Projected Generator Coordinate Method (PGCM)

• PGCM is a symmetry-restored, multi-reference framework that represents nuclear states as superpositions of nonorthogonal mean-field vacua along collective coordinates.
• It employs projection techniques via the Hill–Wheeler–Griffin equation to restore broken symmetries and capture key static correlations such as deformation and pairing.
• PGCM has been applied in ab initio nuclear spectroscopy and giant-resonance studies, offering insights into collective responses and shape fluctuations.

Projected Generator Coordinate Method (PGCM) is a symmetry-restored multi-reference many-body framework in which nuclear states of good quantum numbers are represented as superpositions of nonorthogonal, symmetry-breaking mean-field vacua generated along one or several collective coordinates. In contemporary nuclear-structure work, PGCM has been developed as an ab initio tool for low-lying spectroscopy and collective response, including giant resonances, and as a beyond-mean-field framework that combines exact restoration of broken symmetries with non-perturbative mixing of large-amplitude collective configurations. In the formulation used for giant monopole resonances, a PGCM state of good symmetry $\sigma\equiv\{J\,M\,\Pi\,N\,Z\}$ is written as a superposition of symmetry-projected Hartree–Fock–Bogoliubov (HFB) vacua, and the variational problem reduces to the Hill–Wheeler–Griffin (HWG) equation for projected norm and Hamiltonian kernels (Porro et al., 2024). In ab initio applications to neon isotopes, giant monopole resonances, odd-mass nuclei, magnetic-dipole strength, and Gamow–Teller transitions, PGCM is used as the component of the many-body description that captures strong static correlations associated with deformation, pairing, and shape fluctuations, while exact projection enforces the required quantum numbers (Frosini et al., 2021).

1. Formal definition and variational structure

The defining PGCM ansatz is a projected superposition of reference vacua labeled by a collective coordinate $q$. In one common notation,

$$\Psi_\nu^\sigma = \int dq\, f_\nu^\sigma(q)\,P^\sigma\,|\Phi(q)\rangle,$$

where $P^\sigma$ restores the broken symmetries of the underlying HFB vacua and $f_\nu^\sigma(q)$ is determined variationally. After discretization of the generator coordinate and of the group integrals entering the projector, the Ritz variational principle yields the HWG equation

$$\sum_{q'} \big[H^\sigma(q,q')-E_\nu^\sigma N^\sigma(q,q')\big]\,f_\nu^\sigma(q')=0,$$

with projected kernels

$$N^\sigma(q,q')\equiv \langle \Phi(q)|P^\sigma|\Phi(q')\rangle,\qquad H^\sigma(q,q')\equiv \langle \Phi(q)|H\,P^\sigma|\Phi(q')\rangle.$$

An equivalent representation expands the projected state over symmetry-group rotations $R(\phi)$ weighted by the appropriate irreducible-representation coefficient $g^\sigma(\phi)$, so that the method can be viewed simultaneously as configuration mixing in $q$-space and as exact contraction over symmetry manifolds (Porro et al., 2024).

For low-lying spectroscopy, the same formal structure is commonly written with explicit particle-number, angular-momentum, and parity projectors. In the notation used for ab initio neon calculations,

$q$0

or, on a discrete mesh,

$q$1

For odd-mass and odd-odd systems the same variational logic is retained, but the basis states are built from one-quasiparticle or two-quasiparticle excitations before projection; in the odd-mass $q$2Ne application the collective index is $q$3, while in the Gamow–Teller extension the odd–odd states are expanded over projected proton–neutron two-quasiparticle configurations (Lin et al., 2024).

The formal role of PGCM is therefore twofold. First, it restores symmetries that are deliberately broken at the mean-field level in order to reveal collective minima. Second, it mixes distinct mean-field configurations non-perturbatively, thereby generating collective wave functions that spread over deformation, pairing, radius, or other selected coordinates. In the language used in the multi-reference perturbation-theory series, PGCM provides a zeroth-order approximation that captures strong static correlations, while missing dynamical correlations can subsequently be added perturbatively through PGCM-PT (Frosini et al., 2021).

2. Generator-coordinate manifolds and reference states

The generator coordinates are chosen to span the dominant collective degrees of freedom of the problem under study. In monopole-resonance calculations, the coordinate set is explicitly two-dimensional,

$q$4

where the root-mean-square radius is

$q$5

and the axial quadrupole deformation is

$q$6

Each reference state $q$7 is obtained from a constrained HFB problem,

$q$8

so that the expectation values of the selected collective operators match the target point in generator-coordinate space (Porro et al., 2024).

Across the cited literature, the reference manifold varies with the physics of interest. In the ab initio neon study, representative coordinates include quadrupole deformation $q$9, triaxiality $$\Psi_\nu^\sigma = \int dq\, f_\nu^\sigma(q)\,P^\sigma\,|\Phi(q)\rangle,$$0, octupole deformation $$\Psi_\nu^\sigma = \int dq\, f_\nu^\sigma(q)\,P^\sigma\,|\Phi(q)\rangle,$$1, pairing gaps, and rms radius; the method is described there as “ideally suited to capturing strong static correlations associated with shape deformation and fluctuations” (Frosini et al., 2021). In the sd-shell M1-strength benchmark, two alternative generator-coordinate sets are tested. Set A constrains the three $$\Psi_\nu^\sigma = \int dq\, f_\nu^\sigma(q)\,P^\sigma\,|\Phi(q)\rangle,$$2 and $$\Psi_\nu^\sigma = \int dq\, f_\nu^\sigma(q)\,P^\sigma\,|\Phi(q)\rangle,$$3 components of the M1 operator together with isovector cranking, while Set B constrains isovector and isoscalar proton–neutron pairing together with isoscalar cranking. These two coordinate manifolds are designed to span both vibrational and rotational modes of the $$\Psi_\nu^\sigma = \int dq\, f_\nu^\sigma(q)\,P^\sigma\,|\Phi(q)\rangle,$$4 spectrum (Bofos et al., 16 Jul 2025).

A distinct extension is the dynamical GCM (DGCM), in which a collective coordinate and its canonically conjugate momentum are used simultaneously as generator coordinates. In the application to particle-number projection, the pair $$\Psi_\nu^\sigma = \int dq\, f_\nu^\sigma(q)\,P^\sigma\,|\Phi(q)\rangle,$$5 is treated as a conjugate set, combining a superposition of BCS states with different mean particle numbers and the usual integration over the gauge angle. The corresponding nonorthogonal dynamical states satisfy the Peierls–Thouless conjugacy condition, and the resulting formalism is presented as a simpler alternative to variation after projection for particle number (Hizawa et al., 2020).

For odd-mass systems, the reference manifold is enlarged from vacua to blocked quasiparticle configurations. In the quantum-number-projected GCM calculation of $$\Psi_\nu^\sigma = \int dq\, f_\nu^\sigma(q)\,P^\sigma\,|\Phi(q)\rangle,$$6Ne, one starts from axially deformed HFB wave functions with different quadrupole deformations and constructs one-quasiparticle “Odd A” states. The final PGCM basis then carries the blocked quasiparticle label $$\Psi_\nu^\sigma = \int dq\, f_\nu^\sigma(q)\,P^\sigma\,|\Phi(q)\rangle,$$7, the intrinsic angular-momentum component $$\Psi_\nu^\sigma = \int dq\, f_\nu^\sigma(q)\,P^\sigma\,|\Phi(q)\rangle,$$8, and the deformation $$\Psi_\nu^\sigma = \int dq\, f_\nu^\sigma(q)\,P^\sigma\,|\Phi(q)\rangle,$$9 (Lin et al., 2024). For odd–odd systems relevant to Gamow–Teller transitions and $P^\sigma$0 decay, the basis is similarly enlarged to projected proton–neutron two-quasiparticle configurations built on an HFB vacuum with odd average neutron and proton numbers (Chen et al., 8 Jan 2026).

These constructions share a common principle: the reference manifold is not arbitrary, but deliberately chosen so that the projected superposition resolves the collective subspace expected to dominate the targeted observables. This suggests that the accuracy and efficiency of PGCM depend less on a universal coordinate choice than on a physically guided selection of coordinates that break, and subsequently restore, the symmetries relevant to the problem.

3. Symmetry restoration, kernels, and numerical realization

PGCM relies on exact restoration of the symmetries broken by the reference vacua. In the standard nuclear formulation, the particle-number and angular-momentum projectors are written as

$P^\sigma$1

and

$P^\sigma$2

with parity projection included when needed through

$P^\sigma$3

In practice, the group integrals are replaced by discrete quadratures over gauge angles and Euler angles, so that the projected kernels are assembled as finite sums of rotated overlaps and rotated Hamiltonian matrix elements (Frosini et al., 2021).

The rotated norm kernels

$P^\sigma$4

are evaluated through the Onishi formula, Robledo’s Pfaffian, or integral representations. The rotated Hamiltonian kernels

$P^\sigma$5

are obtained via the off-diagonal Wick theorem in the Balian–Brézin form, which expresses them in terms of generalized density matrices. Projection then amounts to summing these rotated kernels with the appropriate irrep weights:

$P^\sigma$6

In the odd-mass $P^\sigma$7Ne study, each kernel is computed by three-fold integration over Euler angles and neutron/proton gauge angles, while in the ab initio neon calculations Pfaffians of the generalized density matrix are used in a numerically stabilized form (Porro et al., 2024).

Once the overlap and Hamiltonian kernels have been assembled, the HWG equation is solved by first diagonalizing the positive-definite overlap matrix and constructing the Hermitian natural basis. Near-linear-dependent combinations are removed by discarding norm eigenmodes below a threshold. In the ab initio monopole-resonance work this reduced problem has dimension $P^\sigma$8, and the diagonalization step scales as $P^\sigma$9 with $f_\nu^\sigma(q)$0, making it very cheap relative to kernel construction (Porro et al., 2024). The same strategy appears in sd-shell and odd-mass applications, where one first diagonalizes the norm matrix, truncates small natural orbitals or norm eigenvalues, and then solves the secular equation in the reduced space (Bofos et al., 16 Jul 2025).

The dominant computational cost lies in constructing kernels between all pairs of generator points. In the monopole-resonance implementation, the number of generator points is of order 10–50 in each dimension and the total number of kernels scales as $f_\nu^\sigma(q)$1. Each unprojected kernel involves operations scaling nominally like $f_\nu^\sigma(q)$2 with the one-body basis dimension, although block structure and rank-reduction of three-body forces are exploited to keep mid-mass calculations tractable (Porro et al., 2024). In the sd-shell M1 benchmark, the number of intrinsic vacua is of order several hundred, and after norm diagonalization the natural-basis dimension remains substantially smaller than the full shell-model dimension (Bofos et al., 16 Jul 2025).

A separate line of development, due to Stuber, reformulates projection algebraically via the invariant mean rather than explicit numerical group integration. In that framework the projected GCM wave function

$f_\nu^\sigma(q)$3

is mapped onto a coupled-cluster-like exponential $f_\nu^\sigma(q)$4, with the symmetry-adapted cluster operators extracted from the invariant mean of $f_\nu^\sigma(q)$5. This is not the dominant implementation in the ab initio nuclear calculations cited above, but it identifies an alternative route to symmetry projection and connects PGCM directly to coupled-cluster theory (Stuber, 2015).

4. Correlation content and relation to adjacent methods

The central distinction between PGCM and QRPA-type approaches is the type of collective correlations they encode. PGCM includes large-amplitude, static, anharmonic fluctuations of collective coordinates non-perturbatively, whereas QRPA is limited to small-amplitude, harmonic oscillations about a single reference. Because PGCM restores exact quantum numbers $f_\nu^\sigma(q)$6, it also prevents spurious admixtures and permits simultaneous treatment of pairing, deformation, and shape coexistence effects in closed- and open-shell nuclei without ad hoc adjustments (Porro et al., 2024).

This difference is especially explicit in giant-resonance applications. In the comparison performed for isoscalar monopole response, QRPA solves the familiar quasiparticle matrix equation with $f_\nu^\sigma(q)$7 matrices around a single HFB minimum, while PGCM diagonalizes the full multi-reference Hamiltonian kernel in the nonorthogonal basis of projected vacua. In $f_\nu^\sigma(q)$8O, PGCM predicts the GMR centroid at $f_\nu^\sigma(q)$9, whereas QRPA(FAM) gives $$\sum_{q'} \big[H^\sigma(q,q')-E_\nu^\sigma N^\sigma(q,q')\big]\,f_\nu^\sigma(q')=0,$$0, corresponding to an anharmonic shift of approximately $$\sum_{q'} \big[H^\sigma(q,q')-E_\nu^\sigma N^\sigma(q,q')\big]\,f_\nu^\sigma(q')=0,$$1. In deformed $$\sum_{q'} \big[H^\sigma(q,q')-E_\nu^\sigma N^\sigma(q,q')\big]\,f_\nu^\sigma(q')=0,$$2Ti, the main breathing-mode peak is nearly unchanged, with QRPA at $$\sum_{q'} \big[H^\sigma(q,q')-E_\nu^\sigma N^\sigma(q,q')\big]\,f_\nu^\sigma(q')=0,$$3 and PGCM at $$\sum_{q'} \big[H^\sigma(q,q')-E_\nu^\sigma N^\sigma(q,q')\big]\,f_\nu^\sigma(q')=0,$$4, but the GMR–GQR coupling peak is shifted downward by approximately $$\sum_{q'} \big[H^\sigma(q,q')-E_\nu^\sigma N^\sigma(q,q')\big]\,f_\nu^\sigma(q')=0,$$5 by anharmonic and projection effects (Porro et al., 2024).

The multi-reference perturbation-theory series sharpens the conceptual division between static and dynamical correlations. In that formulation, PGCM is the symmetry-conserving variational core that captures rotational, vibrational, and pairing-fluctuation effects, while PGCM-PT adds the remaining weak dynamical correlations perturbatively in a state-specific way. The formalism reduces to PGCM at lowest order and to single-reference perturbation theory in appropriate limits (Frosini et al., 2021). In proof-of-principle calculations, strict PGCM reproduces the dominant static physics but can leave excitation spectra too dilated and binding energies significantly underbound, while PGCM-PT(2) brings the results closer to exact solutions (Frosini et al., 2021).

DGCM occupies a related but distinct position in this landscape. By treating a collective coordinate and its conjugate momentum simultaneously, it extends the usual projected mixing space and can mimic some of the energy lowering otherwise associated with variation after projection. In the particle-number application, the method lowers the ground-state energy significantly, especially for magic nuclei for which the pairing gap is zero in the BCS approximation; the reported gain is approximately $$\sum_{q'} \big[H^\sigma(q,q')-E_\nu^\sigma N^\sigma(q,q')\big]\,f_\nu^\sigma(q')=0,$$6–$$\sum_{q'} \big[H^\sigma(q,q')-E_\nu^\sigma N^\sigma(q,q')\big]\,f_\nu^\sigma(q')=0,$$7 relative to variation before projection in open-shell nuclei such as $$\sum_{q'} \big[H^\sigma(q,q')-E_\nu^\sigma N^\sigma(q,q')\big]\,f_\nu^\sigma(q')=0,$$8Ni, and approximately $$\sum_{q'} \big[H^\sigma(q,q')-E_\nu^\sigma N^\sigma(q,q')\big]\,f_\nu^\sigma(q')=0,$$9–$$N^\sigma(q,q')\equiv \langle \Phi(q)|P^\sigma|\Phi(q')\rangle,\qquad H^\sigma(q,q')\equiv \langle \Phi(q)|H\,P^\sigma|\Phi(q')\rangle.$$0 even with DGCM(3) in closed-shell $$N^\sigma(q,q')\equiv \langle \Phi(q)|P^\sigma|\Phi(q')\rangle,\qquad H^\sigma(q,q')\equiv \langle \Phi(q)|H\,P^\sigma|\Phi(q')\rangle.$$1Ni (Hizawa et al., 2020).

A common misconception is that symmetry restoration alone exhausts the beyond-mean-field physics. The cited studies do not support that view. They repeatedly distinguish the non-perturbative static correlations captured by PGCM from the dynamical correlations that remain outside the strict PGCM manifold and can require either perturbative corrections, enlarged coordinate spaces, or both (Frosini et al., 2021).

5. Applications across spectroscopy and response

PGCM has been applied to a broad set of observables spanning low-lying spectra, giant resonances, odd-mass rotational structure, magnetic-dipole strength, and Gamow–Teller transitions. In the ab initio study of the neon isotopic chain, Frosini and collaborators use PGCM together with the quasi-exact in-medium no-core shell model. There PGCM is shown to be a suitable method to tackle the low-lying spectroscopy of complex nuclei and to describe shape deformation and fluctuations, although the island-of-inversion region remains challenging and appears to require dynamical correlations beyond strict PGCM (Frosini et al., 2021).

For giant monopole resonances, the first two papers of the 2024 series establish PGCM as an ab initio alternative to QRPA-based descriptions. In the second paper, PGCM is applied to the doubly open-shell nuclei $$N^\sigma(q,q')\equiv \langle \Phi(q)|P^\sigma|\Phi(q')\rangle,\qquad H^\sigma(q,q')\equiv \langle \Phi(q)|H\,P^\sigma|\Phi(q')\rangle.$$2Ti, $$N^\sigma(q,q')\equiv \langle \Phi(q)|P^\sigma|\Phi(q')\rangle,\qquad H^\sigma(q,q')\equiv \langle \Phi(q)|H\,P^\sigma|\Phi(q')\rangle.$$3Si, and $$N^\sigma(q,q')\equiv \langle \Phi(q)|P^\sigma|\Phi(q')\rangle,\qquad H^\sigma(q,q')\equiv \langle \Phi(q)|H\,P^\sigma|\Phi(q')\rangle.$$4Mg. The reported centroid energies are $$N^\sigma(q,q')\equiv \langle \Phi(q)|P^\sigma|\Phi(q')\rangle,\qquad H^\sigma(q,q')\equiv \langle \Phi(q)|H\,P^\sigma|\Phi(q')\rangle.$$5 for $$N^\sigma(q,q')\equiv \langle \Phi(q)|P^\sigma|\Phi(q')\rangle,\qquad H^\sigma(q,q')\equiv \langle \Phi(q)|H\,P^\sigma|\Phi(q')\rangle.$$6Ti, $$N^\sigma(q,q')\equiv \langle \Phi(q)|P^\sigma|\Phi(q')\rangle,\qquad H^\sigma(q,q')\equiv \langle \Phi(q)|H\,P^\sigma|\Phi(q')\rangle.$$7 for $$N^\sigma(q,q')\equiv \langle \Phi(q)|P^\sigma|\Phi(q')\rangle,\qquad H^\sigma(q,q')\equiv \langle \Phi(q)|H\,P^\sigma|\Phi(q')\rangle.$$8Si, and a dominant $$N^\sigma(q,q')\equiv \langle \Phi(q)|P^\sigma|\Phi(q')\rangle,\qquad H^\sigma(q,q')\equiv \langle \Phi(q)|H\,P^\sigma|\Phi(q')\rangle.$$9 structure for $R(\phi)$0Mg. The comparison to experiment is given as $R(\phi)$1 for $R(\phi)$2Ti, approximately $R(\phi)$3 for $R(\phi)$4Si, and $R(\phi)$5 for $R(\phi)$6Mg. In $R(\phi)$7Ti the method isolates a GMR–GQR coupling peak at $R(\phi)$8 with $R(\phi)$9 and a main breathing-mode peak at $g^\sigma(\phi)$0 with $g^\sigma(\phi)$1, while in $g^\sigma(\phi)$2Mg it predicts extremely fragmented monopole strength between $g^\sigma(\phi)$3 and $g^\sigma(\phi)$4 (Porro et al., 2024). The stated interpretation is that PGCM captures fragmentation and monopole–quadrupole coupling that are difficult to represent within harmonic approaches.

Odd-mass structure has also been addressed. In the quantum-number-projected GCM calculation of $g^\sigma(\phi)$5Ne with a chiral NN+3N interaction, the low-lying spectrum consists of two rotational-like bands built on $g^\sigma(\phi)$6 and $g^\sigma(\phi)$7 intrinsic configurations. The collective wave functions peak at $g^\sigma(\phi)$8, and the inclusion of the 3N interaction is reported to be crucial for realistic quadrupole deformation and rotational spacings (Lin et al., 2024).

In valence-space M1-strength calculations, PGCM has been benchmarked directly against exact diagonalization using the USDB interaction in the sd shell. In $g^\sigma(\phi)$9Mg, exact diagonalization yields two strong low-energy peaks at $q$0 and $q$1, a quasi-continuum up to approximately $q$2, and a summed strength $q$3, corresponding to approximately $q$4 of the total model-space sum rule $q$5. Lorentz-folded results with $q$6 give an exact centroid $q$7 and $q$8, compared with PGCM Set A values $q$9 and $q$00, and PGCM Set B values $q$01 and $q$02. Both PGCM sets recover the low-energy enhancement in the de-excitation strength (Bofos et al., 16 Jul 2025).

A minimal extension of PGCM to Gamow–Teller transitions in even–even nuclei and to odd–odd intermediate states has been benchmarked in the fp shell. For calcium and titanium isotopes, the method reproduces low-lying transitions and the onset of the giant-resonance region quite well, with performance comparable to CI(2p2h). For $q$03Ca$q$04Ti $q$05, the exact shell-model matrix element is reported as approximately $q$06, while PGCM saturates at approximately $q$07, i.e. about $q$08 above the shell-model value; the dominant source of the discrepancy is the first intermediate $q$09 transition (Chen et al., 8 Jan 2026).

PGCM methods have also entered double-beta-decay theory through a projected GCM derivation of PQRPA equations for axially deformed nuclei. In that setting, the odd–odd intermediate states are built from proton–neutron quasiparticle pairs on top of an HFB vacuum, and the GCM/GOA treatment reduces the full HWG problem to a QRPA-like equation for projected phonons. The resulting formalism is designed for non-closure calculations of $q$10 nuclear matrix elements with exact particle-number and angular-momentum projection (Bobyk et al., 2014).

6. Uncertainties, limitations, and development paths

The 2024 ab initio monopole-resonance paper provides one of the most explicit uncertainty analyses for PGCM calculations. It separates theoretical uncertainties into three categories: Hamiltonian modeling, basis representation, and many-body solution. The Hamiltonian-modeling category includes chiral EFT truncation from NLO to $q$11LO to $q$12LO, low-energy-constant fit uncertainties that are “not yet propagated,” and SRG truncation through induced many-body terms. The basis-representation category includes one-body basis truncation $q$13 and three-body basis truncation $q$14. The many-body-solution category includes three-body rank-reduction error of approximately $q$15–$q$16, choice of generator coordinates, sampling density and interval in $q$17, and omission of dynamical correlations beyond strict PGCM, denoted there as PGCM-PT truncation (Porro et al., 2024).

For the $q$18 monopole resonance in $q$19Ti, the reported quantitative findings are specific. One-body basis convergence in $q$20 yields less than $q$21 uncertainty on the centroid and at most approximately $q$22 on the width. Dependence on oscillator frequency $q$23 induces approximately $q$24–$q$25 uncertainty on the centroid and approximately $q$26 on the width. Chiral truncation uncertainty is at most approximately $q$27 on the centroid and approximately $q$28 on the width, with non-monotonic behavior for the latter. Variation of the SRG flow parameter shifts centroids by approximately $q$29, while reasonable generator-coordinate meshes in one and two dimensions produce negligible changes, below $q$30 on centroids (Porro et al., 2024).

The most persistent limitation identified across the cited studies is the absence of dynamical correlations in strict PGCM. In the neon-isotope work, PGCM binding energies are described as underbound by $q$31–$q$32 and rms radii as often overestimated, with the island-of-inversion physics not fully reproduced (Frosini et al., 2021). In the PGCM-PT study, strict PGCM underbinds full configuration interaction by roughly $q$33–$q$34 in small model spaces and produces low-lying spectra that are too dilated; in $q$35Ne, PGCM-PT(2) improves the first $q$36 and $q$37 excitations but still leaves them approximately $q$38–$q$39 away from FCI, indicating missing higher-rank correlations or missing generator coordinates such as octupole or triaxial modes (Frosini et al., 2021).

Several development paths are already explicit in the literature. One is the perturbative completion of PGCM through PGCM-PT, including the use of MR-IMSRG pre-processing to absorb part of the dynamical correlations and improve convergence (Frosini et al., 2021). Another is enrichment of the variational manifold through additional generator coordinates, including pairing amplitudes, quasi-particle excitations, full triaxial-octupole grids, and cranking modes (Frosini et al., 2021). A third is extension to odd-mass and odd–odd systems, now demonstrated for $q$40Ne and for GT transitions (Lin et al., 2024). The DGCM work suggests an additional direction in which conjugate collective momenta are incorporated directly into the generator-coordinate manifold (Hizawa et al., 2020).

Taken together, these results define PGCM as a symmetry-conserving, non-perturbative framework for static collective correlations rather than as a complete many-body theory by itself. Its established strengths are exact restoration of broken symmetries, flexible handling of deformation and pairing, and mean-field-like scaling with system size. Its established limitations are sensitivity to the chosen collective manifold and incomplete treatment of dynamical correlations unless supplemented by perturbative, enlarged-manifold, or Hamiltonian-preprocessing schemes.