All-orders status of the $\phi^4$ random-field Ising model universality class at glassy terminal critical points in replica field theory

Lines of discontinuous glass transitions in constrained glass-forming systems (a finite fraction of particles frozen at equilibrium positions, a pinning field coupled to the overlap, or an adsorbing porous matrix) terminate at terminal critical points where the glass transition becomes continuous. Franz, Parisi and Ricci-Tersenghi (2013) showed at one-loop order that the replica field theory of these points maps onto the $\phi^4$ Random Field Ising Model: at the terminal point, where the leading mass $m_1\to 0$ and the leading cubic couplings coincide ($\omega_1=\omega_2$), the potentially relevant sub-leading cubic-vertex contributions cancel unexpectedly, leaving the quartic RFIM operators in control. The unresolved problem is whether this cancellation, and hence the $\phi^4$-RFIM universality class, holds to all orders in perturbation theory (or equivalently whether the $\phi^4$-RFIM fixed point is stable in the full replica field theory), and what algebraic mechanism — conjecturally a fermionic-like structure of the constrained replica fields $\chi_{ab}$ at $n\to 1$ — generates the cancellations. Adjacent cases (the quenched-potential and lower-reference-temperature pinned constructions, with $m_1>0$) were later treated at all orders and confirmed numerically, but for the degenerate equal-temperature case studied in the source paper the analysis remains at one loop.

Background

In mean-field treatments of glass formation — Mode-Coupling Theory (MCT, a self-consistent theory of the slow dynamics of dense liquids) and Random First Order Transition (RFOT) theory — the glass transition is discontinuous: the system jumps from an ergodic liquid to an amorphous glassy state, and a suitable order parameter is the overlap $Q$ between two equilibrium configurations. A standard way to study this transition is to constrain the system by coupling it to a fixed reference equilibrium configuration: through a pinning field conjugate to the overlap (the Franz-Parisi potential), by freezing (pinning) a finite fraction of the particles at positions drawn from an equilibrium configuration, or by adsorbing the liquid in a porous matrix. In such constrained systems the mean-field phase diagrams generically contain lines of discontinuous glass transitions that terminate at a terminal critical point where the transition becomes continuous — the glassy analogue of the liquid-gas critical point. Since this point also terminates a thermodynamic (not merely dynamical) transition line, it can survive in finite dimension, which makes its universality class a sharp and experimentally testable question.

The natural theoretical language is replica field theory (the replica trick: introduce $n$ copies of the system and analytically continue $n$ to a non-integer value to handle quenched disorder or constrained ensembles). The fluctuations of the overlap around its plateau value are described by an $n\times n$ matrix field $\phi_{ab}(x)$ with vanishing diagonal entries, whose effective Lagrangian near the dynamical transition retains quadratic mass terms and two cubic replica invariants (cubic combinations of the $\phi_{ab}$ that are invariant under permutation of the replica indices). Franz, Parisi, Ricci-Tersenghi and Rizzo (Eur. Phys. J. E 2011) showed that the most singular part of this cubic replica theory maps onto a $\phi^3$ theory in a quenched (spatially frozen) random magnetic field (the $\phi^3$-RFIM), whose upper critical dimension — the dimension above which interactions are irrelevant and Gaussian criticality holds — is $D=8$. At a terminal critical point the leading cubic coupling $g=\omega_1-\omega_2$ vanishes and the leading mass $m_1\to 0$, so the expectation is that the next relevant operators are the quartic replica-symmetric invariants, giving the standard $\phi^4$ Random Field Ising Model (RFIM: Ising spins in a quenched random field, upper critical dimension $D=6$).

The subtlety that makes the question nontrivial is that the components of $\phi_{ab}$ do not share a single scaling dimension (the power of length a field carries under a scale transformation). One must change basis to a soft field $\phi$, a field $\omega$, and constrained fields $\chi_{ab}$ (whose replica row and column sums vanish). Dimensional analysis then shows that, besides the quartic invariants that generate the $\phi^4$-RFIM vertices (relevant — growing under coarse-graining — below $D=6$), some sub-leading cubic invariants built on the $\chi_{ab}$ fields are also superficially relevant below $D=6$; power counting alone therefore cannot decide the universality class. Franz, Parisi and Ricci-Tersenghi (J. Stat. Mech. 2013) computed the one-loop self-energy corrections (the leading perturbative corrections to the propagator) generated by the residual cubic vertices and found unexpected cancellations: at $\omega_1=\omega_2$ both the leading corrections of order $m_1^{D/2-6}$ and the next ones of order $m_1^{D/2-5}$ vanish (they are proportional to $\omega_1-\omega_2$), the first nonvanishing contribution being of order $m_1^{D/2-4}$, which is irrelevant down to $D=4$. At one loop the $\phi^4$-RFIM mapping thus holds, and finite-size scaling (the system-size scaling of observables measured in simulation) of a diluted 3-spin model on a random graph with a fraction of frozen spins confirmed RFIM critical behavior numerically.

Subsequent work clarified the neighboring cases. Franz and Parisi (J. Stat. Mech. 2013, P11012) analyzed the annealed potential (Ising universality class) and, at all orders of perturbation theory, the quenched potential and the pinned construction with reference temperature below the working temperature — the cases in which the leading mass $m_1$ stays positive at criticality and only the longitudinal and anomalous modes (the replica-symmetric sectors of the fluctuation matrix) soften, while the replicon sector (the sector associated with replica-symmetry breaking) stays massive — establishing the RFIM class there. Biroli, Cammarota, Tarjus and Tarzia (Phys. Rev. Lett. 2014) independently derived the RFIM mapping for the terminal point of a liquid coupled to a reference configuration by a source $\epsilon$, where the replicated pair-overlap fields stay massive. Numerical simulations of three-dimensional liquids (Berthier and Jack 2015; Guiselin, Berthier and Tarjus 2020, 2022) found RFIM-like criticality at the overlap terminal point, although Küchler and Horbach (2024) cautioned that one popular numerical diagnostic of RFIM scaling can also arise generically. On the methodological side, Rizzo (Phys. Rev. B 2016) developed all-orders diagrammatic selection rules for the related replica theory of the dynamical crossover, and Kaviraj and Trevisani (JHEP 2022) constructed a modern renormalization-group framework for replica disordered theories (the RFIM and random-field $\phi^3$ model), explicitly proposing its extension to replica Lagrangians of glassy continuous transitions as an open direction. None of these, however, settles the degenerate case of the source paper: the equal-temperature pinned-particle terminal point, where $m_1\to 0$ and $\omega_1=\omega_2$, has been analyzed only at one loop. The authors conjectured that the cancellations reflect a fermionic-like algebraic structure of the constrained fields $\chi_{ab}$ in the $n\to 1$ limit — going beyond the identification of the replica determinant with a fermionic determinant that underlies Parisi-Sourlas dimensional reduction (the perturbative equivalence of the RFIM with a pure model in two fewer dimensions, rooted in a hidden supersymmetry) — but no such structure has been identified.

Problem Statement

Consider the replica field theory governing long-wavelength fluctuations at the terminal critical point of a constrained glassy system in which a finite fraction of degrees of freedom is frozen at positions drawn from an equilibrium configuration at the same temperature as the free degrees of freedom (the symmetric pinned-particle construction; the same replica-symmetric theory also describes higher-order glass singularities). The fluctuation field is an $n\times n$ matrix $\phi_{ab}(x)$ (indices $a,b=1,\dots,n$ label the $n$ replicas) with vanishing diagonal entries, and the effective Lagrangian in $D$ spatial dimensions is that of Franz, Parisi and Ricci-Tersenghi (J. Stat. Mech. 2013),

$\mathcal{L}=\frac{1}{2}\int d^{D}x\,\Big(\sum_{ab}(\nabla\phi_{ab})^{2}+m_{1}\sum_{ab}\phi_{ab}^{2}+m_{2}\sum_{abc}\phi_{ab}\phi_{ac}+m_{3}\sum_{abcd}\phi_{ab}\phi_{cd}\Big)-\frac{1}{6}\int d^{D}x\,\Big(\omega_{1}\,\mathrm{Tr}\,\phi^{3}+\omega_{2}\sum_{ab}\phi_{ab}^{3}\Big),$

where $\mathrm{Tr}\,\phi^{3}\equiv\sum_{abc}\phi_{ab}\phi_{bc}\phi_{ca}$, the sums over replica indices are unrestricted, and the constraint $\phi_{aa}=0$ enforces $a\neq b$ inside each field. (This is the convention of the source paper, whose bare propagator carries explicit factors $(1-\delta_{ab})$; the cubic terms stand outside the factor $\frac{1}{2}$, which fixes the normalization of the couplings $\omega_{1},\omega_{2}$.) The theory is defined at integer $n\ge 2$ and the physical replica limit $n\to 1$ is taken after the computation of observables. At the terminal critical point the leading mass vanishes, $m_{1}\to 0$ (while $m_{2}$ and $m_{3}$ remain finite), and the two cubic couplings coincide, $\omega_{1}=\omega_{2}$, so the leading cubic structure — the $\phi^3$ theory in a quenched random field that governs the dynamical transition — degenerates, and the leading allowed interactions are the quartic replica-symmetric invariants (in the convention of the source paper, $u_{1}\,\mathrm{Tr}\,\phi^{4}+u_{2}\sum_{ab}\phi_{ab}^{4}+u_{3}\sum_{abc}\phi_{ab}^{2}\phi_{ac}\phi_{cb}$), which generate the $\phi^4$ Random Field Ising Model (RFIM: Ising spins coupled to a spatially frozen random field).

The components of $\phi_{ab}$ do not share a single scaling dimension (the power of length a field carries under a scale transformation). One must change basis to a soft field $\phi$, a field $\omega$, and constrained fields $\chi_{ab}$, through $\phi_{ab}=(\phi-\frac{1}{2}\omega)(1-\delta_{ab})+U_{ab}\omega+\chi_{ab}$, where $U_{ab}=\delta_{a,\,a-(-1)^{a}}$ is a fixed band matrix of band size two and the $\chi_{ab}$ are symmetric, with vanishing diagonal, vanishing replica row sums (hence vanishing column sums), and orthogonality to $U$ ($\sum_{ab}U_{ab}\chi_{ab}=0$). Fixing the scaling dimension of the mass to one, the fields scale as $D_{\phi}=\frac{D}{4}-1$, $D_{\omega}=D_{\phi}+1$, $D_{\chi}=D_{\phi}+\frac{1}{2}$. The quartic invariants generate the $\phi^4$-RFIM vertices, of scaling dimension $\frac{D}{2}-3$ and relevant below $D=6$; but the sub-leading cubic invariants built on the constrained fields — the combination $\omega_{1}(\mathrm{Tr}\,\chi^{3}+\sum_{ab}\chi_{ab}^{3})$, of superficial dimension $\frac{D}{4}-\frac{3}{2}$ — are, in the absence of cancellations, also relevant below $D=6$, so the universality class cannot be decided by power counting. At one loop the cubic-vertex contributions to the self-energy (the leading perturbative correction to the two-point function) cancel unexpectedly at $\omega_{1}=\omega_{2}$: the leading corrections of order $m_{1}^{D/2-6}$ and the next ones of order $m_{1}^{D/2-5}$ both vanish (they are proportional to $\omega_{1}-\omega_{2}$), the first nonvanishing contribution being of order $m_{1}^{D/2-4}$, which is irrelevant down to $D=4$.

The open problem, in two tightly linked parts:

(1) Prove, or disprove, that at this terminal point the sub-leading cubic replica invariants are irrelevant to all orders in perturbation theory relative to the quartic ones — equivalently, that the $\phi^4$-RFIM fixed point is stable in the full replica field theory (all replica-symmetric and non-symmetric invariants retained), so that the $\phi^4$-RFIM universality class is exact for this replica field theory in general dimension. A disproof must exhibit the specific loop order and replica invariant for which a cancellation fails, or for which a sub-leading cubic operator is more relevant than the quartic terms, with the explicit computation.

(2) Identify the algebraic mechanism responsible for the cancellations already observed at one loop, and show that it implies the cancellations order by order.

The problem is settled when the universality class of the equal-temperature pinned terminal critical point is established beyond one loop — by an all-orders proof together with its mechanism, or by an explicit breakdown at a definite loop order and operator.

References

1. S. Franz, G. Parisi, F. Ricci-Tersenghi, Glassy critical points and the random field Ising model, J. Stat. Mech. (2013) L02001, DOI 10.1088/1742-5468/2013/02/l02001, https://doi.org/10.1088/1742-5468/2013/02/l02001 (arXiv:1203.4849).
2. S. Franz, G. Parisi, Universality Classes of Critical Points in Constrained Glasses, J. Stat. Mech. (2013) P11012, DOI 10.1088/1742-5468/2013/11/P11012, https://doi.org/10.1088/1742-5468/2013/11/P11012 (arXiv:1307.4955).
3. G. Biroli, C. Cammarota, G. Tarjus, M. Tarzia, Random-Field-like Criticality in Glass-Forming Liquids, Phys. Rev. Lett. 112, 175701 (2014), DOI 10.1103/PhysRevLett.112.175701, https://doi.org/10.1103/PhysRevLett.112.175701 (arXiv:1309.3194).
4. S. Franz, G. Parisi, F. Ricci-Tersenghi, T. Rizzo, Field Theory of Fluctuations in Glasses, Eur. Phys. J. E 34, 102 (2011), DOI 10.1140/epje/i2011-11102-0, https://doi.org/10.1140/epje/i2011-11102-0 (arXiv:1109.4759).
5. T. Rizzo, Dynamical Landau theory of the glass crossover, Phys. Rev. B 94, 014202 (2016), DOI 10.1103/PhysRevB.94.014202, https://doi.org/10.1103/PhysRevB.94.014202 (arXiv:1603.02925).
6. L. Berthier, R. L. Jack, Evidence for a Disordered Critical Point in a Glass-Forming Liquid, Phys. Rev. Lett. 114, 205701 (2015), DOI 10.1103/PhysRevLett.114.205701, https://doi.org/10.1103/PhysRevLett.114.205701 (arXiv:1503.08576).
7. R. L. Jack, J. P. Garrahan, Phase Transition for Quenched Coupled Replicas in a Plaquette Spin Model of Glasses, Phys. Rev. Lett. 116, 055702 (2016), DOI 10.1103/PhysRevLett.116.055702, https://doi.org/10.1103/PhysRevLett.116.055702 (arXiv:1508.06470).
8. B. Guiselin, L. Berthier, G. Tarjus, Random-field Ising model criticality in a glass-forming liquid, Phys. Rev. E 102, 042129 (2020), DOI 10.1103/PhysRevE.102.042129, https://doi.org/10.1103/PhysRevE.102.042129 (arXiv:2004.10555).
9. B. Guiselin, L. Berthier, G. Tarjus, Statistical mechanics of coupled supercooled liquids in finite dimensions, SciPost Phys. 12, 091 (2022), DOI 10.21468/SciPostPhys.12.3.091, https://doi.org/10.21468/SciPostPhys.12.3.091 (arXiv:2105.08946).
10. N. Küchler, J. Horbach, On the growing length scale in a replica-coupled glassforming liquid, arXiv:2402.04205 (2024), https://arxiv.org/abs/2402.04205.
11. L. Leuzzi, T. Rizzo, Logarithmic critical slowing down in complex systems: from statics to dynamics, Phys. Rev. B 109, 174211 (2024), DOI 10.1103/PhysRevB.109.174211, https://doi.org/10.1103/PhysRevB.109.174211 (arXiv:2403.07565).
12. A. Kaviraj, E. Trevisani, Random Field $\phi^3$ Model and Parisi-Sourlas Supersymmetry, JHEP 08 (2022) 290, DOI 10.1007/JHEP08(2022)290, https://doi.org/10.1007/JHEP08(2022)290 (arXiv:2203.12629).
13. S. Franz, G. Parisi, F. Ricci-Tersenghi, Overlap locking and non-perturbative effects in spin glasses, arXiv:2602.18399 (2026), https://arxiv.org/abs/2602.18399.
14. G. Biroli, C. Cammarota, Fluctuations and Shape of Cooperative Rearranging Regions in Glass-Forming Liquids, Phys. Rev. X 7, 011011 (2017), DOI 10.1103/PhysRevX.7.011011, https://doi.org/10.1103/PhysRevX.7.011011.