Hilltop-Squared Inflation: Model Dynamics

• Hilltop-squared inflation is a class of single-field models defined by a squared hilltop potential that shapes early universe dynamics.
• The analysis distinguishes between stabilized formulations and unphysical truncations, which critically affect predictions for r and nₛ.
• Generalized models with m ≥ 3, along with non-minimal or loop-generated deformations, remain viable under current observational bounds.

Searching arXiv for recent and foundational papers on hilltop-squared inflation and close variants. Search query: hilltop-squared inflation n=2 hilltop quartic hilltop-squared ACT Planck non-minimal coupling microcanonical density matrix Hilltop-squared inflation is a class of single-field inflationary models in which the inflaton potential is the square of a hilltop factor, most commonly

$$V(\phi)=V_{0}\Bigl[1-\Bigl(\frac{\phi}{\mu}\Bigr)^{p}\Bigr]^{2}.$$

In the literature, the label is used in two related but non-identical ways: for the specific quadratic case $p=2$, and for the broader $n=2$ subclass of generalized hilltop potentials $V_{0}[1-(\phi/\mu)^{m}]^{n}$ with $n=2$. That distinction is central. The minimal stabilized quadratic model $V_{0}(1-\phi^{2}/m^{2})^{2}$ has a sharply constrained and, in the simplest one-parameter form, observationally excluded status, whereas the generalized $n=2$ family with $m\ge 3$ remained viable under Planck-era constraints and is only more recently being squeezed by ACT-era determinations of the scalar tilt (Kallosh et al., 2019, Lillepalu et al., 2022, Lynker et al., 20 Jul 2025).

1. Definition and nomenclature

The generalized hilltop family is usually written as

$$V(\phi)=V_{0}\Bigl[1-\Bigl(\frac{\phi}{\mu}\Bigr)^{m}\Bigr]^{n},$$

with $V_{0}>0$, $p=2$0, $p=2$1, and $p=2$2. Hilltop-squared inflation is the specialization $p=2$3,

$p=2$4

or, in an alternative notation common in later phenomenological studies,

$p=2$5

Here $p=2$6 is the overall energy scale fixed by the amplitude of scalar perturbations, $p=2$7 is the free inflaton scale, and $p=2$8 or $p=2$9 controls the steepness of the hilltop (Lillepalu et al., 2022, Lynker et al., 20 Jul 2025).

The simplest quadratic member of the class is the stabilized potential

$n=2$0

Its physical role is best understood in contrast with the “sloppy” form $n=2$1, which crosses through zero at $n=2$2 and then descends to $n=2$3. In the stabilized version, $n=2$4 at $n=2$5, the potential vanishes at $n=2$6, it is positive definite, and near $n=2$7 it has a quadratic minimum (Kallosh et al., 2019).

This stabilization issue is not a minor technicality. A major part of the subsequent literature on hilltop-squared inflation concerns whether one is analyzing the genuine stabilized theory or an inconsistent truncation whose large-field behavior produces misleading predictions for $n=2$8 (Kallosh et al., 2019).

2. Slow-roll structure

For the generalized $n=2$9 family, setting $V_{0}[1-(\phi/\mu)^{m}]^{n}$0 gives

$V_{0}[1-(\phi/\mu)^{m}]^{n}$1

with derivatives

$V_{0}[1-(\phi/\mu)^{m}]^{n}$2

$V_{0}[1-(\phi/\mu)^{m}]^{n}$3

The slow-roll parameters are therefore

$V_{0}[1-(\phi/\mu)^{m}]^{n}$4

$V_{0}[1-(\phi/\mu)^{m}]^{n}$5

At horizon exit, the leading-order observables are

$V_{0}[1-(\phi/\mu)^{m}]^{n}$6

where $V_{0}[1-(\phi/\mu)^{m}]^{n}$7 is the field value $V_{0}[1-(\phi/\mu)^{m}]^{n}$8 e-folds before the end of inflation (Lillepalu et al., 2022).

The e-fold integral is

$V_{0}[1-(\phi/\mu)^{m}]^{n}$9

For $n=2$0 and $n=2$1, the plateau approximation $n=2$2 yields

$n=2$3

so that $n=2$4. This is the origin of the characteristic power-law dependence of $n=2$5 and $n=2$6 on $n=2$7 across the viable part of the parameter space (Lillepalu et al., 2022).

For the special quadratic case $n=2$8, one can write the slow-roll functions in closed form as

$n=2$9

with $V_{0}(1-\phi^{2}/m^{2})^{2}$0. This gives

$V_{0}(1-\phi^{2}/m^{2})^{2}$1

In this form, all explicit $V_{0}(1-\phi^{2}/m^{2})^{2}$2-dependence appears through $V_{0}(1-\phi^{2}/m^{2})^{2}$3, while the implicit dependence enters through the horizon-exit field value fixed by the e-fold integral (Lynker et al., 2023).

3. The special quadratic model and the $V_{0}(1-\phi^{2}/m^{2})^{2}$4 versus $V_{0}(1-\phi^{2}/m^{2})^{2}$5 issue

The quadratic hilltop-squared model occupies a singular position in the literature because its observational fate depends on whether one studies the inconsistent linear-tail potential $V_{0}(1-\phi^{2}/m^{2})^{2}$6 or the stabilized form $V_{0}(1-\phi^{2}/m^{2})^{2}$7. In the small-$V_{0}(1-\phi^{2}/m^{2})^{2}$8 hilltop regime, $V_{0}(1-\phi^{2}/m^{2})^{2}$9, inflation occurs very near $n=2$0, where

$n=2$1

Hence

$n=2$2

Matching $n=2$3 would require $n=2$4, whereas $n=2$5 gives $n=2$6. In this sense the genuinely small-field quadratic hilltop regime is excluded by CMB data (Kallosh et al., 2019).

In the opposite large-$n=2$7 limit, the unbounded potential $n=2$8 gives the familiar result

$n=2$9

identical to linear chaotic inflation. Numerically, $m\ge 3$0 for $m\ge 3$1 and $m\ge 3$2 for $m\ge 3$3. However, this limit is an artifact of the unphysical linear tail at $m\ge 3$4, because inflation ends while the field is effectively rolling on

$m\ge 3$5

Once the potential is completed so that it has a true minimum at $m\ge 3$6, the local shape near the end of inflation becomes quadratic rather than linear, and the attractor changes to

$m\ge 3$7

which means $m\ge 3$8 at $m\ge 3$9 or $$V(\phi)=V_{0}\Bigl[1-\Bigl(\frac{\phi}{\mu}\Bigr)^{m}\Bigr]^{n},$$0 at $$V(\phi)=V_{0}\Bigl[1-\Bigl(\frac{\phi}{\mu}\Bigr)^{m}\Bigr]^{n},$$1 (Kallosh et al., 2019).

The conceptual lesson is precise. If the potential near the exit patch behaves as $$V(\phi)=V_{0}\Bigl[1-\Bigl(\frac{\phi}{\mu}\Bigr)^{m}\Bigr]^{n},$$2, then $$V(\phi)=V_{0}\Bigl[1-\Bigl(\frac{\phi}{\mu}\Bigr)^{m}\Bigr]^{n},$$3 gives $$V(\phi)=V_{0}\Bigl[1-\Bigl(\frac{\phi}{\mu}\Bigr)^{m}\Bigr]^{n},$$4, whereas $$V(\phi)=V_{0}\Bigl[1-\Bigl(\frac{\phi}{\mu}\Bigr)^{m}\Bigr]^{n},$$5 gives $$V(\phi)=V_{0}\Bigl[1-\Bigl(\frac{\phi}{\mu}\Bigr)^{m}\Bigr]^{n},$$6. The often-quoted $$V(\phi)=V_{0}\Bigl[1-\Bigl(\frac{\phi}{\mu}\Bigr)^{m}\Bigr]^{n},$$7 prediction therefore does not characterize the consistent quadratic hilltop-squared model; it tracks the unphysical linear tail of an unbounded truncation. In the stabilized one-parameter $$V(\phi)=V_{0}\Bigl[1-\Bigl(\frac{\phi}{\mu}\Bigr)^{m}\Bigr]^{n},$$8 theory, the small-$$V(\phi)=V_{0}\Bigl[1-\Bigl(\frac{\phi}{\mu}\Bigr)^{m}\Bigr]^{n},$$9 regime fails on $V_{0}>0$0, and the large-$V_{0}>0$1 regime fails on $V_{0}>0$2. The conclusion drawn in the post-Planck analysis is that the simplest one-parameter hilltop-squared model is ruled out for all $V_{0}>0$3 (Kallosh et al., 2019).

4. Generalized canonical hilltop-squared inflation

The broader $V_{0}>0$4 family with $V_{0}>0$5 has a qualitatively different phenomenology. A numerical survey of

$V_{0}>0$6

subject to Planck 2018 plus BICEP/Keck 2021 constraints found that the original “sombrero” model $V_{0}>0$7 predicts $V_{0}>0$8 even for $V_{0}>0$9 and is disfavoured, while increasing $p=2$00 reduces $p=2$01, with compatibility appearing for $p=2$02. In the viable region the required scale is roughly

$p=2$03

which also ensures the self-consistency of the plateau approximation $p=2$04 (Lillepalu et al., 2022).

The same study fixes the overall amplitude through

$p=2$05

with $p=2$06–$p=2$07. A representative viable point,

$p=2$08

gives

$p=2$09

and therefore

$p=2$10

Similar values were reported for $p=2$11 with $p=2$12, and the inflationary scale was found throughout the scanned parameter space to sit around the grand unification scale (Lillepalu et al., 2022).

The later scaling analysis recast the same class in terms of $p=2$13-dependent fingerprints. At fixed $p=2$14, the observables were found numerically to obey approximate power laws over the viable region,

$p=2$15

For the benchmark $p=2$16 hilltop-squared case, scanning $p=2$17 with $p=2$18 gave approximate local fits

$p=2$19

The same analysis identified a bifurcation in the $p=2$20-plane at a critical $p=2$21 for $p=2$22: below $p=2$23, increasing $p=2$24 moves the low-$p=2$25 end of the family toward higher $p=2$26, whereas above $p=2$27 the ordering reverses (Lynker et al., 2023).

Reheating analyses also differ within the class. In the general scaling study, imposing $p=2$28, $p=2$29, and $p=2$30, together with matter-dominated reheating $p=2$31, further shrank the viable region and typically enforced $p=2$32–$p=2$33 and $p=2$34–$p=2$35 (Lynker et al., 2023). In the specific squared-quartic model

$p=2$36

a more detailed reheating analysis found that the Planck-allowed region survives only in a narrow strip with

$p=2$37

and that viable reheating requires $p=2$38; for $p=2$39 there is no viable reheating once $p=2$40 (Hoffmann et al., 2021).

5. Loop-generated hilltops, microcanonical initial conditions, and non-minimal coupling

A distinct line of work derives hilltop inflation not from a prescribed polynomial potential but from CFT-driven cosmology with microcanonical initial conditions. In that framework, the microcanonical density matrix admits a Euclidean path-integral representation over fields periodic on $p=2$41. After integrating out a large conformal sector, one obtains garland instantons with periodic Euclidean oscillations of both the scale factor and the inflaton. Periodicity imposes

$p=2$42

so $p=2$43 must possess at least one extremum. The relevant solutions are under-barrier oscillations near a local maximum of the potential, and after analytic continuation they provide initial data for ordinary Lorentzian slow roll (Barvinsky et al., 2015).

Near the maximum,

$p=2$44

At the Euclidean-Lorentzian junction one finds

$p=2$45

so that characteristically $p=2$46. This yields

$p=2$47

with typical predictions $p=2$48 and $p=2$49–$p=2$50, plus a thermal correction $p=2$51. In this approach the hill-like potential is loop-generated: logarithmic running in non-minimal Higgs inflation or in $p=2$52 gravity turns an asymptotically shift-invariant Einstein-frame plateau into a genuine local maximum (Barvinsky et al., 2015).

The companion derivation emphasized two additional structural points. First, the same loop mechanism in non-minimal Higgs or scalaron models generates a local maximum in the Einstein frame from otherwise flat large-field potentials. Second, an $p=2$53 term functions as an indispensable finite renormalization device in CFT-driven cosmology: it removes the ghost-inducing $p=2$54 term from the conformal anomaly action and simultaneously provides the scalaron degree of freedom whose Einstein-frame potential acquires the hilltop form. In that construction one again obtains $p=2$55, $p=2$56, and $p=2$57 (Barvinsky et al., 2015).

A different deformation of hilltop-squared inflation supplements the squared-quartic potential with a non-minimal coupling $p=2$58. In the Jordan frame,

$p=2$59

and after the conformal transformation to the Einstein frame the potential becomes

$p=2$60

For $p=2$61, perturbative corrections slightly increase $p=2$62 and suppress $p=2$63; a representative point with $p=2$64, $p=2$65, and $p=2$66 gives

$p=2$67

For $p=2$68, the conformal rescaling produces an exponentially flat plateau and the model enters a universal-attractor regime,

$p=2$69

with $p=2$70–$p=2$71 yielding $p=2$72 and $p=2$73. The associated energy scale is $p=2$74–$p=2$75 (Yuennan et al., 21 Nov 2025).

6. Observational status and interpretive issues

The observational status of hilltop-squared inflation depends on which member of the family is meant. For the minimal canonical quadratic model, the post-Planck verdict is negative: the small-$p=2$76 regime yields unacceptably small $p=2$77, and the properly stabilized large-$p=2$78 regime yields $p=2$79, which exceeds the Planck-era upper bounds. The $p=2$80 alternative does not rescue the model because it is tied to an inconsistent potential unbounded from below (Kallosh et al., 2019).

For the broader canonical $p=2$81 family with $p=2$82, the status was more favorable under Planck 2018 plus BICEP/Keck 2021. Those datasets allowed $p=2$83, $p=2$84, compatibility for $p=2$85, and $p=2$86, with predictions $p=2$87–$p=2$88 and $p=2$89 in representative cases (Lillepalu et al., 2022).

The ACT DR6 shift in the scalar tilt tightened this substantially. Using the combined P-ACT-LB constraint

$p=2$90

the systematic analysis of hilltop-squared inflation found that $p=2$91 is effectively excluded, that only $p=2$92 retains surviving bands, and that these require super-Planckian $p=2$93. For $p=2$94, the allowed range moves from $p=2$95 under Planck to $p=2$96 under ACT, with the upper end extending to $p=2$97. In the same analysis there are no viable points with $p=2$98, typical solutions have $p=2$99, and matter-dominated reheating is entirely absent in the hilltop-squared class under ACT DR6 (Lynker et al., 20 Jul 2025).

Two misconceptions recur in discussions of the subject. The first is that “hilltop-squared inflation” denotes a single model. In fact the data support a sharper statement: the minimal stabilized $n=2$00 model is ruled out, whereas generalized canonical $n=2$01 variants were viable under Planck-era bounds and are now under stronger pressure from ACT, and non-minimally coupled or loop-generated variants can still reach the higher-$n=2$02, low-$n=2$03 region (Kallosh et al., 2019, Lynker et al., 20 Jul 2025, Yuennan et al., 21 Nov 2025). The second is that $n=2$04 is a generic prediction of the $n=2$05 theory; it is not. In the consistent stabilized model the relevant attractor is $n=2$06, while non-minimal coupling or CFT-driven constructions can suppress $n=2$07 down to the $n=2$08–$n=2$09 level (Kallosh et al., 2019, Barvinsky et al., 2015).

A plausible implication is that future measurements of $n=2$10 at the $n=2$11 level will have unusually strong discriminatory power within the hilltop-squared theory space. This suggestion is explicit in the scaling and reheating analyses, which connect improved tensor bounds to tighter determinations of $n=2$12, the scaling exponents $n=2$13 and $n=2$14, and the post-inflationary reheating history (Lynker et al., 2023, Lillepalu et al., 2022).