Beyond first order: the fixed-point structure and eigenvalue spectrum of the Bell-Wilson nonlinear renormalization group for the Gaussian model

In 1974 Bell and Wilson introduced a nonlinear renormalization-group (RG) transformation for continuous-spin systems - a momentum-space block-spin transformation whose kernel contains a term cubic in the old spin variables - and analyzed it about the Gaussian fixed point to first order in the cubic nonlinearity parameter. They found that the unphysical (redundant) eigenvalue is shifted while the physically significant temperature eigenvalue is not, and that in one region of transformation-parameter space a useful fixed point is reached without the parameter tuning that linear transformations require. The exact kernel integrals are analytically intractable, and the analysis has never been extended beyond first order: the explicit form of the nonlinear fixed-point Hamiltonian (whose quartic couplings were only verified to be finite), the higher-order eigenvalue corrections, the boundary between the tuning-free and tuning-required parameter regions, and the borderline-dimensional behavior (the first-order redundant-eigenvalue shift has a pole at d=2, and the finiteness arguments hold only for d>2) all remain undetermined. Later literature resolved the adjacent general questions - redundant-operator eigenvalues are scheme-dependent (Wegner) and nonlinear block-spin RGs for discrete spins were developed in detail - but left this specific continuation open. The problem is to determine the beyond-first-order fixed-point structure and stability spectrum of this concretely defined transformation.

Background

Near a continuous phase transition, the renormalization group (RG) studies a system of spins on a lattice through a transformation $\mathcal{K}' = T[\mathcal{K}]$ acting on Hamiltonians, which must (i) preserve the partition function and (ii) possess a nontrivial fixed point $\mathcal{K}^* = T[\mathcal{K}^*]$ whose long-distance behavior matches that of the critical system. Linearizing the iterated transformation about the fixed point, $T[\mathcal{K}^*+\epsilon\,\delta\mathcal{K}] \approx \mathcal{K}^*+\epsilon\Lambda\,\delta\mathcal{K}$, yields eigenoperators with eigenvalues $\Lambda$; the relevant eigenvalues ($\Lambda>1$) determine critical exponents. A block-spin transformation defines new (block) spin variables $S$ in terms of the old spins $\sigma$. In a linear transformation $S$ is linearly related to $\sigma$ - for example, in momentum space, integrating out Fourier components with $|\vec q|>1/2$ and rescaling the remainder - while in a nonlinear transformation the block spin is a nonlinear function of the old spins, such as the majority rule $\sigma' = \mathrm{sgn}(\sum\sigma)$ used by Niemeijer and van Leeuwen for the two-dimensional Ising model on a triangular lattice.

Two structural features of linear transformations motivated the source paper. First, a linear block-spin kernel enforces a linear relation between the spin-spin correlation functions of $\mathcal{K}$ and $\mathcal{K}'$; at a fixed point this fixes a normalization parameter $b$ of the kernel to a special value $b^*$ (otherwise the iterated Hamiltonian trivializes), so a linear RG carries an adjustable parameter that must be tuned. Second, the fixed manifold of the linear transformation is a line parametrized by an unphysical overall spin normalization, giving a marginal operator with eigenvalue exactly 1. Wegner proved in 1974 that eigenvalues of such redundant operators - those associated with field redefinitions rather than physical perturbations - are entirely scheme-dependent, while physically significant eigenvalues are not.

Bell and Wilson asked whether nonlinearity changes this. For the Gaussian model of Berlin and Kac - the exactly solvable continuous-spin model with a purely quadratic Hamiltonian - they modified the momentum-space block-spin kernel by adding a term cubic in the old spin variables, with a momentum-conserving delta function, parametrized by $c$; the transformation is linear at $c=0$. The exact analysis being analytically intractable ("we are unable to carry out analytically the integrals required in Eq. (3.1)"), they worked to first order in $\delta c$ and $\delta b = b-b^*$, finding: (a) a fixed point near the Gaussian one is reached without tuning only within the limits established by their first-order analysis of Section III (Eq. (3.10)), which distinguishes a region I, where every critical Hamiltonian flows to a single fixed point without tuning anything, from a region II, where the spin normalization must be tuned, the precise first-order extent of the two regions being as given in the source; (b) the redundant eigenvalue is shifted from 1 to $1+(6K_d/(d-2))\,\delta c/(b^{*3}z)$, where $K_d$ is the $d$-dimensional solid-angle factor and $z$ the quadratic-coupling parameter of the reference fixed point - a shift that is unphysical and has a pole at $d=2$; (c) the physically significant temperature eigenvalue $\Lambda_0=4$ (corresponding to the Gaussian thermal exponent $y_t=2$) is unchanged at order $\delta c$; and (d) the new fixed point is no longer quadratic in the spins - it contains quartic couplings whose explicit form was never computed ("not actually been calculated, but only verified to be finite"). The finiteness arguments of their Appendix hold only for $d>2$.

Afterwards, the general conceptual questions were settled by other routes. Wegner's theorem (already cited in the source as a preprint) established the scheme-dependence of redundant eigenvalues, and the modern exact-RG literature treats nonlinear field reparameterizations as redundant directions, building schemes that eliminate them (Dietz and Morris; Baldazzi, Ben Alì Zinati and Falls, whose "essential renormalisation group" still credits Bell-Wilson 1974 for the early recognition of this physical/unphysical split). The source's suggested application direction - nonlinear transformations for discrete-spin systems like the Ising model - was developed in detail by the field through third-order cumulant expansions (systematic moment expansions of the block-spin Boltzmann weight) for majority-rule cells (Sudbø and Hemmer), Monte Carlo RG, which numerically computes the exact linearization of nonlinear block-spin transformations (Swendsen), and tensor-network RG linearizations (renormalization methods acting on a tensor representation of the partition function; Lyu, Xu and Kawashima). Nonlinear blocking kernels for continuous spins, by contrast, were abandoned rather than made practical: continuum RG practice uses linear kernels with the reparameterization freedom absorbed into scheme choices. What no later work did is extend the specific perturbative analysis of the Bell-Wilson transformation itself beyond first order; that continuation is the unresolved target.

Problem Statement

Determine, beyond first order in the cubic-kernel nonlinearity, the fixed-point structure and stability spectrum of the nonlinear renormalization-group transformation introduced by Bell and Wilson (Physical Review B 10, 3935 (1974)) for the Gaussian model: the momentum-space block-spin transformation whose kernel (their Eqs. (2.4) and (3.1), parametrized by an overall spin-scale $a$, a linear coefficient $b$, and a cubic coefficient $c$, with $a$ to some extent superfluous) contains, in addition to the linear term $b\,\sigma_{\vec q/2}$, a term cubic in the old spin variables with a momentum-conserving delta function (linear at $c=0$), acting on the Gaussian-model Hamiltonian space with modes $|\vec q|>1/2$ integrated out and the retained modes rescaled to $|\vec q|<1$.

The single overarching objective is the complete quantitative characterization of this transformation's fixed points and linearized spectrum beyond the first order in $\delta c$ (and $\delta b$) to which the source restricted, where "beyond first order" means the next (second) order in $\delta c$ and $\delta b$. Its scientifically linked components, all of which are required, are:

1. Fixed-point Hamiltonian: compute the fixed point $\mathcal{K}^*+\delta\mathcal{K}^*$ explicitly at $c\neq 0$ - the quartic and higher spin couplings that the source only verified to be finite - and determine whether the first-order picture, in which a single fixed point replaces the line of Gaussian fixed points with a well-defined basin of attraction, survives at higher order.

2. Eigenvalue spectrum: determine, beyond first order, (i) the correction to the redundant (formerly marginal) eigenvalue, whose first-order shift $1+(6K_d/(d-2))\,\delta c/(b^{*3}z)$ is scheme-dependent, and (ii) whether the physically significant temperature eigenvalue, equal to $\Lambda_0=4$ at leading order, retains that value.

3. Region I/II boundary: locate, beyond the first-order boundary between regions I and II of the source's Eq. (3.10), the boundary in the $(b,c)$ parameter plane between the region where the iterated transformation reaches a useful fixed point without tuning (region I) and the region where the spin normalization or a transformation parameter must be tuned (region II).

4. Borderline dimensions: determine the fate of the construction for $d\le 2$, where the source's Appendix finiteness arguments fail and the first-order redundant-eigenvalue shift has a pole at $d=2$. This component lies outside the second-order-in-$\delta c$ scope of the other three: determine whether a fixed point of the assumed form exists for $d\le 2$, whether the fixed point persists to the same second order in $\delta c$ in the regime where that expansion remains defined (for example $d=2+\epsilon$), or whether an explicit resummed or non-perturbative construction valid for $d\le 2$ can be given.

References

1. T. L. Bell and K. G. Wilson, "Nonlinear renormalization groups", Physical Review B 10, 3935-3944 (1974), DOI 10.1103/PhysRevB.10.3935, https://doi.org/10.1103/PhysRevB.10.3935
2. F. J. Wegner, "Some invariance properties of the renormalization group", Journal of Physics C: Solid State Physics 7, 2098-2108 (1974), DOI 10.1088/0022-3719/7/12/004, https://doi.org/10.1088/0022-3719/7/12/004
3. Th. Niemeijer and J. M. J. van Leeuwen, "Wilson Theory for Spin Systems on a Triangular Lattice", Physical Review Letters 31, 1411-1414 (1973), DOI 10.1103/PhysRevLett.31.1411, https://doi.org/10.1103/PhysRevLett.31.1411
4. Th. Niemeijer and J. M. J. van Leeuwen, "Wilson theory for 2-dimensional Ising spin systems", Physica 71, 17-40 (1974), DOI 10.1016/0031-8914(74)90044-5, https://doi.org/10.1016/0031-8914(74)90044-5
5. T. H. Berlin and M. Kac, "The Spherical Model of a Ferromagnet", Physical Review 86, 821-835 (1952), DOI 10.1103/PhysRev.86.821, https://doi.org/10.1103/PhysRev.86.821
6. Aa. S. Sudbø and P. C. Hemmer, "Cumulant expansion in renormalization-group transformations on Ising spin systems: A third-order calculation", Physical Review B 13, 980-982 (1976), DOI 10.1103/PhysRevB.13.980, https://doi.org/10.1103/PhysRevB.13.980
7. R. H. Swendsen, "Monte Carlo renormalization-group studies of the d=2 Ising model", Physical Review B 20, 2080-2087 (1979), DOI 10.1103/PhysRevB.20.2080, https://doi.org/10.1103/PhysRevB.20.2080
8. J. S. Høye, "Study of spin-cell transformations for renormalization-group equations of spin systems", Physical Review B 12, 5219-5225 (1975), DOI 10.1103/PhysRevB.12.5219, https://doi.org/10.1103/PhysRevB.12.5219
9. T. L. Bell and K. G. Wilson, "Finite-lattice approximations to renormalization groups", Physical Review B 11, 3431-3444 (1975), DOI 10.1103/PhysRevB.11.3431, https://doi.org/10.1103/PhysRevB.11.3431
10. V. Kushnir and B. Rosenstein, "Renormalization group transformations for spin systems", Physica A: Statistical Mechanics and its Applications 221, 117-124 (1995), DOI 10.1016/0378-4371(95)00244-2, https://doi.org/10.1016/0378-4371(95)00244-2
11. V. Kushnir and B. Rosenstein, "Decimation in more than one dimension", Physics Letters B 357, 169-176 (1995), DOI 10.1016/0370-2693(95)00876-M, https://doi.org/10.1016/0370-2693(95)00876-M, arXiv:cond-mat/9506060
12. J. A. Dietz and T. R. Morris, "Redundant operators in the exact renormalisation group and in the f(R) approximation to asymptotic safety", Journal of High Energy Physics 2013(7), 64 (2013), DOI 10.1007/JHEP07(2013)064, https://doi.org/10.1007/JHEP07(2013)064
13. A. Baldazzi, R. Ben Alì Zinati and K. Falls, "Essential renormalisation group", SciPost Physics 13, 085 (2022), DOI 10.21468/SciPostPhys.13.4.085, https://doi.org/10.21468/SciPostPhys.13.4.085, arXiv:2105.11482
14. X. Lyu, R. G. Xu and N. Kawashima, "Scaling dimensions from linearized tensor renormalization group transformations", Physical Review Research 3, 023048 (2021), DOI 10.1103/PhysRevResearch.3.023048, https://doi.org/10.1103/PhysRevResearch.3.023048, arXiv:2102.08136
15. K. G. Wilson, "The renormalization group and critical phenomena", Reviews of Modern Physics 55, 583-600 (1983), DOI 10.1103/RevModPhys.55.583, https://doi.org/10.1103/RevModPhys.55.583