How to find the Gateaux differential of this functional?

Math100
Messages
813
Reaction score
229
Homework Statement
Let ## a<b, A ## and ## B ## be constants.
Find the Gateaux differential of the functional ## S[y]=\int_{a}^{b}(y'^{2}+\omega^{2}y^{2}+2yx^{4})dx, y(a)=A, y(b)=B ##, where ## \omega ## is a positive constant.
Relevant Equations
Suppose ## X ## and ## Y ## are locally convex topological vector spaces, ## U\subseteq X ## is open, and ## F: X\rightarrow Y ##. The Gateaux differential ## dF(u; \psi) ## of ## F ## at ## u\in U ## in the direction ## \psi\in X ## is defined as ## dF(u; \psi)=\lim_{\tau\rightarrow 0}\frac{F(u+\tau\psi)-F(u)}{\tau}=\frac{d}{d\tau}F(u+\tau\psi)\biggr\rvert_{\tau=0} ##. If the limit exists for all ## \psi\in X ##, then one says that ## F ## is Gateaux differentiable at ## u ##.
I am not sure if this is correct, but here is my work by using the definition of the Gateaux differential:
\begin{align*}
&dS(y; \psi)=\lim_{\tau\rightarrow 0}\frac{S(y+\tau\psi)-S(y)}{\tau}=\frac{d}{d\tau}S(y+\tau\psi)\biggr\rvert_{\tau=0}\\
&S(y+\tau\psi)=\int_{a}^{b}[(y+\tau\psi)'^{2}+\omega^{2}(y+\tau\psi)^{2}+2(y+\tau\psi)x^{4}]dx\\
&=[(y+\tau\psi)'^{2}x+\omega^{2}(y+\tau\psi)^{2}x+\frac{2}{5}(y+\tau\psi)x^{5}]_{a}^{b}\\
&=(y+\tau\psi)'^{2}(b-a)+\omega^{2}(y+\tau\psi)^{2}(b-a)+[\frac{2}{5}(y+\tau\psi)](b^{5}-a^{5})\\
\end{align*}
Thus ## S(y+\tau\psi)\biggr\rvert_{\tau=0}=y'^{2}(b-a)+\omega^{2}y^{2}(b-a)+\frac{2}{5}y(b^{5}-a^{5}) ##.
So ## \frac{d}{d\tau}S(y+\tau\psi)\biggr\rvert_{\tau=0}=0 ##.

Can anyone please check/verify/confirm this?
 
Physics news on Phys.org
Math100 said:
Homework Statement:: Let ## a<b, A ## and ## B ## be constants.
Find the Gateaux differential of the functional ## S[y]=\int_{a}^{b}(y'^{2}+\omega^{2}y^{2}+2yx^{4})dx, y(a)=A, y(b)=B ##, where ## \omega ## is a positive constant.
Relevant Equations:: Suppose ## X ## and ## Y ## are locally convex topological vector spaces, ## U\subseteq X ## is open, and ## F: X\rightarrow Y ##. The Gateaux differential ## dF(u; \psi) ## of ## F ## at ## u\in U ## in the direction ## \psi\in X ## is defined as ## dF(u; \psi)=\lim_{\tau\rightarrow 0}\frac{F(u+\tau\psi)-F(u)}{\tau}=\frac{d}{d\tau}F(u+\tau\psi)\biggr\rvert_{\tau=0} ##. If the limit exists for all ## \psi\in X ##, then one says that ## F ## is Gateaux differentiable at ## u ##.

I am not sure if this is correct, but here is my work by using the definition of the Gateaux differential:
\begin{align*}
&dS(y; \psi)=\lim_{\tau\rightarrow 0}\frac{S(y+\tau\psi)-S(y)}{\tau}=\frac{d}{d\tau}S(y+\tau\psi)\biggr\rvert_{\tau=0}\\
&S(y+\tau\psi)=\int_{a}^{b}[(y+\tau\psi)'^{2}+\omega^{2}(y+\tau\psi)^{2}+2(y+\tau\psi)x^{4}]dx\\
&=[(y+\tau\psi)'^{2}x+\omega^{2}(y+\tau\psi)^{2}x+\frac{2}{5}(y+\tau\psi)x^{5}]_{a}^{b}\\
&=(y+\tau\psi)'^{2}(b-a)+\omega^{2}(y+\tau\psi)^{2}(b-a)+[\frac{2}{5}(y+\tau\psi)](b^{5}-a^{5})\\
\end{align*}
Thus ## S(y+\tau\psi)\biggr\rvert_{\tau=0}=y'^{2}(b-a)+\omega^{2}y^{2}(b-a)+\frac{2}{5}y(b^{5}-a^{5}) ##.
So ## \frac{d}{d\tau}S(y+\tau\psi)\biggr\rvert_{\tau=0}=0 ##.

Can anyone please check/verify/confirm this?
I'm not all that familiar with the Gateaux differential.

It seems to me that you need to first do the differentiation, ##\dfrac{d}{d\tau} S(y+\tau\psi)##, then evaluate the result at ##\tau =0##.
 
Math100 said:
Homework Statement:: Let ## a<b, A ## and ## B ## be constants.
Find the Gateaux differential of the functional ## S[y]=\int_{a}^{b}(y'^{2}+\omega^{2}y^{2}+2yx^{4})dx, y(a)=A, y(b)=B ##, where ## \omega ## is a positive constant.
Relevant Equations:: Suppose ## X ## and ## Y ## are locally convex topological vector spaces, ## U\subseteq X ## is open, and ## F: X\rightarrow Y ##. The Gateaux differential ## dF(u; \psi) ## of ## F ## at ## u\in U ## in the direction ## \psi\in X ## is defined as ## dF(u; \psi)=\lim_{\tau\rightarrow 0}\frac{F(u+\tau\psi)-F(u)}{\tau}=\frac{d}{d\tau}F(u+\tau\psi)\biggr\rvert_{\tau=0} ##. If the limit exists for all ## \psi\in X ##, then one says that ## F ## is Gateaux differentiable at ## u ##.

I am not sure if this is correct, but here is my work by using the definition of the Gateaux differential:<br /> \begin{align*}<br /> &amp;dS(y; \psi)=\lim_{\tau\rightarrow 0}\frac{S(y+\tau\psi)-S(y)}{\tau}=\frac{d}{d\tau}S(y+\tau\psi)\biggr\rvert_{\tau=0}\\<br /> &amp;S(y+\tau\psi)=\int_{a}^{b}[(y+\tau\psi)&#039;^{2}+\omega^{2}(y+\tau\psi)^{2}+2(y+\tau\psi)x^{4}]dx\\<br /> &amp;=[(y+\tau\psi)&#039;^{2}x+\omega^{2}(y+\tau\psi)^{2}x+\frac{2}{5}(y+\tau\psi)x^{5}]_{a}^{b}<br /> \end{align*}

No. y and \psi are arbitrary functions of x satisfying y(a) = A, y(b) = B and \psi(a) = \psi~(b) = 0. You can't actually evaluate the integral because you don't know what y and \psi are. Instead you must bring the differentiation with respect to \tau inside the integral, where it becomes a partial derivative at constant x.
 
pasmith said:
No. y and \psi are arbitrary functions of x satisfying y(a) = A, y(b) = B and \psi(a) = \psi~(b) = 0. You can't actually evaluate the integral because you don't know what y and \psi are. Instead you must bring the differentiation with respect to \tau inside the integral, where it becomes a partial derivative at constant x.
Do you mean ## S(y+\tau\psi)=\int_{a}^{b}\frac{d}{d\tau}(y'^{2}+\omega^{2}y^{2}+2yx^{4})dx ##? If so, then how to evaluate this?
 
Math100 said:
Do you mean ## S(y+\tau\psi)=\int_{a}^{b}\frac{d}{d\tau}(y'^{2}+\omega^{2}y^{2}+2yx^{4})dx ##? If so, then how to evaluate this?

\begin{split}<br /> S[y + \tau \psi] &amp;= \int_a^b (y(x)&#039; + \tau \psi(x)&#039;)^2 + \omega^2(y(x) + \tau\psi(x))^2 + 2(y(x) + \tau \psi(x))x^4\,dx \\<br /> \left.\frac{d}{d\tau} S[y + \tau \psi]\right|_{\tau=0} &amp;= <br /> \int_a^b \left.\frac{\partial}{\partial \tau}\left(((y&#039;(x) + \tau\psi&#039;(x))^2 + \omega^2(y(x) + \tau\psi(x))^2 + 2(y(x) + \tau\psi(x))x^4\right)\right|_{\tau=0}\,dx. \end{split} Again, you need to calculate the (partial) derivative with respect to \tau and then set \tau = 0. \left.\dfrac{df}{dt}\right|_{t=t_0} is how we indicate where the derivative is to be evaluated using Leibnitz's notation; with Newton's notation we would just write f&#039;(t_0).
 
Last edited:
pasmith said:
\begin{split}<br /> S[y + \tau \psi] &amp;= \int_a^b (y(x)&#039; + \tau \psi(x)&#039;)^2 + \omega^2(y(x) + \tau\psi(x))^2 + 2(y(x) + \tau \psi(x))x^4\,dx \\<br /> \left.\frac{d}{d\tau} S[y + \tau \psi]\right|_{\tau=0} &amp;=<br /> \int_a^b \left.\frac{\partial}{\partial \tau}\left(((y&#039;(x) + \tau\psi&#039;(x))^2 + \omega^2(y(x) + \tau\psi(x))^2 + 2(y(x) + \tau\psi(x))x^4\right)\right|_{\tau=0}\,dx. \end{split} Again, you need to calculate the (partial) derivative with respect to \tau and then set \tau = 0. \left.\dfrac{df}{dt}\right|_{t=t_0} is how we indicate where the derivative is to be evaluated using Leibnitz's notation; with Newton's notation we would just write f&#039;(t_0).
After breaking down into smaller pieces, I got the following:
\begin{align*}
&(y'(x)+\tau\psi'(x))^{2}=y'^{2}(x)+2y'(x)\tau\psi'(x)+\tau^{2}\psi'^{2}(x)\\
&\omega^{2}(y(x)+\tau\psi(x))^{2}=\omega^{2}y^{2}(x)+2\tau\psi(x)\omega^{2}+\omega^{2}\tau^{2}\psi^{2}(x)\\
&2(y(x)+\tau\psi(x))x^{4}=2x^{4}y(x)+2x^{4}\tau\psi(x)\\
&\frac{d}{d\tau}[(y(x)'+\tau\psi(x)')^{2}]=2y'(x)\psi'(x)+2\tau\psi'^{2}(x)\\
&\frac{d}{d\tau}[\omega^{2}(y(x)+\tau\psi(x))^{2}]=2\psi(x)\omega^{2}+2\omega^{2}\tau\psi^{2}(x)\\
&\frac{d}{d\tau}[2(y(x)+\tau\psi(x))x^{4}]=2x^{4}\psi(x)\\
\end{align*}
So
\begin{align*}
&\frac{d}{d\tau}S[y+\tau\psi]_{\tau=0}\\
&=[2y'(x)\psi'(x)+2\tau\psi'^{2}(x)+2\psi(x)\omega^{2}+2\omega^{2}\tau\psi^{2}(x)+2x^{4}\psi(x)]_{\tau=0}\\
&=2y'(x)\psi'(x)+2\psi(x)\omega^{2}+2x^{4}\psi(x).\\
\end{align*}
Thus
\begin{align*}
&\int_{a}^{b}[2y'(x)\psi'(x)+2\psi(x)\omega^{2}+2x^{4}\psi(x)]dx\\
&=[2xy'(x)\psi'(x)+2x\psi(x)\omega^{2}+\frac{2}{5}x^{5}\psi(x)]_{a}^{b}\\
&=[2by'(x)\psi'(x)+2b\psi(x)\omega^{2}+\frac{2}{5}b^{5}\psi(x)]-[2ay'(x)\psi'(x)+2a\psi(x)\omega^{2}+\frac{2}{5}a^{5}\psi(x)]\\
&=(b-a)(2y'(x)\psi'(x)+2\psi(x)\omega^{2})+(b^{5}-a^{5})(\frac{2}{5}\psi(x)).\\
\end{align*}
Thus
\begin{align*}
&\frac{d}{d\tau}S[y+\tau\psi]_{\tau=0}=(b-a)(2y'(x)\psi'(x)+2\psi(x)\omega^{2})+(b^{5}-a^{5})(\frac{2}{5}\psi(x)).\\
\end{align*}
Is this correct?
 
Math100 said:
After breaking down into smaller pieces, I got the following:
\begin{align*}
&(y'(x)+\tau\psi'(x))^{2}=y'^{2}(x)+2y'(x)\tau\psi'(x)+\tau^{2}\psi'^{2}(x)\\
&\omega^{2}(y(x)+\tau\psi(x))^{2}=\omega^{2}y^{2}(x)+2\tau\psi(x)\omega^{2}+\omega^{2}\tau^{2}\psi^{2}(x)\end{align*}

You are missing a y(x) from the second term on the right on the last line. But it is easier to use the chain rule rather than multiply everything out: \frac{\partial}{\partial \tau} (y(x) + \tau \psi(x)&#039;)^2 = 2(y&#039;(x) + \tau \psi&#039;(x))\psi&#039;(x) etc.

Thus
\begin{align*}

&\int_{a}^{b}[2y'(x)\psi'(x)+2\psi(x)\omega^{2}+2x^{4}\psi(x)]dx\\

&=[2xy'(x)\psi'(x)+2x\psi(x)\omega^{2}+\frac{2}{5}x^{5}\psi(x)]_{a}^{b}\\

&=[2by'(x)\psi'(x)+2b\psi(x)\omega^{2}+\frac{2}{5}b^{5}\psi(x)]-[2ay'(x)\psi'(x)+2a\psi(x)\omega^{2}+\frac{2}{5}a^{5}\psi(x)]\\

&=(b-a)(2y'(x)\psi'(x)+2\psi(x)\omega^{2})+(b^{5}-a^{5})(\frac{2}{5}\psi(x)).\\

\end{align*}

Aside from the factor of y(x) missing from the second term, the first line is essentially correct. But the rest is wrong. You have a definite integral with respect to x; its value should therefore not depend on x, but should be a function of the limits a and b. However, you can't evaluate the integral since you don't know how the integrand depends on x (because you don't know what y and \psi are). It is not in general true that \int_a^b xy(x)\,dx = by(b) - ay(a).

The most you can do is integrate the first term by parts to get <br /> \int_a^b 2y&#039;(x)\psi&#039;(x) + 2\omega^2y(x)\psi(x) + 2x^4 \psi(x)\,dx = 2 \int_a^b \left(-y&#039;&#039;(x) + \omega^2y(x) + x^4\right)\psi(x)\,dx + 2\left[y&#039;(x)\psi(x)\right]_a^b but that is as far as you can go.
 
Back
Top