In Kopec and Usadel's Phys. Rev. Lett. 78.1988, a spin glass Hamiltonian is introduced in the form:

$$ H = \frac{\Delta}{2}\sum_i \Pi^2_i - \sum_{i<j}J_{ij}\sigma_i \sigma_j, $$ where the variables $ \sigma_i (i = 1, \ldots, N) $ are associated with spin degrees of freedom [...] and canonically conjugated to the "momentum" operators $ \Pi_i $ such that $ [\sigma_i, \Pi_j] = i \delta_{ij} $.

Now, I am accustomed to writing the "kinetic" term in a transverse-field Ising-like Hamiltonian as $ \propto \sum_i \sigma^x_i $ (working in the standard basis of $ \{\sigma^z_i\} $), so this passage is raising some questions for me.

What are these $ \Pi_i $ operators? If $ \Pi_i^2 = \sigma^x_i $, like I initially believed, then they cannot be observables, for the square of a self-adjoint operator is positive semidefinite (which $ \sigma^x_i $ is not). In fact, if one restricts to the $ i $-th spin and takes $ i = j $, one can easily prove that $$ [\sigma^z, \Pi] = \sigma^z \Pi - \Pi^\dagger \sigma^z = i \mathbb 1 $$ is satisfied for $$ \Pi = \begin{pmatrix}i/2&b\\-\bar{b}&-i/2\end{pmatrix} $$ with $ b \in \mathbb C $. This squares to a multiple of the identity matrix, which seems like an odd choice for a kinetic term. I feel I am missing something here.

More generally speaking, can one even define a momentum "canonically conjugate" to $ \sigma^z $, or any other spin operator for that matter? As far as I understand, in classical mechanics the variables conjugate to physical rotations are angles, but this cannot be ported over to QM in any obvious way.

  • $\begingroup$ have a look at JxJy .... uncertainty relations en.wikipedia.org/wiki/… . $\endgroup$
    – anna v
    Feb 14, 2016 at 11:51
  • $\begingroup$ Minor comment to the post (v1): In the future please link to abstract pages rather than pdf files. $\endgroup$
    – Qmechanic
    Mar 21, 2018 at 17:46
  • $\begingroup$ @Qmechanic Duly noted. Thank you for pointing that out. $\endgroup$
    – derpy
    Mar 21, 2018 at 20:40

2 Answers 2


Even though the explicit commutator you wrote is wrong--you should not have conjugated $\Pi$ in the second term-- your conclusion is sound that you cannot possibly satisfy the Born-Heisenberg commutation relation with 2x2 matrices.

In fact, there is a general Theorem: The Heisenberg algebra does not admit faithful finite-dimensional (matrix) representations. So, whatever else they might be, your variables $\sigma, \Pi$ are not bounded operators---and so cannot be the 2x2 matrices you are considering.

This observation was first made by P Jordan, Zeits. f. Phys. 44 1 (1927).


First of all, there is a very elementary reason why $[\sigma_i,\Pi_j] = i \delta_{ij}$ is impossible for finite dimensional matrices. Because that identity would result in the following contradiction: $$ \text{Tr}[A,B] = 0 \neq \text{Tr} (i \cdot \mathbb{1}) = i \cdot n$$ Where $A,B$ are $n \times n$ matrices.

However, if we read the paper by Kopec and Usadel carefully, we notice a key phrase that is not mentioned in the question:

To capture the essential physics of the problem we consider a quantized spherical model on the Bethe lattice given by the Hamiltonian: $$ H = \frac{\Delta}{2} \sum_i \Pi_i^2 - \sum_{i,j} J_{ij} \sigma_i \sigma_j $$

The highlighted word "spherical model" means that we have the spherical constraint $\sum_i \sigma_i^2 = 1$ instead of the Ising constraint $\sigma_i^2 = 1$ for all $i$. So the $\sigma_i$ in this model is in fact a constrained position operator acting on an infinite dimensional Hilbert space, and a commutation relation like $[\sigma_i,\Pi_j] = i \delta_{ij}$ becomes conceivable.

However, I have one lingering confusion. When considering spherical constraints, I am used to quantizing the Dirac brackets instead of Poisson brackets (The Dirac bracket is a modification of Poisson brackets, discussed, for example, in Weinberg QFT section 7.6). Using Dirac brackets: $$ [\Pi_i, \sigma_j] \neq i \mathbb{1}$$ So I don't quite understand why the author used the standard canonical commutation relation to quantize the system. I hope someone else will clear up this point.

  • 1
    $\begingroup$ You didn't quote enough from the paper. The very next sentence clears up your point: ''located on the Bethe lattice with coordination z". Thus there is no spherical constraint imposed; the $\sigma_j$ just run over the lattice. Instead, it is said a few lines later, that only the mean of the spherical relation is enforced to be reproduced. $\endgroup$ Mar 28, 2018 at 18:44
  • $\begingroup$ @ArnoldNeumaier Thanks for pointing that out. I in fact did not notice this statement. However this statement makes the paper more confusing for me. If $\sigma_j$'s are discrete variables, then the commutation relation is impossible. Now, if we take $\sigma_j$'s to be continuous variables satisfying the averaged constraint, then I don't understand equation (2) in the paper, which clearly uses a delta function to impose the exact spherical constraint! In either case I don't understand why we are allowed to do canonical quantization instead of Dirac quantization. $\endgroup$ Mar 29, 2018 at 3:09
  • $\begingroup$ Yes, they don't do what they say and in fact enforce in the path integral (2) not the averaged constraint but the exact constraint. Using the integral representation of the delta-constraint introduced after (3) converts this (nonrigorously) to an integral over a path integral of the unconstrained system. This allows them to evaluate the path integral using the standard commutation rules. $\endgroup$ Apr 1, 2018 at 17:04

Your Answer

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

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