In Yang-Mills theory on a Euclidean Cauchy surface, the physical gauge group is often taken to be \mathcal{G}^I/\mathcal{G}^\infty_0, where \mathcal{G}^I consists of boundary-preserving gauge transformations asymptoting to a constant, and \mathcal{G}^\infty_0 consists of transformations generated by the Gauss law constraint. We rigorously derive this physical gauge group for both Abelian and non-Abelian theories. A key result is that restricting to \mathcal{G}^I follows from the structure of the instantaneous state space on which the instantaneous Lagrangian is defined. We extend our analysis to Yang-Mills-Higgs theory, showing that boundary conditions and the physical gauge group differ between the unbroken and broken phases.
Let \pi\colon P\to \Sigma be a principal G-bundle. Then \Omega^k_\text{hor}(P,\mathfrak{g})^\text{Ad} and \Omega^k(\Sigma,\text{Ad}(P)) are canonically isomorphic as vector spaces through the pullback(Recall that for a fiber bundle E\to N any map f\colon M\to N induces a pullback bundle f^*E\to M. In this case the pullback (of the adjoint bundle) is the trivial vector bundle P\times \mathfrak{g}.) \pi^*.
Two connections A,A'\in\Omega^1(P,\mathfrak{g}) cannot be added or subtracted. Of course, we can define A+A' and A-A' as forms, but these forms do not reproduce the generators \xi\in\mathfrak{g} of fundamental vector fields X_\xi, since (A+A')(X_\xi)=2\xi and (A-A')(X_\xi)=0. Thus connection 1-forms cannot be added or subtracted to produce new connection 1-forms. By Theorem 2.1, however, differences of connections w.r.t. a fixed reference connection A_\text{ref} form a vector space. Therefore differences of connections w.r.t. A_\text{ref} can indeed be added and subtracted. When one trivializes P with a section s, there is a preferred reference connection, namely the zero connection on the trivial bundle. On an untrivialized bundle, however, there is no notion of “zero”, implying that the space of connections is an affine space. Thus, T\text{Conn}(P)\cong T\Omega^1(\Sigma,\text{Ad}(P)) is really an identification of the tangent bundles of affine spaces, where the origin of \Omega^1(\Sigma,\text{Ad}(P) is “forgotten”.
The instantaneous Lagrangian of Yang-Mills theory in temporal gauge is [armsLinearizationStabilityGravitational1979]
Here F(A) denotes the curvature 2-form of the connection 1-form A, which is the magnetic field, and \left\lVert\cdot\right\rVert is the usual norm on forms:
where * denotes the Hodge star operator.(It is important to remember for our conformal analysis in Section ? that the Hodge star operator “contains” the metric.)
We can derive expression (3.1) for the Lagrangian from the usual covariant action on spacetime M=\Sigma\times\mathbb{R}:
Here we have written \tilde{A} to stress that this is a gauge field on spacetime M instead of space \Sigma. Denoting coordinates on \Sigma by x^i and the coordinate on \mathbb{R} by t=x^0, it is not difficult to show that the action in coordinates becomes the usual [hamiltonMathematicalGaugeTheory2017]
where \mu=0,1,2,3, i=1,2,3 and F_{\mu\nu}=\partial_\mu A_\nu-\partial_\nu A_\mu+[A_\mu,A_\nu] is the antisymmetric field strength tensor (which clearly satisfies F_{00}=0). The term \text{Tr}(F_{0i}F^{0i}) is (minus) the energy of the electric field (the “kinetic” energy), and the term \text{Tr}(F_{ij}F^{ij}) equals twice the energy of the magnetic field (the “potential” energy).
If we now impose temporal gauge A_0=0 we obtain F_{0i}=\partial_0 A_i=\dot{A}_i. We can then rewrite the action as(We use (-,+,+,+) signature for the metric, which explains the minus sign in -\dot{A}_i\dot{A}^i.)
But F_{ij} is just the curvature of the connection A_i on three-dimensional space \Sigma, so in coordinate-free notation we find, with slight abuse of notation:
where it is understood that A\in Q\subset \Omega^1(\Sigma,\text{Ad}(P)) signifies the spatial part of A_\mu. Technically there is no sense in which the spatial gauge field A has a time-derivative \dot{A}. What is really meant by this expression is that \dot{A} should be viewed as a tangent vector at the point A, obtained as a derivative along a curve \mathbb{R}\to Q. Thus we replace \dot{A}\to \alpha_A\in T_AQ and we obtain the Lagrangian in Eq. (3.1). \Box
The condition \hat{\alpha}|_{\partial\hat{\Sigma}}=0 guarantees square-integrability.
Since \hat{\alpha} is smooth even on the boundary \partial\hat{\Sigma}, we can perform a Taylor expansion at a point on the boundary in the coordinate \rho=\pi-R\geq 0 around \rho=0. Expanding
we see that the requirement \hat{\alpha}|_{\partial\hat{\Sigma}}=0 implies a_0=b_0=0, so that to lowest order \hat{\alpha}_\theta and \hat{\alpha}_\phi are linear in \rho. But for large r:
Since \hat{\alpha}_\theta,\hat{\alpha}_\phi\sim \rho close to the boundary \partial\hat{\Sigma}, this implies that f^*\hat{\alpha}_\theta,f^*\hat{\alpha}_\phi\to 0+\mathcal{O}(r^{-1}). So the requirement that a 1-form \alpha on \Sigma equals the pullback of some 1-form \hat{\alpha} on \hat{\Sigma} that vanishes on the boundary, gives the conditions \alpha_r\to 0+\mathcal{O}(r^{-2}) and \alpha_\theta,\alpha_\phi\to 0+\mathcal{O}(r^{-1}), which are strong enough to guarantee square-integrability. \Box
A curvature 2-form on \Sigma obtained as f^*\hat{F}, where \hat{F} extends smoothly to \partial\hat{\Sigma}, is square-integrable.
Since F_{ii} components vanish by antisymmetricity, it suffices to check F_{r\theta},F_{r\phi} and F_{\theta\phi}. As the radial component is the only one whose exterior derivative transforms with a conformal factor (i.e. dR=K\,dr), we find that f^*\hat{F}_{r\theta}=K\,F_{r\theta}, and the same for F_{r\phi}. The component F_{\theta\phi} is left invariant. Thus, if a 2-form F is assumed to equal a pullback f^*\hat{F}, then we automatically find that F_{r\theta},F_{r\phi}\to 0+\mathcal{O}(r^{-2}) and F_{\theta\phi}=\mathcal{O}(1). Comparing to Section ? we find that this is easily enough to guarantee that F is square-integrable. Thus, if one assumes a 1-form A on \Sigma to equal the pullback of a 1-form \hat{A} on \hat{\Sigma}, then the curvature F(\hat{A}) is a well-defined 2-form on \hat{\Sigma} and therefore one automatically finds that F(A) is square-integrable. \Box
When choosing the zero connection as the Dirichlet boundary condition on the conformal boundary, we are already viewing the space \Omega^1(\partial\hat{\Sigma},\mathfrak{g}) of connections on the boundary as a vector space rather than an affine space. Indeed, as an affine space, this space of connections has no zero element. It is therefore more precise to say that one first chooses a Dirichlet boundary condition as an element \hat{A}_\infty\in \text{Conn}(\hat{P}_{\partial\hat{ \Sigma}}), i.e. as a connection 1-form over the boundary, and then one uses this connection as the reference appearing in Theorem 2.1. This ensures that, on the boundary, the Dirichlet boundary condition appears as the zero connection. On the interior of space, the space of connections must then still be viewed as an affine space (cf. Remark 2.2), but this procedure nonetheless gives a trivialization-independent definition of the dynamical sector defined by one’s choice of Dirichlet boundary condition.
Let (\mathcal{P},\omega) be a symplectic manifold and H a Lie group that acts on \mathcal{P} by symplectomorphisms.(I.e. the action of H preserves the symplectic form \omega.) Let \mathfrak{h} denote the Lie algebra of H with dual \mathfrak{h}^*, and write \langle\cdot,\cdot\rangle\colon \mathfrak{h}^*\times\mathfrak{h}\to\mathbb{R} for the pairing of the algebra and its dual. Then a momentum map for the H-action on \mathcal{P} is an equivariant(With respect to the H-action on \mathcal{P} and the coadjoint action on \mathfrak{h}^*.) map \mu\colon \mathcal{P}\to\mathfrak{h}^* such that, for all \xi\in\mathfrak{h}, we have:
Here, \mathbb{X}_\xi denotes the fundamental vector field(In this definition we use the double-slashed notation because this agrees with our subsequent calculations, but of course this definition of the momentum map is also valid for finite-dimensional symplectic manifolds.) generated by \xi, and \langle\mu,\xi\rangle is understood as a function \langle\mu,\xi\rangle\colon \mathcal{P}\to\mathbb{R}, defined as follows: \langle\mu,\xi\rangle(x)=\langle\mu(x),\xi\rangle.
If one requires E\to 0+\mathcal{O}(r^{-2}), then gauge parameters \xi need to approach zero without any particular fall-off rate in order for the Gauss constraint to be the momentum map. If instead we demand E\to 0+\mathcal{O}(r^{-3/2-\epsilon}), then we must have \xi\to 0+\mathcal{O}(r^{-1/2}).
For any \xi\in C_I^\infty(\Sigma,\mathfrak{g}) (and using Stokes’ theorem/partial integration), we have:
where we have used the ad-invariance of the trace, i.e. \text{Tr}\left([E,A]\wedge \xi\right)=\text{Tr}\left(E\wedge[A,\xi]\right). Note that for consistency we have used the \wedge symbol even on \xi\in C^\infty_I(\Sigma,\mathfrak{g}), even though it is a 0-form.
But, if \mu(A,E)=D_AE really is to define the momentum map for the action of \mathcal{G}^I, then by definition it must satisfy the property
where \mathbb{X}_\xi\in\mathfrak{X}(\mathcal{P}) denotes the fundamental vector field on \mathcal{P} generated by the Lie algebra element \xi. We will now check what assumption on the asymptotic behavior of the gauge transformation parameter is required for the above condition to hold.
To this end we first calculate the right- and left-hand sides of Eq. (?) separately and then compare them. We begin with the right-hand side, i.e. \omega(\mathbb{X}_\xi,\cdot). By definition, for any function \mathbb{F}\in C^\infty(\mathcal{P}), we have:
For the functions \mathbb{F}=A and E, this simply gives:
Thus if we put \mathbb{X}_\xi in the first slot of the symplectic form \omega=\int_\Sigma\text{Tr }\mathbbm{d}A\wedge\mathbbm{d}E, i.e. the right-hand side of Eq. (?), we get:
The left hand-side of Eq. (?) gives:
However, we cannot immediately see how this agrees with the expression in Eq. (?), because Eq. \eqref{RHS} contains a term linear in D_A\xi, while the above result has a term that is linear in \xi. Thus we need to do the partial integration in Eq. (Proof), which gives:
The second term in the first integral can be rewritten as:
Thus, the first integral in Eq. (?) equals:
But the trace of the full commutator term gives zero,(Or, since the trace is ad-invariant, we could also immediately have rewritten \text{Tr }E\wedge[\mathbbm{d}A,\xi]=-\text{Tr }[E,\xi]\wedge\mathbbm{d}A.) so we obtain precisely the final expression in Eq. (?)! This implies that from requiring that the Gauss constraint \mu(A,E)=D_AE is the momentum map for the action of the gauge group, it follows that the boundary term in Eq. (?) must be zero. Since that boundary term is an integral of E\wedge\xi, the condition that needs to be imposed on \xi to ensure that this term vanishes depends on the asymptotic fall-off behavior of E. If we take E\to 0+\mathcal{O}(r^{-2}), then \xi just needs to go to zero without any particular fall-off rate (since the integral is over a 2-sphere with radius sent to infinity, which grows as r^2). If instead we demand E\to 0+\mathcal{O}(r^{-3/2-\epsilon}), then we must have \xi\to 0+\mathcal{O}(r^{-1/2}) to guarantee that there is no boundary term. \Box
In our framework, the fall-off conditions ensuring existence of the instantaneous Lagrangian do not define a second constraint imposed on top of a larger unconstrained phase space. Rather, they define the function spaces that constitute the phase space \mathcal{P}:=Q_{A_\infty}\times\Omega^2_\infty(\Sigma,\mathfrak{g}) itself where the square-integrability of fields is built into the definitions of Q_{A_\infty} and \Omega^2_\infty(\Sigma,\mathfrak{g}) from the outset. This is necessary in particular to ensure that the canonical symplectic form \omega=\int_\Sigma\text{Tr }\mathbbm{d}A\wedge\mathbbm{d}E is well-defined. The constraint surface \mathcal{C} is then defined within \mathcal{P} solely by the Gauss constraint, i.e.
In this framework, “allowed” gauge transformations are simply those whose action on instantaneous state space as well as phase space is well-defined, ensuring also that the presymplectic form \omega|_\mathcal{C} is inherited in a well-defined way. Thus, in this respect, the situation is not essentially different from a system without boundary conditions. In particular, the constraint surface is known to be coisotropic if 0 is a regular value of the momentum map, as is the case in electromagnetism. In non-Abelian Yang-Mills theory, however, the momentum map is singular and only regular at the irreducible connections, on which the gauge group acts freely with trivial stabilizer [singerGeometryOrbitSpace1981, Mitter:1979un, Kondracki1986Stratification, ARMS199043]. Thus the failure of the constraint surface to be coisotropic at singular points is a general feature of non-Abelian Yang-Mills theory, regardless of whether we are working on a space with or without boundaries. In this sense the coisotropic/singular nature of the constraint surface is an independent question from the one we are aiming to answer in this article, as we are aiming to find the asymptotic fall-off conditions required to yield a well-defined instantaneous classical field theory in the first place.
Gauge parameters \xi which preserve \mathcal{O}(r^{-3/2-\epsilon}) fall-off, i.e. which are such that \partial_i\xi\to 0+\mathcal{O}(r^{-3/2-\epsilon}), must satisfy \xi\to\text{const}+\mathcal{O}(r^{-1/2-\epsilon}) (although the converse does not hold, think e.g. of \sin(r)/r).
Taking \xi to be real-valued for simplicity (but with obvious generalization to cases other than G=U(1)), by the fundamental theorem of calculus:
From this it follows that \lim_{s\to\infty}\xi(s,\theta,\phi)=:c_{\theta,\phi} exists and that |\xi(r,\theta,\phi)-c_{\theta,\phi}|=\mathcal{O}(r^{-1/2-\epsilon}). A priori it could be the case that the constant c_{\theta,\phi} is different in each angular direction, but it is straightforward to show that this cannot be so. At fixed r we can connect any two points x,y by a curve \gamma of length at most \pi r. This factor of r is cancelled by the 1/r that appears in the angular components of the gradient \nabla\xi expressed in spherical coordinates. Integrating over \gamma we find that |\xi(x)-\xi(y)|\leq \pi\tilde{C}r^{-1/2-\epsilon}. This shows that c_{\theta,\phi}=c_{\theta',\phi'}. Thus we find that any gauge parameter \xi whose derivative d\xi is square-integrable satisfies \xi\to\text{const}+\mathcal{O}(r^{-1/2-\epsilon}). \Box
An infinitesimal symmetry \xi\in\text{Lie}(\mathcal{G}^I)=C^\infty_I(\Sigma,\mathfrak{g}) is called localizable if it vanishes on the (asymptotic) boundary of \Sigma and if for any pair of open sets U,V\subset \Sigma with disjoint closures, there exists a \xi'\in\text{Lie}(\mathcal{G}^I) such that
All gauge symmetries except the global ones are localizable.
This is done most easily by working on \hat{\Sigma}. There \text{Lie}(\mathcal{G}^I) consists of all maps \hat{\xi}\colon \hat{\Sigma}\to\mathfrak{g} that are constant on \partial\hat{\Sigma}. We note that \text{Lie}(\mathcal{G}^I)/\mathfrak{g}, with \mathfrak{g} viewed as the constant maps in C^\infty_I(\Sigma,\mathfrak{g}), is isomorphic to the algebra \mathfrak{G}_\infty of all maps \hat{\xi}\colon \hat{\Sigma}\to\mathfrak{g} that vanish on \partial\hat{\Sigma}. We need to show that \mathfrak{G}_\infty=\mathfrak{G}, i.e. that the algebra of infinitesimal localizable symmetry consists precisely of the gauge parameters that vanish on \partial\hat{\Sigma}.
There are two situations to consider: if U,V are the open subsets from the above definition, such that \hat{\xi}\in\text{Lie}(\mathcal{G}^I) must be localized on U relative to V, then either U could lie in the interior of \hat{\Sigma} or contain (part of) the boundary \partial\hat{\Sigma}. In the first case it is obvious that \hat{\xi} can be localized: we just use a \mathfrak{g}-valued bump function \hat{f} that is the identity on U and becomes zero very quickly outside of U, in particular on V. It is then clear that \hat{f}\cdot\hat{\xi} will be the required element of \text{Lie}(\mathcal{G}^I) that agrees with \hat{\xi} on U and is zero on V. In the case in which U contains part of the boundary it is not immediately clear whether \hat{f}\cdot\hat{\xi}\in\text{Lie}(\mathcal{G}^I). But since \hat{\xi} is zero on \partial\hat{\Sigma}, so is the product \hat{f}\cdot\hat{\xi}. This means \hat{f}\cdot\hat{\xi}\in\mathfrak{G}_\infty\subset\text{Lie}(\mathcal{G}^I). We conclude that the algebra of infinitesimal localizable symmetries \mathfrak{G} is indeed \mathfrak{G}_\infty. \Box
The text on this page is from arXiv:2502.16151v4 by Silvester G. A. Borsboom and Hessel B. Posthuma, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.
The record of each item is in analysis/objects/2502/2502.16151.jsonl.