2
$\begingroup$

I am reading the book "Electronic Structure" by Richard Martin which poses the following problem:

Show that the expectation value of an operator $\hat O$ in a system of identical, non-interacting Fermions (i.e. electrons with the independent particle approximation) has the following form:

$$ \left<\hat O\right> = \sum_{i,\sigma} f_i^\sigma \left<\psi_i^\sigma|\hat O|\psi_i^\sigma\right>$$

Where, $f_i^\sigma = \frac{1}{e^{\beta(\epsilon_i^\sigma - \mu)} + 1}$ and $\psi_i^\sigma$ is the eigenvector of the ith single particle state with spin $\sigma$.

I have begun the derivation thusly (j is the jth eigenstate of the multibody system):

$$ \left<\hat O\right> = Tr\left(\hat \rho \hat O \right) = \sum_j \left<\Psi_j|\hat \rho \hat O |\Psi_j \right>$$

Since $\hat \rho$ is Hermitian:

$$ \sum_j \left<\Psi_j|\hat \rho \hat O |\Psi_j \right> = \sum_j \left<\hat \rho\Psi_j| \hat O |\Psi_j \right> $$

In the Grand Canoncial Ensemble: $\hat \rho = \frac{1}{Z}e^{-\beta(\hat H - \mu\hat N)} $

Further, considering that the jth eigenstate of the multibody system will be composed of N single particle wave functions, one can write:

$$ \sum_j \left<\hat \rho\Psi_j| \hat O |\Psi_j \right> = \sum_{\{n_i,\sigma_i\}} \frac{1}{Z}e^{-\beta(\sum_i n_i\epsilon_i^{\sigma_i}-\mu \sum_i n_i)}\left<\Psi_j| \hat O |\Psi_j \right> $$

Based on my understanding of the other quantum mechanics reference I have been using, if $\hat O$ was simply the operator which returns the number of particles in the ith single particle state with spin $\sigma$, the sum reduces to the $f_i^\sigma$. So my guess as how to obtain the general result is to suppose that $\hat O$ is the combination of N single particle operators $\hat O = \sum_i \hat O_i$.

Hence,

$$\left<\Psi_j| \hat O |\Psi_j \right> = \left<\Psi_j| \sum_i \hat O_i |\Psi_j \right> = \sum_i \left<\psi_i^\sigma | \hat O_i | \psi_i^\sigma \right>$$

After this point, I am stuck and I do not see how this could reduce down to the given result. Further it would seem to me that the author should have placed a restriction on $\hat O$ to be only a single particle operator on the rhs of the target result.

$\endgroup$
2
  • 1
    $\begingroup$ Consider to use \langle and \rangle instead of < and >; this looks much better IMO. $\endgroup$ Commented Jul 26, 2021 at 7:56
  • $\begingroup$ Note: In general, the (non-interacting grand canonical) equilibrium expectation values of n-body operators can be evaluated with the help of Wick's theorem, c.f. e.g. Reichl: 'A Modern Course in Statistical Physics', 4th edition, Appendix D, section 2.3. $\endgroup$ Commented Jul 26, 2021 at 11:57

2 Answers 2

2
$\begingroup$

I think this is a situation where it is much easier to derive the desired result using second quantization. To start, let us consider a system of identical particles and let $O = \sum\limits_{k} o_k $ denote a generic one-body operator on the respective Fock space. In the language of second quantization, we can express this operator as $$ O = \sum\limits_{ij} \langle i|o|j\rangle \,a_i^\dagger a_j \quad . $$ The expectation value of an (one-body) operator in the state $\rho$ is defined by $\langle O \rangle_\rho \equiv \mathrm{Tr} \, \rho \, O$, where the trace is performed on the Fock space. By defining the elements of the one-body reduced density matrix in the state $\rho$ as $$\gamma_{ij} \equiv \mathrm{Tr}\, \rho\, a_j^\dagger a_i \quad ,$$ we see that we can write the expectation value of $O$ as $$\langle O \rangle _\rho = \mathrm{tr}\, \gamma \,o \quad , $$ where now the trace is performed on the single-particle Hilbert space.


In the following, $i,j$ denote elements of the basis in which the single-particle Hamiltonian is diagonal.

For a system of non-interacting fermions and in equilibrium in the grand canonical ensemble, it follows that

$$ \gamma_{ij} = \delta_{ij} \,\langle n_i \rangle_\rho \quad . $$ This can be derived by e.g. applying Wick's theorem. Here, $$\langle n_i \rangle_\rho = \frac{1}{e^{(\epsilon_i - \mu)/{k_{\mathrm{B}}T}}+1} $$ is the well-known expression for the average occupation number of the single-particle state $i$.

Finally, this shows that indeed $$ \langle O \rangle_\rho = \sum\limits_i \frac{1}{e^{(\epsilon_i - \mu)/{k_{\mathrm{B}}T}}+1}\, \langle i|o|i\rangle \quad \quad ,$$ for a system of non-interacting fermions in grand canonical equilibrium.

As a last point, note that an analogous result holds for the case of non-interacting identical bosons.

$\endgroup$
4
  • $\begingroup$ Thanks, I am not very familiar with second quantization so I don't quite follow the first half of the derivation completely. Though I think I have some ideas now to try to reduce the operator expectation value to $tr\gamma o$ in 1st quantization. If I can get there, then as you have shown, the rest of the derivation follows from the fact that $\gamma_{ij} = \delta_{ij} \langle n_i \rangle$ $\endgroup$
    – gigo318
    Commented Jul 26, 2021 at 17:56
  • 1
    $\begingroup$ @gigo318 Okay, great. A good introduction for the case of first quantization is given in Parr and Yang: 'Density-Functional Theory of Atoms and Molecules', 1989, section 2.3. The derivation of the fact that one can express expectation values with the help of the 1-RDM is similar as the derivation that for local one-body operators you can use the density to express the expectation value. Anyway, it is always good to learn some second quantization. $\endgroup$ Commented Jul 26, 2021 at 18:19
  • 1
    $\begingroup$ @gigo318 Actually, there are better sources for the derivation. Take a look here or here, section 3.52. starting at p. 21. Hope this helps. $\endgroup$ Commented Jul 26, 2021 at 18:58
  • 1
    $\begingroup$ Cool, I will definitely give these a read! I feel like many DFT sources I have read are either too cursory or jump straight into more advanced aspects of the theory. $\endgroup$
    – gigo318
    Commented Jul 26, 2021 at 19:06
2
$\begingroup$

So I believe it can also be derived using 1st quantization as follows:

$$\langle \hat O \rangle = \sum_{\left\{ n \right\}_j \in N^{\infty}} \langle \Psi_j|\hat \rho \hat O |\Psi_j\rangle = \sum_{\left\{ n \right\}_j \in N^{\infty}} \langle \hat \rho \Psi_j|\hat O |\Psi_j\rangle =\sum_{\left\{ n \right\}_j \in N^{\infty}} \tilde \rho_j \langle \Psi_j| \hat O |\Psi_j\rangle$$

Where $\tilde \rho_j$ is the eigenvalue of $\hat \rho$ associated with state j.

Presuming that $\hat O$ can be written as the sum of one body operators:

$$\hat O = \sum_k \hat o_k$$

Where k is the index of particle k. One can now write:

$$\langle \hat O \rangle = \sum_{\left\{ n \right\}_j \in N^{\infty}} \tilde \rho_j \langle \Psi_j| \sum_k \hat o_k |\Psi_j\rangle = \sum_{\left\{ n \right\}_j \in N^{\infty}} \tilde \rho_j \sum_i n_i\langle \psi_i| \hat o |\psi_i\rangle$$

Where I have used the fact that the expectation value of the single body operator $\hat o_k$ acting on the the multibody wave function $\Psi_j$ is: $$ \langle \Psi_j | \hat o_k | \Psi_j \rangle = \sum_i n_i \frac{(N-1)!}{N!} \langle\psi_i|\hat o |\psi_i \rangle$$

Where i sums over all the single-body eigenstates present in $\Psi_j$, the $N!$ in the denominator comes from the fact that $\Psi_j$ is represented by a Slater determinant of single-body eigenstates. The $(N-1)!$ in the numerator represents the fact that each particle index is repeated $N!/N$ times. Finally, $n_i$ represents the number of times that eigenstate i is repeated.

Thus when this sum is performed N times, the factorials are eliminated.

Returning to $\langle \hat O \rangle$, one has:

$$\langle \hat O \rangle = \sum_{\left\{ n \right\}_j \in N^{\infty}} \tilde \rho_j \sum_i n_i\langle \psi_i| \hat o |\psi_i\rangle = \sum_{\left\{ n \right\}_j \in N^{\infty}} \sum_i \tilde \rho_j n_i\langle \psi_i| \hat o |\psi_i\rangle = \sum_{i} \sum_{\left\{ n \right\}_j \in N^{\infty}} \tilde \rho_j n_i\langle \psi_i| \hat o |\psi_i\rangle $$

Which can be achieved by re-indexing the sum appropriately (I think?).

Continuing:

$$\langle \hat O \rangle = \sum_i \left( \sum_{\left\{ n \right\}_j \in N^{\infty}} \tilde \rho_j n_i \right) \langle \psi_i| \hat o |\psi_i\rangle = \sum_i \langle n_i \rangle \langle \psi_i| \hat o |\psi_i\rangle$$

Using the fact that: $$ \langle n_i \rangle = \frac{1}{e^{\beta(\epsilon_i-\mu)} \pm 1} = f_i$$

One has:

$$\langle \hat O \rangle = \sum_i f_i \langle \psi_i| \hat o |\psi_i\rangle $$

As originally desired.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.