Ashtekar Variables and Connection Dynamics

Canonical general relativity, as the ADM programme leaves it, is geometrodynamics: the phase space is a Riemannian 3-metric $q_{ab}$ and its conjugate momentum $\pi^{ab}$ , and the dynamics is the deformation of one spatial slice into the next. It is exact and it is beautiful, but it is also stubborn. The Hamiltonian constraint that generates time evolution is a non-polynomial function of $q_{ab}$ — it carries the factor $\sqrt{\det q}$ and the intrinsic scalar curvature $R^{(3)}[q]$ , a second-order differential operator on the metric — and forty years of effort could not turn that expression into a well-defined quantum operator. In 1986 Abhay Ashtekar performed a canonical transformation that changed the character of the problem entirely. His move was to stop describing the slice by its metric and describe it instead by a connection. Package the geometry into a densitized triad $E^a_i$ , the square-root-of-the-metric object that plays the role of a Yang-Mills electric field, and package the extrinsic curvature into an $SU(2)$ connection $A_a^i=\Gamma_a^i+\beta K_a^i$ that plays the role of the vector potential. The two are canonically conjugate with an ultralocal bracket, exactly like the fields of a gauge theory. In these variables the Hamiltonian constraint of general relativity — with Ashtekar's original complex self-dual connection — collapses to a polynomial: the bilinear $\epsilon_{ijk}E^a_iE^b_jF_{ab}^k$ , quadratic in the triad and linear in the curvature. Gravity, on a spatial slice, has been made to look like $SU(2)$ Yang-Mills theory without a background metric — the doorway to loop quantum gravity, spin networks, and the discrete quantum geometry of area and volume that occupies everything downstream of this article.

Key takeaways

Ashtekar variables recast canonical gravity from the ADM pair $(q_{ab},\pi^{ab})$ into a gauge-theory pair — an $SU(2)$ connection $A_a^i$ and a densitized triad $E^a_i$ — with the ultralocal bracket $\{A_a^i(x),E^b_j(y)\}=\kappa\beta\,\delta_a^b\delta^i_j\delta^3(x,y)$ .
The triad carries the geometry, $E^a_iE^b_i=q\,q^{ab}$ , while the connection $A_a^i=\Gamma_a^i+\beta K_a^i$ blends the spin connection $\Gamma$ with the extrinsic curvature $K$ ; the Barbero-Immirzi parameter $\beta$ dials their mixture.
With Ashtekar's self-dual choice $\beta=\pm i$ the factor $1+\beta^2$ vanishes and the Hamiltonian constraint becomes the polynomial $\mathcal{S}\propto\epsilon_{ijk}E^a_iE^b_jF_{ab}^k$ — vastly simpler than the ADM form, at the cost of complex variables and reality conditions.
Barbero's real connection ( $\beta=\gamma\in\mathbb{R}$ ) avoids reality conditions and is the backbone of loop quantum gravity, reintroducing a non-polynomial term $\propto(1+\gamma^2)$ in the scalar constraint.
A genuinely new Gauss constraint $\mathcal{G}_i=D_aE^a_i\approx0$ joins the ADM vector and scalar constraints, generating $SU(2)$ frame rotations; all three are first-class and close on the hypersurface deformation algebra.
Quantizing holonomies of $A$ and fluxes of $E$ yields discrete area and volume spectra, e.g. the smallest area quantum $\approx 2.77\,\ell_P^2\approx7.2\times10^{-70}\ \mathrm{m^2}$ , and fixes $\beta\approx0.127$ from black-hole entropy.
Key idea: Trade the spatial metric and its momentum for a gauge-theoretic pair — an $SU(2)$ connection $A_a^i=\Gamma_a^i+\beta K_a^i$ and a densitized triad $E^a_i$ with $E^a_iE^b_i=q\,q^{ab}$ — so that gravity looks like a background-independent Yang-Mills theory and the constraints become polynomial.

Where this lives today

Ashtekar variables are not a historical curiosity — they are the live substrate of an entire quantum-gravity programme. In loop quantum gravity, the real Ashtekar-Barbero connection is quantized through holonomies $h_e=\mathcal{P}\exp\int_e A$ along edges and fluxes $\int_S E$ through surfaces; the resulting spin-network Hilbert space carries the discrete area and volume operators of equation (6), the setting in which "space is granular at $10^{-70}\ \mathrm{m^2}$ " is a precise statement rather than a slogan. In loop quantum cosmology, the same variables reduced to a homogeneous isotropic model replace the classical Big Bang with a non-singular bounce at Planckian density — holonomy quantization caps the curvature and turns the singularity into a smooth transition from a prior contracting phase, now studied for its imprint on the CMB. The black-hole entropy calculation of Worked Example 1 remains a signature result, deriving $S=A/(4\ell_P^2)$ from microstate counting and fixing $\gamma$ . On the covariant side, spin-foam models (the EPRL amplitude and relatives) implement the scalar constraint as a sum over histories of spin networks. And the original self-dual variables have seen a revival: because they make the constraint truly polynomial and connect to twistor methods, several groups now pursue the reality conditions through analytic continuation in $\gamma$ . The open frontier — recovering smooth classical spacetime from the discrete quantum geometry — is precisely a question about these variables.

The physical picture

Start where aq-gr-adm-hamiltonian-formulation leaves off. A spatial slice $\Sigma$ carries a metric $q_{ab}$ and a momentum $\pi^{ab}$ that encodes how fast the slice is bending inside spacetime, and the whole content of Einstein's equations is repackaged into constraints on that pair, developed in aq-gr-hamiltonian-momentum-constraints. The trouble is the scalar constraint: it contains $\sqrt{q}\,R^{(3)}[q]$ , and $R^{(3)}$ is built from two derivatives of the metric with the metric itself hidden inside inverse-metric factors — a non-polynomial mess that resists quantization. Ashtekar's reframing hides the geometry inside a different object. Introduce an orthonormal triad $e^a_i$ — three vector fields that at each point form a right-handed frame, so that $q_{ab}=e_a^ie_b^i$ — and densitize it, $E^a_i=\sqrt{q}\,e^a_i$ . This $E^a_i$ is the gravitational twin of a Yang-Mills electric field: a vector density carrying an internal $SU(2)$ index. Its conjugate partner is a connection $A_a^i$ , and the metric is now a derived quantity, reconstructed from $E$ through $E^a_iE^b_i=q\,q^{ab}$ . Curvature of space no longer appears as $R^{(3)}$ ; it appears as the field strength $F_{ab}^i$ of a gauge field living on the slice. The first figure sets the two descriptions side by side and shows what the trade buys.

The connection itself is a blend, $A=\Gamma+\beta K$ , of two geometric ingredients. The spin connection $\Gamma_a^i$ is a function of $E$ alone and knows the intrinsic twist of the slice — how the triad rotates as you parallel-transport it, the same object that appears in the Gauss-Codazzi relations of aq-gr-gauss-codazzi-mainardi-equations. The term $\beta K_a^i$ carries the extrinsic curvature, how the slice bends inside spacetime. The Barbero-Immirzi parameter $\beta$ is the dial that sets their relative weight. Classically it is invisible — a rescaling $\beta\to\beta'$ is a canonical transformation that leaves every equation of motion unchanged — but quantum mechanically it survives as a genuine coupling, setting the scale of the area and volume spectra and appearing in black-hole entropy. Why does the trade simplify anything? For the special value $\beta=\pm i$ — the self-dual connection Ashtekar first wrote down — the four-dimensional curvature seen by $A$ is the self-dual (left-handed) half of the Riemann tensor, and the awkward extrinsic-curvature term in the scalar constraint carries a prefactor $1+\beta^2$ that vanishes identically. What remains is polynomial. The price is that $A_a^i$ is then complex, and one must bolt on reality conditions to guarantee the recovered metric is real and Lorentzian — conditions that are hard to impose in the quantum theory, and the reason Barbero's later real connection became the practical choice. The genuinely new structural feature is the Gauss constraint. Choosing a triad instead of a metric introduces three redundant degrees of freedom per point — you can rotate the frame $e^a_i$ at each point without touching $q_{ab}$ — so a new constraint $D_aE^a_i=0$ must appear to kill that redundancy. It is word-for-word the Gauss law of an $SU(2)$ Yang-Mills theory, and it is exactly this gauge structure that makes holonomies of $A$ and fluxes of $E$ the natural quantum variables. Those variables have discrete geometric spectra, the content of the second figure.

Figure

Quantum geometry from the connection: a surface Σ pierced by spin-network edges carrying half-integer spins j, each puncture contributing a quantum of area A = 8π γ ℓ_P² √(j(j+1)). The area operator has a discrete ladder of eigenvalues; the smallest nonzero quantum is 2.77 ℓ_P² ≈ 7.2×10⁻⁷⁰ m² for γ ≈ 0.127

The core equations

The first pair of relations fix what the new variables are: how the geometry is stored in the triad, and how the connection is assembled. Everything downstream — the bracket, the curvature, the constraints — follows from these definitions plus the demand that $A$ and $E$ be canonically conjugate.

E^a_i\,E^b_i \;=\; q\,q^{ab},\qquad E^a_i=\sqrt{q}\,e^a_i,\qquad q=\det(q_{ab}) \qquad (1)

where: $E^a_i$ — densitized triad, a vector density of weight $+1$ carrying internal $SU(2)$ index $i=1,2,3$ ; $e^a_i$ — orthonormal triad (frame field) with $q_{ab}=e_a^ie_b^i$ ; $q^{ab}$ — inverse spatial metric; $q$ — metric determinant. Reading: the metric is recovered from $E$ , not fundamental; the triad is the primary field, so local frame rotations are a gauge symmetry. The configuration variable is a connection built by adding the extrinsic curvature to the spin connection. The mixing coefficient $\beta$ is the one free number in the construction, and its value selects between two inequivalent quantum theories.

A_a^i \;=\; \Gamma_a^i \;+\; \beta\,K_a^i,\qquad K_a^i=K_{ab}\,e^{b\,i} \qquad (2)

where: $A_a^i$ — the $\mathfrak{su}(2)$ -valued connection; $\Gamma_a^i$ — spin connection compatible with the triad, a function of $E$ alone; $K_a^i$ — extrinsic curvature projected on the triad; $\beta$ — Barbero-Immirzi parameter, $\beta=\pm i$ self-dual, $\beta=\gamma\in\mathbb{R}$ real. Reading: $A$ is a dynamical variable, not a geometric object read off the metric — it contains the momentum information $K$ . The pair is canonically conjugate with an ultralocal Poisson bracket — the same structure as the potential and electric field of a gauge theory, with no metric factors on the right-hand side.

\{A_a^i(x),\,E^b_j(y)\} \;=\; \kappa\,\beta\,\delta_a^b\,\delta^i_j\,\delta^{3}(x,y) \qquad (3)

where: $\kappa=8\pi G$ — gravitational coupling; $\beta$ — rescales the bracket; $\delta^3(x,y)$ — Dirac delta on $\Sigma$ . Reading: "ultralocal" means the bracket is a delta function with no derivatives — the hallmark of a gauge-theory phase space and what makes holonomy-flux quantization possible. The curvature of $A$ is the non-abelian field strength, and it is the object the constraints are built from. It is quadratic in $A$ because $SU(2)$ is non-abelian.

F_{ab}^i \;=\; \partial_a A_b^i - \partial_b A_a^i + \epsilon^{ijk}A_a^j A_b^k \qquad (4)

where: $F_{ab}^i$ — field strength 2-form of $A_a^i$ ; $\epsilon^{ijk}$ — $SU(2)$ structure constants; $\partial_a$ — spatial partial derivative. Reading: $F$ is the Yang-Mills "magnetic field" of the gravitational connection; where $R^{(3)}$ stood in ADM, $F$ now stands. The dynamics is three sets of first-class constraints. The Gauss constraint is new relative to ADM; the vector constraint generates spatial diffeomorphisms; the scalar constraint generates time evolution and carries the whole difficulty of the theory.

\mathcal{G}_i = D_aE^a_i \approx 0,\qquad \mathcal{V}_a = F_{ab}^i\,E^b_i \approx 0,\qquad \mathcal{S} = \frac{\epsilon_{ijk}\,E^a_i E^b_j}{\sqrt{\det E}}\Big(F_{ab}^k - (1+\beta^2)\,\epsilon^{k}{}_{lm}K_a^l K_b^m\Big) \approx 0 \qquad (5)

where: $\mathcal{G}_i$ — Gauss constraint, generates internal $SU(2)$ rotations; $\mathcal{V}_a$ — vector (diffeomorphism) constraint, up to a Gauss piece; $\mathcal{S}$ — scalar (Hamiltonian) constraint; $\det E=q$ ; $(1+\beta^2)$ — the coefficient that vanishes at $\beta=\pm i$ ; $\approx$ — weak equality, holds on the constraint surface. Reading: the entire content of Einstein's equations is "all three vanish"; at $\beta=\pm i$ the last term drops, and the densitized constraint $\sqrt{\det E}\,\mathcal{S}=\epsilon_{ijk}E^a_iE^b_jF_{ab}^k$ — equivalently $\mathcal{S}$ smeared with a densitized lapse $\tilde N=N/\sqrt{\det E}$ — is then polynomial in $(A,E)$ , which is the sense in which the self-dual theory is polynomial. Quantizing the holonomy-flux algebra turns the classical geometry into operators with discrete spectra. The area of a surface pierced by spin-network edges is the cleanest example.

\hat A(S)\,|s\rangle \;=\; 8\pi\gamma\,\ell_P^2\sum_{p\in S}\sqrt{j_p(j_p+1)}\;|s\rangle,\qquad \ell_P^2=\frac{\hbar G}{c^3} \qquad (6)

where: $\hat A(S)$ — area operator for surface $S$ ; $j_p$ — half-integer spin on the edge piercing $S$ at puncture $p$ ; $\gamma$ — real Barbero-Immirzi parameter; $\ell_P$ — Planck length, $\ell_P\approx1.616\times10^{-35}\ \mathrm{m}$ . Reading: area is quantized in grains proportional to $\gamma\ell_P^2$ ; space has a smallest nonzero area, and $\gamma$ sets its size. Finally, matching that quantum geometry to the Bekenstein-Hawking entropy of a horizon fixes the otherwise-free parameter to a definite number.

\gamma \;=\; \frac{\ln(2j+1)}{2\pi\sqrt{j(j+1)}}\;\xrightarrow{\ j=1/2\ }\;\frac{\ln 2}{\pi\sqrt{3}}\;\approx\;0.127 \qquad (7)

where: $\gamma$ — value fixed by horizon microstate counting; $j$ — dominant puncture spin, minimized at $j=1/2$ ; $\ln(2j+1)$ — logarithm of the number of states per puncture. Reading: a classically invisible parameter is pinned to a pure number by demanding $S=A/(4\ell_P^2)$ — but this is the simplified all-punctures-same- $j$ count; the refined Domagala-Lewandowski and Meissner counting shifts it to $\gamma\approx0.2375$ , and the later $SU(2)$ Chern-Simons treatment of the isolated horizon to $\gamma\approx0.274$ .

Derivation ACanonical conjugacy of the Ashtekar pair

The goal is to show, starting from the ADM symplectic structure that aq-gr-adm-hamiltonian-formulation establishes, that the connection $A_a^i=\Gamma_a^i+\beta K_a^i$ and the densitized triad $E^a_i$ are genuinely canonically conjugate — that their Poisson bracket is the ultralocal delta of equation (3) — and to make explicit why one is free to add the spin connection $\Gamma$ without spoiling that conjugacy. The whole reformulation rests on this being true, so we track the symplectic potential term by term from $\pi^{ab}\delta q_{ab}$ to $E^a_i\delta A_a^i$ .

Step 1. Write the ADM symplectic potential $\Theta=\frac{1}{2\kappa}\int_\Sigma \pi^{ab}\,\delta q_{ab}\,\mathrm{d}^3x$ , with $\pi^{ab}=\sqrt{q}\,(K^{ab}-q^{ab}K)$ the momentum conjugate to the metric. Why: the symplectic potential is where the Poisson brackets live — reading off conjugate pairs is reading off which $\delta$ -of-what multiplies which field in $\Theta$ .

Step 2. Trade the metric for the triad. Because $q_{ab}=e_a^ie_b^i$ , a variation of the metric is $\delta q_{ab}=2e_{(a}^i\,\delta e_{b)}^i$ , one finds $\pi^{ab}\delta q_{ab}=-2K_a^i\,\delta E^a_i$ up to a total derivative, where $K_a^i\equiv K_{ab}e^{b\,i}$ is the (undensitized) extrinsic-curvature triad and $E^a_i=\sqrt q\,e^a_i$ carries the density weight. Why: this rewrites the geometry so that the conjugate of the triad $E$ is (up to $-2$ ) the extrinsic curvature — the momentum now lives in a triad-indexed object, ready to be promoted to a connection.

Step 3. Isolate the redundancy. The step above is exact only modulo a piece proportional to $\partial_aE^a_i+\epsilon_{ijk}\Gamma_a^jE^a_k$ , which is the covariant divergence of $E$ — the future Gauss constraint. On gauge-invariant (rotation-invariant) variations this term vanishes, so it does not affect the symplectic form on the physical phase space. Why: the extra frame-rotation freedom of the triad shows up here as a Gauss term; it is precisely the redundancy the new Gauss constraint will remove.

Step 4. Shift the momentum by the spin connection. Since $\Gamma_a^i$ is a functional of $E$ alone, $\int_\Sigma \Gamma_a^i\,\delta E^a_i\,\mathrm{d}^3x$ is (after integration by parts) an exact variation, so adding it changes $\Theta$ only by a boundary term and leaves all brackets untouched; the extrinsic-curvature variable is thereby promoted to $A_a^i=\Gamma_a^i+\beta K_a^i$ . Why: this is the crux — you can absorb $\Gamma$ into the momentum "for free" because it carries no independent symplectic content, converting the bare extrinsic curvature into a true gauge connection whose curvature enters the constraints polynomially.

Step 5. Read off the bracket. Collecting Steps 1-4 gives $\Theta=-\frac{1}{\kappa\beta}\int_\Sigma A_a^i\,\delta E^a_i\,\mathrm{d}^3x$ , which is $\Theta=\frac{1}{\kappa\beta}\int_\Sigma E^a_i\,\delta A_a^i\,\mathrm{d}^3x$ up to the exact variation $\delta\!\int_\Sigma E^a_iA_a^i$ , and the coefficient $1/(\kappa\beta)$ inverts to give $\{A_a^i(x),E^b_j(y)\}=\kappa\beta\,\delta_a^b\delta_j^i\delta^3(x,y)$ . Why: the form $\int(\text{momentum})\,\delta(\text{configuration})$ is the definition of a conjugate pair; $E$ is the momentum, $A$ the configuration variable, exactly as for a Yang-Mills potential and its electric field.

Go deeper · Derivation B · 2 minWhy the self-dual choice makes the scalar constraint polynomial

Derivation BWhy the self-dual choice makes the scalar constraint polynomial

The aim is to demonstrate concretely that Ashtekar's value $\beta=\pm i$ collapses the Hamiltonian constraint of general relativity into the polynomial bilinear $\epsilon_{ijk}E^a_iE^b_jF_{ab}^k$ , to identify exactly which term dies and why, and to be honest about the compensating cost — the reality conditions that make the complex formulation awkward to quantize. This is the single result that motivated the entire connection-dynamics programme, so it is worth seeing the cancellation happen explicitly.

Step 1. Start from the general- $\beta$ scalar constraint of equation (5), $\mathcal{S}=\frac{\epsilon_{ijk}E^a_iE^b_j}{\sqrt{\det E}}\big(F_{ab}^k-(1+\beta^2)\,\epsilon^{k}{}_{lm}K_a^lK_b^m\big)$ . The first term is polynomial in $(A,E)$ ; the second, built from two powers of the extrinsic curvature, is the non-polynomial nuisance inherited from the ADM $\sqrt{q}R^{(3)}$ . Why: separating the constraint into a "curvature" piece and a " $K\!K$ " piece isolates precisely the part we want to eliminate.

Step 2. Insert the self-dual value. With $\beta=i$ the prefactor of the second term is $1+\beta^2=1+i^2=1-1=0$ , so the entire extrinsic-curvature contribution vanishes identically. Why: this is a genuine algebraic cancellation, not an approximation — it holds pointwise for every field configuration, which is why the simplification is exact rather than perturbative.

Step 3. Collect what survives: $\mathcal{S}=\frac{1}{\sqrt{\det E}}\,\epsilon_{ijk}E^a_iE^b_jF_{ab}^k$ , quadratic in the triad and linear in the connection's curvature — a polynomial (up to the overall density weight $1/\sqrt{\det E}$ ). Why: a polynomial constraint is the object one can hope to promote to a well-defined quantum operator; the ADM form never offered this.

Step 4. Interpret geometrically. For the self-dual connection $A^+=\Gamma+iK$ , the field strength $F[A^+]$ is the self-dual projection of the four-dimensional Riemann tensor, so $\mathcal{S}=0$ is a repackaging of the Gauss-Codazzi relation of aq-gr-gauss-codazzi-mainardi-equations in self-dual language. Why: the polynomial form is not a trick — it reflects the algebraic fact that the self-dual half of curvature carries all of Einstein's equations on a slice.

Step 5. Pay the price. Because $A^+$ is complex, the recovered metric is only guaranteed real and Lorentzian if one imposes the reality conditions $A_a^i+\overline{A_a^i}=2\Gamma_a^i[E]$ and a condition on $\dot E$ . These are straightforward classically but notoriously hard to implement as operator statements, which is why Barbero's real $\gamma\in\mathbb{R}$ connection — reinstating the $K\!K$ term but keeping everything real — became the version used in loop quantum gravity. Why: honesty about the trade-off explains the historical fork: self-dual buys polynomiality but owes reality conditions; real Barbero pays with a harder constraint but no reality conditions.

Go deeper · Derivation C · 2 minClosure of the constraint algebra and the count of physical degrees of freedom

Derivation CClosure of the constraint algebra and the count of physical degrees of freedom

The goal is to verify that the Gauss, vector, and scalar constraints are first-class — that their mutual Poisson brackets close on the constraints themselves, so the theory is consistent and the constraints generate gauge — and then to use that closure to count the true physical degrees of freedom, confirming the expected two per point of the graviton. Closure is what guarantees that time evolution preserves the constraint surface; the counting is the sanity check that no physics was lost or gained in the reformulation.

Step 1. Smear each constraint with a test field: $\mathcal{G}[\Lambda]=\int\Lambda^i\mathcal{G}_i$ , $\mathcal{V}[\vec N]=\int N^a\mathcal{V}_a$ , $\mathcal{S}[N]=\int N\mathcal{S}$ . The Gauss constraints close into an $\mathfrak{su}(2)$ algebra, $\{\mathcal{G}[\Lambda],\mathcal{G}[\Lambda']\}=\mathcal{G}[[\Lambda,\Lambda']]$ . Why: this is literally the Yang-Mills gauge algebra — the internal-rotation sector of gravity in triad variables behaves exactly like an $SU(2)$ gauge theory.

Step 2. Check the diffeomorphism sector. Gauss transforms covariantly under diffeos, $\{\mathcal{V}[\vec N],\mathcal{G}[\Lambda]\}=\mathcal{G}[\mathcal{L}_{\vec N}\Lambda]$ , and the vector constraints close among themselves, $\{\mathcal{V}[\vec N],\mathcal{V}[\vec M]\}=\mathcal{V}[[\vec N,\vec M]]$ , the Lie algebra of vector fields on $\Sigma$ . Why: spatial diffeomorphisms must form a subalgebra for the theory to be generally covariant on the slice, and they do.

Step 3. Compute the hard bracket, scalar with scalar: $\{\mathcal{S}[N],\mathcal{S}[M]\}=\mathcal{V}[\,q^{ab}(N\partial_bM-M\partial_bN)\,]$ , where the smearing carries the inverse metric $q^{ab}=E^a_iE^b_i/\det E$ . Why: the right-hand side is a diffeomorphism, so the bracket closes — but the smearing is field-dependent, a structure function rather than a constant, and that is the crux of the whole quantization difficulty.

Step 4. Name the algebra. Steps 1-3 reproduce the hypersurface deformation (Dirac) algebra of ADM with an $\mathfrak{su}(2)$ factor adjoined; all brackets close on constraints, so the system is first-class with no secondary constraints beyond these. Why: first-class means every constraint generates a gauge transformation and none is a genuine equation of motion in disguise — the reformulation has not introduced spurious second-class constraints.

Step 5. Count. The phase space has $18$ real dimensions per point ( $A_a^i$ and $E^a_i$ each have $3\times3=9$ components). There are $3+3+1=7$ first-class constraints; each first-class constraint removes two phase-space dimensions (one for the constraint, one for the gauge orbit it generates). So the physical phase space is $18-2\times7=4$ dimensions per point, i.e. $2$ configuration degrees of freedom — the two polarizations of the graviton. Why: recovering exactly two matches ADM and linearized gravity, confirming the connection variables describe the same physics with no lost or added modes.

Limiting cases

The character of the theory is controlled almost entirely by the value of $\beta$ , with the linearized and metric limits recovering familiar physics.

Case	Result	Meaning
Self-dual $\beta=i$	$\mathcal{S}\propto\epsilon_{ijk}E^a_iE^b_jF_{ab}^k$	Ashtekar's original choice: the scalar constraint is strictly polynomial and $A^+=\Gamma+iK$ is the self-dual connection, but $A_a^i$ is complex and reality conditions must be imposed to recover Lorentzian gravity.
Real Barbero $\beta=\gamma\in\mathbb{R}$	$\mathcal{S}\propto\epsilon_{ijk}E^a_iE^b_jF_{ab}^k-2(1+\gamma^2)E^a_{[i}E^b_{j]}K_a^iK_b^j$	Real $SU(2)$ connection, no reality conditions; the extra $(1+\gamma^2)$ term makes the constraint non-polynomial, but this is the version underlying loop quantum gravity.
Anti-/self-dual pair	$A_a^{i\,\pm}=\Gamma_a^i\pm iK_a^i$	Only the $\pm i$ combinations make the four-dimensional curvature (anti-)self-dual; this algebraic fact is the origin of the polynomial simplification.
Weak-field / linearized	$A_a^i\approx\Gamma_a^i,\quad E^a_i\approx\delta^a_i+\tfrac12 h^a{}_i$	For nearly flat slices the connection reduces to the spin connection of small metric perturbations $h_{ab}$ , recovering linearized gravity in triad form with two graviton polarizations.
Classical observables	$\gamma$ drops out	Classically $\gamma$ is pure gauge (a canonical rescaling), so no classical measurement sees it; only the quantum area/volume spectra and black-hole entropy depend on it.

Common trapConfusing the self-dual and real connections, or mistaking A for the Levi-Civita spin connection

Two traps, one paragraph. First, the self-dual ( $\beta=i$ ) and real ( $\beta=\gamma\in\mathbb{R}$ ) formulations are not interchangeable: the self-dual connection makes the Hamiltonian constraint genuinely polynomial but forces a complex phase space plus reality conditions, whereas the real connection keeps everything real at the price of the non-polynomial $(1+\gamma^2)$ term. Loop quantum gravity uses the real version; Ashtekar's original 1986 papers use the self-dual one. Second, $A_a^i$ is not simply the metric-compatible spin connection $\Gamma_a^i$ . It contains the extrinsic-curvature piece $\beta K_a^i$ , so it is a genuine dynamical configuration variable carrying momentum information — treating it as a derived geometric object (the way $\Gamma$ is derived from $E$ ) throws away exactly the data that makes it half of a canonical pair.

Worked examples

1 · Fixing the Barbero-Immirzi parameter from black-hole entropy (fully worked)

A horizon of area $A$ is punctured by $N$ spin-network edges, all carrying the same spin $j$ . Each puncture contributes area $8\pi\gamma\ell_P^2\sqrt{j(j+1)}$ and carries $2j+1$ internal states. Demand that the microstate count reproduce the Bekenstein-Hawking entropy $S=A/(4\ell_P^2)$ and find the value of $\gamma$ this forces.

Step 1. Write the area in terms of the punctures: $A=N\cdot 8\pi\gamma\ell_P^2\sqrt{j(j+1)}$ . The number of microstates is $2j+1$ per puncture, independent across punctures, so the statistical entropy is $S=\ln\big[(2j+1)^N\big]=N\ln(2j+1)$ .

Step 2. Impose the Bekenstein-Hawking value $S=A/(4\ell_P^2)$ . Substitute both expressions:

N\ln(2j+1)=\frac{N\cdot 8\pi\gamma\ell_P^2\sqrt{j(j+1)}}{4\ell_P^2}=2\pi\gamma N\sqrt{j(j+1)}.

The puncture count $N$ and the $\ell_P^2$ cancel, leaving a pure-number condition.

Step 3. Solve for $\gamma$ : $\gamma=\dfrac{\ln(2j+1)}{2\pi\sqrt{j(j+1)}}$ . This must hold for the dominant spin, and one checks that the right-hand side is largest at $j=1/2$ (each additional unit of area is cheapest in entropy for the smallest spin), so $j=1/2$ punctures dominate the counting.

Step 4. Evaluate at $j=1/2$ : $2j+1=2$ and $\sqrt{j(j+1)}=\sqrt{\tfrac12\cdot\tfrac32}=\sqrt{3}/2$ , hence

\gamma=\frac{\ln 2}{2\pi\cdot(\sqrt3/2)}=\frac{\ln 2}{\pi\sqrt3}=\frac{0.6931}{5.441}=\mathbf{0.1274}.

A classically invisible parameter is pinned to $\gamma\approx0.127$ purely by matching quantum-geometry state-counting to the Bekenstein-Hawking area law. (This is a simplified toy counting. The refined counting of Domagala-Lewandowski and Meissner, which keeps the full multiplicity of puncture configurations rather than only $2j+1$ per puncture, shifts this to $\gamma\approx0.2375$ ; the later $SU(2)$ Chern-Simons treatment of the isolated horizon gives instead $\gamma\approx0.274$ . The mechanism is identical in each case, so the $2.77\,\ell_P^2$ area gap computed below is tied to the $\gamma\approx0.127$ toy value, not a universal number.)

2 · The smallest quantum of area (fully worked)

Using the area spectrum $\hat A=8\pi\gamma\ell_P^2\sqrt{j(j+1)}$ with $\gamma=0.127$ and $\ell_P=1.616\times10^{-35}\ \mathrm{m}$ , find the smallest nonzero area a surface can have.

Step 1. The smallest nonzero eigenvalue comes from a single puncture of the smallest spin, $j=1/2$ , giving $\sqrt{j(j+1)}=\sqrt{3}/2=0.8660$ .

Step 2. Assemble the numerical coefficient: $8\pi\gamma\sqrt{j(j+1)}=8\pi(0.127)(0.8660)=(3.192)(0.8660)=\mathbf{2.77}$ , so the area gap is $\Delta A=2.77\,\ell_P^2$ .

Step 3. Convert to SI. With $\ell_P^2=(1.616\times10^{-35})^2=2.612\times10^{-70}\ \mathrm{m^2}$ ,

\Delta A=2.77\times2.612\times10^{-70}=\mathbf{7.2\times10^{-70}\ m^2}.

Space has a smallest nonzero area of order $10^{-69}\ \mathrm{m^2}$ — a definite prediction of the connection quantization, some seventy orders of magnitude below anything a laboratory can resolve, which is why the discreteness is invisible at accessible scales yet sharp in principle.

Physics of the portrait

The connection variables are the point where canonical gravity stopped being an exercise in metric wrangling and became a gauge theory. One person made that turn.

Origins · 1 minAbhay Ashtekar and how this was discovered

Abhay Ashtekar

b. 1949, Indian-American theoretical physicist · canonical gravity and loop quantum gravity

Abhay Ashtekar was working within John Wheeler and Bryce DeWitt's tradition of canonical quantum gravity — the geometrodynamics of the ADM phase space — when, in 1986, he found the variables that now bear his name. The ADM Hamiltonian constraint had resisted quantization for two decades precisely because of its non-polynomial dependence on the 3-metric; Ashtekar's insight was that by trading the metric for a densitized triad and packaging the extrinsic curvature into a self-dual $SU(2)$ connection $A_a^i=\Gamma_a^i+iK_a^i$ , the constraint becomes a low-order polynomial. The reformulation revealed gravity, on a spatial slice, as a background-independent Yang-Mills-like theory — a structural unification that had been invisible in metric language. Within two years Carlo Rovelli and Lee Smolin built the loop representation on Ashtekar's phase space, and by the mid-1990s the programme had produced discrete spectra for area and volume — a concrete, calculable quantum geometry. The complex self-dual connection carried a cost, the reality conditions, which J. Fernando Barbero addressed in 1995 with a real $SU(2)$ connection at the price of a messier constraint, and Giorgio Immirzi highlighted the free parameter that now labels the two of them. Ashtekar went on to co-develop loop quantum cosmology, where the same variables replace the Big Bang singularity with a bounce. His variables remain the kinematical foundation of the entire loop and spin-foam approach to quantum gravity.

History

1962

— Arnowitt, Deser and Misner cast general relativity as geometrodynamics on the metric phase space

1986

— Ashtekar introduces the self-dual connection variables and the polynomial Hamiltonian constraint

1988

— Rovelli and Smolin construct the loop representation on Ashtekar's phase space

1994

— Rovelli and Smolin derive discrete area and volume spectra from the connection quantization

1995

— Barbero constructs the real SU(2) connection for Lorentzian signature, avoiding reality conditions

1997

— Immirzi isolates the free parameter now bearing his name

Going further

References. Ashtekar, A. (1986). New variables for classical and quantum gravity. Phys. Rev. Lett. 57, 2244. · Ashtekar, A. (1987). New Hamiltonian formulation of general relativity. Phys. Rev. D 36, 1587. · Barbero G., J.F. (1995). Real Ashtekar variables for Lorentzian signature space-times. Phys. Rev. D 51, 5507. · Immirzi, G. (1997). Real and complex connections for canonical gravity. Class. Quantum Grav. 14, L177. · Thiemann, T. (2007). Modern Canonical Quantum General Relativity. Cambridge University Press. · Rovelli, C. (2004). Quantum Gravity. Cambridge University Press. · Ashtekar, A. & Lewandowski, J. (2004). Background independent quantum gravity: a status report. Class. Quantum Grav. 21, R53. · Wald, R.M. (1984). General Relativity, Appendix E. University of Chicago Press. Key numerical anchors used throughout: $\gamma=\ln2/(\pi\sqrt3)\approx0.127$ , the area gap $2.77\,\ell_P^2\approx7.2\times10^{-70}\ \mathrm{m^2}$ , and $S=A/(4\ell_P^2)$ — all standard results of loop quantum gravity. The degrees-of-freedom count is $18-2\times7=4$ , and the $(1+\beta^2)$ term vanishes identically at $\beta=i$ .

Prerequisites. ADM 3+1 Decomposition of Spacetime · ADM Hamiltonian Formulation of General Relativity · Hamiltonian and Momentum Constraints · Gauss-Codazzi-Mainardi Equations. Leads to. Loop Quantum Gravity · Initial-Value (Cauchy) Problem in General Relativity · Conformal Method for Constraint Solving.

Ask the tutor: Ask me to walk through why $1+\beta^2=0$ collapses the Hamiltonian constraint, or how holonomies and fluxes turn the Ashtekar connection into the discrete area spectrum of loop quantum gravity.

In one sentence, what did Ashtekar actually change?

He performed a canonical transformation on the ADM phase space, replacing the 3-metric $q_{ab}$ and its momentum $\pi^{ab}$ with a densitized triad $E^a_i$ (which stores the geometry) and an $SU(2)$ connection $A_a^i=\Gamma_a^i+\beta K_a^i$ (which stores the extrinsic curvature). The point is not new physics — the two descriptions are classically equivalent — but new form: in the connection variables the Hamiltonian constraint of general relativity becomes polynomial (for the self-dual choice), so gravity on a slice reads like a background-independent Yang-Mills theory. That change of form is what made a serious attack on quantization possible, and it launched loop quantum gravity.

What is the Barbero-Immirzi parameter, and why does it matter if it is classically invisible?

It is the coefficient $\beta$ mixing the spin connection and the extrinsic curvature in $A=\Gamma+\beta K$ . Classically it is pure gauge: any two values are related by a canonical transformation, so no classical experiment can detect it — the equations of motion, the geodesics, the light bending are all independent of $\beta$ . Quantum mechanically it refuses to disappear. It multiplies the area and volume eigenvalues (equation 6), so it sets the scale of the discrete quantum geometry, and matching black-hole microstate counting to the Bekenstein-Hawking entropy pins it to a definite number, $\gamma\approx0.127$ in the simplest counting (or $\approx0.2375$ in the fuller $SU(2)$ version). A parameter that is invisible classically but physical quantum-mechanically is a striking feature — it is a place where quantization genuinely adds content.

Why does using a triad introduce a Gauss constraint that the metric formulation lacks?

Because a triad carries more information than a metric. The metric $q_{ab}$ has six independent components at each point; an orthonormal triad $e^a_i$ has nine. The extra three are the freedom to rotate the frame at each point — an internal $SU(2)$ rotation — which changes the triad but leaves $q_{ab}=e_a^ie_b^i$ untouched. That redundancy must be removed, or the triad theory would have three spurious degrees of freedom per point that the metric theory does not. The Gauss constraint $D_aE^a_i\approx0$ is exactly the generator that quotients out those frame rotations, and it is word-for-word the Gauss law of an $SU(2)$ gauge theory — which is the deep reason the connection formulation is a gauge theory at all.

What are the reality conditions, and why are they such a headache?

In the self-dual formulation the connection $A^+=\Gamma+iK$ is genuinely complex, so the phase space has been complexified. Real Lorentzian general relativity is only recovered on the subspace where the reality conditions hold: $A_a^i+\overline{A_a^i}=2\Gamma_a^i[E]$ (the connection's real part is the spin connection of the triad) together with a condition ensuring $\dot E$ is real. Classically these are easy — you just restrict to the real slice. Quantum mechanically they are hard, because they must be promoted to operator conditions or adjointness requirements on the Hilbert space, and it is not obvious how to do that consistently while keeping the polynomial constraint manageable. This difficulty is the main reason the field switched to Barbero's real connection for loop quantum gravity, accepting a non-polynomial constraint in exchange for a real phase space with no reality conditions.

If the metric is derived from the triad, is the triad more fundamental than the geometry?

Within this formulation, operationally yes. The densitized triad $E^a_i$ is a canonical variable — half of the fundamental Poisson-bracket pair — while the metric $q_{ab}$ is a composite, reconstructed through $E^a_iE^b_i=q\,q^{ab}$ ; the quantum theory quantizes $E$ via fluxes, not $q$ directly. But "fundamental" is a statement about the choice of variables, not about nature: the metric and triad descriptions are classically equivalent, and the triad only looks more basic because it is the object with the clean gauge-theoretic bracket. The payoff of treating $E$ as primary is that the frame-rotation gauge symmetry becomes explicit and holonomy-flux quantization becomes available.

How does discreteness of area follow from this, and is it a prediction or an input?

It is a derived result, not an assumption. One quantizes the flux $\int_S E$ of the triad through a surface, and the area operator is built from these fluxes. Acting on a spin-network state — a graph whose edges carry $SU(2)$ spins $j$ — the operator is diagonal, and its eigenvalues are the sum $8\pi\gamma\ell_P^2\sum_p\sqrt{j_p(j_p+1)}$ over the edges piercing the surface. Because the spins are half-integers, the spectrum is discrete, with a smallest nonzero gap of about $2.77\,\ell_P^2$ . Nothing discrete was put in by hand: the discreteness comes from the compactness of $SU(2)$ (whose representations are labelled by discrete spins), inherited directly from the gauge structure of the Ashtekar connection. That is why the connection reformulation, and not some separate postulate, is credited with predicting granular space.