Cosmological Attractors: Inflation and Beyond

• Cosmological attractors are model families where late-time inflationary and dynamical behaviors emerge robustly, independent of detailed potentials, couplings, or initial conditions.
• They employ mechanisms like pole-driven kinetic terms and hyperbolic moduli-space geometry to yield universal observables such as nₛ and r, unifying diverse inflationary models.
• The framework extends across single-field, multifield, dark energy, and modified gravity models, linking early-universe inflation with late-time cosmic evolution and observational diagnostics.

Cosmological attractors are solution classes or model families in which late-time cosmological behavior becomes insensitive to large classes of microscopic potentials, couplings, or initial conditions. In inflationary theory the term usually denotes universal plateau predictions, often with $n_s=1-\frac{2}{N}$ and a tensor amplitude fixed by a geometric parameter; in dynamical-systems cosmology it can denote stable fixed points, scaling solutions, general-relativity limits, or Lorenz-type strange attractors. The common theme is asymptotic loss of sensitivity, but the operative mechanism depends on context: hyperbolic moduli-space geometry, poles in kinetic terms, conserved nontrivial measures on reduced phase space, or genuine dissipation in autonomous flows (Galante et al., 2014, Remmen et al., 2013, Russo, 2022).

1. Inflationary universality and the pole mechanism

In the inflationary literature, cosmological attractors are most commonly defined by the universality of observables rather than by literal phase-space contraction. A broad Einstein-frame class is

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$

with Laurent expansions near a pole,

$$K_E(\rho)=\frac{a_p}{\rho^p}+\ldots,\qquad V_E(\rho)=V_0(1+c\rho+\ldots).$$

For $p>1$, the pole controls the leading large-$N$ observables; the distinguished case is $p=2$, for which

$$n_s = 1-\frac{2}{N},\qquad r\simeq \frac{8a_2}{N^2}.$$

For $\alpha$-attractors, $a_2=\frac{3\alpha}{2}$, giving

$$n_s \simeq 1-\frac{2}{N},\qquad r\simeq \frac{12\alpha}{N^2}.$$

The order of the pole fixes $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$0, while the residue fixes $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$1 (Galante et al., 2014, Fumagalli, 2016).

A complementary formulation uses conformal variables. In multi-field conformal attractors, the boundary of the moduli space in the original variables becomes infinitely far away after gauge fixing conformal symmetry and passing to canonical Einstein-frame variables. This produces exponential stretching and flattening of the potential near the boundary, so even steep functions of the original fields generate slow-roll plateaus. In the single-field prototype, a generic function of the conformal ratio becomes

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$2

and the associated $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$3-models

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$4

share the leading predictions

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$5

The attractor behavior is therefore geometric: distinct potentials become equivalent near the boundary after canonical normalization (Kallosh et al., 2013).

The generalized $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$6-attractor formulation packages the same phenomenon into plateau potentials written directly in terms of the canonical inflaton. A representative bosonic Lagrangian is

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$7

For $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$8 not too far from $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$9, the predictions are

$$K_E(\rho)=\frac{a_p}{\rho^p}+\ldots,\qquad V_E(\rho)=V_0(1+c\rho+\ldots).$$0

whereas large $$K_E(\rho)=\frac{a_p}{\rho^p}+\ldots,\qquad V_E(\rho)=V_0(1+c\rho+\ldots).$$1 interpolates toward chaotic inflation as the inflaton moduli space becomes flat (Cecotti et al., 2014). This same boundary/pole interpretation also underlies the later supergravity $$K_E(\rho)=\frac{a_p}{\rho^p}+\ldots,\qquad V_E(\rho)=V_0(1+c\rho+\ldots).$$2- and $$K_E(\rho)=\frac{a_p}{\rho^p}+\ldots,\qquad V_E(\rho)=V_0(1+c\rho+\ldots).$$3-model constructions, where different choices of $$K_E(\rho)=\frac{a_p}{\rho^p}+\ldots,\qquad V_E(\rho)=V_0(1+c\rho+\ldots).$$4 preserve the leading observables because the inflationary regime is controlled primarily by the kinetic singularity rather than by the detailed functional form of the potential (Kallosh et al., 2015).

2. Hyperbolic geometry, Kähler curvature, and supergravity dualities

A central development in the subject is the identification of the attractor parameter $$K_E(\rho)=\frac{a_p}{\rho^p}+\ldots,\qquad V_E(\rho)=V_0(1+c\rho+\ldots).$$5 with the geometry of the scalar manifold. In curved $$K_E(\rho)=\frac{a_p}{\rho^p}+\ldots,\qquad V_E(\rho)=V_0(1+c\rho+\ldots).$$6-scale supergravity, the Kähler potential

$$K_E(\rho)=\frac{a_p}{\rho^p}+\ldots,\qquad V_E(\rho)=V_0(1+c\rho+\ldots).$$7

defines the symmetric manifold

$$K_E(\rho)=\frac{a_p}{\rho^p}+\ldots,\qquad V_E(\rho)=V_0(1+c\rho+\ldots).$$8

with curvature

$$K_E(\rho)=\frac{a_p}{\rho^p}+\ldots,\qquad V_E(\rho)=V_0(1+c\rho+\ldots).$$9

Along the real trajectory,

$p>1$0

large $p>1$1 corresponds to the boundary $p>1$2, where the plateau emerges. Small $p>1$3 therefore corresponds to large negative curvature, while $p>1$4 is a singular flat-geometry limit (Roest et al., 2015).

The hyperbolic geometry can be written in half-plane or disk coordinates, related by the Cayley transform

$p>1$5

These coordinate systems are isometric under Möbius transformations. A notable refinement is the Kähler frame

$p>1$6

which makes the Abelian dilatation symmetry explicit; in that frame the inflaton shift symmetry is manifest, so the inflaton remains naturally light. Higher-order curvature deformations with coefficients $p>1$7 and $p>1$8 preserve the relevant symmetries and can stabilize the orthogonal direction. Along the inflationary trajectory the curvature becomes

$p>1$9

with $N$0 a necessary condition and $N$1 a sufficient condition for stabilization during inflation (Carrasco et al., 2015).

Supergravity embeddings then separate inflationary dynamics from supersymmetry breaking and uplifting. In the two-superfield nilpotent construction one uses

$N$2

or a flat Kähler analogue. Along the inflationary path the potential depends only on the first terms in the expansions of $N$3 and $N$4, so generic deformations and arbitrary uplifting to a de Sitter vacuum do not spoil the attractor form. At the minimum,

$N$5

which cleanly separates supersymmetry breaking, vacuum energy, and inflationary plateau physics (Scalisi, 2015).

A further geometric development is the higher-curvature dual of these models. In the $N$6 superconformal/supergravity theory with an inflaton multiplet $N$7 and stabilizer/goldstino multiplet $N$8, the standard two-derivative theory is dual to a four-derivative higher-curvature supergravity in which one chiral multiplet is traded for the curvature superfield

$N$9

In the degenerate linear case $p=2$0, both matter multiplets become non-dynamical, leaving pure higher-derivative supergravity; for generic $p=2$1, at least one matter multiplet remains, so the dual is a higher-curvature theory coupled to scalars rather than a pure curvature theory (Cecotti et al., 2014).

3. Multifield, modular, and generalized attractor families

Attractor universality is not restricted to single-field inflation. In the multi-field $p=2$2 model with complex disk variable $p=2$3, both the radial field and the axion remain light during inflation. The hyperbolic metric suppresses angular motion strongly enough that large deformations of both the radial and angular potentials preserve the leading predictions. For $p=2$4,

$p=2$5

and for general $p=2$6 the standard formula

$p=2$7

is recovered. The relevant slow-roll condition is

$p=2$8

which expresses the “double attractor” suppression of both radial and angular sensitivity (Achúcarro et al., 2017).

An exact discrete-symmetry realization is provided by $p=2$9 cosmological attractors. The basic field is

$$n_s = 1-\frac{2}{N},\qquad r\simeq \frac{8a_2}{N^2}.$$0

with Lagrangian

$$n_s = 1-\frac{2}{N},\qquad r\simeq \frac{8a_2}{N^2}.$$1

The kinetic term retains continuous hyperbolic $$n_s = 1-\frac{2}{N},\qquad r\simeq \frac{8a_2}{N^2}.$$2 symmetry, जबकि the potential is modular invariant. In the parametrization

$$n_s = 1-\frac{2}{N},\qquad r\simeq \frac{8a_2}{N^2}.$$3

the derivatives $$n_s = 1-\frac{2}{N},\qquad r\simeq \frac{8a_2}{N^2}.$$4 and $$n_s = 1-\frac{2}{N},\qquad r\simeq \frac{8a_2}{N^2}.$$5 are double-exponentially suppressed at large $$n_s = 1-\frac{2}{N},\qquad r\simeq \frac{8a_2}{N^2}.$$6, so the axion becomes effectively frozen on the inflationary plateau. Modular invariants built from the Klein $$n_s = 1-\frac{2}{N},\qquad r\simeq \frac{8a_2}{N^2}.$$7-function, the Dedekind $$n_s = 1-\frac{2}{N},\qquad r\simeq \frac{8a_2}{N^2}.$$8-function, and related quantities therefore reproduce the standard attractor predictions

$$n_s = 1-\frac{2}{N},\qquad r\simeq \frac{8a_2}{N^2}.$$9

while tying $\alpha$0 to discrete values determined by compactification data (Kallosh et al., 2024).

The $\alpha$1-attractor framework generalizes these constructions by working backwards from the Einstein frame. Starting with

$\alpha$2

the Weyl rescaling gives

$\alpha$3

For the hyperbolic case

$\alpha$4

one has the identification

$\alpha$5

Exponential $\alpha$6- and $\alpha$7-model potentials then reproduce the ordinary $\alpha$8-attractor plateau behavior, while new polynomial attractors with KKLTI-type large-field tails satisfy

$\alpha$9

spanning the interval

$a_2=\frac{3\alpha}{2}$0

with

$a_2=\frac{3\alpha}{2}$1

This yields a unified family of exponential and polynomial attractors with a supergravity embedding (Kallosh et al., 6 May 2026).

The same hyperbolic mechanism extends into late-time cosmology. In $a_2=\frac{3\alpha}{2}$2-attractor dark-energy models with $a_2=\frac{3\alpha}{2}$3, the asymptotic equation of state is

$a_2=\frac{3\alpha}{2}$4

while the inflationary tensor signal remains

$a_2=\frac{3\alpha}{2}$5

For $a_2=\frac{3\alpha}{2}$6, the models generically converge to the de Sitter limit $a_2=\frac{3\alpha}{2}$7, although some viable thawing models with $a_2=\frac{3\alpha}{2}$8 remain compatible with current observations (Akrami et al., 2017).

4. Quantum robustness and observational diagnostics

A major question is whether attractor universality survives quantum corrections. In the renormalization-group improved treatment, the effective action replaces fixed couplings by running couplings $a_2=\frac{3\alpha}{2}$9 and chooses the RG time $$n_s \simeq 1-\frac{2}{N},\qquad r\simeq \frac{12\alpha}{N^2}.$$0 so that the leading Coleman–Weinberg logarithm vanishes,

$$n_s \simeq 1-\frac{2}{N},\qquad r\simeq \frac{12\alpha}{N^2}.$$1

For the generic plateau action

$$n_s \simeq 1-\frac{2}{N},\qquad r\simeq \frac{12\alpha}{N^2}.$$2

with

$$n_s \simeq 1-\frac{2}{N},\qquad r\simeq \frac{12\alpha}{N^2}.$$3

the leading predictions remain unchanged provided the field dependence of the RG scale is smooth near the inflationary region,

$$n_s \simeq 1-\frac{2}{N},\qquad r\simeq \frac{12\alpha}{N^2}.$$4

Under this condition, the RG flow modifies horizon-exit quantities but cancels out of the leading observables, leaving

$$n_s \simeq 1-\frac{2}{N},\qquad r\simeq \frac{12\alpha}{N^2}.$$5

or equivalently $$n_s \simeq 1-\frac{2}{N},\qquad r\simeq \frac{12\alpha}{N^2}.$$6 for $$n_s \simeq 1-\frac{2}{N},\qquad r\simeq \frac{12\alpha}{N^2}.$$7-attractors. The exception is the tuned regime in which the RG-corrected coefficient $$n_s \simeq 1-\frac{2}{N},\qquad r\simeq \frac{12\alpha}{N^2}.$$8 crosses zero, allowing hilltop or critical inflation (Fumagalli, 2016).

Observation enters the attractor program through geometry. In the supergravity realization, the tensor amplitude determines $$n_s \simeq 1-\frac{2}{N},\qquad r\simeq \frac{12\alpha}{N^2}.$$9, and $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$00 determines the curvature in the inflaton direction,

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$01

This means that a measurement of primordial gravity waves would not only constrain the inflationary energy scale but also probe the curvature of the inflaton sector and, in the special $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$02 duality class, whether the inflaton is best described as a fundamental matter multiplet or as a higher-curvature excitation (Cecotti et al., 2014).

The same parameter can correlate early- and late-universe observables. In the dark-energy extension, future B-mode detectors test $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$03 and thus $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$04, while large-scale-structure surveys constrain the present and asymptotic dark-energy equation of state. In quintessential inflation, gravitational reheating leads to a long kination phase with $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$05, increases the required number of inflationary e-folds by about

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$06

and shifts the spectral index upward by roughly

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$07

This provides an observational discriminator between conventional reheating histories and quintessential-inflation realizations, even when the late-time universe is close to $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$08 (Akrami et al., 2017).

5. Phase-space attractors, conserved measures, and fixed points

Outside the inflationary-universality sense, cosmological attractors are often defined through explicit phase-space dynamics. A conceptual subtlety arises already in single-field flat FRW cosmology. The full minisuperspace phase space is four-dimensional, but for $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$09 the scalar dynamics can be projected to an effective phase space

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$10

with vector field

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$11

However, $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$12 are not canonical coordinates. The physically relevant measure is

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$13

with conservation law

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$14

For the quadratic potential $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$15, a unique analytic conserved measure exists on the punctured plane and diverges near the apparent attractor curves. Apparent convergence in $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$16 is therefore compatible with Liouville’s theorem: the “attractor” is a coordinate-dependent manifestation of a nontrivial conserved measure, not a true Hamiltonian sink (Remmen et al., 2013).

In multi-component FRW systems, by contrast, genuine fixed-point analyses are standard. For the flat three-fluid model consisting of a barotropic fluid, pressureless dark matter, and a scalar with exponential potential

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$17

the autonomous system has ten critical points. The usual potential-kinetic-matter scaling solution is not the unique late-time attractor once dark matter is included. New asymptotic states include a potential-kinetic-dark-matter attractor $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$18 for $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$19 and a one-parameter family $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$20 for $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$21 in which dark energy, matter, and dark matter coexist. The model also admits solutions with two transient periods of acceleration and two transient periods of deceleration (Azreg-Aïnou, 2013).

Noncanonical alternatives to inflation can also exhibit dynamical attractors. In tachyacoustic cosmology, the universe undergoes decelerating expansion,

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$22

while a superluminal sound speed decreases with time so that the sound horizon shrinks. For both a cuscuton-like Lagrangian and a DBI Lagrangian admitting exact power-law solutions, numerical phase-space evolution shows convergence toward the analytic Hamilton–Jacobi trajectory. In this setting “attractor” has the conventional dynamical meaning: nearby homogeneous initial conditions evolve toward the same background solution (Bessada et al., 2012).

A modified-gravity example arises in newer general relativity (NGR), where vacuum spatially flat FLRW cosmology reduces to a homogeneous quadratic dynamical system. Besides the nonhyperbolic saddle at the origin, the homogeneous structure produces projective fixed points with radial evolution

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$23

In the viable type-2 subclass, the effective dark-energy barotropic index

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$24

controls the asymptotics: $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$25 gives phantom behavior $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$26 and big-rip-type future attractors, whereas $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$27 gives non-phantom behavior $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$28 with either eternal expansion $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$29 or turnaround followed by big crunch (Hohmann et al., 22 May 2025).

6. Strange attractors, modified-gravity limits, and broader usages

Some uses of the term refer to genuinely dissipative chaos. In the model of Einstein gravity coupled to three self-interacting scalars with target-space metric

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$30

one field has a negative kinetic term. With superpotential

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$31

the cosmological equations reduce to the quadratic autonomous system

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$32

When

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$33

the flow is dissipative, and the model admits Lorenz-like strange attractors, including double-scroll and multi-scroll cases. The resulting cosmologies alternate chaotically between contraction and expansion because $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$34 changes sign with $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$35, while long quasi-de Sitter phases occur near

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$36

where $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$37 (Russo, 2022).

A different attractor notion appears in scalar-tensor gravity with disformal matter coupling,

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$38

This theory can admit a late-time attractor toward general relativity, but only if the disformal scale satisfies roughly

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$39

That scale is far larger than the scale relevant for neutron-star spontaneous scalarization, $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$40, and the required large values can suppress scalarization and induce ghost instabilities in scalar perturbations. Here the existence of an attractor does not imply phenomenological viability (Silva et al., 2019).

String cosmology supplies yet another variant. In heterotic models with finite temperature and Scherk–Schwarz supersymmetry breaking, the dynamics is drawn toward a radiation-like critical solution

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$41

The order-parameter modulus $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$42 is stabilized or destabilized by the one-loop thermal free energy. When the attractor drives the ratio $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$43 into the critical regime, $${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$44 condenses, previously massless states acquire masses

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$45

and a relic density of non-relativistic dark matter can be generated. In this usage, the attractor is a thermal-moduli trajectory organizing a cosmological phase transition (Coudarchet et al., 2018).

The attractor terminology also appears in exact off-diagonal constructions. Using the anholonomic frame deformation method in two-measure theories and effective Einstein–Yang–Mills–Higgs systems, one can generate anisotropic and inhomogeneous FLRW-, Bianchi-, and Kasner-like solutions whose asymptotic scalar sector reproduces plateau inflation. After suitable nonholonomic deformations and field redefinitions, these configurations yield the standard attractor observables

$${\cal L}=\sqrt{-g}\left[\frac12 R-\frac12 K_E(\rho)(\partial\rho)^2 -V_E(\rho)\right],$$46

interpolating between plateau, Starobinsky/Higgs-type, and quadratic regimes (Rajpoot et al., 2016).

Taken together, these works show that “cosmological attractor” is not a single formal category. In one line of research it denotes universality of inflationary observables controlled by poles and hyperbolic geometry; in another it denotes stable fixed points or scaling solutions of autonomous cosmological systems; in a third it refers to dissipative strange attractors or late-time attractors toward general relativity. This suggests that the term has a family resemblance rather than a unique definition, and that its precise meaning must be read from the geometric, dynamical, or measure-theoretic structure of the model under discussion.