Adiabatic Regularization in Quantum Cosmology

• Adiabatic Regularization is a renormalization method that isolates ultraviolet divergences in quantum fields on curved, time-dependent spacetimes using a systematic WKB expansion.
• The technique involves subtracting local high-frequency terms mode by mode up to a specified adiabatic order, ensuring a conserved and locally finite stress tensor in semiclassical gravity.
• Extensions to fermionic, gauge, and interacting fields, along with its role in inflationary spectra and running coupling analyses, demonstrate its versatility in quantum cosmology.

Adiabatic regularization is a renormalization scheme for quantum fields in curved and time-dependent spacetimes, especially Friedmann–Lemaître–Robertson–Walker (FLRW) cosmologies, in which ultraviolet divergences are isolated by expanding exact mode functions in an adiabatic or WKB series and subtracting the corresponding local high-frequency terms mode by mode from quantities such as $\langle \phi^2\rangle$, $\langle T_{\mu\nu}\rangle$, induced currents, and power spectra (Landete et al., 2013, Marañón-González et al., 2024). Originally developed by Parker and Fulling for scalar fields in expanding universes, the method has since been generalized to spin-$\tfrac12$, gauge, and spin-$1$ fields, to interacting backgrounds, and to subtraction schemes with explicit renormalization scales, making its status as a full renormalization prescription rather than a mere regulator increasingly explicit (Rio et al., 2024, Chu et al., 2016, Ferreiro et al., 2018).

1. Formal construction

For scalar fields in spatially flat FLRW spacetime, the basic structure is a mode equation of the form

$$\ddot h_k+\left(\omega_k^2+\sigma\right)h_k=0,\qquad \omega_k=\sqrt{\frac{k^2}{a^2}+m^2},$$

or, after a field redefinition, a conformal-time equation

$\chi_k''+\Omega^2(\eta)\chi_k=0.$

The adiabatic ansatz is then written in WKB form,

$$h_k(t)\sim \frac{1}{\sqrt{W_k(t)}}\,\exp\!\left(-i\int^t W_k(t')\,dt'\right),$$

or equivalently

$$\chi_k^{(2n)}(\eta)=\frac{1}{\sqrt{2W_{2n}(\eta)}}\exp\!\left(-i\int^\eta W_{2n}(\eta')\,d\eta'\right),$$

with $W$ determined recursively from the mode equation (Nakanishi, 2015, Marañón-González et al., 2024).

The adiabatic hierarchy is defined by counting derivatives of the background. In the standard scalar construction, $a$ is adiabatic order $\langle T_{\mu\nu}\rangle$0, $\langle T_{\mu\nu}\rangle$1 is order $\langle T_{\mu\nu}\rangle$2, $\langle T_{\mu\nu}\rangle$3 is order $\langle T_{\mu\nu}\rangle$4, and higher derivatives follow similarly. For the usual scalar WKB expansion, odd orders vanish. This derivative counting organizes the ultraviolet asymptotics of the exact modes and thereby identifies the subtraction terms required for renormalization (Marañón-González et al., 2023, Marañón-González et al., 2024).

The number of subtracted orders depends on the observable. For scalar fields, $\langle T_{\mu\nu}\rangle$5 requires subtraction through second adiabatic order, whereas $\langle T_{\mu\nu}\rangle$6 requires subtraction through fourth order. In the generic single-field inflationary formulation, this appears as subtracting the zeroth and second adiabatic pieces of the two-point function,

$\langle T_{\mu\nu}\rangle$7

while the stress tensor requires fourth-order subtraction because it is an operator of adiabatic order $\langle T_{\mu\nu}\rangle$8 (Nakanishi, 2015, Marañón-González et al., 2024).

2. Renormalized observables and semiclassical backreaction

Adiabatic regularization is implemented directly inside momentum integrals. In cosmological scalar-field applications one writes, for example,

$\langle T_{\mu\nu}\rangle$9

$\tfrac12$0

with subtraction terms built from the adiabatic expansion of the modes. In this way the renormalized stress tensor remains conserved and local, and the subtraction terms can be interpreted as the local counterterms required by semiclassical gravity (Ferreiro et al., 2023).

This counterterm interpretation is central. In the effective Einstein equations, changing the subtraction prescription changes the renormalized stress tensor only by local geometric tensors. One explicit form is

$\tfrac12$1

where $\tfrac12$2 is the Einstein tensor and $\tfrac12$3 is a higher-curvature conserved tensor. This means that different admissible adiabatic prescriptions correspond to different renormalizations of $\tfrac12$4, $\tfrac12$5, and higher-curvature couplings rather than to different physical theories (Ferreiro et al., 2023).

The same logic appears in functional-determinant calculations. For a scalar field integrated out in a cosmological background, the one-loop effective potential can be written in terms of mode functions and then rendered finite by subtracting the second-order adiabatic approximation,

$\tfrac12$6

In that setting the subtraction is interpreted as fixing counterterms so that the effective potential vanishes in flat spacetime, isolating the part of the quantum correction due specifically to cosmic expansion (Kaya et al., 2015).

3. Extensions beyond free scalar fields

The scalar WKB template does not generalize unchanged to fermions. For spin-$\tfrac12$7 fields in FLRW, the mode functions are two coupled amplitudes $\tfrac12$8 and $\tfrac12$9 obeying the Dirac normalization

$1$0

so the scalar WKB ansatz fails to preserve the correct spinorial inner product. The fermionic adiabatic expansion therefore uses a different template,

$1$1

$1$2

with explicit first and second adiabatic orders fixed by the Dirac equation and normalization. This construction reproduces the conformal and axial anomalies and yields a renormalized fermionic stress tensor in de Sitter space (Landete et al., 2013).

Interacting fermionic backgrounds require an additional adiabatic assignment. For a Yukawa interaction with a homogeneous scalar background, $1$3 is counted as adiabatic order $1$4. In that case the renormalized condensate requires subtraction through third adiabatic order,

$1$5

while the stress tensor requires subtraction through fourth order. The subtraction terms match the covariant counterterms needed in the semiclassical Einstein and scalar equations, and the Yukawa contribution to the conformal anomaly agrees with the heat-kernel method (Ferreiro et al., 2019, Rio et al., 2017).

Gauge and spin-$1$6 fields require further modifications. For a $1$7 gauge field in a conformally flat spacetime, the adiabatic scheme is built after covariant gauge fixing and the introduction of mass terms for the gauge and ghost fields; fourth-order subtraction of the stress tensor reproduces the known conformal anomaly (Chu et al., 2016). For massive Proca fields in FLRW, the transverse polarizations behave like conformally coupled scalar modes of mass $1$8, while the longitudinal mode has a different ultraviolet structure. In the massless limit the renormalized Proca stress tensor is not Maxwell alone; rather, the Proca result is related to Maxwell theory by subtracting a minimally coupled scalar, which can be interpreted as a Stueckelberg-type field (Marañón-González et al., 2023).

A related extension concerns Dirac fields in time-varying electric backgrounds. There the electromagnetic potential must be counted as adiabatic order $1$9,

$$\ddot h_k+\left(\omega_k^2+\sigma\right)h_k=0,\qquad \omega_k=\sqrt{\frac{k^2}{a^2}+m^2},$$0

if the current renormalization is to remain compatible with gravity and with the Schwinger–DeWitt expansion. The renormalized induced current is then obtained by subtracting the adiabatic expansion of the current integrand through third order (Beltrán-Palau et al., 2020).

4. Power spectra, horizon crossing, and the infrared problem

Adiabatic regularization has been applied extensively to inflationary perturbations. For general single-field inflation, including $$\ddot h_k+\left(\omega_k^2+\sigma\right)h_k=0,\qquad \omega_k=\sqrt{\frac{k^2}{a^2}+m^2},$$1-inflation, the curvature perturbation subtraction term takes the form

$$\ddot h_k+\left(\omega_k^2+\sigma\right)h_k=0,\qquad \omega_k=\sqrt{\frac{k^2}{a^2}+m^2},$$2

and in non-minimally coupled models the subtraction term is frame independent because the comoving curvature perturbation is conformally invariant and the effective Jordan-frame quantities satisfy $$\ddot h_k+\left(\omega_k^2+\sigma\right)h_k=0,\qquad \omega_k=\sqrt{\frac{k^2}{a^2}+m^2},$$3 and $$\ddot h_k+\left(\omega_k^2+\sigma\right)h_k=0,\qquad \omega_k=\sqrt{\frac{k^2}{a^2}+m^2},$$4 (Nakanishi, 2015).

Within the Urakawa–Starobinsky viewpoint, the subtraction term is evolved after horizon or sound-horizon crossing rather than being frozen at crossing. Under slow-roll and near scale invariance, the subtraction term then decays with the number of $$\ddot h_k+\left(\omega_k^2+\sigma\right)h_k=0,\qquad \omega_k=\sqrt{\frac{k^2}{a^2}+m^2},$$5-folds, so that the regularized spectrum approaches the bare spectrum. In nonminimally coupled single-field inflation with varying sound speed, this decay is exponential in the number of $$\ddot h_k+\left(\omega_k^2+\sigma\right)h_k=0,\qquad \omega_k=\sqrt{\frac{k^2}{a^2}+m^2},$$6-folds under the assumptions used there (Alinea et al., 2017).

A major counterposition is that adiabatic subtraction is fundamentally a ultraviolet procedure and should not be used as a physical modification of the super-Hubble spectrum. In slow-roll inflation one can parameterize the relevant scale by

$$\ddot h_k+\left(\omega_k^2+\sigma\right)h_k=0,\qquad \omega_k=\sqrt{\frac{k^2}{a^2}+m^2},$$7

with $$\ddot h_k+\left(\omega_k^2+\sigma\right)h_k=0,\qquad \omega_k=\sqrt{\frac{k^2}{a^2}+m^2},$$8 in the far ultraviolet, $$\ddot h_k+\left(\omega_k^2+\sigma\right)h_k=0,\qquad \omega_k=\sqrt{\frac{k^2}{a^2}+m^2},$$9 at horizon crossing, and $\chi_k''+\Omega^2(\eta)\chi_k=0.$0 in the far infrared. The critique is that different adiabatic subtraction schemes agree only in the far ultraviolet, become scheme dependent near horizon crossing, and cease to be reliable in the far infrared, where the modes are no longer oscillatory. On that view, the unrenormalized inflationary spectra are the physical, measurable quantities in the far infrared (0906.4772).

Related infrared objections arise for relic gravitational waves and for massless scalar fields in Robertson–Walker spacetimes. In those cases, subtracting the adiabatic counterterms for all $\chi_k''+\Omega^2(\eta)\chi_k=0.$1-modes can distort the low-frequency spectrum, suppress observable bands, or even induce infrared divergences that were absent in the unregularized spectrum. To avoid this, an inside-horizon prescription has been proposed: only the modes with $\chi_k''+\Omega^2(\eta)\chi_k=0.$2 at a chosen time $\chi_k''+\Omega^2(\eta)\chi_k=0.$3 are regularized, while long-wavelength modes are left unchanged. In these treatments the inside-horizon prescription yields ultraviolet-convergent spectra without low-frequency distortion and can also restore nonnegative, ultraviolet- and infrared-convergent spectra for massless fields (Zhang et al., 2018, Zhang et al., 2019).

5. Scale-dependent schemes and running couplings

A major development is the introduction of an arbitrary subtraction scale $\chi_k''+\Omega^2(\eta)\chi_k=0.$4 into adiabatic regularization. In that generalized scheme the leading frequency is taken to be

$\chi_k''+\Omega^2(\eta)\chi_k=0.$5

while the difference

$\chi_k''+\Omega^2(\eta)\chi_k=0.$6

is treated as adiabatic order $\chi_k''+\Omega^2(\eta)\chi_k=0.$7. The resulting subtraction terms remain local and covariant, but the renormalized current and stress tensor acquire explicit $\chi_k''+\Omega^2(\eta)\chi_k=0.$8-dependence. Demanding that the effective semiclassical equations be independent of the arbitrary scale then yields renormalization-group running directly from adiabatic subtraction (Ferreiro et al., 2018).

For scalar QED this procedure reproduces the standard logarithmic running of the electric charge,

$\chi_k''+\Omega^2(\eta)\chi_k=0.$9

but the gravitational couplings exhibit additional power-law terms. In particular,

$$h_k(t)\sim \frac{1}{\sqrt{W_k(t)}}\,\exp\!\left(-i\int^t W_k(t')\,dt'\right),$$0

and

$$h_k(t)\sim \frac{1}{\sqrt{W_k(t)}}\,\exp\!\left(-i\int^t W_k(t')\,dt'\right),$$1

The dimensionless couplings therefore run logarithmically as in dimensional regularization, while the dimensionful couplings $$h_k(t)\sim \frac{1}{\sqrt{W_k(t)}}\,\exp\!\left(-i\int^t W_k(t')\,dt'\right),$$2 and $$h_k(t)\sim \frac{1}{\sqrt{W_k(t)}}\,\exp\!\left(-i\int^t W_k(t')\,dt'\right),$$3 acquire non-logarithmic contributions associated with quartic and quadratic ultraviolet structure (Ferreiro et al., 2018).

The same scale logic has been extended to an interacting $$h_k(t)\sim \frac{1}{\sqrt{W_k(t)}}\,\exp\!\left(-i\int^t W_k(t')\,dt'\right),$$4 theory at one loop. There the effective time-dependent mass is $$h_k(t)\sim \frac{1}{\sqrt{W_k(t)}}\,\exp\!\left(-i\int^t W_k(t')\,dt'\right),$$5, the zeroth-order frequency is replaced by $$h_k(t)\sim \frac{1}{\sqrt{W_k(t)}}\,\exp\!\left(-i\int^t W_k(t')\,dt'\right),$$6, and the $$h_k(t)\sim \frac{1}{\sqrt{W_k(t)}}\,\exp\!\left(-i\int^t W_k(t')\,dt'\right),$$7-dependent subtraction terms are used to define a preferred instantaneous vacuum at $$h_k(t)\sim \frac{1}{\sqrt{W_k(t)}}\,\exp\!\left(-i\int^t W_k(t')\,dt'\right),$$8 by requiring the renormalized energy density and pressure to vanish mode by mode at that time. In those constructions $$h_k(t)\sim \frac{1}{\sqrt{W_k(t)}}\,\exp\!\left(-i\int^t W_k(t')\,dt'\right),$$9 is fixed once and for all as a renormalization choice, and in practice a minimum admissible value is often required for the vacuum to exist (Ferreiro et al., 2022).

A different modification is physical scale adiabatic regularization (PSAR). Instead of one arbitrary renormalization scale, PSAR introduces mass scales adapted to the physical amplification scale of the problem. For the stress tensor one uses $$\chi_k^{(2n)}(\eta)=\frac{1}{\sqrt{2W_{2n}(\eta)}}\exp\!\left(-i\int^\eta W_{2n}(\eta')\,d\eta'\right),$$0, $$\chi_k^{(2n)}(\eta)=\frac{1}{\sqrt{2W_{2n}(\eta)}}\exp\!\left(-i\int^\eta W_{2n}(\eta')\,d\eta'\right),$$1, and $$\chi_k^{(2n)}(\eta)=\frac{1}{\sqrt{2W_{2n}(\eta)}}\exp\!\left(-i\int^\eta W_{2n}(\eta')\,d\eta'\right),$$2, while for the two-point function one extra scale $$\chi_k^{(2n)}(\eta)=\frac{1}{\sqrt{2W_{2n}(\eta)}}\exp\!\left(-i\int^\eta W_{2n}(\eta')\,d\eta'\right),$$3 suffices. The rationale is that standard adiabatic subtraction can distort infrared-amplified spectra when the physical mass $$\chi_k^{(2n)}(\eta)=\frac{1}{\sqrt{2W_{2n}(\eta)}}\exp\!\left(-i\int^\eta W_{2n}(\eta')\,d\eta'\right),$$4 enters the subtraction terms too directly; choosing $$\chi_k^{(2n)}(\eta)=\frac{1}{\sqrt{2W_{2n}(\eta)}}\exp\!\left(-i\int^\eta W_{2n}(\eta')\,d\eta'\right),$$5 and $$\chi_k^{(2n)}(\eta)=\frac{1}{\sqrt{2W_{2n}(\eta)}}\exp\!\left(-i\int^\eta W_{2n}(\eta')\,d\eta'\right),$$6 to be of the order of the relevant physical scale, such as $$\chi_k^{(2n)}(\eta)=\frac{1}{\sqrt{2W_{2n}(\eta)}}\exp\!\left(-i\int^\eta W_{2n}(\eta')\,d\eta'\right),$$7 in de Sitter or $$\chi_k^{(2n)}(\eta)=\frac{1}{\sqrt{2W_{2n}(\eta)}}\exp\!\left(-i\int^\eta W_{2n}(\eta')\,d\eta'\right),$$8 in geometric reheating, suppresses this infrared distortion while preserving ultraviolet cancellation and stress-tensor conservation. In de Sitter, for a light field,

$$\chi_k^{(2n)}(\eta)=\frac{1}{\sqrt{2W_{2n}(\eta)}}\exp\!\left(-i\int^\eta W_{2n}(\eta')\,d\eta'\right),$$9

so the regularized spectrum retains the expected scale-invariant infrared plateau (Ferreiro et al., 2023).

6. Relation to other renormalization methods and continuing debates

Adiabatic regularization belongs to the same renormalization equivalence class as several covariant coordinate-space methods. For Dirac fields in expanding universes, a recent review sketches the proof that adiabatic and DeWitt–Schwinger point-splitting give the same renormalized stress tensor,

$W$0

with the coincidence limits of the adiabatic subtraction terms matching those of the DeWitt–Schwinger expansion (Rio et al., 2024). For Yukawa-coupled fermions, the conformal anomaly obtained by adiabatic subtraction agrees in its scheme-independent terms with the heat-kernel effective-action computation (Rio et al., 2017). These results support the view that adiabatic regularization is not merely computationally convenient but a fully legitimate local renormalization prescription.

Coordinate-space checks also exist in special cases. For a massless conformally coupled scalar field in flat Robertson–Walker spacetimes, direct Green’s-function subtraction yields the same vanishing regularized power spectrum and stress tensor as the adiabatic method, and for several minimally coupled cases—radiation domination, matter domination, and de Sitter—the regularized spectra likewise vanish (Zhang et al., 2019). In de Sitter space, recent work on a spin-$W$1 field compares adiabatic and point-splitting regularization and reports agreement between the two approaches for the regularized correlation function and stress tensor (Ye et al., 27 Sep 2025).

The method has also been compared conceptually with momentum-space BPHZ subtraction. In that analogy the external momentum expansion of BPHZ is replaced by an expansion around a slowly varying geometry, with the scale factor and its derivatives playing the role of external data. This suggests that adiabatic regularization can be viewed as a momentum-space renormalization prescription tailored to homogeneous cosmological backgrounds (Marañón-González et al., 2024).

Several points nevertheless remain contested. The infrared status of subtraction terms in inflationary and gravitational-wave spectra is disputed, as are the physically appropriate subtraction domains and scales. A further recent controversy concerns subtraction order for fermions in de Sitter space: one analysis argues that second-order subtraction is sufficient for the spin-$W$2 spectral stress tensor and power spectrum, while fourth-order subtraction is an oversubtraction that changes the sign of the vacuum energy density and produces a spurious massless-limit anomaly (Ye et al., 27 Sep 2025). This differs from the standard fourth-order prescription for local fermionic stress tensors in earlier adiabatic constructions (Landete et al., 2013). A plausible implication is that the minimal subtraction order may depend not only on the field type but also on the precise observable and on whether one is renormalizing spectral quantities or fully local tensor expectation values.