What is the spectrum of the spherical harmonic oscillator?

Why the spherical oscillator does not factor

A d-dimensional quantum harmonic oscillator on flat space is characterized as the unique SO(d)-invariant self-adjoint second-order differential operator whose principal symbol is the Euclidean metric, whose ground state is Gaussian, and whose ground-state energy vanishes. Because it splits as a sum of independent one-dimensional oscillators, it can be solved with annihilation and creation operators. The paper asks what happens when the flat metric is replaced by the round metric of the d-sphere, written in stereographic coordinates, while keeping a quadratic potential ω²r².

The resulting operator is L_ω = Δ_S^d + ω²r², where Δ_S^d is the nonnegative Laplace operator on the sphere expressed in stereographic coordinates. The authors note that this operator is essentially self-adjoint on compactly supported smooth functions on R^d, and that its spectrum is routine to find. The important structural point is negative: the operator does not factor, and annihilation and creation operators are not available. That is why the paper has to compute the spectrum directly rather than borrow the flat-space algebraic solution.

The quadratic potential is singular from the point of view of the original d-sphere and breaks the O(d) symmetry down to O(d−1). The domain of the operator also depends sensitively on ω: the constant function 1 lies in the domain when ω = 0 but not when ω > 0. These are not technical asides; they are the reason the authors expect quantities to depend discontinuously on the dimension parameter d.

The two-dimensional ground state is Cauchy type

For a linear oscillator the ground state is a Gaussian. The paper asks whether the Cauchy-type density (1+r²)^(-ω), taken with respect to the sphere volume form in stereographic coordinates, can be the ground state of a natural geometric operator. The answer is yes exactly when d = 2: there is a unique SO(d)-invariant self-adjoint second-order differential operator whose principal symbol comes from the round metric on the d-sphere, whose ground state is (1+r²)^(-ω), and whose ground-state energy vanishes. The operator is L_ω = Δ_S^d + ω²r² − ω.

In the general dimension d, the ground state of L_ω for ω > 0 is (1+r²)^(-η) with η = (√((d−2)²+16ω²) − (d−2))/4, and the ground-state energy is d(√((d−2)²+16ω²) − (d−2))/8, choosing the positive square root. The proof substitutes δ(r) = (1+r²)^(-η) into L_ω and finds that δ is an eigenfunction precisely when 2η² + (d−2)η − 2ω² = 0, with the sign of the root fixed by requiring η ≥ 0, η vanishing at ω = 0, and δ lying in the domain.

The special role of d = 2 is visible in the same formula: only there does the exponent reduce to ω itself, which is what makes the Cauchy-type density the natural ground state and connects the construction to the two-dimensional setting discussed later. The authors describe d = 2 as critical, and they support that label with the observation that the radial eigenfunction equation is regular at r = 0 when d = 1 and singular otherwise.

The two-dimensional spectrum in closed form

For d = 2 and ω ≥ 0, the paper proves that the ground state of L_ω is (1+r²)^(-ω) with eigenvalue ω, and that the spectrum consists of λ(m,n) = (m + ½(n+1+√(n²+4ω²)))² − ω² − ¼ for integers m, n ≥ 0. The multiplicity is 1 when n = 0 and 2 otherwise, corresponding to the cos(nθ) and sin(nθ) angular dependence. The eigenfunctions have the form r^n (1+r²)^((√(4ω²+4λ+1)−1)/2) F(−m−√(4ω²+n²), −m, n+1, −r²) f(θ), where F is the hypergeometric function, an even polynomial of degree 2m in r in this case, and f(θ) is a linear combination of cos(nθ) and sin(nθ).

The derivation separates variables in the eigenvalue equation, reduces the radial part to a hypergeometric equation with regular singular points, and requires the radial solution to be square integrable with respect to (1+r²)^(-2) r dr on (0,∞). A necessary condition is that one of the hypergeometric parameters be a non-positive integer. The authors check that the branch with b = −m gives square-integrable eigenfunctions and that the alternative branch with a = −m and b > 0 forces n = 0 and then contradicts b > 0, so it contributes no eigenvalues. A separate argument using ellipticity and the angular expansion shows that separation of variables yields a complete set of eigenfunctions.

The ω = 0 limit is a useful check rather than a separate theory: the spectrum becomes N(N+1) with multiplicity 2N+1 for N = 0, 1, …, which is the standard spectrum of the positive Laplacian on the 2-sphere. The paper also warns that the λ(m,n) values are not necessarily distinct even when ω ≠ 0. For example, when ω = 2√3 one has λ(3,1) = 44 = λ(1,4), and when ω = 4 one has λ(9,6) = λ(1,15). The multiplicity count in the theorem is still correct because it counts eigenfunctions, not distinct numerical values, but the authors state that they have not found a robust deterministic condition ruling out degeneracy.

Dimension enters through the angular quantum number

For general d ≥ 2 and ω > 0, the ground state is (1+r²)^(-|√((d−2)²+16ω²) − (d−2)|/4) with ground-state energy d|√((d−2)²+16ω²) − (d−2)|/8. The excited spectrum is λ(m,n) = (m + ¼(√((2n+d−2)²+16ω²) + (d−2+2(n+1))))² − ω² − ¼(d−1)², again for m, n ≥ 0. The multiplicity is 1 when n = 0 and, when n > 0, equals the binomial difference C(N+d−1, N) − C(N+d−3, N−3) with N = n(n+d−2).

The structure of this formula explains where the dimension actually enters. The angular part of the eigenfunction is an eigenfunction of Δ_S^(d−1) with eigenvalue n(n+d−2), so the integer n is not merely a label for the angular frequency as in the two-dimensional case; it indexes the spherical harmonic level on the (d−1)-sphere. The radial equation then has indicial roots n and −(n+d−2) at r = 0, and regularity forces the solution to behave like r^n there. The hypergeometric parameters carry both n and d, and the square-integrability condition is checked to hold automatically for the branch that produces λ(m,n).

The case d = 1 is genuinely different and the paper treats it separately. There r is not restricted to be positive, the radial equation has no singularity at r = 0, and the solutions are written in terms of Legendre P and Q functions. The spectrum is λ_m = (m + ¼(1 − |√(1+16ω²)|))² − ω² for m ≥ 0, and for ω > 0 it is multiplicity free. The authors rule out the Legendre P branch by showing it is not square integrable except when m = 0. This is a concrete instance of the paper's broader point that quantities depend discontinuously on the dimension parameter.

Shifting by the ground-state energy

The paper also records the spectrum of L_ω with its ground-state energy subtracted, so that the lowest eigenvalue is zero. In that normalization the ground state is (1+r²)^(-(d−2+|√((d−2)²+16ω²)|)/4). For d ≥ 2 the spectrum becomes λ(m,n) = (m + ¼(|√((2n+d−2)²+16ω²)| + (d−2+2(n+1))))² − (¼(|√((d−2)²+16ω²)| + d))², with the same multiplicity rule as before. For d = 1 it becomes λ_m = (m + ¼(1 − |√(1+16ω²)|))² − (¼(1 − |√(1+16ω²)|))², with multiplicity 1.

This shift matters for reading the formulas correctly. The unshifted eigenvalues of L_ω are not the energies one would compare with a Hamiltonian normalized to vanishing vacuum energy, and the paper's motivation for the spherical oscillator is precisely a setting in which vanishing ground-state energy is part of the characterization. The two families of formulas therefore answer different questions: one gives the spectrum of L_ω, the other gives the spectrum after the constant offset has been removed. Mixing them would misstate both the ground-state energy and the excited-level spacing.

What the partition function is for

The abstract states that the paper computes the spectrum and partition function in the two-dimensional case, and the motivation given is that the spherical oscillator is potentially relevant to chiral models in two-dimensional quantum field theory. The concrete target is the principal chiral model, the two-dimensional quantum sigma model whose target is a compact semisimple Lie group. The paper notes that it is not known how to rigorously formulate this model, although there is a conjecture for its scattering theory.

The proposed connection is a factorization hypothesis, not an established result. The authors suggest it is possible that the Hamiltonian in finite volume, when space is a circle S¹, factors as a sum of ordinary harmonic oscillators and spherical oscillators. They hope to use the partition functions associated to spherical oscillators as a means to test this possibility. In the discussion they take the target to be SU(2), i.e. the 3-sphere, for simplicity, and note that in that case the classical action does not factor, renormalization group techniques are needed even at the effective field theory level, and conformal invariance is broken and replaced by asymptotic freedom at the quantum level.

The discussion then writes a speculative partition function for a Hamiltonian denoted H(1), where the argument records the assumption that the circle radius is one. The formula combines a factor built from the spherical-oscillator levels with a factor 1/η(e^(−4t))² coming from the Gaussian χ degrees of freedom, and it uses λ(k,m,n) = m² + m + m(n+√(n²+16k²)) + (n+1)(n+√(n²+16k²))/2 − 2k with multiplicity 1 when n = 0 and 2 otherwise. The authors explicitly flag this as speculative and note that the Hamiltonian should depend on the radius, since conformal invariance is broken, which they say they do not fully understand. They also state that they do not know how to compare the spectrum of H(1) with the conjectural scattering solution, and that they are seeking a numerical test for the formula.

Where the exact results stop

The rigorous content of the paper is the spectral theory of L_ω: self-adjointness with compact resolvent, hence discrete spectrum, the ground state and its energy, and the closed-form spectrum with multiplicities in the two-dimensional case and in general dimension. The paper proves these results and checks the square-integrability conditions that make the hypergeometric candidates genuine eigenfunctions in the domain. The principal chiral model discussion is presented as motivation and as a possible application, with the factorization of the finite-volume Hamiltonian described as a possibility and the partition function for H(1) described as speculative.

The most decision-relevant distinction for a reader is therefore between the two-dimensional spectrum, which is exact and explicit, and the use of that spectrum as a test of a conjectured factorization in the chiral model, which is not established. The paper's own framing supports this split: it separates the computation of the spectrum and partition function from the hope that these partition functions can test whether the finite-volume Hamiltonian factors. The authors also identify the concrete obstacle to closing that gap, namely that the radius dependence of the Hamiltonian is not fully understood, so the spectrum of H(1) cannot yet be compared with the conjectural scattering solution.

Within the exact results, the priority is the d = 2 case. It is the case where the Cauchy-type density is the ground state with vanishing ground-state energy, it is the case the authors call critical, and it is the case whose spectrum and partition function are computed for the chiral-model motivation. The general-d formulas and the d = 1 multiplicity-free result are supporting structure that shows how the two-dimensional case sits inside a family whose behavior changes discontinuously with dimension.

Sources used in this answer