Status of local gauge invariance in axiomatic quantum field theory In his recent review...


*

*Sergio Doplicher, The principle of locality: Effectiveness, fate, and challenges, J. Math. Phys. 51, 015218 (2010), doi
...Sergio Doplicher mentions an important open problem in the program of axiomatic quantum field theory, quotes:

In the physical Minkowski space, the study of possible extensions of superselection theory
  and statistics to theories with massless particles such as QED is still a fundamental open problem.

...

More generally the algebraic meaning of quantum gauge theories in terms local observables is
  missing.
  This is disappointing since the success of the standard model showed the key role of the gauge
  principle in the description of the physical world; and because the validity of the principle of
  locality itself might be thought to have a dynamical origin in the local nature of the fundamental
  interactions, which is dictated by the gauge principle combined with the principle of minimal
  coupling.

While it is usually hard enough to understand the definite results of a research program, it is even harder if not impossible, as an outsider, to understand the so far unsuccessful attempts to solve important open problems.
So my question is: 


*

*Is it possible to decribe the walls that attempts to incorporate local gauge invariance in AQFT have hit?

*What about the possibility that this is the wrong question and there is and should not be a place of local gauge invariance in AQFT?
Edit: For this question let's assume that a central goal of the reasearch program of AQFT is to be able to describe the same phenomena as the standard model of particle physics. 
 A: I think the open question here should be formulated -- and usually is formulated -- not as whether Yang-Mills theory "has a place" in AQFT, but whether one can abstractly characterize those local nets that arise from quantization of Yang-Mills-type lagrangians. In other words: since AQFT provides an axiomatics for QFT independently of a quantization process starting form an action functional: can one detect from the end result of quantization that it started out from a Yang-Mills-type Lagrangian?
On the other hand, we certainly expect that the quantization of any Yang-Mills-type action functional yields something that satisfies the axioms of AQFT. While for a long time there was no good suggestion for how to demonstrate this, Fredenhagen et al. have more recently been discussing how all the standard techniques of perturbative QFT serve to provide (perturbative) constructions of local nets of observables. References are collected here, see in particular the last one there on the perturbative construction of local nets of observables for QED.
In this respect, concerning Kelly's comment, one should remember that also the construction of gauge theory examples in axiomatic TQFT is not completely solved. One expects that the Reshetikhin-Turaev+construction for the modular category of $\Omega G$-representations gives the quantization of G-Chern-Simons theory, but I am not aware that this has been fully proven. And for Chern-Simons theory as an extended TQFT there has only recently be only a partial proposal for the abelian case FHLT. Finally, notice also that non-finite degrees of freedom can be incorporated here, if one passes to non-compact cobordisms (see the end of Lurie's), which in 2-dimensions are "TCFTs" that contain all the 2d TQFT models that physicists care about, such as the A-model and the B-model.
Concerning Moshe's comment: the known dualities between gauge and non-gauge theories usually involve a shift in dimension. This would still seem to allow for the question whether a net in a fixed dimension is that of a Yang-Mills-type theory.
But even if it turns out that Yang-Mills-type QFTs do not have an intrinsic characterization. their important invariant porperties should have. For instance it should be possible to tell from a local net of observables whether the theory is asymptotically free. I guess?
A: This is more of an extended comment, rather than an answer completely separate from what Urs and Moshe have already said. The axioms of AQFT are designed to capture a mathematical model of the physical observables of a theory, while OTOH gauge invariance is a feature of a formulation of a theory, though perhaps an especially convenient one. Yours and related questions are somewhat muddied by the fact that one physical theory may have several equivalent, but distinct formulations, which may also have different gauge symmetries. One example of this phenomenon is gravity, consider the metric and frame-field formulations, and another one according to Moshe is Seiberg duality. Another confounding factor is that some physical theories are only known in a formulation involving gauge symmetries (automatically rendering such formulations "especially convenient"), which naturally leads to your second question. However, one must remember that by design the gauge formulation should be visible in the AQFT framework only if it is detectable through physical observables.
Now, to be honest, I really have no idea about what is the state of the art in AQFT of figuring out when a given net of local algebras of observables admits an "especially convenient" formulation involving gauge symmetry. But I believe answering this kind of question will remain difficult until the notion of "especially convenient" is made mathematically precise. I don't know how much progress has been made on that front either. But I think a prototype of this kind of question can be analyzed, though somewhat sketchily, in the simplified case of classical electrodynamics.
Suppose we are given a local net of Poisson algebras of physical observables (the quantum counterpart would have *-algebras, but other than that, the geometry of the theory is very similar). The first step is to somehow recognize that this net of algebras is generated by polynomials in smeared fields, $\int f(F,x) g(x)$, where $g(x)$ is a test volume form, and $f(F,x)$ is some function of $F$ and its derivatives at $x$, with $F(x)$ a 2-form satisfying Maxwell's equations $dF=0$ and $d({*}F)=0$. Since we were handed the net of algebras with a given Poisson structure, as a second step we can compute the Poisson bracket $\{F(x),F(y)\}=(\cdots)$. The answer for electrodynamics would be the well known Pauli-Jordan / Lichnerowicz / causal propagator, which I will not reproduce here. Very roughly speaking, the components of $F(x)$ and the expression for the Pauli-Jordan propagator give a set of local "coordinates" on the phase space of the theory and an expression for the Poisson tensor on it. In the third step we can compute the inverse of the Poisson tensor, which if it exists would be a symplectic form. The answer for electrodynamics is well known and what's important is that the symplectic form is not given by some local expression like $\Omega(\delta F_1,\delta F_2) = \int \omega(\delta F_1, \delta F_2, x)$, where $\omega$ is a form depending only on the values of $\delta F_{1,2}(x)$ and their derivatives at $x$. Step four would consist of asking the question whether there is another choice of local "coordinates" on the phase space in which the symplectic form is local. The answer is again well known: extend the phase space by introducing the 1-form field $A(x)$ such that $F=dA$. The pullback of the symplectic form to the extended phase space now has a local expression $\Omega(\delta A_1,\delta A_2) = \int_\Sigma [{*}d(\delta A_1)(x)\wedge (\delta A_2)(x) - (1\leftrightarrow 2)]$, up to some constant factors, with $\Sigma$ some Cauchy surface. Note that $\Omega$ is no longer symplectic on the extended phase space, but only presymplectic, while its projection back to the physical phase space is. As a last step, one might try to solve the inverse problem of the calculus of variations and come up with a local action principle reproducing the the equations of motion for $A$ and the presymplectic structure $\Omega$.
Let me summarize. (1) Obtain fundamental local fields and their equations of motion. (2) Express the Poisson tensor and symplectic form in terms of local fields. (3) Introduce new fields to make the expression for the (pre)symplectic form local. (4) Obtain local action principle in the new fields. Note that gauge symmetry and all the issues associated with it appear precisely in step (3). In my limited understanding of it, the literature on AQFT has spent a significant amount of time on step (1), but perhaps not enough time on steps (2) and (3) even to formulate these problems precisely.
Finally, I should emphasize that the idea that redundant gauge degrees of freedom are introduced principally to give local structure to the (pre)symplectic structure on phase space is somewhat speculative. But it seems to fit in the field theories I am familiar with and I've not been able to identify a different yet equally competitive one.
A: 
What about the possibility that this is the wrong question and there is and should not be a place of (sic) local gauge invariance in AQFT?

I'd guess this hinges upon how one views AQFT. One can view AQFT in one of two ways:


*

*AQFT as a theory needs to correspond to nature.

*AQFT as a theory need not correspond to nature.


If AQFT needs to correspond to nature, then it should incorporate local gauge invariance, as nature incorporates local gauge invariance. (Note, "incorporate" here could mean include a mechanism which at "low energies" looks like local gauge invariance.)
If AQFT need not correspond to nature, then it need not incorporate local gauge invariance.
With that in mind I would also add that axiomatic TQFT includes local gauge symmetries without problems. In fact, axiomatic TQFT local gauge symmetries are so "strong" that they remove all local degrees of freedom.
