Toward a Quantum Variance Formula on Compact Arithmetic Hyperbolic 3-Manifolds: Coadjoint Orbit Geometry and a Conditional Trace Formula

The paper addresses a question that lies at the intersection of quantum chaos, the arithmetic theory of automorphic forms, and the orbit method for real reductive groups. The motivating problem is the quantum variance problem: for a compact Riemannian manifold Y with ergodic geodesic flow and a sequence of Laplace eigenfunctions φ_j with eigenvalues μ_j tending to infinity, one studies the second-order statistics of the matrix elements ⟨φ_j, Ψ φ_j⟩ for a fixed smooth observable Ψ. Quantum ergodicity, established by Shnirelman, Zelditch, and Colin de Verdière, guarantees that for a density-one subsequence these matrix elements converge to the classical average of Ψ against Liouville measure, and quantum unique ergodicity, when it holds, guarantees that the full sequence converges. Neither statement controls the variance, which is the finer question of the asymptotic size of ∑_j |⟨φ_j, Ψ φ_j⟩|² restricted to a high-energy spectral window. In arithmetic settings, the variance becomes accessible because of Hecke operators, theta correspondences, and L-functions. The foundational results are due to Luo and Sarnak, who proved the quantum variance asymptotic for Hecke–Maass forms on the modular surface SL₂(ℤ)\ℍ², and to Luo–Sarnak–Zhao, who extended it to holomorphic forms and to other arithmetic surfaces. Sarnak and Zhao later gave a representation-theoretic proof via theta correspondence, and Nelson subsequently established the analogue for compact arithmetic hyperbolic surfaces associated with quaternion algebras over number fields. In each of these cases the asymptotic has the shape ∑_π μ_π(Ψ₁) conj(μ_π(Ψ₂)) ∼ c · L(σ, 1/2) · ∫_H ⟨s Ψ₁, Ψ₂⟩ ds, where μ_π are microlocal lifts attached to automorphic representations π, L(σ, 1/2) is a central L-value attached to the automorphic representation generated by Ψ, and the integral runs over a stabilizer H inside an appropriate coadjoint orbit group. The present paper initiates the analogous theory for compact arithmetic hyperbolic 3-manifolds, where the relevant real Lie group is G = PGL₂(ℂ), viewed as a six-dimensional real Lie group with maximal compact subgroup K = PU(2) ≅ SO(3), and the symmetric space is hyperbolic 3-space ℍ³. The arithmetic quotients Y = Γ\ℍ³ arise from maximal orders in quaternion algebras over number fields with exactly one complex archimedean place, with no other complex places and with every real archimedean place being a place where the quaternion algebra is division. The paper is deliberately divided into an unconditional part, which is proved in full, and a conditional part, which isolates the analytic hypotheses required for a genuine quantum variance theorem. The unconditional part determines the geometry of the regular semisimple coadjoint orbits of PGL₂(ℂ) viewed as a real Lie group, and proves the weighted three-dimensional Weyl law for the scalar Laplace spectrum of a compact torsion-free arithmetic hyperbolic 3-manifold. The conditional part formulates a quadratic trace formula for quantum variance in dimension three and isolates seven hypotheses: a normalized archimedean local Rallis identity at the complex place, the existence and localization of microlocal lifts for PGL₂(ℂ) adapted to the representation theory of PU(2) ≅ SO(3), a uniform semiclassical localization theorem for orbit-supported convolution operators, global control of off-diagonal theta contributions, global theta unfolding with diagonal factorization, matching of the orbit integral with the Weyl density, and control of error sums after summation. Under these hypotheses the paper obtains an asymptotic formula of the expected shape, with a main term involving a central L-value L^(S)(σ, 1/2), an integral over the stabilizer ℂˣ, the three-dimensional Weyl density √μ, and a global constant C₀ determined by Tamagawa measures, the local Rallis constant A_ℂ, and the orbit disintegration normalization. The constant C₀ is kept symbolic; its numerical value is not computed. The unconditional orbit geometry is as follows. Let 𝔤 = sl₂(ℂ) viewed as a real Lie algebra, so that dim_ℝ 𝔤 = 6. Choose the standard real basis H = [1 0; 0 −1], E = [0 1; 0 0], F = [0 0; 1 0], together with their imaginary multiples iH, iE, iF. The commutation relations in the real Lie algebra are [H, E] = 2E, [H, F] = −2F, [E, F] = H. The bracket is ℂ-bilinear on 𝔤 = sl₂(ℂ), so the relations extend to the real basis by [iX, Y] = i[X, Y], [X, iY] = i[X, Y], [iX, iY] = −[X, Y], for X, Y ∈ {H, E, F}. For example, [iH, iE] = −2E, not 2iE. This distinction is important for the explicit computation on the negative real slice. We identify the dual 𝔤* with 𝔤 using the real pairing B(X, ξ) = 2 Re tr(Xξ), for X ∈ 𝔤 and ξ ∈ 𝔤*. This pairing is real-valued and non-degenerate. Define p : 𝔤* → ℂ by p(ξ) = det(ξ). Because det is invariant under conjugation, p is invariant under the coadjoint action of G = PGL₂(ℂ). For z ∈ ℂˣ, set O_z = { ξ ∈ 𝔤* : p(ξ) = z }. The fiber over z = 0 consists of nilpotent or zero elements and is singular; the paper focuses on regular semisimple orbits, namely z ≠ 0. Proposition 4.1 states that for every z ∈ ℂˣ, O_z is a single coadjoint orbit. The proof is as follows. Let ξ ∈ O_z. Since ξ is traceless and det(ξ) = z ≠ 0, its characteristic polynomial is t² + z, and its eigenvalues are ±α, where α² = −z. Because z ≠ 0, the eigenvalues ±α are distinct and nonzero, so ξ is diagonalizable over ℂ and conjugate in SL₂(ℂ) to diag(α, −α). Under the identification PGL₂(ℂ) = SL₂(ℂ)/{±1}, this same normal form holds modulo the center. The two possible choices of square root, α and −α, give the diagonal matrices diag(α, −α) and diag(−α, α), which are conjugate in PGL₂(ℂ) by the Weyl element w = [0 1; 1 0]. Hence any two elements of O_z are conjugate in PGL₂(ℂ), and O_z is a single coadjoint orbit. Proposition 4.2 states that the stabilizer of a regular point in O_z is isomorphic to ℂˣ. The proof takes the base point ξ_α = diag(α, −α) with α ≠ 0; an element g ∈ PGL₂(ℂ) stabilizes ξ_α if and only if it commutes with ξ_α. Since ξ_α has distinct eigenvalues, its centralizer in PGL₂(ℂ) is the image of the diagonal torus { diag(a, b) : a, b ∈ ℂˣ } / ℂˣ, which is isomorphic to ℂˣ via the map diag(a, b) ↦ a/b. Corollary 4.3 gives dim_ℝ O_z = dim_ℝ G − dim_ℝ ℂˣ = 6 − 2 = 4 for z ≠ 0. Proposition 4.4 states that the KKS form is non-degenerate on every regular coadjoint orbit. The KKS form is defined by ω_ξ([X, ξ], [Y, ξ]) = B(ξ, [X, Y]) / (2π), for X, Y ∈ 𝔤. This is well-defined because if [X, ξ] = 0, then B(ξ, [X, Y]) = B([ξ, X], Y) = 0 by invariance of the pairing. The proof of non-degeneracy: let v = [X, ξ] be a tangent vector at ξ; suppose that ω_ξ(v, [Y, ξ]) = 0 for all Y ∈ 𝔤. Then B(ξ, [X, Y]) = 0 for all Y. By invariance of the trace pairing, B(ξ, [X, Y]) = B([ξ, X], Y). Since the pairing is non-degenerate, [ξ, X] = 0. Hence v = [X, ξ] = 0. Therefore the kernel of ω_ξ is trivial, so ω_ξ is non-degenerate. The paper then computes the KKS form explicitly at a convenient base point on the negative real slice. Let z = −t² with t > 0, and choose ξ_t = tH = diag(t, −t). Then det(ξ_t) = −t² = z. The tangent space T_{ξ_t} O_z, identified with a subspace of 𝔤 via the pairing B, is spanned over ℝ by v₁ = E, v₂ = iE, v₃ = F, v₄ = iF. These are the images under the coadjoint action of the elements X₁ = −E/(2t), X₂ = −iE/(2t), X₃ = F/(2t), X₄ = iF/(2t), in the sense that [X_j, ξ_t] = λ_j v_j with λ_j ∈ ℝ nonzero. Explicitly, [E, ξ_t] = −2tE, [iE, ξ_t] = −2tiE, [F, ξ_t] = 2tF, [iF, ξ_t] = 2tiF. The KKS pairings are computed as follows. First, [X₁, X₃] = [−E/(2t), F/(2t)] = −H/(4t²). Therefore B(ξ_t, [X₁, X₃]) = B(tH, −H/(4t²)) = 2 Re tr(tH · (−H/(4t²))) = 2 Re(−1/(4t) · tr(H²)) = 2 Re(−1/(4t) · 2) = −1/t. Hence ω(E, F) = (−1/t)/(2π) = −1/(2πt). Second, [X₂, X₄] = [−iE/(2t), iF/(2t)] = H/(4t²). Therefore B(ξ_t, [X₂, X₄]) = B(tH, H/(4t²)) = 1/t, and hence ω(iE, iF) = 1/(2πt). All other pairings among the basis vectors vanish. Thus, in the ordered basis (E, iE, F, iF), the matrix of ω is [0 0 −a 0; 0 0 0 a; a 0 0 0; 0 −a 0 0], where a = 1/(2πt). The Pfaffian of this matrix is Pf(ω) = a² = 1/(4π²t²). Since t² = |z|, we obtain |Pf(ω_{ξ_t})| = 1/(4π²|z|). This formula is specific to the negative real slice z = −t²; it is not a general formula in z. For example, on the positive real slice z = t², the base point would be ξ = itH, and a direct computation gives a different Pfaffian. The restriction to the negative real slice is essential for the formula. Consequently, the Liouville volume form on the orbit, in the chosen tangent coframe (α¹, α², α³, α⁴), is Ω_z = 1/(4π²|z|) α¹ ∧ α² ∧ α³ ∧ α⁴, up to the global orientation sign. The precise conversion from this coordinate expression to a G-invariant measure on O_z ≅ G/ℂˣ depends on the chosen Haar normalizations, which are absorbed into the local constants in the conditional trace formula. The second unconditional result is the weighted three-dimensional Weyl law. Let Y = Γ\ℍ³ be a compact torsion-free hyperbolic 3-manifold, and let μ_j be the eigenvalues of the non-negative Laplace–Beltrami operator on Y, counted with multiplicity. For every q ∈ C_c^∞(0, ∞), the theorem states that lim_{h→0} h³ ∑_j q(h² μ_j) = vol(Y)/(4π²) ∫₀^∞ q(u) √u du. The proof is as follows. Let N(Λ) = #{ j : μ_j ≤ Λ }. The standard Weyl law for the Laplace–Beltrami operator on a compact three-dimensional Riemannian manifold gives N(Λ) ∼ vol(Y)/(6π²) Λ^(3/2) as Λ → ∞; indeed, the general leading constant is (2π)^(−3) vol(B³) vol(Y), where vol(B³) = 4π/3, which gives (1/(8π³)) · (4π/3) · vol(Y) = vol(Y)/(6π²). Fix q ∈ C_c^∞(0, ∞). Put R = h^(−2). Then S_R(q) := ∑_j q(μ_j/R) can be written as a Stieltjes integral S_R(q) = ∫₀^∞ q(x/R) dN(x). Since q is supported away from 0 and has compact support, integration by parts gives S_R(q) = −∫₀^∞ N(Ru) q′(u) du. Using the Weyl asymptotic N(Ru) = vol(Y)/(6π²) (Ru)^(3/2) + o(R^(3/2)), uniformly for u in the compact support of q, we get S_R(q) = −vol(