Broeckhove-Deumens Theorem in Shell Models

• Broeckhove-Deumens Theorem is an existence theorem asserting that a continuous manifold of non-orthogonal states can be discretized to span a Hilbert space.
• It underpins symmetry-restored methods in nuclear shell-model calculations by justifying the use of discrete non-orthogonal Slater determinants over vast, orthonormal configuration spaces.
• Numerical implementations using PAV and VAP approaches demonstrate near-exact reproduction of shell-model energies in systems such as sd-shell nuclei, 48Cr, and 78Ni.

Searching arXiv for the specified paper and closely related background on non-orthogonal shell-model / symmetry-restored methods. The Broeckhove–Deumens theorem, as stated in recent shell-model work, is an existence theorem about the discretization of continuous non-orthogonal manifolds in Hilbert space. For a separable Hilbert space $\mathscr L^2$ and a closed subspace $$\mathscr H = \overline{\mathrm{span}\,\Gamma} \subseteq \mathscr L^2$$, where $\Gamma$ is a continuous, dense set of non-orthogonal states, the theorem asserts the existence of a discrete countable non-orthogonal subset $\Gamma_0 \subset \Gamma$ such that $\mathscr H = \overline{\mathrm{span}\,\Gamma_0}$ (Dao et al., 11 Jul 2025). In nuclear-structure applications, this statement underpins the use of discrete sets of symmetry-restored non-orthogonal Slater determinants in place of very large orthonormal shell-model configuration spaces. The 2025 work “Exact solutions of the nuclear shell-model secular problem: Discrete Non-Orthogonal Shell Model within a Variation After Projection approach” formulates this connection explicitly and presents numerical realizations in sd-shell nuclei, in $^{48}$Cr, and in $^{78}$Ni (Dao et al., 11 Jul 2025).

1. Formal statement and Hilbert-space setting

The paper gives the theorem in the following form: for a separable Hilbert space $\mathscr L^2$ and a closed subspace $\mathscr H = \overline{\mathrm{span}\,\Gamma}$, with $\Gamma$ a continuous, dense set of non-orthogonal states, there exists a discrete countable non-orthogonal subset $$\mathscr H = \overline{\mathrm{span}\,\Gamma} \subseteq \mathscr L^2$$0 such that

$$\mathscr H = \overline{\mathrm{span}\,\Gamma} \subseteq \mathscr L^2$$1

Equivalently, any $$\mathscr H = \overline{\mathrm{span}\,\Gamma} \subseteq \mathscr L^2$$2 can be represented as

$$\mathscr H = \overline{\mathrm{span}\,\Gamma} \subseteq \mathscr L^2$$3

with $$\mathscr H = \overline{\mathrm{span}\,\Gamma} \subseteq \mathscr L^2$$4 and convergence in the Hilbert-space norm (Dao et al., 11 Jul 2025).

In the shell-model context, the relevant space is the valence-space many-body Hilbert space $$\mathscr H = \overline{\mathrm{span}\,\Gamma} \subseteq \mathscr L^2$$5. The standard orthonormal basis is the set of spherical Slater determinants, identified in the paper with the usual $$\mathscr H = \overline{\mathrm{span}\,\Gamma} \subseteq \mathscr L^2$$6-scheme basis. Since this space is finite dimensional in the applications considered, the countable subset implied by the theorem becomes finite in practice. The significance of the theorem in this setting is therefore precise: it guarantees the existence of a finite family of non-orthogonal Slater determinants spanning the same space as the standard shell-model basis (Dao et al., 11 Jul 2025).

The paper also emphasizes the theorem’s limitations. It does not specify the nature of the continuous family $$\mathscr H = \overline{\mathrm{span}\,\Gamma} \subseteq \mathscr L^2$$7, does not characterize the discrete subset $$\mathscr H = \overline{\mathrm{span}\,\Gamma} \subseteq \mathscr L^2$$8, and does not provide a constructive procedure for obtaining it. Its content is existential rather than algorithmic. A central contribution of the 2025 work is therefore not a new theorem, but an explicit numerical construction showing how such a subset can be realized for realistic shell-model Hamiltonians (Dao et al., 11 Jul 2025).

2. Shell-model secular problem in a non-orthogonal representation

The shell-model Hamiltonian is written as a standard one- plus two-body operator in a valence space of spherical harmonic-oscillator orbitals,

$$\mathscr H = \overline{\mathrm{span}\,\Gamma} \subseteq \mathscr L^2$$9

with antisymmetry conditions

$\Gamma$0

Here the creation and annihilation operators obey the canonical anticommutation relations, and the indices label spherical single-particle orbitals in the chosen valence shell (Dao et al., 11 Jul 2025).

Rather than expanding eigenstates in an orthonormal configuration basis, the paper uses non-orthogonal Slater determinants $\Gamma$1, optionally supplemented by many-particle–many-hole excitations. Two frameworks are distinguished: projection after variation (PAV), based on constrained Hartree–Fock plus $\Gamma$2 excitations, and variation after projection (VAP), based on Thouless-parameterized determinants (Dao et al., 11 Jul 2025).

States of good $\Gamma$3 are constructed as

$\Gamma$4

and in the extended PAV treatment as

$\Gamma$5

Here $\Gamma$6 is the parity projector and $\Gamma$7 is the angular-momentum projector from intrinsic projection $\Gamma$8 to laboratory projection $\Gamma$9 (Dao et al., 11 Jul 2025).

The basic objects are the norm and Hamiltonian kernels,

$\Gamma_0 \subset \Gamma$0

which define the Hill–Wheeler–Griffin generalized eigenvalue problem

$\Gamma_0 \subset \Gamma$1

Written in standard linear-algebra form, this is the generalized eigenproblem $\Gamma_0 \subset \Gamma$2 in a non-orthogonal basis (Dao et al., 11 Jul 2025).

For the VAP construction, the determinants are generated through the Thouless parameterization

$\Gamma_0 \subset \Gamma$3

with $\Gamma_0 \subset \Gamma$4 a skew-symmetric complex matrix and $\Gamma_0 \subset \Gamma$5 a fixed reference Slater determinant for that branch. This parameterization provides the variational degrees of freedom used to optimize projected states directly (Dao et al., 11 Jul 2025).

3. Numerical realization in the discrete non-orthogonal shell model

The paper interprets “realization” of the theorem as the explicit construction, for a given realistic shell-model Hamiltonian, of a finite discrete set $\Gamma_0 \subset \Gamma$6 of non-orthogonal Slater determinants such that the generalized eigenvalue problem in that space reproduces the exact shell-model eigenvalues and eigenstates for low-lying states of interest (Dao et al., 11 Jul 2025).

In the PAV implementation, the starting point is a continuous generator-coordinate set of triaxial Hartree–Fock states labeled by quadrupole deformation coordinates $\Gamma_0 \subset \Gamma$7. A basis-selection procedure identified as the Caurier basis-selection technique is then used to extract a minimal discrete subset. The procedure begins with an HF minimum, selects subsequent states by minimizing the lowest eigenvalue of the generalized eigenproblem in the enlarged space, and continues iteratively until convergence. After this HF stage, the basis is enlarged by selected $\Gamma_0 \subset \Gamma$8 excitations that further improve the variational energy (Dao et al., 11 Jul 2025).

This yields the PAV realization

$\Gamma_0 \subset \Gamma$9

The importance of this construction lies in its diagnostic role. The paper shows that a manifold restricted to $\mathscr H = \overline{\mathrm{span}\,\Gamma_0}$0-constrained HF determinants is generally insufficient: it spans a proper subspace of the shell-model space and leaves residual correlation energy unrecovered even when the number of HF generator points becomes very large. By contrast, the enlarged HF plus $\mathscr H = \overline{\mathrm{span}\,\Gamma_0}$1 set can reach the exact shell-model ground-state energies in the test cases considered (Dao et al., 11 Jul 2025).

The VAP implementation is more compact. It constructs generic non-orthogonal determinants through direct minimization of the projected energy, thereby generating states that effectively encode the physics of many $\mathscr H = \overline{\mathrm{span}\,\Gamma_0}$2 configurations. The paper explicitly states that VAP is “strictly equivalent to performing $\mathscr H = \overline{\mathrm{span}\,\Gamma_0}$3 excitations” in the full space, and presents VAP as the practical device that finds a discrete subset satisfying the theorem’s completeness requirement in concrete shell-model calculations (Dao et al., 11 Jul 2025).

This suggests a useful conceptual distinction between theorem and realization. The theorem asserts only the existence of a discrete spanning subset. The discrete non-orthogonal shell model provides an explicit algorithmic route to such a subset, with the quality of the realization judged by whether the resulting Ritz eigenvalues coincide with exact shell-model energies to numerical precision.

4. Variation after projection as the constructive mechanism

In the VAP framework, the projected energy functional is

$\mathscr H = \overline{\mathrm{span}\,\Gamma_0}$4

with

$\mathscr H = \overline{\mathrm{span}\,\Gamma_0}$5

This is the Ritz variational principle applied to symmetry-restored states (Dao et al., 11 Jul 2025).

Variation with respect to the mixing amplitudes $\mathscr H = \overline{\mathrm{span}\,\Gamma_0}$6 produces the generalized eigenvalue problem. Variation with respect to the intrinsic Slater determinants, using the Thouless parameterization, yields the projected Brillouin condition

$\mathscr H = \overline{\mathrm{span}\,\Gamma_0}$7

The paper identifies this as the resonating Hartree–Fock condition of Fukutome, generalized to symmetry projection (Dao et al., 11 Jul 2025).

The implementation uses a hybrid optimization strategy. The first $\mathscr H = \overline{\mathrm{span}\,\Gamma_0}$8 Slater determinants are kept fixed once optimized, the full Brillouin condition is applied only to the most recently added determinants, and a quasi-Newton algorithm of L-BFGS type is used to minimize the projected energy (Dao et al., 11 Jul 2025). The practical consequence is that VAP searches directly in the manifold of symmetry-restored non-orthogonal determinants rather than relying on an a priori deformation mesh plus explicit combinatorial excitation generation.

Within the logic of the theorem, this role is decisive. PAV with deformation coordinates alone fails because the generator manifold is not complete for the shell-model space under study. PAV plus $\mathscr H = \overline{\mathrm{span}\,\Gamma_0}$9 corrections can restore completeness, but the enlargement can become combinatorially large. VAP bypasses this by optimizing the determinants in the projected space itself, so that each determinant functions as a highly compressed carrier of many-shell-model configurations (Dao et al., 11 Jul 2025).

5. Empirical demonstrations in sd-shell nuclei, $^{48}$0Cr, and $^{48}$1Ni

The paper tests the theorem’s realization in several nuclear systems. In sd-shell nuclei with the USDB interaction, it examines $^{48}$2Ne, $^{48}$3Mg, $^{48}$4Si, and $^{48}$5Al. For $^{48}$6Mg, DNO-SM(PAV) with only $^{48}$7-constrained HF determinants gives a ground-state energy of $^{48}$8 MeV, whereas the exact shell-model result is $^{48}$9 MeV, a difference of approximately $^{78}$0 MeV. The paper reports that this discrepancy does not disappear even when approximately 6000 HF points are included across the $^{78}$1 plane, indicating that the pure deformation manifold spans only a proper subspace of the full shell-model space (Dao et al., 11 Jul 2025).

When $^{78}$2 excitations are added in PAV, the agreement becomes essentially exact. For the same nucleus, the enlarged calculation yields $^{78}$3 MeV versus the exact $^{78}$4 MeV. Similar exact agreement is reported for $^{78}$5Ne and $^{78}$6Si. The dominant excitations are described as $^{78}$7 particle–hole pairs coupled to $^{78}$8, that is, pair-condensate-like configurations (Dao et al., 11 Jul 2025).

The VAP results are more compressed still. For $^{78}$9Ne, $\mathscr L^2$0Mg, $\mathscr L^2$1Si, and $\mathscr L^2$2Al, the differences between DNO-SM(VAP) and exact shell-model energies are at the level of a few $\mathscr L^2$3 MeV. Two representative basis-size comparisons are highlighted: $\mathscr L^2$4Si is reproduced with 45 VAP Slater determinants versus 93,710 standard shell-model configurations, and $\mathscr L^2$5Mg with 16 determinants versus 28,503 shell-model configurations (Dao et al., 11 Jul 2025). The paper presents these calculations as numerical evidence that a finite set of projected non-orthogonal Slater determinants can span the full valence space for the low-lying states considered.

In $\mathscr L^2$6Cr with the KB3 interaction, the VAP approach is applied state by state along the yrast band. The resulting energies for $\mathscr L^2$7 agree with exact shell-model values to within $\mathscr L^2$8 MeV, often exactly within the quoted digits, with basis sizes ranging from 50 determinants for $\mathscr L^2$9 down to 12 determinants for $\mathscr H = \overline{\mathrm{span}\,\Gamma}$0 (Dao et al., 11 Jul 2025). The $\mathscr H = \overline{\mathrm{span}\,\Gamma}$1-ray transitions,

$\mathscr H = \overline{\mathrm{span}\,\Gamma}$2

reproduce the backbending pattern exactly, including the impact of proton–neutron pairing correlations. In the paper’s interpretation, this shows that number-conserving non-orthogonal Slater determinants under VAP fully capture pairing correlations in this rotational band (Dao et al., 11 Jul 2025).

For $\mathscr H = \overline{\mathrm{span}\,\Gamma}$3Ni in the pf-sdg space with the PFSDG-U interaction, the shell-model dimension is reported as approximately $\mathscr H = \overline{\mathrm{span}\,\Gamma}$4 in the $\mathscr H = \overline{\mathrm{span}\,\Gamma}$5-scheme, beyond direct full diagonalization. Truncated shell-model calculations up to 10p–10h yield a ground-state energy of $\mathscr H = \overline{\mathrm{span}\,\Gamma}$6 MeV and an extrapolated value of $\mathscr H = \overline{\mathrm{span}\,\Gamma}$7 MeV. DNO-SM(VAP), starting from a spherical HF reference and adding optimized non-orthogonal determinants, converges to $\mathscr H = \overline{\mathrm{span}\,\Gamma}$8 MeV, lower than the extrapolated shell-model energy and presented as essentially the best variational bound in the full valence space (Dao et al., 11 Jul 2025). The paper further reports exponential-like convergence with the number of determinants and saturation at the quoted value.

6. Conceptual scope, limitations, and broader connections

The theorem’s conceptual content is straightforward in functional-analytic terms: a dense continuous manifold of non-orthogonal vectors admits a discrete countable subset with the same closed span. In finite-dimensional shell-model spaces, this reduces to the statement that the full valence-space Hilbert space can be spanned by a finite family of non-orthogonal Slater determinants, provided that the family is sufficiently rich (Dao et al., 11 Jul 2025).

The limitations arise at the level of realization rather than theorem. The paper notes that the completeness of a particular continuous generator manifold is not automatic. For the triaxial deformation manifold $\mathscr H = \overline{\mathrm{span}\,\Gamma}$9, separability and completeness are not trivial to prove in general, and the explicit calculations show that the subspace generated by $\Gamma$0-HF states is not the full shell-model space. Persistent energy differences, even with a dense deformation mesh, provide the practical evidence for this restriction (Dao et al., 11 Jul 2025). A common misconception is therefore that any continuous generator-coordinate manifold is automatically complete; the paper explicitly argues against this.

The non-orthogonal basis must also be numerically well behaved. Because overlap matrices $\Gamma$1 can become ill-conditioned when determinants are nearly linearly dependent, basis selection and stabilization are required. The use of Caurier’s selection method in PAV and controlled determinant addition in VAP addresses this issue operationally (Dao et al., 11 Jul 2025).

The results are also scoped to low-lying states. The paper presents exact or near-exact reproduction for ground states and yrast bands, and describes this as a realization of the theorem for “low-lying states of interest” (Dao et al., 11 Jul 2025). This suggests that the practical algorithm is optimized for the physically relevant low-energy sector, even though the theorem itself is formulated at the level of the entire closed span.

The broader theoretical placement given in the paper links the theorem to generator coordinate methods, the Hill–Wheeler–Griffin framework, Peierls–Yoccoz symmetry restoration, non-orthogonal configuration interaction, resonating Hartree–Fock, and VAMPIR-like strategies (Dao et al., 11 Jul 2025). In this perspective, the Broeckhove–Deumens theorem serves as the mathematical foundation for replacing continuous generator-coordinate superpositions by discrete non-orthogonal expansions without loss of completeness, while the discrete non-orthogonal shell model shows that this replacement can be made numerically exact for realistic shell-model Hamiltonians.

A plausible implication is that the theorem’s practical value is greatest not as a standalone formal result, but as a justification for aggressively compressed, symmetry-restored non-orthogonal representations of shell-model eigenstates. In the examples studied, the compression is extreme, yet the resulting energies coincide with exact diagonalization where available, or improve upon the best truncated-extrapolated shell-model estimates in spaces near current computational limits (Dao et al., 11 Jul 2025).