Approximate Bosonization in Quantum Many-Body Systems

• Approximate bosonization is a framework that recasts fermionic, anyonic, or gauge-coupled many-body systems into bosonic effective models through controlled limits and truncations.
• It employs techniques such as mean-field scaling, quadratic truncation, derivative expansions, and large-N limits to capture collective modes and self-generated gauge interactions.
• These methods are applied in diverse regimes—from two-dimensional anyon systems to three-dimensional Fermi gases and 2+1D dualities—offering rigorous control and insights into domain validity and limitations.

Approximate bosonization denotes a family of controlled reductions in which a fermionic, anyonic, or gauge-coupled many-body problem is represented by bosonic degrees of freedom only in a specified limit, sector, or truncation. In the literature this includes the average-field description of two-dimensional extended anyons in an almost-bosonic scaling regime (Lundholm et al., 2015), quasifree bosonic dynamics for collective particle-hole excitations of a three-dimensional Fermi gas (Benedikter et al., 2021), boson bilinear Hamiltonians for the $\mathbf Q=0$ neutral continuum of Dirac fermions (Mantilla et al., 2020), scale-dependent bosonic variables in nonperturbative FRG treatments of Luttinger liquids (Daviet et al., 2021), and several $2+1$-dimensional duality constructions in which exact lattice identities, infrared matching, large-mass expansions, or higher-spin Ward identities produce bosonic descriptions that are exact only in special regimes and approximate elsewhere (Turker et al., 2020, Santos et al., 2019, Li, 2019). The common feature is that fermionic or anyonic statistics are not discarded; they are reorganized into currents, gauge fields, collective modes, or bosonic effective actions.

1. Forms of approximation

Approximate bosonization is not a single procedure. In the almost-bosonic anyon problem one takes

$\alpha=\frac{\beta}{N-1},\qquad \beta>0,$

with $N\to\infty$ and $R=R(N)\to0$ not too fast, so that the gauge interactions are weak enough that genuine bosonic condensation still occurs, but strong enough to leave a finite mean-field imprint (Lundholm et al., 2015). In the three-dimensional Fermi-gas construction, localized particle-hole pair operators satisfy approximate canonical commutation relations on few-excitation states because the Pauli principle becomes negligible when the number of excitations is $o(N^{2/3})$ (Benedikter et al., 2021). In the $\mathbf Q=0$ Dirac problem, neutral interband bilinears are treated as bosons after projection onto the singly occupied sector and truncation to quadratic order (Mantilla et al., 2020).

Other implementations are field-theoretic. In FRG-based flowing bosonization, the approximation is a derivative expansion of the scale-dependent effective action to second order, together with a scale-dependent linear redefinition of the phase field $\vartheta$ so that $\varphi$ and $\vartheta$ remain conjugate at all scales (Daviet et al., 2021). In higher-dimensional Fermi-surface bosonization, the long-wavelength limit replaces the Moyal bracket by a Poisson bracket, while RPA appears when one truncates the exact bosonic action to its quadratic piece plus the two-point density-density interaction (Park et al., 2023). In the $2+1$0 WZW formulation of bosonized Fermi surfaces, the exact description holds in a particular $2+1$1 limit, and deviations are organized as a tower of irrelevant corrections suppressed by powers of $2+1$2 (Chen et al., 8 Oct 2025).

This variety of mechanisms is essential. The word “approximate” may refer to mean-field scaling, low-energy matching, quadratic truncation, large-mass expansion, low-excitation sectors, long-wavelength limits, or controlled $2+1$3 expansions, rather than to a single universal loss of information.

2. Average-field approximation for almost-bosonic anyons

For two-dimensional extended anyons in the magnetic gauge picture, the microscopic Hamiltonian is written on the bosonic subspace $2+1$4 as

$2+1$5

with smeared Aharonov-Bohm fluxes of radius $2+1$6. The smeared Coulomb kernel is

$2+1$7

and the two-body vector potential is

$2+1$8

with $2+1$9 (Lundholm et al., 2015).

The almost-bosonic regime is defined by

$\alpha=\frac{\beta}{N-1},\qquad \beta>0,$0

for a certain $\alpha=\frac{\beta}{N-1},\qquad \beta>0,$1 depending on the growth of $\alpha=\frac{\beta}{N-1},\qquad \beta>0,$2 at infinity. If $\alpha=\frac{\beta}{N-1},\qquad \beta>0,$3 is fixed and $\alpha=\frac{\beta}{N-1},\qquad \beta>0,$4, one recovers ordinary bosons. If instead $\alpha=\frac{\beta}{N-1},\qquad \beta>0,$5 and $\alpha=\frac{\beta}{N-1},\qquad \beta>0,$6, the singular two- and three-body gauge interactions each contribute at leading order $\alpha=\frac{\beta}{N-1},\qquad \beta>0,$7 to the energy per particle. Under the ansatz of complete Bose-Einstein condensation,

$\alpha=\frac{\beta}{N-1},\qquad \beta>0,$8

the energy per particle converges formally to the average-field functional

$\alpha=\frac{\beta}{N-1},\qquad \beta>0,$9

where

$N\to\infty$0

so that

$N\to\infty$1

The main theorem states that, for a confining potential $N\to\infty$2 with sufficiently rapid growth at infinity and $N\to\infty$3 not too fast,

$N\to\infty$4

Moreover, if $N\to\infty$5 is a ground state of $N\to\infty$6 and $N\to\infty$7 its $N\to\infty$8-particle reduced density matrix, then for each fixed $N\to\infty$9,

$R=R(N)\to0$0

In the formulation of the paper, one thereby “bosonizes” the system approximately: the limiting theory is just ordinary bosons moving independently in a self-consistent magnetic field of strength $R=R(N)\to0$1, and the statistics of anyons becomes encoded in a self-generated gauge field (Lundholm et al., 2015).

3. Collective particle-hole bosons in fermionic many-body systems

In the rigorous treatment of fermionic many-body dynamics on the three-dimensional torus, the starting Hamiltonian on fermionic Fock space is

$R=R(N)\to0$2

in the coupled mean-field / semiclassical regime

$R=R(N)\to0$3

After the particle-hole transform $R=R(N)\to0$4, the correlation Hamiltonian contains a bosonizable quartic part $R=R(N)\to0$5 built from delocalized pair operators

$R=R(N)\to0$6

A patch decomposition of the Fermi-surface shell with

$R=R(N)\to0$7

produces localized pair operators $R=R(N)\to0$8 satisfying the approximate CCR

$R=R(N)\to0$9

on few-excitation states (Benedikter et al., 2021).

The quadratic bosonized Hamiltonian consists of a linearized kinetic part $o(N^{2/3})$0 and an interaction part already quadratic in the $o(N^{2/3})$1 variables. Each finite-dimensional quadratic block is diagonalized by a bosonic Bogoliubov transformation $o(N^{2/3})$2, yielding the RPA ground-state energy

$o(N^{2/3})$3

and an effective free boson Hamiltonian

$o(N^{2/3})$4

For initial data consisting of $o(N^{2/3})$5 collective excitations on top of the Bogoliubov-rotated sea, the norm-approximation theorem states

$o(N^{2/3})$6

where $o(N^{2/3})$7 evolves under $o(N^{2/3})$8. The paper describes this as a rigorous version of approximate bosonization and as the first non-perturbative, norm-level justification of the Random Phase Approximation for the dynamics of a three-dimensional Fermi gas (Benedikter et al., 2021).

A related construction for the $o(N^{2/3})$9 continuum of two-dimensional Dirac fermions projects onto the single-occupancy subspace, where each singly occupied momentum state is equivalent to an interband particle-hole pair of total momentum $\mathbf Q=0$0. After a Holstein-Primakoff mapping, or directly through

$\mathbf Q=0$1

one obtains a boson-bilinear Hamiltonian

$\mathbf Q=0$2

Its exact diagonalization is exactly equivalent to summing the infinite SCHF-Kadanoff-Baym ladder for the particle-hole propagator $\mathbf Q=0$3. Applied to optical conductivity, the construction reproduces the perturbative RG coefficient

$\mathbf Q=0$4

and, after numerical diagonalization up to $\mathbf Q=0$5, finds only $\mathbf Q=0$6 deviations from $\mathbf Q=0$7 (Mantilla et al., 2020).

4. Dualities and bosonization in $\mathbf Q=0$8 dimensions

One line of work formulates bosonization through loop representations and Chern-Simons couplings. For lattice Wilson fermions on a cubic lattice, the exact partition function can be written as a sum over closed, nonbacktracking loops $\mathbf Q=0$9, each carrying a spin-factor $\vartheta$0, where $\vartheta$1 is the writhe. In the low-energy sector, restricted to single-winding loops, the continuum-like form is

$\vartheta$2

For the Villain form of the Abelian Chern-Simons-Higgs model, duality maps the theory to closed vortex loops, each with phase $\vartheta$3 and Boltzmann weight $\vartheta$4. At $\vartheta$5 and single winding,

$\vartheta$6

and after identifying $\vartheta$7 the two partition sums coincide exactly: $\vartheta$8. Away from the exact lattice mapping, the same work describes approximate bosonization in the infrared through large-$\vartheta$9 or large-$\varphi$0 expansions, weak gauge coupling, suppression of amplitude fluctuations, and matching of induced Chern-Simons and Maxwell terms (Turker et al., 2020).

The bosonization of the $\varphi$1-dimensional Thirring model uses a deformation of a basic fermion-boson duality relation. The bosonized model interpolates between the ultraviolet and infrared regimes, and the current map is

$\varphi$2

In the formulation summarized for the single-species case, approximate control is available in the critical $\varphi$3 regime, in the large-mass expansion $\varphi$4, and in the strong-coupling limit on one side versus a weak Maxwell sector on the other. The duality predicts the existence of a nontrivial ultraviolet fixed point in the Thirring model, and the strict $\varphi$5 limit recovers the Fradkin-Schaposnik mapping to Maxwell-Chern-Simons theory (Santos et al., 2019).

A different notion of approximate bosonization appears in Chern-Simons vector models with slightly broken higher-spin symmetry. There the approximate higher-spin Ward identity for a non-conserved spin-4 current $\varphi$6 yields a bootstrap equation that splits into parity-even and parity-odd sectors under $\varphi$7. Writing

$\varphi$8

one obtains a triangular recursion in the ’t Hooft coupling $\varphi$9, and the recursion closes at second order: $\vartheta$0 for all $\vartheta$1. The $\vartheta$2 odd term in the spinning four-point function is shown to coincide with the Legendre transform of the free-boson correlator, providing new evidence for $\vartheta$3 bosonization at the spinning four-point level (Li, 2019).

5. Flowing, microscopic, and exact bosonic formulations

In one-dimensional quantum fluids, bosonization is usually formulated in terms of canonically conjugate fields $\vartheta$4 and $\vartheta$5. The FRG construction starts from the Euclidean action

$\vartheta$6

and a scale-dependent effective action $\vartheta$7 satisfying Wetterich’s equation. A derivative expansion to second order reproduces the sine-Gordon flow in the $\vartheta$8-sector, but, if $\vartheta$9 is held fixed, it fails to capture the renormalization of the superfluid stiffness. The remedy is a scale-dependent linear redefinition

$2+1$00

chosen so that the mixed term remains exactly $2+1$01 and no $2+1$02 term arises at $2+1$03. Only with this flowing reparametrization does one obtain the standard phenomenology of the Luttinger liquid, characterized by the sound velocity and the renormalized Luttinger parameter $2+1$04 (Daviet et al., 2021).

Microscopic bosonization of one-dimensional band structures extends the method away from the usual low-energy window. Snyman and Florens reformulate the kinetic term in the energy basis, extend the spectrum by spectator fermions, and define bosonic modes

$2+1$05

The free kinetic energy then becomes

$2+1$06

Local one-body scattering is mapped to exponentials of bosonic fields, and expanding those exponentials generates an infinite hierarchy of anharmonic boson-boson interactions even in the forward channel. At weak coupling a first-order coherent-state displacement reproduces the x-ray edge power law; at stronger coupling a variational coherent-state ansatz, and then a superposed ansatz, gives results that agree quantitatively with exact diagonalization. The paper reports near-perfect agreement up to $2+1$07 and a computational scaling of $2+1$08 versus $2+1$09 for exact fermionic methods (1705.01280).

For non-interacting fermions with a Fermi surface in arbitrary dimensions, Park and Balents derive an exact bosonic action by coupling a source to the one-body density matrix, performing a Legendre transform, and parametrizing

$2+1$10

with a Hermitian field $2+1$11 that mixes occupied and unoccupied states. The resulting action

$2+1$12

is exact, and the physical density correlators are obtained by summing only tree-level $2+1$13-diagrams. In the long-wavelength limit, the Moyal bracket reduces to a Poisson bracket and the action agrees with the form obtained by Delacretaz and collaborators. When a density-density interaction is added, truncating the bosonic action to its quadratic part plus the two-point interaction produces exactly the RPA series, with

$2+1$14

The same framework reproduces the three- and four-point density correlators of the fermionic theory (Park et al., 2023).

The quantization problem for bosonized Fermi surfaces is treated differently in the $2+1$15 WZW approach. There the semiclassical phase-space field is embedded into a matrix-valued description in which a general Fermi surface is exactly described by a particular $2+1$16 limit of a $2+1$17 WZW model plus a tower of irrelevant corrections. The noncommutative phase-space algebra is represented by the $2+1$18 algebra in the ’t Hooft basis, recovering $2+1$19 as $2+1$20. The formulation resolves the naive overcounting of patch bosons: although there are $2+1$21 bosonic components $2+1$22, the exact central charge is

$2+1$23

so only $2+1$24 of the would-be $2+1$25 modes remain gapless. In this language, cutting the Fermi surface into patches emerges as selecting a maximal torus $2+1$26 inside $2+1$27, described as “abelianization,” rather than as a fundamental necessity (Chen et al., 8 Oct 2025).

6. Validity regimes, limitations, and recurrent issues

A recurrent misconception is that bosonization is either exact everywhere or merely heuristic everywhere. The cited literature instead divides sharply between exact statements and controlled approximations. The average-field limit for extended anyons is rigorous only when $2+1$28, $2+1$29, and $2+1$30 not too fast (Lundholm et al., 2015). The norm-approximation theorem for fermionic many-body dynamics holds for a specific class of low-energy initial data built from collective particle-hole excitations and uses compact support and positivity assumptions on $2+1$31 (Benedikter et al., 2021). The $2+1$32 Dirac construction assumes sector decoupling, bilinear truncation of the Holstein-Primakoff expansion, and self-consistent Hartree-Fock rather than vertex corrections beyond the Kadanoff-Baym ladder (Mantilla et al., 2020).

A second recurrent issue concerns the role of patching and collective variables. In some approaches patches are a technical device required to make approximate bosonic commutators transparent, as in the three-dimensional Fermi gas (Benedikter et al., 2021). In the $2+1$33 WZW approach, by contrast, patching is not fundamental: it reappears only as abelianization inside a matrix-valued theory that already resolves the overcounting problem (Chen et al., 8 Oct 2025). This suggests that different bosonizations may organize the same collective physics with very different auxiliary structures.

A third issue is whether bosonic fields can be kept fixed under coarse-graining. The FRG analysis shows that this fails for the phase field $2+1$34: without a flowing reparametrization one would obtain an unrenormalized superfluid stiffness $2+1$35, in conflict with known superfluid physics. After the passive frame change, the infrared action becomes the familiar two-parameter Luttinger-liquid action with $2+1$36 (Daviet et al., 2021). In this setting, approximate bosonization is inseparable from the choice of renormalization variables.

Finally, the literature distinguishes carefully between exact dualities and approximate infrared matching in $2+1$37 dimensions. The loop-writhe identity between lattice fermions and Abelian Chern-Simons-Higgs bosons is exact only for the specified lattice theory, $2+1$38, and single-winding loops; the continuum correspondence then uses large-mass or Higgs-phase approximations (Turker et al., 2020). The bosonization of the Thirring model is under control in the critical, large-mass, and strong-coupling regimes rather than as an exact analogue of $2+1$39-dimensional bosonization (Santos et al., 2019). In higher-spin Chern-Simons vector models, the word “approximate” refers instead to slightly broken higher-spin symmetry and to a bootstrap recursion that nonetheless terminates at $2+1$40 (Li, 2019). Across these examples, approximate bosonization is best understood not as a weakened version of an exact dictionary, but as a structured hierarchy of bosonic reductions whose domain of validity is part of the statement itself.