Consolidating the ensemble renormalization group: validity for spin glasses, extension beyond hierarchical lattices, and the random-field hierarchical model at $\rho=3/2$

The ensemble renormalization group (ERG) is a real-space renormalization-group (RG) scheme for systems with quenched disorder in which the transformation maps the distribution of couplings of an ensemble of systems of size $N$ to that of an ensemble of size $N/2$, fixing the image distribution so that a set of disorder-averaged observables stays invariant. After the method reproduced the phase diagram and critical exponents of the random-field hierarchical model (RFHM) at zero temperature (Decelle, Parisi and Rocchi, 2014), the originating paper left three issues open, addressed here as one coherent problem — consolidating the ERG into a validated, general real-space RG for quenched disordered systems. Its linked components are: (1) implementing the ERG for a spin-glass model at large transformation sizes and settling the contested question of whether its critical fixed point is marginally stable, which would invalidate exponent extraction below the upper critical dimension; (2) extending the ERG prescription to non-hierarchical (e.g., Euclidean) topologies; and (3) determining whether the RFHM has a ferromagnetic transition at its marginal decay exponent $\rho=3/2$, and its critical behavior there. All three components are open as of September 2026.

Background

Real-space renormalization-group (RG) methods transform a statistical-mechanical model on a lattice into a similar model on a coarser lattice; iterating the map reconstructs the flow of couplings and yields critical exponents such as $\nu$, which governs the divergence of the correlation length at a critical point. For systems with quenched disorder — random couplings or random fields that are fixed once and for all, with only thermal fluctuations averaged — real-space RG is notoriously delicate: transformations applied sample by sample, separately to each realization of the disorder, have been found to give incorrect critical exponents for spin glasses away from the mean-field regime. The ensemble renormalization group (ERG), introduced by Angelini, Parisi and Ricci-Tersenghi (2013), takes a different route: instead of transforming individual samples, it maps the distribution of couplings of an ensemble of systems of size $N=2^n$ to the corresponding distribution at size $2^{n-1}$, choosing the image distribution so that a chosen set of observables averaged over the disorder — typically scale-invariant ratios of correlation functions — is kept invariant. Iterating this map reconstructs an RG flow in the space of disorder distributions, from which critical points and exponents are estimated at a computational cost far below that of large-scale Monte Carlo (MC) simulation.

The random-field hierarchical model (RFHM) is the model on which the ERG has been most stringently validated. It is the Dyson hierarchical Ising ferromagnet — a chain of Ising spins with a recursively nested block structure, in which two blocks of $2^{k-1}$ spins are coupled by a uniform interaction whose strength decays with the block level $k$ as $J_k\propto 2^{-\rho k}$ — supplemented by independent Gaussian random fields of variance $h^2$ acting on each spin. The continuous exponent $\rho$ plays the role of an inverse dimension: smaller $\rho$ means longer-ranged interactions and more mean-field-like behavior. In the pure model a ferromagnetic transition occurs for $1<\rho<2$ and is mean-field-like for $\rho<3/2$. With random fields the transition survives only in the window $1\le\rho<3/2$: it is mean-field-like for $\rho<4/3$ (the analog of the upper critical dimension $D=6$ of the short-range random-field Ising model) and non-mean-field for $4/3<\rho<3/2$. The endpoint $\rho=3/2$ is the marginal case, the hierarchical analog of the lower-critical-dimension question for random-field systems, where the Imry-Ma argument — the comparison of domain-wall and random-field energies that normally fixes the lower critical dimension — ceases to decide the outcome. This picture comes from early RG work (Rodgers and Bray, 1988) and subsequent numerical studies: Parisi and Rocchi (2014) computed the critical exponents at $T=0$ in the non-mean-field window from exact ground-state computations combined with finite-size scaling (FSS), and Okuyama and Ohzeki (2025) rigorously proved that long-range order exists for $1<\alpha<3/2$ (their $\alpha$ is the same decay exponent as $\rho$) at low temperature, including zero temperature, for sufficiently weak random fields — a proof that stops strictly short of the marginal point.

Decelle, Parisi and Rocchi (2014) applied the ERG to the RFHM at $T=0$, performing the ensemble transformation for systems of up to $n=11$ levels (i.e., $2^{11}$ spins) with up to $10^7$ disorder samples at the smallest sizes, and recovered the critical line $(h/J)_c(\rho)$ and the exponent $\nu$ in excellent agreement with the independent ground-state FSS values — a validation of the method on an unfrustrated disordered model. Its conclusions pose as perspectives: finding a spin-glass (SG) model for which the ERG can be implemented at large system sizes, which would also confirm the choice of matched observables made in the original ERG study (where SG transformations were limited to about $n=4$ levels, with exponent errors of order 20%); and finding an implementation for other topologies, such as Euclidean lattices. The model section additionally notes that the situation at $\rho=3/2$ is 'still not clear'.

The SG question has since become a genuine controversy. Castellana (2019) re-analyzed the ERG for the hierarchical Edwards-Anderson spin glass (Ising spins with independent random couplings) and argued that the ensemble transformation has a marginally stable critical fixed point — the largest eigenvalue of the linearized flow is $\lambda_{\max}=1$, forcing the prediction $2^{1/\nu}=1$, i.e., trivially wrong exponents — and concluded that finding a real-space RG method that yields correct universal quantities for spin glasses below the upper critical dimension remains an open problem. Angelini, Parisi and Ricci-Tersenghi (2020) replied that this rests on a misinterpretation of finite-size effects and that the ERG remains sound if properly used; Castellana's reply (2020) maintained the original claim, and the exchange closed without resolution. No subsequent work has implemented the ERG for a spin glass at larger sizes, extended it to Euclidean topologies, or systematically quantified its finite-size biases.

Problem Statement

The problem is to consolidate the ensemble renormalization group (ERG) into a validated, general real-space RG method for quenched disordered systems, by resolving the three open issues left by the ERG study of the random-field hierarchical model (RFHM). The components are linked by this common objective: components (1) and (2) establish, respectively, the method's validity in frustrated systems and its transferability across lattice topologies, while component (3) is the outstanding physical question of the very model on which the method was validated.

(1) ERG for spin glasses. Identify a spin-glass model for which the ERG can be carried out at transformation sizes substantially beyond the roughly four levels of the original 2013 implementation (comparable, for instance, to the eleven levels reached for the RFHM), implement the ensemble transformation on it with an explicitly justified set of matched disorder-averaged observables, and compare the resulting critical points and universal quantities with independent Monte Carlo or exact benchmarks. This includes settling the contested stability question: determine whether the linearized ERG flow at the spin-glass critical fixed point has largest eigenvalue $\lambda_{\max}=1$ (marginal stability, in which case exponent extraction from the flow is invalid below the upper critical dimension and the 2013 spin-glass exponent estimates are artifacts of a pre-asymptotic regime) or $\lambda_{\max}=2^{1/\nu}>1$, with those estimates asymptotically correct.

(2) ERG beyond hierarchical lattices. Develop an ensemble-matching ERG prescription for a non-hierarchical topology — concretely, for a Euclidean-lattice model such as the two- or three-dimensional Edwards-Anderson spin glass or random-field Ising model — or demonstrate a definite obstruction that prevents the construction from being transferred.

(3) RFHM at the marginal point. For the RFHM with Gaussian random fields at zero temperature, determine whether a ferromagnetic phase transition exists at the marginal decay exponent $\rho=3/2$ — that is, whether the critical field strength $(h/J)_c$ is nonzero there — and, if it exists, establish its critical behavior including exponents and possible logarithmic corrections; if it does not, characterize the criticality that remains (for example, the nature of the zero-field transition and the scaling of the approach to it).

Progress on each of the components is scientifically valuable in its own right.

References

1. A. Decelle, G. Parisi, J. Rocchi, Ensemble renormalization group for the random-field hierarchical model, Phys. Rev. E 89, 032132 (2014). DOI: 10.1103/PhysRevE.89.032132; arXiv:1401.1751; https://doi.org/10.1103/PhysRevE.89.032132
2. M. C. Angelini, G. Parisi, F. Ricci-Tersenghi, Ensemble renormalization group for disordered systems, Phys. Rev. B 87, 134201 (2013). DOI: 10.1103/PhysRevB.87.134201; arXiv:1111.6869; https://doi.org/10.1103/PhysRevB.87.134201
3. M. Castellana, Real-space renormalization-group methods for hierarchical spin glasses, J. Phys. A: Math. Theor. 52, 445002 (2019). DOI: 10.1088/1751-8121/ab2f46; arXiv:1910.04521; https://doi.org/10.1088/1751-8121/ab2f46
4. M. C. Angelini, G. Parisi, F. Ricci-Tersenghi, Comment on 'Real-space renormalization-group methods for hierarchical spin glasses', J. Phys. A: Math. Theor. 53, 418001 (2020). DOI: 10.1088/1751-8121/ab8414; arXiv:1911.02323; https://doi.org/10.1088/1751-8121/ab8414
5. M. Castellana, Reply to comment on 'Real-space renormalization-group methods for hierarchical spin glasses', J. Phys. A: Math. Theor. 53, 418002 (2020). DOI: 10.1088/1751-8121/ab8671; arXiv:2102.06263; https://doi.org/10.1088/1751-8121/ab8671
6. G. Parisi, J. Rocchi, Critical exponents of the random field hierarchical model, Phys. Rev. B 90, 024203 (2014). DOI: 10.1103/PhysRevB.90.024203; arXiv:1309.7470; https://doi.org/10.1103/PhysRevB.90.024203
7. M. Okuyama, M. Ohzeki, Existence of long-range order in random-field Ising model on Dyson hierarchical lattice, J. Stat. Phys. 192, 16 (2025). DOI: 10.1007/s10955-025-03399-9; arXiv:2410.11515; https://doi.org/10.1007/s10955-025-03399-9
8. G. J. Rodgers, A. J. Bray, Critical behaviour of Dyson's hierarchical model with a random field, J. Phys. A: Math. Gen. 21, 2177 (1988). DOI: 10.1088/0305-4470/21/9/030; https://doi.org/10.1088/0305-4470/21/9/030
9. F. L. Metz, J. Rocchi, P. Urbani, Statistical mechanics of the spherical hierarchical model with random fields, J. Stat. Mech. (2014) P09018. DOI: 10.1088/1742-5468/2014/09/P09018; arXiv:1406.1539; https://doi.org/10.1088/1742-5468/2014/09/P09018
10. I. Balog, G. Tarjus, M. Tissier, Critical behaviour of the random-field Ising model with long-range interactions in one dimension, J. Stat. Mech. (2014) P10017. DOI: 10.1088/1742-5468/2014/10/P10017; arXiv:1407.4891; https://doi.org/10.1088/1742-5468/2014/10/P10017
11. D. Bachtis, Overlap renormalization group transformations for disordered systems, J. Phys. A: Math. Theor. 57, 235003 (2024). DOI: 10.1088/1751-8121/ad4c2e; arXiv:2302.08459; https://doi.org/10.1088/1751-8121/ad4c2e
12. M. C. Angelini, G. Biroli, Real space renormalization group theory of disordered models of glasses, Proc. Natl. Acad. Sci. USA 114, 3328 (2017). DOI: 10.1073/pnas.1613126114; arXiv:1604.03717; https://doi.org/10.1073/pnas.1613126114
13. M. C. Angelini, S. Palazzi, G. Parisi, T. Rizzo, Critical exponents of the spin glass transition in a field at zero temperature, Proc. Natl. Acad. Sci. USA 122, e2511882122 (2025). DOI: 10.1073/pnas.2511882122; arXiv:2502.21089; https://doi.org/10.1073/pnas.2511882122
14. M. C. Angelini, S. Palazzi, G. Parisi, T. Rizzo, Long-range spin glass in a field at zero temperature, Phys. Rev. B 114, 014410 (2026). DOI: 10.1103/kfpr-xh2l; arXiv:2602.03488; https://doi.org/10.1103/kfpr-xh2l
15. M. Castellana, C. Barbieri, Hierarchical spin glasses in a magnetic field: A renormalization-group study, Phys. Rev. B 91, 024202 (2015). DOI: 10.1103/PhysRevB.91.024202; arXiv:1503.00066; https://doi.org/10.1103/PhysRevB.91.024202
16. G. S. Hartnett, The hierarchical parity model, Physica A 617, 128679 (2023). DOI: 10.1016/j.physa.2023.128679; arXiv:2208.13316; https://doi.org/10.1016/j.physa.2023.128679
17. T. C. Lubensky, T. Temesvári, I. Kondor, M. C. Angelini, Renormalization Group in Spin Glasses, in Spin Glass Theory and Far Beyond: Replica Symmetry Breaking after 40 Years (World Scientific, 2023). DOI: 10.1142/9789811273926_0004; arXiv:2302.05292; https://doi.org/10.1142/9789811273926_0004