# Doubt regarding functional derivative for the Thomas Fermi kinetic energy

• JhonDoe
In summary, the conversation discusses the procedure for obtaining the functional derivative of the kinetic energy in density functional theory, specifically the result of ##\frac{\delta E_k}{\delta \rho(\mathbf{r})}=-\frac{1}{4} \frac{\nabla^2 \rho(\mathbf{r})}{\rho(\mathbf{r})}+\frac{1}{8}\frac{\nabla \rho(\mathbf{r}) \cdot \nabla \rho(\mathbf{r})}{\rho(\mathbf{r})^2}##. The conversation delves into the derivation of this result, with the use of Taylor expansion and the functional derivative
JhonDoe
Homework Statement
This is not homework
Relevant Equations
##E_k=\int \frac{\nabla \rho(\mathbf{r}) \cdot \nabla \rho(\mathbf{r})}{\rho(\mathbf{r})}d\mathbf{r}##
I have some doubts with respect on how the functional derivative for the kinetic energy in density functional theory is obtained.

In particular, I'm interested in how to get the result##\frac{\delta E_k}{\delta \rho(\mathbf{r})}=-\frac{1}{4} \frac{\nabla^2 \rho(\mathbf{r})}{\rho(\mathbf{r})}+\frac{1}{8}\frac{\nabla \rho(\mathbf{r}) \cdot \nabla \rho(\mathbf{r})}{\rho(\mathbf{r})^2}##

I understand that the procedure starts by Taylor expansion of the argument in the integral.

Taylor expanding at ##\mathbf{r}## gives, for a point ##\mathbf{r}'## close to ##\mathbf{r}##

##\frac{\nabla \rho(\mathbf{r}') \cdot \nabla \rho(\mathbf{r}')}{\rho(\mathbf{r}')} \simeq \frac{\nabla \rho(\mathbf{r}) \cdot \nabla \rho(\mathbf{r})}{\rho(\mathbf{r})} - \frac{1}{4} \frac{\nabla^2 \rho(\mathbf{r})}{\rho(\mathbf{r})} \delta \rho(\mathbf{r}') + \frac{1}{8}\frac{\nabla \rho(\mathbf{r}) \cdot \nabla \rho(\mathbf{r})}{\rho(\mathbf{r})^2}\delta \rho(\mathbf{r}')##

That is the result reported everywhere, including wikipedia, so I know it is correct. However, I have no problem with the first linear term, this is to say the term ##\frac{1}{4} \frac{\nabla^2 \rho(\mathbf{r})}{\rho(\mathbf{r})} \delta \rho## which is obtained by simply taking the derivative with respect to ##\rho## for the denominator. The term I have trouble with is the one that involves the derivative ##\frac{\partial }{\partial \rho} \nabla \rho(\mathbf{r}) \cdot \nabla \rho(\mathbf{r})## I don't really understand how to get the derivative of the gradient with respect to the function ##\rho##.

Starting from$$E_k[\rho] = \int d\mathbf{r} \frac{\nabla \rho \cdot \nabla \rho}{8\rho} : = \int d\mathbf{r} f \quad \mathrm{where} \quad f := \frac{\nabla \rho \cdot \nabla \rho }{8\rho}$$Using the result in the Wikipedia article for the functional derivative of a functional of the form ##F[\rho] = \int d\mathbf{r} f(\mathbf{r}, \rho{(\mathbf{r})}, \nabla \rho(\mathbf{r}))##, you have$$\frac{\delta E_k}{\delta \rho} = \frac{\partial f}{\partial \rho} - \nabla \cdot \frac{\partial f}{\partial \nabla \rho} \ \ \ (1)$$You then have for the partial derivative with respect to ##\rho##\begin{align*}

\frac{\partial f}{\partial \rho} = \frac{- \nabla \rho \cdot \nabla \rho}{8 \rho^2}\end{align*}and with respect to ##\nabla \rho##,\begin{align*}

\frac{\partial f}{\partial \nabla \rho} = \frac{1}{8 \rho} \left( 2\nabla \rho \right) = \frac{\nabla \rho}{4 \rho}\end{align*}from which it follows that, denoting ##\Delta : = \nabla^2##,\begin{align*}

\nabla \cdot \frac{\partial f}{\partial \nabla \rho} = \nabla \cdot \frac{\nabla \rho}{4 \rho} &= \frac{\rho \Delta \rho - \nabla \rho \cdot \nabla \rho}{4\rho^2}\end{align*}Inserting those results in ##(1)## gives the quoted functional derivative

Last edited by a moderator:
JhonDoe
Hi. Thank you for your useful reply. I realized that I made a wrong statement in the first post. The term that gives me confusion is the one that involves the Laplacian, the term ##\frac{1}{4} \frac{\nabla^2 \rho(\mathbf{r})}{\rho(\mathbf{r})} \delta \rho##. I see now that it can be obtained with the identity given in wikipedia, your post clarified it.

How should I proceed to obtain the derivative ##\frac{\partial }{\partial \rho} \nabla \rho(\mathbf{r}) \cdot \nabla \rho(\mathbf{r})## if I wanted to obtain the result in that way?

No problem. I'm not completely sure what you're trying to do in your last sentence, because ##\frac{\partial}{ \partial \rho} \nabla \rho \cdot \nabla \rho## ought to just be zero. Perhaps someone else understands better what your method is, and will chip in

JhonDoe
You are absolutely right, let me explain. I was just following the book "Density functional theory of atoms and molecules" by Parr and Weitago. In the appendix of that book the authors obtain the functional derivatives by making a Taylor expansion of the function in the kernel ##f## in ##E_k[\rho]=\int f d\mathbf{r}##. Given that wikipedia uses that formula, which I don't know how to demonstrate, I wanted to follow their approach by Taylor expanding the kernel of the integral to the first order. The Taylor expansion should give:

##\frac{\nabla \rho(\mathbf{r}') \cdot \nabla \rho(\mathbf{r}')}{\rho(\mathbf{r}')} \simeq \frac{\nabla \rho(\mathbf{r}) \cdot \nabla \rho(\mathbf{r})}{\rho(\mathbf{r})} - \frac{1}{4} \frac{\nabla^2 \rho(\mathbf{r})}{\rho(\mathbf{r})} \delta \rho(\mathbf{r}') + \frac{1}{8}\frac{\nabla \rho(\mathbf{r}) \cdot \nabla \rho(\mathbf{r})}{\rho(\mathbf{r})^2}\delta \rho(\mathbf{r}')+\ldots##
In that way, keeping the taylor expansion to first order gives the desired result for the functional derivative. The first term in the right hand side goes to the lhs, and the desired result is obtained.

So, the term involving the Laplacian comes from the derivative ##\frac{\partial}{ \partial \rho} \nabla \rho \cdot \nabla \rho##. But now, as you pointed out, that can not be, because that partial derivative should be zero, so, maybe in the Taylor expansion I should consider ##\nabla \rho## as an independent variable. I think that may give the correct result. However, the book doesn't make explicit the dependence on ##\nabla \rho## for the functional. It gives the functional ##E_k[\rho]##, which only depends explicitly on ##\rho##. So, I am trying to guess, but maybe instead of the partial derivative I should have the total derivative ##\frac{d}{ d \rho} \nabla \rho \cdot \nabla \rho= \frac{\partial}{ \partial \nabla \rho} \nabla \rho \cdot \nabla \rho \frac{\partial \nabla \rho}{ \partial \rho} ##, the term ##\frac{ \partial \nabla \rho} { \partial \rho} ## must have a dependence on ##\rho##, doesn't it? how should I evaluate that term? I think it just has to give the divergence appearing at the end in the formula. Maybe using index notation for the i-th component of the gradient ##\frac{ \partial \nabla \rho} { \partial \rho}_i ## should give in some way the divergence, but I'm a bit confused.

By the way, how do you evaluate the derivative ##\frac{\partial}{ \partial \nabla \rho} \nabla \rho \cdot \nabla \rho##? In index notation should it be ##\frac{\partial}{ \partial \nabla \rho} \nabla \rho \cdot \nabla \rho|_i=\frac{\partial}{ \partial \nabla \rho_i} \frac{\partial \rho }{ \partial x_j} \frac{\partial \rho }{ \partial x_j}## ? how do you treat the derivative in the denominator with respect to the gradient?

Last edited:
What is the result for this derivative: ##\frac{d }{d\rho} (\nabla \rho)^2##?

I have trouble when deriving inside the gradient: ##\frac{d }{d\rho} (\nabla \rho)^2=2\nabla \rho \times \frac{d }{d\rho} (\nabla \rho)=2\nabla \rho \times(\nabla \frac{d }{d\rho} \rho) ## ?

I would have thought that's again zero, and it's probably best to use the "##\cdot##" product instead of a ##\times## symbol, considering that ##(\nabla \rho)^2## is shorthand for ##\nabla \rho \cdot \nabla \rho## and is a scalar.

I'll admit that I'm not completely following what you're trying to do here, so I'm going to ask and see if anyone else can help. Hold tight!

JhonDoe
Ok. Thanks. I agree its a scalar, but the argument has a dependence on ##\rho##, so I think it shouldn't be zero.

JhonDoe said:
Ok. Thanks, but I think the gradient has a dependence on ##\rho##, so I think it shouldn't be zero.
In this context however, it looks like we're considering a function of both ##\rho## and ##\nabla \rho##, which are themselves both functions of ##\mathbf{r}##, in which case we ought to treat them as independent variables when taking partial derivatives.

Last edited by a moderator:
I think that shouldn't be necessary. I've tried this. In index notation:

##\frac{d }{d\rho} (\nabla \rho)^2=\frac{d }{d\rho} \frac{\partial}{\partial x_i} \rho \frac{\partial}{\partial x_i} \rho##

Then
##\frac{d }{d\rho} \frac{\partial}{\partial x_i} \rho \frac{\partial}{\partial x_i} \rho=\frac{\partial}{\partial x_i} \frac{d \rho}{d\rho} \times \frac{\partial}{\partial x_i} \rho +\frac{\partial}{\partial x_i} \frac{d \rho}{d\rho} \times \frac{\partial}{\partial x_i} \rho ##

Then
##\frac{\partial}{\partial x_i} \frac{d }{d\rho} \rho \frac{\partial}{\partial x_i} \rho=2 \frac{\partial}{\partial x_i} \frac{\partial}{\partial x_i}\rho##?

Which means: ##\frac{d }{d\rho} (\nabla \rho)^2=2\nabla^2 \rho##

I would like to know if that is correct. And if it isn't, where I am committing the mistake.

I really can't follow what you did there, I'm not sure why you kept interchanging the orders of the ##d/d \rho## and ##\partial / \partial x^i## operators. I don't think you can simplify past ##\frac{d}{d\rho} (\nabla \rho)^2 = \frac{d}{d\rho} \left( \frac{\partial \rho}{\partial x^i} \frac{\partial \rho}{\partial x^i} \right) = 2 \frac{\partial \rho}{\partial x^i} \frac{d}{d\rho} \left(\frac{\partial \rho}{\partial x^i} \right)##, with summation understood over ##i##.

JhonDoe
I'm not sure if this is of any help, but in Section 3 of this article there's a justification for that kinetic energy functional in the case of 1-electron system, i.e. when ##\rho (\mathbb{r}) = |\psi (\mathbb{r})|^2##.

http://przyrbwn.icm.edu.pl/APP/PDF/115/a115z310.pdf

JhonDoe
etotheipi said:
I really can't follow what you did there, I'm not sure why you kept interchanging the orders of the ##d/d \rho## and ##\partial / \partial x^i## operators. I don't think you can simplify past ##\frac{d}{d\rho} (\nabla \rho)^2 = \frac{d}{d\rho} \left( \frac{\partial \rho}{\partial x^i} \frac{\partial \rho}{\partial x^i} \right) = 2 \frac{\partial \rho}{\partial x^i} \frac{d}{d\rho} \left(\frac{\partial \rho}{\partial x^i} \right)##, with summation understood over ##i##.
Why not? its a linear operator. Maybe you are right, but I'm not really sure about that.

I’m not a mathematician, so this argument will be VERY heuristic, but nonentheless, here’s how I justify to myself that a function and its derivative in a functional are to be treated as independent variables:

We start with ##\frac{d}{d\rho} = \frac{d}{dx}\frac{dx}{d\rho}## and when we plug this back into our original expression, we find that we have pieces that look like ##\frac{d}{dx}\frac{dx}{d\rho}\frac{d\rho}{dx}##. Continuing to be VERY HEURISTIC AND NOT AT ALL RIGOROUS, we note that with the right assumptions and well-behaved enough functions, ##\frac{dx}{d\rho}\frac{d\rho}{dx} =1##. Then taking the derivative of 1 gives you the zero that you need.

Edit: here’s a much more rigorous explanation for your question:
https://math.stackexchange.com/ques...-and-y-be-independent-in-variational-calculus

atyy, JhonDoe and etotheipi
JhonDoe said:
Homework Statement:: This is not homework
Relevant Equations:: ##E_k=\int \frac{\nabla \rho(\mathbf{r}) \cdot \nabla \rho(\mathbf{r})}{\rho(\mathbf{r})}d\mathbf{r}##

I have some doubts with respect on how the functional derivative for the kinetic energy in density functional theory is obtained.

In particular, I'm interested in how to get the result##\frac{\delta E_k}{\delta \rho(\mathbf{r})}=-\frac{1}{4} \frac{\nabla^2 \rho(\mathbf{r})}{\rho(\mathbf{r})}+\frac{1}{8}\frac{\nabla \rho(\mathbf{r}) \cdot \nabla \rho(\mathbf{r})}{\rho(\mathbf{r})^2}##

I understand that the procedure starts by Taylor expansion of the argument in the integral.

Taylor expanding at ##\mathbf{r}## gives, for a point ##\mathbf{r}'## close to ##\mathbf{r}##

##\frac{\nabla \rho(\mathbf{r}') \cdot \nabla \rho(\mathbf{r}')}{\rho(\mathbf{r}')} \simeq \frac{\nabla \rho(\mathbf{r}) \cdot \nabla \rho(\mathbf{r})}{\rho(\mathbf{r})} - \frac{1}{4} \frac{\nabla^2 \rho(\mathbf{r})}{\rho(\mathbf{r})} \delta \rho(\mathbf{r}') + \frac{1}{8}\frac{\nabla \rho(\mathbf{r}) \cdot \nabla \rho(\mathbf{r})}{\rho(\mathbf{r})^2}\delta \rho(\mathbf{r}')##

That is the result reported everywhere, including wikipedia, so I know it is correct. However, I have no problem with the first linear term, this is to say the term ##\frac{1}{4} \frac{\nabla^2 \rho(\mathbf{r})}{\rho(\mathbf{r})} \delta \rho## which is obtained by simply taking the derivative with respect to ##\rho## for the denominator. The term I have trouble with is the one that involves the derivative ##\frac{\partial }{\partial \rho} \nabla \rho(\mathbf{r}) \cdot \nabla \rho(\mathbf{r})## I don't really understand how to get the derivative of the gradient with respect to the function ##\rho##.
I haven't read this thread in too much detail, but it really seems like there's some confusing language in this thread about how functional derivatives are distinguished from normal derivatives. In particular, I don't think Taylor expansion is the way to go. What you probably want to do is use the formula described at this section of the linked Wikipedia page:
https://en.wikipedia.org/wiki/Functional_derivative#Determining_functional_derivatives

The key point is to consider the integral to be a function of two independent parameters: ##\rho## and ##\nabla \rho##. Then the functional derivative becomes:
$${\delta f(\rho, \nabla \rho) \over \delta \rho} = {\partial f \over \partial \rho} - \nabla \cdot {\partial f \over \partial \nabla \rho}$$

So to answer your question at the bottom, ##\frac{\partial }{\partial \rho} \nabla \rho(\mathbf{r}) \cdot \nabla \rho(\mathbf{r}) = 0##. The other term comes from taking the partial with respect to ##\nabla \rho##.

jim mcnamara and etotheipi

## 1. What is the functional derivative for the Thomas Fermi kinetic energy?

The functional derivative for the Thomas Fermi kinetic energy is a mathematical tool used to determine how a functional (a function of a function) changes with respect to its input function. In the context of the Thomas Fermi kinetic energy, it is used to calculate the variation of the kinetic energy functional with respect to the electron density function.

## 2. Why is the functional derivative important in the Thomas Fermi model?

The functional derivative is important in the Thomas Fermi model because it allows us to determine the electron density function that minimizes the total energy of the system. This is necessary for accurately predicting the behavior of a system of electrons in the presence of an external potential.

## 3. How is the functional derivative calculated for the Thomas Fermi kinetic energy?

The functional derivative for the Thomas Fermi kinetic energy is calculated by taking the derivative of the kinetic energy functional with respect to the electron density function and then setting it equal to zero. This results in an equation that can be solved for the electron density function.

## 4. What are the assumptions made in using the functional derivative in the Thomas Fermi model?

The functional derivative in the Thomas Fermi model makes several assumptions, including the assumption that the electrons are in a thermal equilibrium and that the system is in a state of local thermodynamic equilibrium. It also assumes that the electron density function is smooth and varies slowly compared to the size of the system.

## 5. Can the functional derivative be used for other kinetic energy functionals?

Yes, the functional derivative can be used for other kinetic energy functionals. However, it may not always result in an equation that can be easily solved for the electron density function, as it does in the Thomas Fermi model. In some cases, numerical methods may be necessary to find the optimal electron density function.

Replies
0
Views
271
Replies
1
Views
843
Replies
4
Views
1K
Replies
1
Views
1K
Replies
1
Views
905
Replies
4
Views
1K
Replies
5
Views
2K
Replies
1
Views
1K