I have been told that \[

[\hat x^2,\hat p^2]=2i\hbar (\hat x\hat p+\hat p\hat x)

\] illustrates

*.*

**ordering ambiguity**What does that mean?

I tried googling but to no avail.

Thank you.

**LM:**The ordering ambiguity is the statement – or the "problem" – that for a classical function \(f(x,p)\), or a function of analogous phase space variables, there may exist multiple operators \(\hat f(\hat x,\hat p)\) that represent it i.e. that have the right form in the \(\hbar\to 0\) limit. In particular, the quantum Hamiltonian isn't uniquely determined by the classical limit.

This ambiguity appears even if we require the quantum operator corresponding to a real function to be Hermitian and \(x^2 p^2\) is the simplest demonstration of this "more serious" problem. (The Hermitian parts of \(\hat x\hat x\hat p\) and \(\hat x \hat p \hat x\) or other cubic expressions are the same which is a fact that simplifies a step in the calculation of the hydrogen spectrum using the \(SO(4)\) symmetry.) On one hand, the Hermitian part of \(\hat x^2 \hat p^2\) is\[

\hat x^2 \hat p^2 - [\hat x^2,\hat p^2]/2 = \hat x^2\hat p^2 -i\hbar (\hat x\hat p+\hat p\hat x)

\] where I used your commutator.

On the other hand, we may also classically write the product and add the hats as \(\hat x \hat p^2\hat x\) which is already Hermitian. But\[

\hat x \hat p^2\hat x = \hat x^2 \hat p^2+\hat x[\hat p^2,\hat x] = \hat x^2\hat p^2-2i\hbar\hat x\hat p

\] where you see that the correction is different because \(\hat x\hat p+\hat p\hat x\) isn't quite equal to \(2\hat x\hat p\) (there's another, \(c\)-valued commutator by which they differ). So even when you consider the Hermitian parts of the operators "corresponding" to classical functions, there will be several possible operators that may be the answer. The \(x^2p^2\) is the simplest example and the two answers we got differed by a \(c\)-number. For higher powers or more general functions, the possible quantum operators may differ by \(q\)-numbers, nontrivial operators, too.

This is viewed as a deep problem (perhaps too excessive a description) by the physicists who study various effective quantum mechanical models such as those with a position-dependent mass – where we need \(p^2/2m(x)\) in the kinetic energy and by an expansion of \(m(x)\) around a minimum or a maximum, we may get the \(x^2p^2\) problem suggested above.

But the ambiguity shouldn't really be surprising because it's the quantum mechanics, and not the classical physics, that is fundamental. The quantum Hamiltonian contains all the information, including all the behavior in the classical limit. On the other hand, one can't "reconstruct" the full quantum answer out of its classical limit. If you know the limit \(\lim_{\hbar\to 0} g(\hbar)\) of one variable \(g(\hbar)\), it clearly doesn't mean that you know the whole function \(g(\hbar)\) for any \(\hbar\).

Many people don't get this fundamental point because they think of classical physics as the fundamental theory and they consider quantum mechanics just a confusing cherry on a pie that may nevertheless obtained by quantization, a procedure they consider canonical and unique (just hat addition). It's the other way around, quantum mechanics is fundamental, classical physics is just a derivable approximation valid in a limit, and the process of quantization isn't producing unique results for a sufficiently general classical limit.

**Quantum field theory**

The ordering ambiguity also arises in field theory. In that case, all the ambiguous corrections are actually divergent, due to short-distance singularities, and the proper definition of the quantum theory requires one to understand renormalization. At the end, what we should really be interested in is the space of relevant/consistent quantum theories, not "the right quantum counterpart" of a classical theory (the latter isn't fundamental so it shouldn't stand at the beginning or base of our derivations).

In the path-integral approach, one effectively deals with classical fields and their classical functions so the ordering ambiguities seem to be absent; in reality, all the consequences of these ambiguities reappear anyway due to the UV divergences that must be regularized and renormalized. I have previously explained why the uncertainty principle and nonzero commutators are respected by the path-integral approach due to the non-smoothness of the dominant paths or histories. There are additional things to say once we start to deal with UV-divergent diagrams. The process of regularization and renormalization depends on the subtraction of various divergent counterterms, to get the finite answer, which isn't quite unique, either (the finite leftover coupling may be anything and the value has to be determined by a measurement).

That's why the renormalization ambiguities are just the ordering ambiguities in a different language – one that is more physical and less dependent on the would-be classical formalism, however. Whether we study those things as ordering ambiguities or renormalization ambiguities, the lesson is clear: the space of possible classical theories isn't the same thing as the space of possible quantum theories and we shouldn't think about the classical answers when we actually want to do something else – to solve the problems in quantum mechanics.

Incidentally, for TRF readers only: the "space of possible theories" is similar to the "moduli space" and the Seiberg-Witten analysis of the \(\NNN=2\) gauge theories is the prettiest among the simplest examples of the fact that the quantum moduli space is different than the classical one – often much richer and more interesting. You should never assume the classical answers when you're doing quantum mechanics because quantum mechanics has the right to modify all these answers. And this world is quantum mechanical, stupid.

Oh darn,

ReplyDeletenow I have to login into Physics SE and upvote this nice answer, I cant help it ... :-D!

This is a nice article. You mention that in the path integral where one deals with classical fields the ordering ambiguities seem to be absent but reappear due to the UV divergences. I remember that the path integral can be derived from the Schrodinger equation (plus the identity 1 = \int dq |q> <q| ) through a discretization plus limiting procedure. Aren't there already ambiguities in the discretization procedure and if so are they related to the operators ordering ambiguities?

ReplyDeleteI'm curious about this because for other reasons I was looking at the path integral solutions of stochastic differential equations (Langevin / Fokker-Planck) and there are discretization ambiguities (in the case of multiplicative noise) which lead to different path integrals depending on the discretization procedure (for example Ito or Stratonovitch) you choose. And a Fokker Planck equation is a bit like an Euclidean Schrodinger equation.

Dear Scooby, I wasn't assuming one does any discretization - the "major" ways to evaluate path integrals don't use any - but if one exploits some discretization, all the ambiguities reappear in the precise definition of the discretized theory.

ReplyDeleteThey're short-distance ambiguities; in the continuum, these ambiguities may be linked to "strictly zero" distances. In a discretized picture, these minimal distances are finite but at this level of the lattice spacing etc., one finds all the ambiguities - and perhaps even new ones.

The Ito-Stratonovitch ambiguity is closer to time-ordering ambiguities etc. From some more vague viewpoints, they're all analogous, however.

Indeed there are some papers doing unambiguosly this link between discretization ambiguities and ordering ambiguities, so that different choosings of the derivative map to different orderings. Very old PhysReview and similar, I have not read them since in 20 years.

ReplyDeleteGoing to the speculative thing, and connecting with the remark of Lubos on renormalisation, it is also amusing that Runge-Kutta methods and renormalization are controlled by a same structure, Butcher trees. http://en.wikipedia.org/wiki/Butcher_group