1
$\begingroup$

I'm trying to evaluate Euler Lagrange equation from the following relation: $$ F[f(\vec{r})]=\int_{\vec{r_1}}^{\vec{r_2}} d^n r J[f(\vec{r}),\nabla f (\vec{r}),H f(\vec{r}) ] $$

where $H$ is the hessian and $\vec{r}$ represent $n$ independent variables. Our boundary condition is that $\eta $ and its derivatives shout be zero at end points.

what I tried is:

$$ \left [ \frac{\partial F}{\partial \alpha}\right ]_{\alpha=0} =0 $$ and

$$ f(\vec{r},\alpha)=f(\vec{r},0)+\alpha \eta (\vec{r}) \\ \nabla f(\vec{r},\alpha)=\nabla f(\vec{r},0)+\alpha \nabla \eta (\vec{r}) \\ H f(\vec{r},\alpha)=H f(\vec{r},0)+\alpha H \eta (\vec{r}) $$ so we can write $$ \int_{\vec{r_1}}^{\vec{r_2}} \left [ \frac{\partial J}{\partial f}\frac{\partial f}{\partial \alpha}+\frac{\partial J}{\partial \nabla f} \frac{\partial \nabla f}{\partial \alpha}+\frac{\partial J}{\partial H f}\frac{\partial H f}{\partial \alpha} \right ] d^n r \\ \int \left [ \frac{\partial J}{\partial f}\eta(\vec{r})+\frac{\partial J}{\partial \nabla f} \nabla \eta(\vec{r})+\frac{\partial J}{\partial H f} H \eta(\vec{r}) \right ] d^n r $$ for the second and third terms in bracket I use the integrating by parts. For the second term I wrote: $$ = \eta(\vec{r}) \frac{\partial J}{\partial \nabla f}-\int \eta(\vec{r}) \nabla .\left ( \frac{\partial J}{\partial \nabla f} \right ) d^n r $$ and for the second term including the hessian: $$ =\nabla \eta(\vec{r}) \frac{\partial J}{\partial H f}-\int \nabla \eta(\vec{r}) \nabla .\left ( \frac{\partial J}{\partial \nabla f} \right ) d^n r\\ = \eta(\vec{r})\left (\nabla . \frac{\partial J}{\partial H f} \right )+\int \eta(\vec{r}) \nabla \left ( \nabla .\frac{\partial J}{\partial Hf} \right )d^nr $$ The terms including $\eta$ and its derivative are zero with regard to the boundary conditions. Finally we can write the Euler Lagrange equation as:

$$ \frac{\partial J}{\partial f(\vec{r})}-\nabla .\left(\frac{\partial J}{\partial \nabla f(\vec{r})}\right )+ \nabla\left ( \nabla .\frac{\partial J}{\partial Hf(\vec{r})} \right )=0 $$ I'm wondering is it true? Edit: Is it true to integrate from hessian and get gradient?

$\endgroup$
0
$\begingroup$

Yes it is. You can write it explicitly in terms of derivatives, i.e. by seeing the hessian as $\frac{n(n+1)}2$ parameters :

$$ J(Hf)=J(\{\partial _i \partial _j f\}) $$ hence (with summation over repeated indices) $$\delta J =\frac{\partial^2 J}{\partial _i\partial _j f } \delta(\partial _i\partial _j f ) = \frac{\partial^2 J}{\partial _i\partial _j f } \delta\alpha\partial _i\partial _j \eta$$ Integrating by parts twice yields your correct result.

$\endgroup$
  • $\begingroup$ note enough reputation to vote up. Thanks. $\endgroup$ – Abolfazl Oct 23 '14 at 19:55

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.