Let's start from a very simple argument: If $A$ and $B$ are some operators, then I can write their product as
$$AB = (A-\langle A\rangle)(B - \langle B \rangle) + \langle A \rangle B + A \langle B \rangle + \langle A \rangle \langle B \rangle$$ In a mean-field approximation, I would then go on to neglect the first product on the right hand side under the assumption that fluctuations around the mean are small and thus terms quadratic in fluctuations can be neglected.

So far, so good. But then I look at the typical electron-electron interaction term: $$H_{int} = \frac{1}{2}\sum_{k,p,q,\alpha,\beta} V(q) c_{k+q,\alpha}^\dagger c_{p-q,\beta}^\dagger c_{p\beta} c_{k\alpha}$$ To form the operators $A$ and $B$, I have to pair a creation and destruction operator, but there are two different ways to do so.

For the sake of brevity, let's call the operators $c_1^\dagger c_2^\dagger c_3 c_4$. Then in the mean field approach one gets four different terms, one for each way to pick one creation and one destruction operator and everage over them: $$c_1^\dagger c_2^\dagger c_3 c_4 \approx -\langle c_1^\dagger c_3 \rangle c_2^\dagger c_4 -\langle c_2^\dagger c_4 \rangle c_1^\dagger c_3 +\langle c_1^\dagger c_4 \rangle c_2^\dagger c_3 +\langle c_2^\dagger c_3 \rangle c_1^\dagger c_4$$

Naively, I'd have expected a factor of 2 here, because here I used two ways of forming operators $A$ and $B$ (in the sense from above) to obtain my mean-field approximation.

  • $\begingroup$ There is no paradox, there are two ways to pair the operators, and each way gives you two terms, so four terms altogether. The proper formulation of this question is, choosing only one of the pairings to call operators "A" and "B", why do you still get four terms in the mean field expansion of the product? The answer to this question is best done with diagrams--- the original non-diagrammatic form of the mean field approximation is not the best phrasing, because you have to look at all contractions. $\endgroup$
    – Ron Maimon
    Commented Mar 4, 2012 at 2:33
  • $\begingroup$ Could you expand on the part about the diagrammatic derivation of this operator-approximation? $\endgroup$
    – Lagerbaer
    Commented Mar 4, 2012 at 2:57
  • 1
    $\begingroup$ I'm not sure what definition of V you are using. If you use the antisymmetrized matrix element ⟨c2 c1 V c3† c4†⟩, you will get a factor of 1/4, but if you use the Coulomb overlap integral ∫ φ*(x1) φ*(x2) V(x1 - x2) φ(x1) φ(x2) dx1 dx2 you will get a factor of 1/2. $\endgroup$
    – Rufflewind
    Commented Jan 27, 2015 at 23:06

1 Answer 1


I think I worked it out myself. What I found is that trying to arrive at the Hartree-Fock Hamiltonian via this mean-field procedure isn't really that appropriate.

Instead, one says: Given a two-particle operator $\hat O = c_1^\dagger c_2^\dagger c_3 c_4$, how can I find a single-particle operator $\hat O^{MF}$ that is a good approximation?

If we have a complete set of single-particle basis states that include states $1$ to $4$ of the two-particle operator, we would then look at the basis elements of $\hat O$ and $\hat O^{MF}$ and try to find the best possible agreement.

Now, the two-particle operator can do three different things: 1. Leave a state as is. This is the case if either $1 = 3$ and $2 = 4$ or $1 = 4$ and $2 = 3$. 2. Move one particle from a previously occupied state into a previously unoccupied state, e.g. if $1 = 3$ but $2 \not = 4$ or $1 = 4$ and $2 \not= 3$. 3. Move two particles into new states, i.e. if $1$ and $2$ are different from $3$ and $4$.

With our single particle operator, we can only hope to somehow reproduce the behavior of case 1 and 2, but not case 3.

So, given states of the type $|\psi_0\rangle$ that can be written as $\prod_i c_{n_i}^\dagger |0\rangle$, we look at all matrix elements of $\hat O$ between $|\psi_0\rangle$ and states of the type $c_i^\dagger c_j |\psi_0\rangle$, $i \not= j$, that arise from $|\psi_0\rangle$ by exchanging two particles.

It is a bit tedious to work these matrix elements out in detail, but one can then indeed confirm that by chosing $$\hat O^{MF} = \langle c_1^\dagger c_4\rangle c_2^\dagger c_3 + \langle c_2^\dagger c_3\rangle c_1^\dagger c_4 - \langle c_1^\dagger c_3\rangle c_2^\dagger c_4 - \langle c_2^\dagger c_4\rangle c_1^\dagger c_3$$ the matrix elements for the two-particle operator $\hat O$ and the mean-field operator $\hat O^{MF}$ will turn out to be the same (for the states of the type introduced above).


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.