An Introduction to the Nonperturbative Renormalization Group

Bertrand Delamotte

Source-linked AI summary

An Introduction to the Nonperturbative Renormalization Group

Bertrand Delamotte

arXiv:cond-mat/0702365v1Feb 15, 2007cond-mat.stat-mechhep-th

TL;DR

The paper introduces the non-perturbative renormalization group as an elementary framework for understanding scale-dependent physics in statistical mechanics. It explains the method’s conceptual and technical foundations, emphasizing effective actions and approximation procedures. The authors report that short derivative-expansion ansätze can yield quantitatively accurate results, while controlled RG flows remain difficult in gauge theories.

Problem
The paper addresses how to formulate and compute renormalization-group flows for interacting systems when the NPRG equation is an extremely complicated functional partial differential equation.
Method
The paper develops the effective average action method and uses approximation procedures, including derivative expansions, to study NPRG flows.
Results
Short ansätze produce qualitatively good long-distance results, while derivative-expansion series appear to converge rapidly for the three-dimensional Ising model and critical exponents reach best-known values at the stated order.
Takeaways & Limitations
NPRG can connect microphysics to macrophysics and serve as a quantitative tool for strongly correlated systems, not only a qualitative one.
Takeaways & Limitations
For gauge theories, symmetry may be broken at finite k and recovered only as k →0, and completely controlled RG flows remain difficult to compute.

Abstract

from arXiv · show

An elementary introduction to the non-perturbative renormalization group is presented mainly in the context of statistical mechanics. No prior knowledge of field theory is necessary. The aim is this article is not to give an extensive overview of the subject but rather to insist on conceptual aspects and to explain in detail the main technical steps. It should be taken as an introduction to more advanced readings.

Wilson’s renormalization group

The passage presents two lines asking what can be done or written against nightfall.

The speaker asks what can be done and what can be written.
Together, the lines connect action and writing with the onset of darkness.
The question is framed against the fall of night.

1.1 Introduction

The article introduces Wilson’s RG and its modern NPRG implementation, beginning with elementary material before developing the general framework and detailed O(N) calculations.

The article presents Wilson’s renormalization group and its concrete implementation as the non-perturbative renormalization group.
Prior knowledge of perturbative field theory is not required to understand the article’s central ideas.
The opening material is elementary and establishes a language and way of thinking about renormalization group methods.
The article compares perturbative and Wilson’s RG before presenting the Kadanoff-Wilson implementation and the modern framework.
Detailed calculations are given for O(N) models within the modern implementation of Wilson’s ideas.

1.2 The perturbative method in field theory

The perturbative method expands an interacting theory around a solvable reference model, but fluctuation integrals can become large, cutoff-sensitive, and divergent. Renormalization reorganizes the expansion to remove this cutoff dependence and recover universal behavior.

Perturbation theory adds interaction terms to a solvable Gaussian or mean-field reference model through a perturbative expansion.
The expansion generates Feynman diagrams for Green functions, with loop integrals summing fluctuations across wavelengths.
The ultraviolet cutoff Λ regulates these integrals, which can diverge as Λ →∞ and remain strongly dependent on Λ when it is finite but large.
Cutoff-sensitive integrals can invalidate the perturbation expansion even when the bare interaction u0 is small.
Perturbative renormalization reparametrizes the expansion to eliminate sensitive dependence on Λ, while the renormalization group partially resums it to compute universal behavior.
The discussion raises how ultraviolet divergences relate to fluctuations, field products, continuum limits, and the infrared usefulness of fixed points.

1.3 Coarse-graining and effective theories

Wilson’s RG organizes fluctuations by scale, integrating out short-distance modes to construct effective theories for long-distance behavior. Its coupling flows provide a framework for fixed points and avoid the usual summation over all length scales, although exact integration is generally impossible.

Strongly correlated systems near criticality contain fluctuations across wavelengths between the microscopic scale a ∼Λ−1 and the correlation length ξ.
Wilson’s method integrates out short-distance or rapid modes to build an effective theory for the long-distance degrees of freedom.
The rapid modes occupy a momentum shell, while the remaining slow modes define the effective Hamiltonian at a lower scale.
Exact integration would yield an exact solution, but it is usually impossible, so the method’s practical value lies in approximation schemes beyond ordinary perturbation theory.
Integrating out rapid modes generates all couplings compatible with the system’s symmetries, even when many vanish in the initial Hamiltonian.
Because each step integrates over a momentum shell, Wilson’s method avoids summing over all length scales and requires no usual subtraction of divergences.
The scale-dependent coupling flow contains information about the system, including the fixed points central to statistical mechanics.

1.4 Renormalization group transformations

The renormalization-group transformation coarse-grains blocks of spins, integrates out short-distance degrees of freedom, and rescales the lattice while preserving long-distance physics. Its formulation requires generalized couplings because block transformations generate infinitely many interactions, while approximations are safest for small blocks or infinitesimal steps.

Blocks of spins: Wilson’s method illustrates RG in real space by partitioning a triangular Ising lattice into plaquettes and replacing each block with a block-spin variable.The construction separates block-spin configurations from short-distance spin configurations inside each plaquette.
Blocks of spins: The majority rule keeps the block spin Ising-valued but makes its relation to microscopic spins nonlinear and difficult to generalize.The text identifies this nonlinearity as a source of difficulty beyond the triangular-lattice Ising model and for continuous, N-component spins.
Blocks of spins: Integrating out short-distance spins defines an effective Hamiltonian, but even a nearest-neighbor Hamiltonian generates infinitely many interaction terms.Consequently, a finite-coupling Hamiltonian is not stable under block-spin transformations.
Blocks of spins: A linear block-spin transformation and a Hamiltonian containing all Z2-symmetric couplings restore a form-invariant description, although the block variable is no longer an Ising spin.The enlarged coupling vector includes all interactions generated by integrating out fluctuations.
RG transformations: After each coarse-graining step, rescaling the lattice spacing together with the transformation produces an RG transformation that preserves the partition function and long-distance physics.The construction also preserves singularities, critical behavior, and thermodynamical quantities according to the text.
RG transformations: The full microscopic calculation contains more information than needed for universal critical behavior, while qualitative RG-flow behavior can predict non-trivial behavior near second-order transitions.Near a fixed point, coarse-grained trajectories approach a regime where power laws and universal critical exponents can be obtained; approximations remain controlled for small blocks or infinitesimal steps.

1.5 Properties of the RG flow: fixed points, critical surface, relevant directions

The RG flow organizes critical behavior through a stable critical surface and fixed points. Near a fixed point, eigendirections produce power-law scaling, universality, and relations among critical exponents.

Critical surface: The RG flow acts in the coupling-constant space, where the critical surface consists of systems with infinite correlation length and has co-dimension one.For a second-order transition, one parameter, such as temperature, must be fine-tuned to reach criticality.
Fixed points and universality: RG trajectories on the critical surface can converge to a fixed point whose basin of attraction contains systems sharing the same long-distance physics.A physical temperature line is distinct from an RG trajectory, which preserves the partition function rather than representing a direct physical transformation.
Fixed points and universality: The fixed point depends on the chosen RG transformation, while its role is that of a particular critical point governing the associated long-distance behavior.Different points on one RG trajectory may be microscopically different yet have the same partition function and long-distance physics.
Consequences: A fixed point explains universal critical exponents, scaling of thermodynamic quantities, exponent relations, and the irrelevance of infinitely many couplings.The correlation length’s temperature scaling drives the scaling of many other thermodynamic quantities.
Relevant directions: Near the fixed point, RG flow follows power laws along eigendirections, with logarithmic behavior when an eigenvalue is zero.The temperature has a nonvanishing projection onto a relevant eigendirection, although it need not coincide with that eigendirection.
Example and approximations: In the two-dimensional Ising example, RG results can be systematically improved by retaining more couplings, with rapid improvement at the next approximation orders.The practical RG construction uses block spins and truncates the infinite-dimensional space of symmetry-preserving couplings.

The non-perturbative renormalization group

The passage addresses an interlocutor as foolish for believing that the speaker is not identical with them.

The speaker calls the interlocutor foolish.
The speaker rejects the belief that they are not the interlocutor.
The line asserts an identity between speaker and interlocutor.

2.1 Introduction

The section introduces NPRG through Kadanoff–Wilson coarse-graining, contrasting Wilson–Polchinski and effective average action formulations while developing the former's basic construction. It explains mode separation, the running potential, and limitations that motivate the effective average action approach.

NPRG implementations share block spins, coarse-graining, and effective long-distance theories but differ substantially in formulation.
2.1.1 The Wilson-Polchinski approach: The Wilson–Polchinski construction separates rapid and slow Fourier modes, with slow modes representing block spins and rapid modes representing fluctuations within blocks.The split is implemented using complementary propagators and Gaussian integration over rapid modes.
2.1.1 The Wilson-Polchinski approach: Integrating out rapid modes defines a scale-dependent potential Vk, initialized by VΛ = V, whose evolution obeys the Wilson–Polchinski equation.Although called a potential, Vk generically contains derivative terms and arbitrary powers of field derivatives once k < Λ.
2.1.1 The Wilson-Polchinski approach: The slow field at finite k is a stochastic spatially averaged mode rather than a thermal average or straightforward precursor of the order parameter.At k = 0, the zero-momentum mode is not itself the magnetization; the true magnetization is its thermal average.
Block-spins, coarse-graining, Legendre transform, etc: The Wilson–Polchinski potential flow omits rapid-mode correlation functions, whereas the effective average action packages RG flow, fixed points, and correlation-function information in Γk[M].Recovering correlations across the full momentum range requires following an additional source term; the effective average action is presented as mathematically equivalent and more convenient for approximations.
2.1.1 The Wilson-Polchinski approach: Early practical use of NPRG was limited by beliefs that its approximations were uncontrolled and that infinitely many couplings required numerical methods, although crude truncations can remain analytically tractable.The framework has nevertheless been used to study O(N) models in all dimensions, including two, and to recover one-loop results in several limits.

2.2 The exact RG equation and its properties

The exact RG equation governs the scale evolution of the effective average action and has a one-loop-like diagrammatic structure. Its regulator produces regular flows, decoupling, symmetry constraints, and fixed-point-based universality.

Exact RG equation: The RG equation ∂kΓk=f(Γk), also called Wetterich’s equation, is derived by first obtaining an evolution equation for Zk.Its diagrammatic form contains the full propagator, ∂kRk(q), and an integral over q.
Properties: If the regulator respects a microscopic symmetry, Γk preserves it for all k; gauge symmetry is a stated exception because it is recovered only as k → 0.Modified Ward identities control the finite-k symmetry-breaking term, but completely controlled gauge-theory flows remain difficult.
Exact RG equation: The equation resembles a one-loop result, but the M-dependent propagator makes that one-loop structure exact.Figures 2.3 and 2.4 provide the corresponding diagrammatic representations.
Properties: For k > 0, the infrared regulator keeps Γk regular, so singularities build up only as k is lowered.Critical behavior can already appear at finite k for momenta |q| ≫ k.
Properties: Dimensionless and renormalized variables remove explicit dependence on k and Λ, enabling fixed-point searches and preserving the RG composition law.This self-similarity supports the analysis of critical exponents and universality.
Properties: Modes with mass m1 decouple once k ≪ m1 because only momenta |q| ≤ k contribute to the flow.Consequently, low-energy flows contain almost no signal of sufficiently massive modes.

2.3 Approximation procedures

The exact RG equation is a functional partial differential equation that generally requires projection onto a restricted functional space. Green-function truncations and derivative expansions provide complementary approximation strategies.

General strategy: The exact NPRG equation is generally unsolvable because it is a functional partial differential equation involving Γk[M] and Γk^(2)[M].Approximations solve it in a restricted functional space rather than as a series in a small parameter.
The Green function approach: The Green-function approach derives an infinite hierarchy of RG equations for correlation functions and closes it by truncating higher-point or field-dependent terms.Momentum dependence can be retained while field dependence is truncated.
The derivative expansion: The derivative expansion keeps low orders in field gradients while retaining all orders in the field, targeting long-distance correlation functions.The local potential approximation omits field renormalization, while O(∂2) includes the next derivative order.
The derivative expansion: The effective potential Uk=0 contains important statistical-mechanics information, including the equation of state, so it must be computed accurately.This motivates retaining all field orders in the derivative-expansion potential when feasible.
Combined truncations: Field expansions can be combined with the derivative expansion, producing ordinary differential equations for the retained couplings.The simplest example uses the LPA and the first two terms of the potential’s expansion in M.
Combined truncations: Keeping scale-dependent couplings in a two-term potential ansatz captures almost all qualitative critical features of Ising and O(N) models in all dimensions.Without scale dependence, the same ansatz would reduce to Landau’s mean-field approximation.

2.4 The local potential approximation for the Ising model

The local potential approximation projects the RG equation onto the effective potential for uniform fields and yields a numerically tractable flow. Dimensionless variables are required to identify fixed-point potentials.

Flow equation: For a Z2-symmetric theory, the potential Uk is defined by evaluating Γk on uniform field configurations.The flow is obtained by acting with ∂t on this definition and evaluating the right-hand side through the exact RG equation.
Flow equation: The LPA makes the required inverse of Γk^(2)[M]+Rk explicit and produces a closed flow equation for the potential.The truncation is what enables this calculation for uniform fields.
Cutoff dependence: Only a momentum window around k contributes appreciably to the flow for the typical cutoff, efficiently integrating out rapid modes.This behavior explains the decoupling of massive modes.
Dimensionless flow: The dimensionful potential flow has no fixed-point potential, so the analysis must first transform to dimensionless variables.The dimensionless flow combines canonical scaling of Uk and ρ with the model’s dynamical contribution.
Dimensionless flow: The dimensionless potential equation is a partial differential equation that can be integrated numerically to study Ising critical behavior and search for fixed points.The transformation measures lengths in units of the running lattice spacing k^-1.

2.5 The critical and non-critical behavior of the Ising model within the LPA

Within the LPA, fluctuations determine whether the running potential describes broken, critical, or symmetric behavior, while the critical point corresponds to a unique globally well-defined fixed-point solution.

RG flow of the potential: A polynomial microscopic potential is not preserved under RG flow because all Z2-invariant couplings are generated, producing a non-polynomial potential at other scales.This non-polynomial evolution is compatible with the NPRG framework.
RG flow of the minimum: Fluctuations can drive a positive initial minimum κΛ to zero, changing the mean-field broken-phase assignment into the symmetric high-temperature phase.The running minimum, rather than its initial value alone, determines the phase in this analysis.
Non-critical phases: In the broken phase, κ(k) diverges as k^(2−d) for d > 2 while the dimensionful minimum approaches the spontaneous magnetization; in the high-temperature phase, κ(k) reaches zero at a finite scale of order ξ^-1.The finite stopping scale reflects when the coarse-grained system detects its finite correlation length.
Critical surface: The critical surface has co-dimension one because one parameter, such as temperature, must be fine-tuned to obtain a second-order critical point, and it is RG-stable.The critical initial value κc depends on the remaining couplings.
Critical point: At criticality, the minimum remains nonzero for every k > 0 and reaches the origin only as k → 0, so fluctuations at all wavelengths are required.The rescaled potential can nevertheless retain a double-well structure even when the dimensionful effective potential has its minimum at the origin.
Critical fixed point and exponents: Among the solutions of the fixed-point differential equation, only one is well defined for all ρ̃ ∈ [0, ∞[, matching the expected three-dimensional Ising universality-class fixed point.The LPA also yields separately computed critical exponents whose scaling relations are very well satisfied, with ν compared against Monte Carlo results ν = 0.6297(5).

2.6 Perturbative renormalizability, RG flows, continuum limit, asymptotic freedom and all that. . .

Wilson’s RG changes renormalizability from a classification of couplings into a picture of flows through coupling space. Long-distance trajectories rapidly approach a low-dimensional “large river,” while NPRG can retain transient and nonuniversal physics that perturbation theory omits.

The cutoff is a fundamental scale where new physics may occur, rather than merely an unphysical regulator that must always be removed.
Renormalizable couplings and large river effect: Once trajectories approach L, all couplings remain present but become functions of the renormalizable coupling, making the theory predictive from finitely many renormalized inputs.
Renormalizable couplings and large river effect: At long distances, RG trajectories in the infinite-dimensional coupling space are attracted to a low-dimensional submanifold; for critical Z2-invariant theories, this becomes a one-dimensional line L.
Renormalizable couplings and large river effect: The flow rapidly collapses onto L and then evolves slowly from the Gaussian fixed point toward the Wilson-Fisher fixed point, so long-distance behavior acts as if driven by one coupling.
Universality: NPRG can follow arbitrary trajectories from a finite cutoff to k = 0 and compute nonuniversal quantities, with accuracy limited by the chosen truncation of Γ[M].
Continuum limit and asymptotic freedom: In d = 4, the Wilson-Fisher fixed point merges with the Gaussian one, the line L disappears, and no continuum limit exists except for the Gaussian model.

2.7 The O(N) models at O(∂2) of the derivative expansion

The O(N) extension of the NPRG introduces Goldstone-mode contributions and a field renormalization within the O(∂2) derivative expansion. Dimensionless renormalized variables enable fixed-point analysis, while truncated flows reproduce controlled results across several limits and describe special transitions.

The O(N) model includes Goldstone modes, the Mermin-Wagner theorem in d = 2, and the Kosterlitz-Thouless transition for N = 2.
At O(∂2), the LPA’ neglects Yk(M^2) and retains the first field-expansion term of Zk(M^2).
The O(N) RG equation differs from the Ising case through an O(N)-index trace and a term proportional to N − 1 representing Goldstone-boson physics.
Dimensionless renormalized variables remove explicit k- and Zk-dependence from the equation for the dimensionless potential.
The truncated NPRG ansatz retrieves one-loop results near d = 4 and d = 2 and the leading large-N limit, providing an interpolation between these regimes.
For N = 2 in d = 2, the approach qualitatively reproduces the Kosterlitz-Thouless transition through a line of slowly flowing quasi-fixed points.

2.8 Other fields of application of the NPRG in statistical mechanics

The NPRG has been applied to diverse statistical-mechanical systems, including frustrated magnets, nonequilibrium reactions, disorder, Lifshitz points, nucleation, fluids, and fermionic crossovers. These applications provide predictions, quantitative comparisons, or results where perturbative methods were insufficient.

For frustrated magnetic systems in d = 3, NPRG predicts generically weak first-order transitions for N = 2 and N = 3, contrasting with perturbative fixed-point predictions.
In a reaction-diffusion model with particle decay, annihilation, diffusion, and birth, NPRG predicts a continuous transition in all dimensions when µ = 0 and σ ≠ 0.
For parity-conserving reaction-diffusion systems, NPRG finds the fixed point and obtains critical exponents in good agreement with numerical data.
NPRG calculations address Lifshitz points, bubble nucleation, Bose-Einstein crossover, and simple fluids, including quantitative comparisons with experiments and critical-temperature calculations.

2.9 Conclusion

The conclusion presents NPRG as a framework linking microscopic and macroscopic physics while retaining a small number of effective couplings. It reports controlled results from short ansätze and rapid derivative-expansion convergence, alongside a limitation in momentum-dependent correlations.

NPRG continuously relates microphysics to macrophysics and explains how infrared behavior can involve only finitely many coupling constants.
Short NPRG ansätze reproduce one-loop results near upper and lower critical dimensions and at large N.
In three-dimensional Ising calculations, derivative-expansion results appear to converge rapidly, with critical exponents reaching best-known values without resummation.
The derivative expansion is inadequate for calculating momentum dependence and is applicable only when external momenta are below the running scale k.

Appendix

The appendix defines Fourier-space conventions, cutoff and threshold functions, correlation functions, and relations between Wilsonian and effective-average-action formulations. It also derives the flow equations used for the dimensionless potential.

Cutoff functions are specified in real and Fourier spaces, with care required over whether Rk is treated as a function of q or q^2.
The appendix defines 1PI correlation functions as functional derivatives of Γ[M] and relates their Fourier-space conventions.
Threshold functions are introduced, and selected forms become analytically computable for a convenient θ-cutoff.
The appendix derives the relation between Wk and Γk and shows that the regulated second derivative of Γk is the inverse of Wk^(2) in the operator sense.
Within the local potential approximation, Uk is obtained from Γk on uniform fields and is converted to a dimensionless potential through a change of variables.