, the Gateaux differential is ## \bigtriangleup S[y, h]=2\int_{1}^{2}(3y'h'-2yh)dx ##.
Consider the functional ## S[y]=\int_{1}^{2}(3y'^2-2y^2)dx, y(1)=1, y(2)=3 ##.
Let ## F(x, y, y')=3y'^2-2y^2 ##.
Then ## \frac{\partial F}{\partial y'}=6y' ## and ## \frac{\partial F}{\partial y}=-4y ##.
Observe that ## \frac{d}{dx}(\frac{\partial F}{\partial y'})-\frac{\partial F}{\partial y}=0\implies 6\frac{d^2y}{dx^2}+4y=0 ##.
Therefore, the Euler-Lagrange equation is ## 6\frac{d^2y}{dx^2}+4y=0, y(1)=1, y(2)=3 ##.

b) We have ## S[y+\epsilon h]=y(0)+\frac{1}{2}\int_{0}^{1}(2(y'+\epsilon h')^2+x^2(y+\epsilon h)^2)dx ##.
Note that ## \frac{d}{d\epsilon}S[y+\epsilon h]=y(0)+\frac{1}{2}\int_{0}^{1}[4(y'+\epsilon h')h'+2x^2(y+\epsilon h)h]dx=y(0)+\int_{0}^{1}[2(y'+\epsilon h')h'+x^2(y+\epsilon h)h]dx ##.
Thus, the Gateaux differential is ## \bigtriangleup S[y, h]=\int_{0}^{1}(2y'h'+x^2yh)dx ##.
Consider the functional ## S[y]=y(0)+\frac{1}{2}\int_{0}^{1}(2y'^2+x^2y^2)dx, y(1)=2 ##.
Let ## F(x, y, y')=2y'^2+x^2y^2 ##.
Then ## \frac{\partial F}{\partial y'}=4y' ## and ## \frac{\partial F}{\partial y}=2x^2y ##.
Observe that ## \frac{d}{dx}(\frac{\partial F}{\partial y'})-\frac{\partial F}{\partial y}=0\implies 4\frac{d^2y}{dx^2}-2x^2y=0 ##.
Therefore, the Euler-Lagrange equation is ## 4\frac{d^2y}{dx^2}-2x^2y=0, y(1)=2 ##.
 
Physics news on Phys.org
Math100 said:
a) We have ## S[y+\epsilon h]=\int_{1}^{2}[3(y'+\epsilon h')^2-2(y+\epsilon h)^2]dx ##.
Note that ## \frac{d}{d\epsilon}S[y+\epsilon h]=\int_{1}^{2}[6(y'+\epsilon h')h'-4(y+\epsilon h)h]dx=2\int_{1}^{2}[3(y'+\epsilon h')h'-2(y+\epsilon h)]dx ##.
There is a factor ##h## missing (only here at the end).
Math100 said:
Then ## \bigtriangleup S[y, h]=\frac{d}{d\epsilon}S[y+\epsilon h]\rvert_{\epsilon=0}=2\int_{1}^{2}(3y'h'-2yh)dx ##.
Thus, the Gateaux differential is ## \bigtriangleup S[y, h]=2\int_{1}^{2}(3y'h'-2yh)dx ##.
Consider the functional ## S[y]=\int_{1}^{2}(3y'^2-2y^2)dx, y(1)=1, y(2)=3 ##.
Let ## F(x, y, y')=3y'^2-2y^2 ##.
Then ## \frac{\partial F}{\partial y'}=6y' ## and ## \frac{\partial F}{\partial y}=-4y ##.
Observe that ## \frac{d}{dx}(\frac{\partial F}{\partial y'})-\frac{\partial F}{\partial y}=0\implies 6\frac{d^2y}{dx^2}+4y=0 ##.
Therefore, the Euler-Lagrange equation is ## 6\frac{d^2y}{dx^2}+4y=0, y(1)=1, y(2)=3 ##.
Are you sure you shouldn't solve this equation? What else should be the initial values for? At least you should divide by ##6.##
Math100 said:
b) We have ## S[y+\epsilon h]=y(0)+\frac{1}{2}\int_{0}^{1}(2(y'+\epsilon h')^2+x^2(y+\epsilon h)^2)dx ##.
Note that ## \frac{d}{d\epsilon}S[y+\epsilon h]=y(0)+\frac{1}{2}\int_{0}^{1}[4(y'+\epsilon h')h'+2x^2(y+\epsilon h)h]dx=y(0)+\int_{0}^{1}[2(y'+\epsilon h')h'+x^2(y+\epsilon h)h]dx ##.
Note that ##\frac{d}{d\epsilon} y(0)=0.##
Math100 said:
Thus, the Gateaux differential is ## \bigtriangleup S[y, h]=\int_{0}^{1}(2y'h'+x^2yh)dx ##.
Consider the functional ## S[y]=y(0)+\frac{1}{2}\int_{0}^{1}(2y'^2+x^2y^2)dx, y(1)=2 ##.
Let ## F(x, y, y')=2y'^2+x^2y^2 ##.
Then ## \frac{\partial F}{\partial y'}=4y' ## and ## \frac{\partial F}{\partial y}=2x^2y ##.
Observe that ## \frac{d}{dx}(\frac{\partial F}{\partial y'})-\frac{\partial F}{\partial y}=0\implies 4\frac{d^2y}{dx^2}-2x^2y=0 ##.
Therefore, the Euler-Lagrange equation is ## 4\frac{d^2y}{dx^2}-2x^2y=0, y(1)=2 ##.
This is definitely harder to solve, so maybe all you actually need to do is note the Lagrangian. Divide it by ##4##.

It looks okay apart from the typo (missing ##h##), ##y(0)## (vanishes under the differentiation along ##\epsilon##) and the missing divisions (to make ##y''## with a coefficient ##1## in the Lagrangian) and the possible missing solution of the differential equations. At least from what I can tell from re-reading
https://www.physicsforums.com/insights/pantheon-derivatives-part-ii/
which I wrote more than six years ago.
 
For (b), \begin{split}<br /> S[y] &amp;= y(0) + \frac12 \int_0^1 2y&#039;^2 + x^2 y^2 \,dx \\<br /> S[y + \epsilon h] &amp;= y(0) + \epsilon h(0) + \frac12 \int_0^1 2(y&#039; + \epsilon h&#039;)^2 + x^2(y + \epsilon h)^2\,dx \\<br /> \frac{d}{d\epsilon}S[y + \epsilon h] &amp;= h(0) + \frac12 \int_0^1 4(y&#039; + \epsilon h&#039;)h&#039; + 2x^2(y + \epsilon h)h\,dx<br /> \end{split} so \begin{split}\Delta S[y,h] &amp;= h(0) + \frac12 \int_0^1 4y&#039;h&#039; + 2x^2yh\,dx\\<br /> &amp;= h(0) + \int_0^1 2y&#039;h&#039; + x^2 yh\,dx.\end{split} Note the extra h(0) term, which is due to the fact that the variation of y(0) is y(0) + \epsilon h(0). You are given that y(1) = 2, so h(1) = 0 is required; you are not given a condition on y at x = 0 so you can't assume h(0) = 0. What condition will you impose on y at x = 0 to guarantee a unique solution?
 
pasmith said:
For (b), \begin{split}<br /> S[y] &amp;= y(0) + \frac12 \int_0^1 2y&#039;^2 + x^2 y^2 \,dx \\<br /> S[y + \epsilon h] &amp;= y(0) + \epsilon h(0) + \frac12 \int_0^1 2(y&#039; + \epsilon h&#039;)^2 + x^2(y + \epsilon h)^2\,dx \\<br /> \frac{d}{d\epsilon}S[y + \epsilon h] &amp;= h(0) + \frac12 \int_0^1 4(y&#039; + \epsilon h&#039;)h&#039; + 2x^2(y + \epsilon h)h\,dx<br /> \end{split} so \begin{split}\Delta S[y,h] &amp;= h(0) + \frac12 \int_0^1 4y&#039;h&#039; + 2x^2yh\,dx\\<br /> &amp;= h(0) + \int_0^1 2y&#039;h&#039; + x^2 yh\,dx.\end{split} Note the extra h(0) term, which is due to the fact that the variation of y(0) is y(0) + \epsilon h(0). You are given that y(1) = 2, so h(1) = 0 is required; you are not given a condition on y at x = 0 so you can't assume h(0) = 0. What condition will you impose on y at x = 0 to guarantee a unique solution?
I don't know. What will the condition be?
 
Hint: y&#039;h&#039; = (y&#039;h)&#039; - y&#039;&#039;h
 
pasmith said:
Hint: y&#039;h&#039; = (y&#039;h)&#039; - y&#039;&#039;h
I still don't understand.
 
For S[y] = \int_0^1 F(y,y&#039;,x)\,dx the Gateux derivative is \begin{split}<br /> \Delta S[y,h] &amp;= \lim_{\epsilon \to 0} \frac{d}{d\epsilon}S[y + \epsilon h] \\<br /> &amp;= \int_0^1 \frac{\partial F}{\partial y}h + \frac{\partial F}{\partial y&#039;}h&#039;\,dx \\<br /> &amp;= \int_0^1 \frac{\partial F}{\partial y}h + \frac{d}{dx}\left( \frac{\partial F}{\partial y&#039;} h \right)- \frac{d}{dx}\left(\frac{\partial F}{\partial y&#039;}\right)h\,dx \\<br /> &amp;= \left[ \frac{\partial F}{\partial y&#039;} h\right]_0^1 <br /> + \int_0^1 \left( \frac{\partial F}{\partial y} - \frac{d}{dx}\left( \frac{\partial F}{\partial y&#039;} \right)\right)h\,dx. \end{split} To determine the optimum solution y : \forall \mbox{ admissible $h$} : \Delta S[y,h] = 0 we impose the conditions <br /> \begin{split}<br /> \frac{\partial F}{\partial y} - \frac{d}{dx}\left( \frac{\partial F}{\partial y&#039;} \right) &amp;= 0 \\<br /> \frac{\partial F}{\partial y&#039;}(y(0),y&#039;(0),0)h(0) &amp;= 0 \\<br /> \frac{\partial F}{\partial y&#039;}(y(1),y&#039;(1),1)h(1) &amp;= 0. \end{split} This has the form of a second order ODE (the Euler-Lagrange equation) for y subject to two boundary conditions: at x = c \in \{0,1\} either y(c) is specified in which case h(c) = 0 so that the corresponding condition is satisfied, or else y(c) is not specified so that h(c) is arbitrary and we must impose <br /> \frac{\partial F}{\partial y&#039;}(y(c),y&#039;(c),c) = 0.<br /> I'm sure your text or your lecturer will have explained this derivation, and in particular the step of integrating \dfrac{\partial F}{\partial y&#039;}h&#039; by parts. In your working, however, instead of following it through completely to obtain not just the Euler-Lagrange equation but also the boundary terms, you reach the second line of my derivation of \Delta S[y,h] and then just substitute F into the Euler-Lagrange equation, which you quote in your Relevant Equations for the case where y is specified on both boundaries. That is not the point of this exercise. In part (a) y is specified at both boundaries so that approach works, but part (b) is an extension to the case <br /> S[y] = y(0) + \int_0^1 F(y,y&#039;,x)\,dx where the Euler-Lagrange equation does not change but the boundary terms do, and you need to obtain those terms because y(0) is not specified.
 
  • Like
Likes Math100 and fresh_42
Back
Top