Lyth Bound in Inflation

• The Lyth bound is a relation connecting the tensor-to-scalar ratio $r$ of primordial perturbations with the field-space distance $\Delta \phi$ traversed during inflation.
• The conventional Lyth bound estimate assumes constant $r$ over 50-60 e-folds and predicts $r \gtrsim 10^{-2}$ for a super-Planckian excursion, while in open-ended inflationary scenarios this threshold can vary and be weaker or stronger based on model parameters.
• Several generalizations of the Lyth bound exist for non-canonical kinetic terms, multifield inflation, and sourced tensor perturbations, which can lead to different interpretations and constraints on the field range.

The Lyth bound is an integral relation connecting the tensor-to-scalar ratio $r$ of primordial perturbations with the field-space distance traversed during inflation. In canonical single-field slow-roll inflation, where $r=16\epsilon$ and $\epsilon$ is the first slow-roll parameter, the canonically normalized inflaton satisfies

$$\frac{\Delta\phi}{M_{\rm Pl}} =\int dN\,\sqrt{\frac{r(N)}{8}}.$$

If $r$ remains approximately constant over the $50$–$60$ e-folds relevant to the cosmic microwave background (CMB), then

$$\frac{\Delta\phi}{M_{\rm Pl}} \simeq N\sqrt{\frac{r}{8}}.$$

Thus, a tensor signal of order $r\gtrsim 10^{-2}$ conventionally indicates an order-$M_{\rm Pl}$ or super-Planckian canonical excursion. The relation is not, however, an unconditional theorem: its interpretation depends on the evolution of $r=16\epsilon$0, the canonical normalization of the inflaton, the origin of scalar and tensor perturbations, the number of e-folds, and the gravitational and field-content assumptions underlying inflation.

1. Canonical derivation and assumptions

For a canonically normalized scalar field in Einstein gravity,

$r=16\epsilon$1

With $r=16\epsilon$2, the field velocity per e-fold is

$r=16\epsilon$3

In canonical single-field slow-roll inflation, the scalar and tensor spectra are

$r=16\epsilon$4

so that

$r=16\epsilon$5

Combining these equations gives the differential Lyth relation

$r=16\epsilon$6

and hence

$r=16\epsilon$7

If $r=16\epsilon$8 is approximately constant over an interval $r=16\epsilon$9, this becomes

$\epsilon$0

For $\epsilon$1, a sub-Planckian excursion gives the familiar estimate

$\epsilon$2

The threshold is often summarized as $\epsilon$3 for an order-$\epsilon$4 excursion.

The conventional inference assumes a single scalar clock, a canonical two-derivative kinetic term, slow-roll or attractor evolution, vacuum-generated tensor modes, the standard scalar normalization, and a sufficiently persistent tensor signal. It also assumes that the field displacement relevant to the background dynamics can be identified with the physical canonical field-space distance. Under these conditions, the tensor amplitude measures the inflaton velocity per e-fold, and the total excursion accumulates over the inflationary duration.

The Lyth relation itself is therefore more fundamental than the constant-$\epsilon$5 estimate. A measured value of $\epsilon$6 at one pivot scale determines the field range only after the behavior of $\epsilon$7 is specified. A large tensor signal confined to a short interval need not produce the same total excursion as a comparable signal maintained over many e-folds.

2. Evolution of $\epsilon$8 and strengthened bounds

The original estimate effectively treats $\epsilon$9 as constant. More generally, its evolution is related to the scalar and tensor tilts:

$$\frac{\Delta\phi}{M_{\rm Pl}} =\int dN\,\sqrt{\frac{r(N)}{8}}.$$0

For canonical single-field slow roll, the consistency relation is

$$\frac{\Delta\phi}{M_{\rm Pl}} =\int dN\,\sqrt{\frac{r(N)}{8}}.$$1

A red scalar spectrum, $$\frac{\Delta\phi}{M_{\rm Pl}} =\int dN\,\sqrt{\frac{r(N)}{8}}.$$2, therefore changes the evolution of $$\frac{\Delta\phi}{M_{\rm Pl}} =\int dN\,\sqrt{\frac{r(N)}{8}}.$$3 and modifies the field-range integral.

For approximately constant $$\frac{\Delta\phi}{M_{\rm Pl}} =\int dN\,\sqrt{\frac{r(N)}{8}}.$$4 and negligible tensor tilt, one obtains an exponentially varying tensor ratio. Writing $$\frac{\Delta\phi}{M_{\rm Pl}} =\int dN\,\sqrt{\frac{r(N)}{8}}.$$5,

$$\frac{\Delta\phi}{M_{\rm Pl}} =\int dN\,\sqrt{\frac{r(N)}{8}}.$$6

with convention-dependent orientation of $$\frac{\Delta\phi}{M_{\rm Pl}} =\int dN\,\sqrt{\frac{r(N)}{8}}.$$7. Substitution into the Lyth integral gives

$$\frac{\Delta\phi}{M_{\rm Pl}} =\int dN\,\sqrt{\frac{r(N)}{8}}.$$8

For $$\frac{\Delta\phi}{M_{\rm Pl}} =\int dN\,\sqrt{\frac{r(N)}{8}}.$$9 and the Planck-era value $r$0, the corresponding sub-Planckian threshold is approximately

$r$1

This is smaller than the scale-invariant estimate $r$2 because the red tilt makes the integrated tensor amplitude larger than the constant-$r$3 approximation suggests (Aravind et al., 2014).

A related analysis based on large-$r$4 inflationary attractors parameterizes

$r$5

For $r$6, the scalar tilt is

$r$7

The regimes $r$8, $r$9, and $50$0 correspond, respectively, to chaotic or monomial, Starobinsky-like or plateau, and hilltop-type asymptotic behavior. The field range grows as a power of $50$1 for $50$2, logarithmically for $50$3, and becomes increasingly sensitive to the endpoint of inflation for $50$4.

Within small-field attractor classes, the measured red tilt can strengthen the constraint substantially. For $50$5, the sub-Planckian condition was estimated as

$50$6

compared with the original approximately $50$7 threshold (Garcia-Bellido et al., 2014). This stronger value is not universal for arbitrary inflationary histories; it relies on the assumed large-$50$8 attractor behavior and suppressed endpoint corrections.

Allowing running changes the conclusion again. If the scalar running $50$9 or running of running $60$0 is positive, the scalar tilt can become less red or blue after the pivot scale exits. The tensor ratio can then decrease rapidly, concentrating the field excursion near the pivot. A canonical single-field slow-roll analysis found

$60$1

for $60$2 with negligible running, while positive running of order $60$3 can raise the allowed value to the order of $60$4–$60$5; a representative positive running-of-running case yielded $60$6 (Huang, 2015). Such large running requires significant higher derivatives of the potential and must be tested using the full perturbation equations rather than a truncated slow-roll expansion.

Using the ACT DR6 central value $60$7 and $60$8 gives an intermediate estimate,

$60$9

which is weaker than the Planck red-tilt estimate but stronger than the scale-invariant result (Yang et al., 15 Jun 2026). The numerical value is a theoretical central-value estimate rather than a complete marginalized observational posterior.

3. Nonmonotonic evolution and engineered small-field models

The conventional full-duration bound assumes that $$\frac{\Delta\phi}{M_{\rm Pl}} \simeq N\sqrt{\frac{r}{8}}.$$0, and therefore $$\frac{\Delta\phi}{M_{\rm Pl}} \simeq N\sqrt{\frac{r}{8}}.$$1, does not decrease substantially after CMB horizon exit. If $$\frac{\Delta\phi}{M_{\rm Pl}} \simeq N\sqrt{\frac{r}{8}}.$$2 evolves non-monotonically, this assumption fails. The exact field-distance relation remains

$$\frac{\Delta\phi}{M_{\rm Pl}} \simeq N\sqrt{\frac{r}{8}}.$$3

but the integral can remain small if $$\frac{\Delta\phi}{M_{\rm Pl}} \simeq N\sqrt{\frac{r}{8}}.$$4 is large only briefly and then decreases sharply.

A canonical single-field model can therefore follow a sequence in which $$\frac{\Delta\phi}{M_{\rm Pl}} \simeq N\sqrt{\frac{r}{8}}.$$5 is moderately large on CMB scales, producing an observable tensor amplitude; increases during the approximately eight e-fold observational window, shaping the scalar spectrum; decreases rapidly shortly afterward, generating many e-folds with little field motion; and is eventually terminated by a rise of $$\frac{\Delta\phi}{M_{\rm Pl}} \simeq N\sqrt{\frac{r}{8}}.$$6 or by a hybrid or waterfall transition. The mechanism requires the CMB-scale value of $$\frac{\Delta\phi}{M_{\rm Pl}} \simeq N\sqrt{\frac{r}{8}}.$$7 to be unrepresentative of the later evolution.

Because

$$\frac{\Delta\phi}{M_{\rm Pl}} \simeq N\sqrt{\frac{r}{8}}.$$8

a sharp decrease in $$\frac{\Delta\phi}{M_{\rm Pl}} \simeq N\sqrt{\frac{r}{8}}.$$9 enhances scalar power unless compensated by the evolution of $r\gtrsim 10^{-2}$0. Such models generically predict enhanced small-scale power, scale-dependent running, and departures from a single power-law scalar spectrum. Excessive enhancement can produce primordial black holes; the model analysis imposed approximately

$r\gtrsim 10^{-2}$1

The scalar spectrum must be computed over the full CMB and large-scale-structure range, since matching the amplitude, tilt, and running at one pivot is insufficient.

An explicit potential illustrating the mechanism is

$r\gtrsim 10^{-2}$2

with

$r\gtrsim 10^{-2}$3

Initially, the nonconstant terms control the slope and $r\gtrsim 10^{-2}$4 increases. Once $r\gtrsim 10^{-2}$5 dominates, the potential becomes vacuum-energy dominated and

$r\gtrsim 10^{-2}$6

can become very small. A representative point with

$r\gtrsim 10^{-2}$7

was obtained with a sub-Planckian excursion and a hybrid transition ending inflation (Hotchkiss et al., 2011).

The same paper emphasized that the scalar consequences are restrictive. Its explicit numerical spectrum differed from the lowest-order slow-roll estimate by order $r\gtrsim 10^{-2}$8 or more even on large scales, and the spectrum had to remain compatible with $r\gtrsim 10^{-2}$9, high-$M_{\rm Pl}$0 CMB measurements, intermediate-scale power near $M_{\rm Pl}$1, and primordial-black-hole limits. The conclusion was that a value such as $M_{\rm Pl}$2 can coexist with $M_{\rm Pl}$3 in a specially structured canonical model, but only with substantial scale dependence and nontrivial scalar-spectrum constraints.

A distinct proposed minimal bound permits nonmonotonic evolution while allowing the inflaton velocity to decrease sharply:

$M_{\rm Pl}$4

For $M_{\rm Pl}$5, this gives $M_{\rm Pl}$6; for $M_{\rm Pl}$7, $M_{\rm Pl}$8. An excursion below $M_{\rm Pl}$9 requires $r=16\epsilon$00 according to the displayed inequality (Gao et al., 2014). This bound is much weaker than the conventional $r=16\epsilon$01 estimate and is partly heuristic because its derivation assumes a sufficiently long inflationary period and a particular expected evolution of $r=16\epsilon$02.

4. Generalizations beyond the canonical single-field relation

The standard relation between $r=16\epsilon$03 and field motion can fail when the scalar and tensor perturbations do not arise from the same canonical single-field sector.

Noncanonical kinetic terms and effective field range

For theories with two-derivative time kinetic terms, the effective field-range diagnostic is defined using the canonically normalized Goldstone mode $r=16\epsilon$04 associated with broken time translations. If

$r=16\epsilon$05

the relevant field-range quantity is

$r=16\epsilon$06

For a dispersion relation with phase velocity $r=16\epsilon$07 at freeze-out, the generalized Lyth relation takes the form

$r=16\epsilon$08

Under the condition $r=16\epsilon$09, this is at least as strong as the standard Lyth estimate and is strictly stronger for $r=16\epsilon$10. In conventional $r=16\epsilon$11 theories, $r=16\epsilon$12, so a small sound speed strengthens rather than weakens the generalized bound (Baumann et al., 2011).

A large noncanonical coefficient can make a coordinate excursion appear small while increasing the invariant canonical range. For a kinetic term $r=16\epsilon$13, a large $r=16\epsilon$14 produces a generalized field range larger than the naive background displacement. Consequently, coordinate smallness is not sufficient evidence for a physically small excursion.

Multifield inflation

With several fields and field-space metric $r=16\epsilon$15, the invariant path length is

$r=16\epsilon$16

An additional light field can suppress the inflaton contribution to the curvature perturbation through post-inflationary conversion. In a curvaton or modulated-reheating scenario, the same inflaton fluctuation can affect both the duration of inflation and the later evolution of the additional field. These contributions can partially cancel. The inflaton excursion can then be sub-Planckian even when $r=16\epsilon$17 is sizable, but the field-range requirement is transferred to the additional light field and generally persists in the total field space (Kobayashi et al., 2013).

In a turning two-field system, entropy-to-curvature transfer enhances the scalar spectrum. Writing

$r=16\epsilon$18

the tensor-to-scalar ratio becomes

$r=16\epsilon$19

The non-geodesic path length therefore satisfies

$r=16\epsilon$20

which is no weaker than the single-field result. The geodesic distance between the endpoints can nevertheless be smaller than the actual path length. In hyperbolic field space,

$r=16\epsilon$21

This distinction permits a sub-Planckian geodesic separation together with a super-Planckian non-geodesic path, relevant to comparisons between the Lyth bound and the Swampland Distance Conjecture (Bravo et al., 2019).

Sourced tensor perturbations

Gauge fields and other matter sectors can generate tensor modes independently of the inflaton’s vacuum fluctuations. In Chromo-Natural Inflation, an oriented $r=16\epsilon$22 gauge background violates parity, amplifies one gauge tensor helicity through a temporary tachyonic instability, and sources chiral gravitational waves. The tensor amplitude is then not determined solely by $r=16\epsilon$23 or by $r=16\epsilon$24. The axion can remain sub-Planckian because its motion is resisted by a Chern–Simons-induced magnetic-drift force (Adshead et al., 2013).

The model provides a counterexample to the naive statement that an observable tensor signal requires a super-Planckian excursion of the particular inflaton field. It does not, however, eliminate scalar-sector constraints: the same gauge dynamics can generate prolonged scalar instabilities, excessive curvature perturbations, strong scale dependence, nonlinearities, and non-Gaussianity. In the studied realization, the combined scalar and tensor requirements rendered the model observationally inviable.

A different sourced mechanism uses parametric resonance of fluctuations of a second scalar $r=16\epsilon$25 while another field $r=16\epsilon$26 generates the observed curvature perturbation. The resonantly amplified $r=16\epsilon$27 modes source gravitational waves through a quadratic stress tensor, while $r=16\epsilon$28 is later stabilized to prevent persistent curvature perturbations. A representative model achieved

$r=16\epsilon$29

with induced tensor peaks below approximately $r=16\epsilon$30, despite a vacuum tensor ratio of only

$r=16\epsilon$31

Its distinctive prediction is a localized, strongly scale-dependent tensor and CMB $r=16\epsilon$32-mode spectrum rather than a nearly scale-invariant vacuum signal (Cai et al., 2021).

Excited initial states and quantum-state modifications

Bogoliubov-excited scalar and tensor modes modify the scalar and tensor spectra by factors

$r=16\epsilon$33

The tensor-to-scalar ratio becomes

$r=16\epsilon$34

and the generalized field-range relation contains the ratio of these excitation factors.

In principle, scalar suppression or tensor enhancement could reduce the field range inferred from a fixed observed $r=16\epsilon$35. Requiring approximately scale-invariant scalar and tensor spectra over three or four decades, together with backreaction constraints, restricts the excitation parameters to roughly

$r=16\epsilon$36

The resulting modification of the Lyth estimate is at most a few percent:

$r=16\epsilon$37

Thus approximately scale-independent Bogoliubov excitations do not appreciably alleviate Planckian evolution (Aravind et al., 2014).

5. Coordinate, canonical, and geometric meanings of “large field”

The phrase “large-field inflation” can refer to distinct quantities. The Lyth bound fundamentally concerns the distance associated with the canonically normalized dynamical mode or, in multifield systems, the accumulated physical path length. It does not necessarily constrain a coordinate amplitude appearing in a particular Lagrangian.

A noncanonical kinetic model can have

$r=16\epsilon$38

with $r=16\epsilon$39. The coordinate excursion $r=16\epsilon$40 may then be sub-Planckian while the canonical excursion $r=16\epsilon$41 is super-Planckian. Quintessential-inflation models with steep exponential potentials use this distinction: a large kinetic prefactor can suppress the excursion of the original coordinate field, while the canonical field retains a large range. In a representative example with $r=16\epsilon$42, $r=16\epsilon$43, and $r=16\epsilon$44, the original coordinate excursion was estimated as $r=16\epsilon$45, whereas the canonical excursion was of order $r=16\epsilon$46 (Hossain et al., 2014).

A geometric construction based on a Riemann surface similarly does not violate the Lyth bound. A pseudo-Nambu–Goldstone boson can wind through multiple branches of a multivalued effective potential while the local amplitudes of the underlying fields remain sub-Planckian. For two fields with charges $r=16\epsilon$47 and $r=16\epsilon$48, the effective axion decay constant is

$r=16\epsilon$49

and can satisfy $r=16\epsilon$50 even when both vacuum expectation values are sub-Planckian. The canonical path length remains large; only the local field amplitudes are small. The construction therefore satisfies rather than evades the Lyth bound (Harigaya et al., 2014).

In multifield theories, the geodesic distance and the non-geodesic path length may differ:

$r=16\epsilon$51

This distinction is central when comparing the Lyth bound with ultraviolet field-distance criteria. A long bent trajectory can have a relatively short endpoint separation, but the Lyth relation still constrains the path length associated with the background field speed.

6. Limits, observations, and complementary developments

The Lyth bound is a diagnostic of inflationary dynamics rather than a direct observational lower bound on $r=16\epsilon$52. Its standard threshold values depend on the duration of inflation, the evolution of $r=16\epsilon$53, the scalar tilt and its running, and the assumptions used to extrapolate CMB information toward the end of inflation.

For canonical single-field slow roll with approximately constant $r=16\epsilon$54 over $r=16\epsilon$55,

$r=16\epsilon$56

corresponds to a sub-Planckian excursion. Including a red scalar tilt gives approximately $r=16\epsilon$57 for the Planck-era value $r=16\epsilon$58, while the ACT DR6 value $r=16\epsilon$59 gives approximately $r=16\epsilon$60. Specific attractor classes can yield characteristic values near $r=16\epsilon$61, whereas positive running can relax the bound to much larger values.

CMB $r=16\epsilon$62-mode experiments are consequently sensitive to theoretically distinct regimes. Forecasts have considered targets near $r=16\epsilon$63, $r=16\epsilon$64, and $r=16\epsilon$65. Foreground separation, delensing, sky coverage, and instrumental noise are decisive at these amplitudes. The tensor tilt is more difficult to measure: the canonical consistency relation

$r=16\epsilon$66

is generally too small to distinguish from a scale-invariant tensor spectrum using CMB $r=16\epsilon$67-modes alone, with cosmic variance producing an uncertainty of order $r=16\epsilon$68 for $r=16\epsilon$69 (Huang et al., 2017).

The Swampland Distance Conjecture provides a conceptually distinct constraint. In the formulation

$r=16\epsilon$70

the conjecture limits the allowed field distance and can therefore impose an upper bound on $r=16\epsilon$71. The Lyth relation and the distance conjecture constrain opposite sides of the $r=16\epsilon$72 plane: the former relates a tensor signal to a minimum excursion, while the latter restricts the maximum excursion. Their intersection can exclude regions of large-field model space independently of current CMB tensor limits (Furuta et al., 31 Jul 2025).

A proposed “inverse Lyth bound” concerns a different phenomenon: continuous canonical non-attractor evolution with

$r=16\epsilon$73

In this regime, the nonconstant superhorizon curvature mode is anti-damped, while the inflaton kinetic energy decreases rapidly. The field distance over a non-attractor interval obeys

$r=16\epsilon$74

For constant $r=16\epsilon$75 with $r=16\epsilon$76,

$r=16\epsilon$77

Ultra-slow-roll inflation has $r=16\epsilon$78, giving an asymptotic distance

$r=16\epsilon$79

This result is complementary to the tensor-based Lyth bound: the ordinary relation gives a lower field-range requirement for sustained tensor production in attractor inflation, whereas the inverse relation gives an upper field-range constraint during a continuous non-attractor phase that amplifies the scalar mode (Kinney, 26 Aug 2026).

The broad conclusion is that a primordial tensor detection robustly indicates a high inflationary energy scale through the tensor spectrum, but does not by itself prove that a particular microscopic field traversed a super-Planckian distance. Such an inference is reliable only when the tensors are vacuum dominated, the scalar is the unique clock, $r=16\epsilon$80 applies, the signal persists over the relevant e-fold interval, and the canonical field-space metric captures the physical trajectory. When these conditions fail, the standard Lyth inference may be weakened, redistributed among fields, reformulated in terms of an effective field range, or replaced by a model-specific analysis of sourced perturbations, noncanonical dynamics, quantum states, or non-attractor evolution.