Introduction
Gravitational waves (GWs) produced by stochastic processes in the early Universe, such as first-order phase transitions [2, 3] and primordial magnetohydrodynamic (MHD) turbulence [4, 5], are a unique probe of physics at high energy scales, during the first microseconds of the evolution of the universe (see refs. [6, 7] and references therein).
Recent observations by Pulsar Timing Arrays (PTA) [8–13] have reported strong evidence for a stochastic GW background at nano-Hertz frequencies that could be compatible with sourcing anisotropic stresses produced around the Quantum Chromodynamics (QCD) scale [14]. On the other hand, the space-based detector LISA is planned to be launched in the next decade [15, 16], and it will be able to probe GW signals produced at the electroweak scale [7, 17, 18]. Therefore, current and forthcoming GW observations make accurate theoretical predictions of the GW spectrum from early-Universe sources increasingly urgent. Otherwise, extracting clean information of the source and, hence, probing early Universe physics from PTA observations and future data from LISA or other GW experiments, remains elusive.
In a first-order phase transition, GWs are directly produced by three different contributions: bubble collisions [22 , 23 , 24 , 25 ] [4, 19, 20, 21], sound waves or acoustic turbulence [26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39 ] , and vortical (M)HD turbulence [40 , 41 , 42 , 43 , 44 , 45 , 46 , 47 , 48 , 49 , 50 ] [1, 4, 5]. Due to the stochastic nature of these sources, the central object characterizing their emitted GW spectrum is the unequal-time correlator (UETC) of the anisotropic stresses of the source, E Π E_{\Pi} E Π . The UETC is in general unknown a priori for a particular source, especially for highly non-linear systems, and its direct determination requires numerical simulations that are expensive and cover only a limited range of scales and durations [23 , 24 , 25 , 26 , 27 , 32 , 33 , 38 , 45 , 46 , 47 , 48 , 49 , 50 ] [1]. It is therefore essential to have analytical models for the UETC that capture the relevant physics and allow for fast, reliable, and general predictions for the GW spectrum.
In particular, reference [1] introduced the “Constant-in-Time” (CiT) model for decaying vortical turbulence, also validated in ref. [50 ] , and later used in refs. [35 , 51 ] [9, 18] for parameter estimation of turbulence production with PTA data and LISA sensitivity. This model assumes that the dynamical evolution of the source is slower than the timescale of GW production, which corresponds, at a wave number k k k , to the light crossing time, δ t G W = 1 / k \delta t_{\mathrm{GW}} = 1/k δ t GW = 1/ k . Therefore, when the timescales over which the UETC evolves are larger than δ t G W \delta t_{\mathrm{GW}} δ t GW for a mode k k k , the UETC can be taken constant in time for an effective duration of the source. The resulting GW spectrum has a universal shape: a causal k 3 k^{3} k 3 tail below the inverse source duration, a linear branch between the inverse duration and the peak, and a power law decay at large k k k that depends on the nature of the source. We argue that the typical spectral shape of the CiT model is observed for a broad range of sources, which motivates its validity for any source of GWs at characteristic wave numbers for which the evolution of the UETC is slower than the light crossing time, δ t G W = 1 / k \delta t_{\mathrm{GW}} = 1/k δ t GW = 1/ k .
In this work, we provide a generic theoretical framework to describe the GW spectrum and its circular degree of polarization following the CiT model for different types of sources related to fluid per turbations, magnetic fields, and scalar fields. We also address the following key question: for which sources, and in which range of wave numbers, is the CiT model a valid description?
For this purpose, we first consider sound waves and stationary turbulence. Since the UETC of sound waves is determined by the speed of sound, c s c_{\mathrm{s}} c s , assessing the validity of the CiT model is independent of the amplitude of the fluid perturbations, which are assumed small for linear sound waves. We indeed find that, as we decrease the speed of sound, c s c_{\mathrm{s}} c s , the resulting GW spectrum reduces to the one found in the CiT model. However, for a radiation-dominated fluid with relativistic c s 2 = 1 / 3 c_{\mathrm{s}}^2 = 1/3 c s 2 = 1/3 , even though the CiT model can describe the GW spectrum at small k k k [34 , 52 ] , the amplitude at the peak grows with the source duration, as found in previous work [30 ] , whenever the sourcing time, δ t f i n \delta t_{\mathrm{fin}} δ t fin , is long c s δ t f i n k ∗ ≫ 1 c_{\mathrm{s}}\delta t_{\mathrm{fin}}k_* \gg1 c s δ t fin k ∗ ≫ 1 [34 ] , where k ∗ k_* k ∗ is the characteristic wave number of the source.
On the other hand, the CiT model predicts an initial quadratic growth of the GW spectrum with the source duration until each mode k k k reaches saturation by the time δ t G W \delta t_{\mathrm{GW}} δ t GW , as was found in numerical simulations of vortical MHD turbulence [45 , 49 ] [1]. This is in contradiction with earlier theoretical studies that considered Kraichnan decorrelation [53 , 54 ] to describe stationary turbulence [40 ] [5]. The authors find that the GW spectrum should instead grow linearly proportional to the source duration in flat Minkowski space-time, as also argued in refs. [26 , 28 , 30 ] for sound waves. As shown in ref. [34 ] , the linear growth with the source duration is a consequence of taking the limit of long durations of the source (both for sound waves and stationary turbulence), which is not valid at small wave numbers, nor for small decorrelation velocities. Therefore, we provide an explanation for the discrepancy between the theoretical and numerical results: for small fluid or Alfvén velocities (hence, long decorrelation times), the linear growth is not present in the GW spectrum, which can be described by the CiT model, agreeing with the simulations. However, as the decorrelation velocity increases, hence reducing the decorrelation timescale, the component proportional to the source duration arises for vortical MHD turbulence, as predicted in refs. [40 ] [5].
In summary, since the linear growth of the GW amplitude is absent for subrelativistic turbulence, this is precisely the reason why previous numerical work found sound waves or acoustic turbulence
with relativistic equation of state, c s 2 = 1 / 3 c_{\mathrm{s}}^2 = 1/3 c s 2 = 1/3 , to be more efficient than subrelativistic vortical turbulence [45 , 49 ] , even if vortical sources are more efficient in producing anisotropic stresses [48 ] . However, we expect the linear growth to also appear for larger decorrelation velocities, u d c o r r δ t f i n k ∗ ≫ 1 u_{\mathrm{dcorr}}\delta t_{\mathrm{fin}}k_* \gg1 u dcorr δ t fin k ∗ ≫ 1 , making turbulent sources as efficient as sound waves in converting anisotropic stresses into GW energy density.
In order to account for the decay of the source, we extend the CiT model by introducing the coherent-decay model, which assumes that the UETC is coherent in time. We show that this assumption leads to modifications of the GW spectral shape in the IR tail (i.e., small k k k ) of the spectrum when the decay time is large. The deviations from the CiT model depend in general on the time evolution of the energy density and the characteristic length scale of the source. When accounting for the decay of the source, we also show that if the decay timescale is longer than the GW light crossing time, we recover the CiT model.
The models presented in this work are completely generic and can be applied to a broad range of GW sources in the early Universe. After this manuscript is published, these models and the computations presented in this work, will be made available as part of the CosmoGW Python package [55 , 56 ] .
Structure of the paper. This paper is structured as follows. In Sec. II, we review GW production in the early Universe, explicitly showing the relation between the GW spectrum and the UETC of the source. We comment on the computation of the GW spectrum from the results of a numerical simulation using the Weinberg’s formula in Sec. II C, generalizing it to an expanding background. Then, in Sec. III, we provide a comprehensive description of statistically homogeneous and isotropic velocity and magnetic fields, as well as scalar fields, and describe them in terms of their spectral components (i.e., longitudinal, normal, and helical components). We introduce a smoothed broken power law model to describe a generic source of GWs in Sec. III C, and present analytical computation of its moments in App. A. The spectra of the anisotropic stresses produced by these sources are then described in Sec. IV, where broken power law fits for them are provided. Their detailed computation is separately presented in App. B. The models for GW production from slowly decaying sources in the early Universe are presented starting from § 0.5 , where we describe in detail the CiT model for GW production. We apply this model to various physical cases in the context of first-order phase transitions and primordial turbulence in Sec. VI. In particular, we deal in Sec. VI A with the applicability of the CiT model to decaying (M)HD turbulence, showing that, when the decay timescale is much longer than the GW saturation timescale, the effects of Kraichnan decorrelation become negligible as the decorrelation velocity becomes much smaller than the speed of light, recovering in this limit the CiT prediction. In Sec. VI B we show how, in a similar way, the GW spectrum from sound waves produced in the radiation-dominated era of the early Universe deviates from the CiT due to the sound speed being ultrarelativistic, recovering the CiT result in the limit when the sound speed is much smaller than the speed of light. Moreover, in Sec. VI C we discuss the potential applicability of the CiT to bubble collisions in first-order phase transitions sourcing GWs via the anisotropic stresses from the gradients of the scalar field. In Sec. VII we then go beyond the CiT model, introducing time decay of the source, in particular time decay of the amplitude and the peak position of the source, under the assumption that the UETC is coherent in time, which again is applicable as long as the decorrelation time is short. The introduction of this model allows us to account for the decay of the source, and it provides a physical description for the effective source duration required in the CiT model. Furthermore, it avoids the appearance of spurious effects in the GW spectrum due to the artificial abrupt switching off of the CiT model. We finally present our conclusions in Sec. VIII.
Conventions. In this paper, we use the following Fourier convention for a generic field u ( x ) u(\boldsymbol{x}) u ( x ) ,
u ( k ) = ∫ u ( x ) e i k ⋅ x d 3 x , (1) u(\boldsymbol{k}) = \int u(\boldsymbol{x}) e^{i\boldsymbol{k}\cdot\boldsymbol{x}} \,\mathrm{d}^{3}x , \tag*{(1)} u ( k ) = ∫ u ( x ) e i k ⋅ x d 3 x , (1)
u ( x ) = ∫ u ( k ) ( 2 π ) 3 e − i k ⋅ x d 3 k , (2) u(\boldsymbol{x}) = \int\frac{u(\boldsymbol{k})}{(2\pi)^{3}} e^{-i\boldsymbol{k}\cdot\boldsymbol{x}} \,\mathrm{d}^{3}k , \tag*{(2)} u ( x ) = ∫ ( 2 π ) 3 u ( k ) e − i k ⋅ x d 3 k , (2)
distinguishing fields in coordinate and Fourier space only by their arguments. We use Greek letters to denote space-time components, e.g., μ = 0 , 1 , 2 , 3 \mu= 0, 1, 2, 3 μ = 0 , 1 , 2 , 3 , and Latin letters for space components, e.g., i = 1 , 2 , 3 i = 1, 2, 3 i = 1 , 2 , 3 .
In general, we use the Friedmann-Lemaître-Robertson-Walker as the background metric tensor that describes the isotropic and homogeneous expanding Universe, such that space-time coordinates x μ = ( τ , x i ) x^{\mu} = (\tau, x^{i}) x μ = ( τ , x i ) refer to conformal time and comoving space coordinates, and k k k is the comoving wave vector. The four-velocity of the fluid in a expanding Universe is
tor. The four-velocity of the fluid in a expanding Universe is
U μ = γ ( 1 , u i ) / a , γ 2 = 1 1 − u 2 , (3) U^{\mu} = \gamma(1,u_i)/a, \qquad\gamma^{2} = \frac{1}{1-u^{2}}, \tag*{(3)} U μ = γ ( 1 , u i ) / a , γ 2 = 1 − u 2 1 , (3)
where u i ≡ d x i d τ u_i \equiv\frac{\mathrm{d}x^i}{\mathrm{d}\tau} u i ≡ d τ d x i is the peculiar velocity and γ \gamma γ is the Lorentz factor. Therefore, u i u_i u i is defined as an Euclidean vector and, hence, it should not be confused with the spatial components of the four-velocity U i U^i U i . We also introduce the Euclidean vector X i ≡ w γ u i X_i \equiv\sqrt{w}\,\gamma u_i X i ≡ w γ u i , relevant to describe the GW production when the fluid fluctuations are large [57 ] . In the limit of small velocities and enthalpy perturbations, X i ≃ w ˉ u i X_i \simeq\sqrt{\bar{w}}u_i X i ≃ w ˉ u i , where w ˉ \bar{w} w ˉ is the mean fluid enthalpy, such that the statistical description of u i u_i u i is directly inherited by X i X_i X i .
Gravitational wave production in the early Universe
In this section, we review the production of GWs from stochastic sources in the early Universe that occurs during the epoch of radiation domination, and show that the GW spectrum at present time can be computed from the unequal-time correlator (UETC) of the source of GWs, E Π ( τ 1 , τ 2 , k ) E_{\Pi}(\tau_1,\tau_2,k) E Π ( τ 1 , τ 2 , k ) , evaluated at conformal times τ 1 \tau_1 τ 1 and τ 2 \tau_2 τ 2 . We show in Sec. III C how the “Weinberg” formula [58 ] , applied to GW computations from phase transitions in Minkowski space-time [22 , 31 , 32 , 33 , 59 , 60 ] [19, 20] (in particular, see App. A of ref. [33] for details), can be applied to compute the GW spectrum from the realizations of the source, modeled as a stochastic variable, in an expanding Universe with different possibilities for the background evolution.
Tensor-mode perturbations
GWs are defined as the tensor perturbations ℓ i j \ell_{ij} ℓ ij over the Friedmann-Lemaître-Robertson-Walker background metric that characterizes the isotropic and homogeneous expanding Universe with the scale factor a a a ,
d s 2 = a 2 ( τ ) [ − d τ 2 + ( δ i j + ℓ i j ) d x i d x j ] , (4) \mathrm{d}s^{2} = a^{2}(\tau)\left[-\mathrm{d}\tau^{2} + (\delta_{ij}+\ell_{ij})\,\mathrm{d}x^{i}\mathrm{d}x^{j}\right], \tag*{(4)} d s 2 = a 2 ( τ ) [ − d τ 2 + ( δ ij + ℓ ij ) d x i d x j ] , (4)
where τ \tau τ corresponds to conformal time. The evolution of the scale factor a a a with conformal time is determined by the background equation of state w w w , which relates the background pressure and energy density p ˉ = ω ρ ˉ \bar{p} = \omega\bar{\rho} p ˉ = ω ρ ˉ . When ω \omega ω is constant in time,
a ( τ ) = a ∗ ( τ τ ∗ ) 1 + ν , ν = 1 − 3 ω 1 + 3 ω . (5) a(\tau) = a_* \left(\frac{\tau}{\tau_*}\right)^{1+\nu}, \qquad\nu= \frac{1-3\omega}{1+3\omega}. \tag*{(5)} a ( τ ) = a ∗ ( τ ∗ τ ) 1 + ν , ν = 1 + 3 ω 1 − 3 ω . (5)
The GW equation for the scaled strains h i j = a ℓ i j h_{ij}=a\ell_{ij} h ij = a ℓ ij in Fourier space1 is [61 ] [6]
( ∂ τ 2 + k 2 − a ′ ′ a ) h i j ( τ , k ) = 6 Π i j ( τ , k ) a H 2 , (6) \left(\partial_\tau^2+k^2-\frac{a''}{a}\right)h_{ij}(\tau,\mathbf{k})=6\Pi_{ij}(\tau,\mathbf{k})a\mathcal{H}^2, \tag*{(6)} ( ∂ τ 2 + k 2 − a a ′′ ) h ij ( τ , k ) = 6 Π ij ( τ , k ) a H 2 , (6)
where we use a prime to denote derivative with respect to conformal time, k k k is the comoving wave vector, H = a ′ / a \mathcal{H}=a'/a H = a ′ / a the conformal Hubble rate, and Π i j \Pi_{ij} Π ij is the normalized anisotropic stress tensor sourcing the GWs. It can be expressed as
Π i j ( τ , k ) = a 2 Λ i j l m ( k ^ ) T l m ( τ , k ^ ) ρ ˉ ( τ ) , (7) \Pi_{ij}(\tau,\mathbf{k})=a^2\Lambda_{ijlm}(\hat{\mathbf{k}})\frac{T^{lm}(\tau,\hat{\mathbf{k}})}{\bar{\rho}(\tau)}, \tag*{(7)} Π ij ( τ , k ) = a 2 Λ ij l m ( k ^ ) ρ ˉ ( τ ) T l m ( τ , k ^ ) , (7)
where ρ ˉ = 3 H 2 M p l 2 \bar{\rho}=3\mathcal{H}^2M_{\mathrm{pl}}^2 ρ ˉ = 3 H 2 M pl 2 is the background energy density of the Universe, H = H / a H=\mathcal{H}/a H = H / a the physical Hubble rate, M p l M_{\mathrm{pl}} M pl is the reduced Planck mass, and Λ i j l m \Lambda_{ijlm} Λ ij l m is the traceless and transverse (TT) projection operator [62 ] ,
Λ i j l m ( k ^ ) = 1 2 ( P i l P j m + P i m P j l − P i j P l m ) , (8) \Lambda_{ijlm}(\hat{\mathbf{k}})=\frac{1}{2}\left(P_{il}P_{jm}+P_{im}P_{jl}-P_{ij}P_{lm}\right), \tag*{(8)} Λ ij l m ( k ^ ) = 2 1 ( P i l P j m + P im P j l − P ij P l m ) , (8)
with P i j = δ i j − k ^ i k ^ j P_{ij}=\delta_{ij}-\hat{k}_i\hat{k}_j P ij = δ ij − k ^ i k ^ j and k ^ = k / ∣ k ∣ \hat{\mathbf{k}}=\mathbf{k}/|\mathbf{k}| k ^ = k /∣ k ∣ . Since h 0 μ = 0 h^{0\mu}=0 h 0 μ = 0 in the TT gauge, we have restricted the TT projector to Euclidean geometry and, hence, we do not distinguish between lower and upper indices in Π i j \Pi_{ij} Π ij .
During the radiation-dominated era, ω = 1 / 3 \omega=1/3 ω = 1/3 and the scale factor evolves linearly with conformal time, a = a ∗ ( τ / τ ∗ ) = a ∗ τ H ∗ a=a_*(\tau/\tau_*)=a_*\tau\mathcal{H}_* a = a ∗ ( τ / τ ∗ ) = a ∗ τ H ∗ , such that a ′ ′ = 0 a''=0 a ′′ = 0 and H = 1 / τ \mathcal{H}=1/\tau H = 1/ τ . Setting a ∗ = 1 a_*=1 a ∗ = 1 , the GW equation during radiation domination becomes
( ∂ τ 2 + k 2 ) h i j ( τ , k ) = 6 Π i j ( τ , k ) H ∗ τ . (9) \left(\partial_\tau^2+k^2\right)h_{ij}(\tau,\mathbf{k})=6\Pi_{ij}(\tau,\mathbf{k})\frac{\mathcal{H}_*}{\tau}. \tag*{(9)} ( ∂ τ 2 + k 2 ) h ij ( τ , k ) = 6 Π ij ( τ , k ) τ H ∗ . (9)
Assuming that the source is active for a finite duration δ τ f i n = τ f i n − τ ∗ \delta\tau_{\mathrm{fin}}=\tau_{\mathrm{fin}}-\tau_* δ τ fin = τ fin − τ ∗ , the solution to Eq. (6) with initial conditions h i j ( τ ∗ , k ) = h i j ′ ( τ ∗ , k ) = 0 h_{ij}(\tau_*,\mathbf{k})=h'_{ij}(\tau_*,\mathbf{k})=0 h ij ( τ ∗ , k ) = h ij ′ ( τ ∗ , k ) = 0 can be expressed as
h i j ( τ < τ f i n , k ) = 6 ∫ τ ∗ τ d τ 1 a ( τ 1 ) H 2 ( τ 1 ) Π i j ( τ 1 , k ) G ( τ , τ 1 , k ) , (10) h_{ij}(\tau<\tau_{\mathrm{fin}},\mathbf{k})=6\int_{\tau_*}^{\tau}\mathrm{d}\tau_1\,a(\tau_1)\mathcal{H}^2(\tau_1)\Pi_{ij}(\tau_1,\mathbf{k})G(\tau,\tau_1,k), \tag*{(10)} h ij ( τ < τ fin , k ) = 6 ∫ τ ∗ τ d τ 1 a ( τ 1 ) H 2 ( τ 1 ) Π ij ( τ 1 , k ) G ( τ , τ 1 , k ) , (10)
where G ( τ , τ 1 , k ) G(\tau,\tau_1,k) G ( τ , τ 1 , k ) is the Green’s function associated to the wave operator in Eq. (6) . Matching this solution to the freely propagating regime at τ > τ f i n \tau>\tau_{\mathrm{fin}} τ > τ fin , the solution to the GW equation at all times becomes
h i j ( τ , k ) = 6 ∫ τ ∗ τ ˉ d τ 1 a ( τ 1 ) H 2 ( τ 1 ) Π i j ( τ 1 , k ) G ( τ , τ 1 , k ) , (11) h_{ij}(\tau,\mathbf{k})=6\int_{\tau_*}^{\bar{\tau}}\mathrm{d}\tau_1\,a(\tau_1)\mathcal{H}^2(\tau_1)\Pi_{ij}(\tau_1,\mathbf{k})G(\tau,\tau_1,k), \tag*{(11)} h ij ( τ , k ) = 6 ∫ τ ∗ τ ˉ d τ 1 a ( τ 1 ) H 2 ( τ 1 ) Π ij ( τ 1 , k ) G ( τ , τ 1 , k ) , (11)
where τ ˉ = min ( τ f i n , τ ) \bar{\tau}=\min(\tau_{\mathrm{fin}},\tau) τ ˉ = min ( τ fin , τ ) . Its derivative with respect to conformal time is
h i j ′ ( τ , k ) = 6 ∫ τ ∗ τ ˉ d τ 1 a ( τ 1 ) H 2 ( τ 1 ) Π i j ( τ 1 , k ) ∂ τ G ( τ , τ 1 , k ) , (12) h'_{ij}(\tau,\mathbf{k})=6\int_{\tau_*}^{\bar{\tau}}\mathrm{d}\tau_1\,a(\tau_1)\mathcal{H}^2(\tau_1)\Pi_{ij}(\tau_1,\mathbf{k})\partial_\tau G(\tau,\tau_1,k), \tag*{(12)} h ij ′ ( τ , k ) = 6 ∫ τ ∗ τ ˉ d τ 1 a ( τ 1 ) H 2 ( τ 1 ) Π ij ( τ 1 , k ) ∂ τ G ( τ , τ 1 , k ) , (12)
since G ( τ , τ , k ) = 0 G(\tau,\tau,k)=0 G ( τ , τ , k ) = 0 (see, e.g., eq. (2.11) of ref. [63 ] ). In Secs. V–VII, we restrict our analysis to sources produced during the radiation-dominated era. However, we note that our description concerns the dynamical modeling of the source and, hence, it can be directly applied to different cosmologies using Eq. (12) . For radiation domination, the Green’s function is G = 1 k sin [ k ( τ − τ 1 ) ] G=\frac{1}{k}\sin[k(\tau-\tau_1)] G = k 1 sin [ k ( τ − τ 1 )] and Eq. (12) becomes
h i j ′ ( τ , k ) = 6 H ∗ ∫ τ ∗ τ ˉ d τ 1 τ 1 Π i j ( τ 1 , k ) cos [ k ( τ − τ 1 ) ] . (13) h'_{ij}(\tau,\mathbf{k})=6\mathcal{H}_*\int_{\tau_*}^{\bar{\tau}}\frac{\mathrm{d}\tau_1}{\tau_1}\Pi_{ij}(\tau_1,\mathbf{k})\cos[k(\tau-\tau_1)]. \tag*{(13)} h ij ′ ( τ , k ) = 6 H ∗ ∫ τ ∗ τ ˉ τ 1 d τ 1 Π ij ( τ 1 , k ) cos [ k ( τ − τ 1 )] . (13)
Gravitational wave energy and polarization spectra
The GW energy density, normalized to the critical energy density at present cosmic time t 0 t_0 t 0 , ρ ˉ 0 = 3 H 0 2 M p l 2 \bar{\rho}_0=3H_0^2M_{\mathrm{pl}}^2 ρ ˉ 0 = 3 H 0 2 M pl 2 , is
Ω G W ( t 0 ) = M p l 2 4 ρ ˉ 0 ⟨ ∂ t ℓ i j ( x , t 0 ) ∂ t ℓ i j ( x , t 0 ) ⟩ , (14) \Omega_{\mathrm{GW}}(t_0)=\frac{M_{\mathrm{pl}}^2}{4\bar{\rho}_0}\left\langle\partial_t\ell_{ij}(\mathbf{x},t_0)\partial_t\ell_{ij}(\mathbf{x},t_0)\right\rangle, \tag*{(14)} Ω GW ( t 0 ) = 4 ρ ˉ 0 M pl 2 ⟨ ∂ t ℓ ij ( x , t 0 ) ∂ t ℓ ij ( x , t 0 ) ⟩ , (14)
where the brackets indicate an ensemble average over stochastic realizations. The GW energy density can be expressed as [64 ] [6]
Ω G W ( t 0 ) = 1 12 H 0 2 ( a ∗ a 0 ) 4 ⟨ [ h i j ′ − H h i j ] 2 ⟩ ≃ 1 12 H 0 2 ( a ∗ a 0 ) 4 ⟨ h i j ′ ( x , t 0 ) h i j ′ ( x , t 0 ) ⟩ , (15) \begin{aligned}
\Omega_{\mathrm{GW}}(t_0)=\frac{1}{12H_0^2}\left(\frac{a_*}{a_0}\right)^4\left\langle\left[h'_{ij}-\mathcal{H}h_{ij}\right]^2\right\rangle\\
&\simeq\frac{1}{12H_0^2}\left(\frac{a_*}{a_0}\right)^4\left\langle h'_{ij}(\mathbf{x},t_0)h'_{ij}(\mathbf{x},t_0)\right\rangle, \tag*{(15)}
\end{aligned} Ω GW ( t 0 ) = 12 H 0 2 1 ( a 0 a ∗ ) 4 ⟨ [ h ij ′ − H h ij ] 2 ⟩ ≃ 12 H 0 2 1 ( a 0 a ∗ ) 4 ⟨ h ij ′ ( x , t 0 ) h ij ′ ( x , t 0 ) ⟩ , (15)
where h i j ′ h'_{ij} h ij ′ corresponds to the solution to the GW Eq. (13) when one sets a ∗ = 1 a_*=1 a ∗ = 1 [44 ] , and we have omitted the terms proportional to H / k \mathcal{H}/k H / k in the second line since, at present time, all relevant wave numbers of causal signals produced in the early Universe are deep inside the horizon, k τ 0 ≫ 1 k\tau_0 \gg1 k τ 0 ≫ 1 .
We express the spectral properties of h i j ′ h'_{ij} h ij ′ through its two-point correlation function in Fourier space, which for a statistically homogeneous and isotropic field, becomes [44 , 49 , 65 , 66 , 67 , 68 ]
⟨ h i j ′ ( k ) h l m ′ ∗ ( k ′ ) ⟩ = ( 2 π ) 6 δ 3 ( k − k ′ ) F i j , l m h ′ ( k ) 4 π k 3 , (16) \left\langle h'_{ij}(\boldsymbol{k}) h_{lm}^{\prime*}(\boldsymbol{k}') \right\rangle= (2\pi)^6 \delta^3(\boldsymbol{k}-\boldsymbol{k}') \frac{\mathcal{F}^{h'}_{ij,lm}(k)}{4\pi k^3}, \tag*{(16)} ⟨ h ij ′ ( k ) h l m ′ ∗ ( k ′ ) ⟩ = ( 2 π ) 6 δ 3 ( k − k ′ ) 4 π k 3 F ij , l m h ′ ( k ) , (16)
where F i j , l m h ′ ( k ) \mathcal{F}^{h'}_{ij,lm}(k) F ij , l m h ′ ( k ) is
F i j , l m h ′ ( k ) = 1 2 [ Λ i j l m ( k ^ ) S h ′ ( k ) + i A i j l m ( k ^ ) A h ′ ( k ) ] . (17) \mathcal{F}^{h'}_{ij,lm}(\boldsymbol{k}) = \frac{1}{2} \left[\Lambda_{ijlm}(\hat{\boldsymbol{k}}) S_{h'}(k) + i A_{ijlm}(\hat{\boldsymbol{k}}) A_{h'}(k)\right]. \tag*{(17)} F ij , l m h ′ ( k ) = 2 1 [ Λ ij l m ( k ^ ) S h ′ ( k ) + i A ij l m ( k ^ ) A h ′ ( k ) ] . (17)
The spectral functions S h ′ ( k ) S_{h'}(k) S h ′ ( k ) and A h ′ ( k ) A_{h'}(k) A h ′ ( k ) are, respectively, the symmetric and antisymmetric (or chiral) spectra per logarithmic unit of k k k of the strain derivatives h i j ′ h'_{ij} h ij ′ . The chiral spectrum A ( k ) A(k) A ( k ) characterizes the asymmetry of the stochastic field h i j ′ h'_{ij} h ij ′ under mirror reflections, and vanishes when its statistical properties satisfy parity symmetry. The projection tensor Λ \Lambda Λ is defined in Eq. (8) , and A \mathcal{A} A is [66 ]
A i j l m ( k ^ ) = 1 4 k ^ q ( P j m ϵ i l q + P i l ϵ j m q + P i m ϵ j l q + P j l ϵ i m q ) . (18) \mathcal{A}_{ijlm}(\hat{\boldsymbol{k}}) = \frac{1}{4} \hat{k}_q \left(P_{jm}\epsilon_{ilq} + P_{il}\epsilon_{jmq} + P_{im}\epsilon_{jlq} + P_{jl}\epsilon_{imq}\right). \tag*{(18)} A ij l m ( k ^ ) = 4 1 k ^ q ( P j m ϵ i l q + P i l ϵ j m q + P im ϵ j l q + P j l ϵ im q ) . (18)
When applied to a generic tensor, both Λ \Lambda Λ and A \mathcal{A} A project onto the corresponding symmetric and antisymmetric TT components of the tensor. Multiplying Eqs. (15) by δ i l δ j m \delta_{il}\delta_{jm} δ i l δ j m to obtain the trace ⟨ h i j ′ h i j ′ ∗ ⟩ \langle h'_{ij}h_{ij}^{\prime*}\rangle ⟨ h ij ′ h ij ′ ∗ ⟩ , which describes the GW energy density [see Eq. (15) ], we find
Ω G W ( τ , k ) ≡ d Ω G W ( τ ) d ln k ≃ 1 12 H 0 2 ( a ∗ a 0 ) 4 S h ′ ( τ , k ) . (19) \Omega_{\mathrm{GW}}(\tau,k) \equiv\frac{\mathrm{d}\Omega_{\mathrm{GW}}(\tau)}{\mathrm{d}\ln k}
\simeq\frac{1}{12H_0^2}\left(\frac{a_*}{a_0}\right)^4 S_{h'}(\tau,k). \tag*{(19)} Ω GW ( τ , k ) ≡ d ln k d Ω GW ( τ ) ≃ 12 H 0 2 1 ( a 0 a ∗ ) 4 S h ′ ( τ , k ) . (19)
The degree of circular polarization of the GW background is determined by the ratio of the chiral to the symmetric GW spectra,
P G W ( τ , k ) = Ξ G W ( τ , k ) Ω G W ( τ , k ) , (20) \mathcal{P}_{\mathrm{GW}}(\tau,k) = \frac{\Xi_{\mathrm{GW}}(\tau,k)}{\Omega_{\mathrm{GW}}(\tau,k)}, \tag*{(20)} P GW ( τ , k ) = Ω GW ( τ , k ) Ξ GW ( τ , k ) , (20)
where Ξ G W ( τ , k ) \Xi_{\mathrm{GW}}(\tau,k) Ξ GW ( τ , k ) is found contracting Eq. (15) with A i j l m ( k ^ ) A_{ijlm}(\hat{\boldsymbol{k}}) A ij l m ( k ^ ) [46 , 49 , 67 , 69 ] ,
Ξ G W ( τ , k ) ≃ 1 12 H 0 2 ( a ∗ a 0 ) 4 A h ′ ( τ , k ) . (21) \Xi_{\mathrm{GW}}(\tau,k) \simeq\frac{1}{12H_0^2}\left(\frac{a_*}{a_0}\right)^4 A_{h'}(\tau,k). \tag*{(21)} Ξ GW ( τ , k ) ≃ 12 H 0 2 1 ( a 0 a ∗ ) 4 A h ′ ( τ , k ) . (21)
We have again assumed k τ ≫ 1 k\tau\gg1 k τ ≫ 1 , such that contributions proportional to 1 / ( k τ ) 1/(k\tau) 1/ ( k τ ) are omitted as they vanish at present time [44 ] .
The UETC of the stochastic anisotropic stresses Π i j \Pi_{ij} Π ij sourcing GWs can also be decomposed into a symmetric, E Π ( k ) E_\Pi(k) E Π ( k ) , and an antisymmetric, H Π ( k ) H_\Pi(k) H Π ( k ) , spectral function with a decomposition equivalent to that of Eqs. (15),
⟨ Π i j ( τ 1 , k ) Π l m ∗ ( τ 2 , k ′ ) ⟩ = ( 2 π ) 6 δ 3 ( k − k ′ ) F i j , l m Π ( τ 1 , τ 2 , k ) 4 π k 3 , (22) \left\langle\Pi_{ij}(\tau_1,\boldsymbol{k})\Pi^*_{lm}(\tau_2,\boldsymbol{k}') \right\rangle= (2\pi)^6 \delta^3(\boldsymbol{k}-\boldsymbol{k}') \frac{\mathcal{F}^{\Pi}_{ij,lm}(\tau_1,\tau_2,k)}{4\pi k^3}, \tag*{(22)} ⟨ Π ij ( τ 1 , k ) Π l m ∗ ( τ 2 , k ′ ) ⟩ = ( 2 π ) 6 δ 3 ( k − k ′ ) 4 π k 3 F ij , l m Π ( τ 1 , τ 2 , k ) , (22)
with
F i j , l m Π ( τ 1 , τ 2 , k ) = 1 2 [ Λ i j l m ( k ^ ) E Π ( τ 1 , τ 2 , k ) + i A i j l m ( k ^ ) H Π ( τ 1 , τ 2 , k ) ] . (23) \mathcal{F}^{\Pi}_{ij,lm}(\tau_1,\tau_2,k) = \frac{1}{2}\left[\Lambda_{ijlm}(\hat{\boldsymbol{k}}) E_\Pi(\tau_1,\tau_2,k) + i A_{ijlm}(\hat{\boldsymbol{k}}) H_\Pi(\tau_1,\tau_2,k)\right]. \tag*{(23)} F ij , l m Π ( τ 1 , τ 2 , k ) = 2 1 [ Λ ij l m ( k ^ ) E Π ( τ 1 , τ 2 , k ) + i A ij l m ( k ^ ) H Π ( τ 1 , τ 2 , k ) ] . (23)
In this case, Π i j \Pi_{ij} Π ij is evaluated at different times τ 1 \tau_1 τ 1 and τ 2 \tau_2 τ 2 , and different k \boldsymbol{k} k and k ′ \boldsymbol{k}' k ′ .
Inserting the solution for GWs produced during radiation domination, given in Eq. (13) and the UETC E Π ( τ 1 , τ 2 , k ) E_\Pi(\tau_1,\tau_2,k) E Π ( τ 1 , τ 2 , k ) of Eqs. (20) into Eqs. (15), we find S h ′ ( τ , k ) S_{h'}(\tau,k) S h ′ ( τ , k ) , which yields the following expression for the GW spectrum [6]
Ω G W ( τ , k ) = 3 F G W 0 ∫ τ ∗ τ ˉ d τ 1 τ 1 ∫ τ ∗ τ ˉ d τ 2 τ 2 E Π ( τ 1 , τ 2 , k ) cos [ k ( τ − τ 1 ) ] cos [ k ( τ − τ 2 ) ] . (24) \Omega_{\mathrm{GW}}(\tau,k) = 3F_{\mathrm{GW}}^0 \int_{\tau_*}^{\bar{\tau}} \frac{\mathrm{d}\tau_1}{\tau_1} \int_{\tau_*}^{\bar{\tau}} \frac{\mathrm{d}\tau_2}{\tau_2} E_\Pi(\tau_1,\tau_2,k)\cos[k(\tau-\tau_1)]\cos[k(\tau-\tau_2)]. \tag*{(24)} Ω GW ( τ , k ) = 3 F GW 0 ∫ τ ∗ τ ˉ τ 1 d τ 1 ∫ τ ∗ τ ˉ τ 2 d τ 2 E Π ( τ 1 , τ 2 , k ) cos [ k ( τ − τ 1 )] cos [ k ( τ − τ 2 )] . (24)
Ξ G W ( τ , k ) \Xi_{\mathrm{GW}}(\tau,k) Ξ GW ( τ , k ) can be computed using Eq. (24) if we substitute E Π → H Π E_\Pi\to H_\Pi E Π → H Π . The transfer function F G W 0 F_{\mathrm{GW}}^0 F GW 0 represents the redshift of the GW spectrum,
F G W 0 = ( a ∗ a 0 ) 4 ( H ∗ H 0 ) 2 ≃ 1.6 × 10 − 5 h 2 ( 100 g ∗ ) 1 3 , (25) F_{\mathrm{GW}}^0 = \left(\frac{a_*}{a_0}\right)^4\left(\frac{H_*}{H_0}\right)^2
\simeq\frac{1.6\times10^{-5}}{h^2}\left(\frac{100}{g_*}\right)^{\frac{1}{3}}, \tag*{(25)} F GW 0 = ( a 0 a ∗ ) 4 ( H 0 H ∗ ) 2 ≃ h 2 1.6 × 1 0 − 5 ( g ∗ 100 ) 3 1 , (25)
where g ∗ g_* g ∗ is the number of relativistic degrees of freedom at the time of generation τ ∗ \tau_* τ ∗ and h h h characterizes the Hubble rate at present time, H 0 = 100 h k m / ( s M p c ) H_0 = 100\,h\ {\rm km}/({\rm s\,Mpc}) H 0 = 100 h km / ( s Mpc ) .
In this section, we provide an expression analogous to the Weinberg’s formula for the power emitted by isolated binaries, following the description of refs. [31 , 32 , 33 ] , where this expression was used in flat Minkowski space-time. We extend their results to include the effects of the expansion of the Universe during the process of GW production. These expressions explicitly show the impact of computing the GW spectrum from numerical simulations and how these can be interpreted to make predictions of the observable GW spectrum.
In first place, we are interested in evaluating the GW spectrum at present time. For subhorizon modes k τ 0 ≫ 1 k\tau_0 \gg1 k τ 0 ≫ 1 , we can average the product of Green’s functions in Eq. (24) over fast frequency oscillations to find
Ω G W 0 ( k ) = 3 2 F G W 0 ∫ τ ∗ τ f i n d τ 1 τ 1 ∫ τ ∗ τ f i n d τ 2 τ 2 E Π ( τ 1 , τ 2 , k ) cos ( k τ − ) , (26) \Omega_{\mathrm{GW}}^{0}(k)=\frac{3}{2}F_{\mathrm{GW}}^{0}\int_{\tau_*}^{\tau_{\mathrm{fin}}}\frac{\mathrm{d}\tau_1}{\tau_1}\int_{\tau_*}^{\tau_{\mathrm{fin}}}\frac{\mathrm{d}\tau_2}{\tau_2}\mathcal{E}_{\Pi}(\tau_1,\tau_2,k)\cos(k\tau_-), \tag*{(26)} Ω GW 0 ( k ) = 2 3 F GW 0 ∫ τ ∗ τ fin τ 1 d τ 1 ∫ τ ∗ τ fin τ 2 d τ 2 E Π ( τ 1 , τ 2 , k ) cos ( k τ − ) , (26)
where τ − = τ 2 − τ 1 \tau_-=\tau_2-\tau_1 τ − = τ 2 − τ 1 . Since Π i j \Pi_{ij} Π ij is a statistically homogeneous and isotropic stochastic field, we can replace its ensemble average, using the ergodic assumption, by an average over the solid angle at fixed k k k in Fourier space, thus obtaining (see App. A of ref. [33 ] for details)
E Π ( τ 1 , τ 2 , k ) = k 3 2 π 2 V ∫ Ω k d Ω k 4 π Π i j ( τ 1 , k ) Π i j ∗ ( τ 2 , k ) . (27) \mathcal{E}_{\Pi}(\tau_1,\tau_2,k)=\frac{k^3}{2\pi^2V}\int_{\Omega_k}\frac{\mathrm{d}\Omega_k}{4\pi}\Pi_{ij}(\tau_1,\boldsymbol{k})\Pi_{ij}^{*}(\tau_2,\boldsymbol{k}). \tag*{(27)} E Π ( τ 1 , τ 2 , k ) = 2 π 2 V k 3 ∫ Ω k 4 π d Ω k Π ij ( τ 1 , k ) Π ij ∗ ( τ 2 , k ) . (27)
Then, substituting Eq. (27) into Eq. (26) , we find
Ω G W 0 ( k ) = 3 k 3 4 π 2 V F G W 0 ∫ Ω k d Ω k 4 π ∫ τ ∗ τ f i n d τ 1 τ 1 Π i j ( τ 1 , k ) e i k τ 1 × ∫ τ ∗ τ f i n d τ 2 τ 2 Π i j ∗ ( τ 2 , k ) e − i k τ 2 . (28) \Omega_{\mathrm{GW}}^{0}(k)=\frac{3k^3}{4\pi^2V}F_{\mathrm{GW}}^{0}\int_{\Omega_k}\frac{\mathrm{d}\Omega_k}{4\pi}\int_{\tau_*}^{\tau_{\mathrm{fin}}}\frac{\mathrm{d}\tau_1}{\tau_1}\Pi_{ij}(\tau_1,\boldsymbol{k})e^{ik\tau_1}
\times\int_{\tau_*}^{\tau_{\mathrm{fin}}}\frac{\mathrm{d}\tau_2}{\tau_2}\Pi_{ij}^{*}(\tau_2,\boldsymbol{k})e^{-ik\tau_2}. \tag*{(28)} Ω GW 0 ( k ) = 4 π 2 V 3 k 3 F GW 0 ∫ Ω k 4 π d Ω k ∫ τ ∗ τ fin τ 1 d τ 1 Π ij ( τ 1 , k ) e ik τ 1 × ∫ τ ∗ τ fin τ 2 d τ 2 Π ij ∗ ( τ 2 , k ) e − ik τ 2 . (28)
Defining the following integral over the stress-energy tensor T i j ( τ , k ) T_{ij}(\tau,\boldsymbol{k}) T ij ( τ , k ) ,
T i j ( q , k ) = ∫ τ ∗ τ f i n T i j ( τ , k ) τ e i q τ d τ , (29) T_{ij}(q,\boldsymbol{k})=\int_{\tau_*}^{\tau_{\mathrm{fin}}}\frac{T_{ij}(\tau,\boldsymbol{k})}{\tau}e^{iq\tau}\,\mathrm{d}\tau, \tag*{(29)} T ij ( q , k ) = ∫ τ ∗ τ fin τ T ij ( τ , k ) e i q τ d τ , (29)
we find
Ω G W 0 ( k ) = 3 k 3 4 π 2 V ρ ˉ 2 F G W 0 ∫ Ω k d Ω k 4 π Λ i j l m ( k ^ ) [ T i j ( q , k ) T l m ∗ ( q , k ) ] q = k . (30) \Omega_{\mathrm{GW}}^{0}(k)=\frac{3k^3}{4\pi^2V\bar{\rho}^{2}}F_{\mathrm{GW}}^{0}\int_{\Omega_k}\frac{\mathrm{d}\Omega_k}{4\pi}\Lambda_{ijlm}(\hat{\boldsymbol{k}})\left[T_{ij}(q,\boldsymbol{k})T_{lm}^{*}(q,\boldsymbol{k})\right]_{q=k}. \tag*{(30)} Ω GW 0 ( k ) = 4 π 2 V ρ ˉ 2 3 k 3 F GW 0 ∫ Ω k 4 π d Ω k Λ ij l m ( k ^ ) [ T ij ( q , k ) T l m ∗ ( q , k ) ] q = k . (30)
The expansion of the Universe is directly incorporated in our expression by including 1 / τ 1/\tau 1/ τ in the proposed definition of T i j ( q , k ) T_{ij}(q,\boldsymbol{k}) T ij ( q , k ) in Eq. (29) , compared to the one used in previous work [22 , 31 , 32 , 33 , 59 , 60 ] [19, 20], which only applied to flat Minkowski space-time.
We note that this expression can be extended to an expanding background described by a generic constant equation of state ω \omega ω , using the solution to Eq. (6) to define T i j ( q , k ) T_{ij}(q,\boldsymbol{k}) T ij ( q , k ) as
T i j ( q , k ) = H ∗ ∫ τ ∗ τ f i n ( τ τ ∗ ) ν − 1 T i j ( τ , k ) e i q τ d τ . (31) T_{ij}(q,\boldsymbol{k})=\mathcal{H}_*\int_{\tau_*}^{\tau_{\mathrm{fin}}}\left(\frac{\tau}{\tau_*}\right)^{\nu-1}T_{ij}(\tau,\boldsymbol{k})e^{iq\tau}\,\mathrm{d}\tau. \tag*{(31)} T ij ( q , k ) = H ∗ ∫ τ ∗ τ fin ( τ ∗ τ ) ν − 1 T ij ( τ , k ) e i q τ d τ . (31)
This expression applies only in the late-time limit k τ ≫ 1 k\tau\gg1 k τ ≫ 1 , when a ′ ′ / a a^{\prime\prime}/a a ′′ / a can be neglected and the Green’s function of Eq. (6) reduces to G = 1 k sin [ k ( τ − τ 1 ) ] G=\frac{1}{k}\sin[k(\tau-\tau_1)] G = k 1 sin [ k ( τ − τ 1 )] (see App. A of ref. [63 ] ).
Statistical description of vector and scalar fields
The stress tensor producing GWs can include, among others, contributions from a perfect fluid, from electromagnetic fields, and from scalar fields [70 ]
T i j ⊃ w U i U j − E i E j − B i B j + ∂ i ϕ ∂ j ϕ , (32) T_{ij}\supset wU_iU_j-E_iE_j-B_iB_j+\partial_i\phi\partial_j\phi, \tag*{(32)} T ij ⊃ w U i U j − E i E j − B i B j + ∂ i ϕ ∂ j ϕ , (32)
where U μ = γ a ( − 1 , u i ) U_\mu=\gamma a(-1,u_i) U μ = γ a ( − 1 , u i ) , u i ≡ d x i / d τ u_i\equiv\mathrm{d}x^i/\mathrm{d}\tau u i ≡ d x i / d τ is the peculiar velocity, γ 2 = 1 / ( 1 − u 2 ) \gamma^2=1/(1-u^2) γ 2 = 1/ ( 1 − u 2 ) the squared Lorentz factor, w w w is the fluid enthalpy, and E i E_i E i , B i B_i B i , and ϕ \phi ϕ respectively indicate the electric, magnetic, and scalar fields.
Furthermore, we have excluded diagonal terms in the stress-energy tensor, as they vanish after applying the TT projection and, hence, they do not produce GWs. The contributions from the velocity field can be expressed as X i X j X_iX_j X i X j , where X i = w γ u i X_i=\sqrt{w}\,\gamma u_i X i = w γ u i [39 , 57 ] . When the enthalpy and velocity perturbations are small, one can express the UETC of X i ≃ w ˉ u i X_i\simeq\sqrt{\bar{w}}u_i X i ≃ w ˉ u i , where w ˉ \bar{w} w ˉ is the mean fluid enthalpy, as being proportional to the UETC of the velocity field.
We note that the fluid enthalpy, w w w , in the case of a radiation-like fluid, and the electromagnetic energy densities, ρ E = 1 2 E i E i \rho_E=\frac{1}{2}E_iE_i ρ E = 2 1 E i E i and ρ B = 1 2 B i B i \rho_B=\frac{1}{2}B_iB_i ρ B = 2 1 B i B i , are all proportional to a − 4 a^{-4} a − 4 if the fields are not dynamically coupled and just evolve following the Universe expansion. Therefore, since we have defined Π i j = a 2 Λ i j l m T l m / ρ ˉ = Λ i j l m T l m / ρ ˉ \Pi_{ij}=a^2\Lambda_{ijlm}T^{lm}/\bar{\rho}=\Lambda_{ijlm}T^{lm}/\bar{\rho} Π ij = a 2 Λ ij l m T l m / ρ ˉ = Λ ij l m T l m / ρ ˉ in Eq. (7) , and we focus our analysis on the radiation-dominated era, such that the average fluid energy density follows ρ ˉ ∼ a − 4 \bar{\rho}\sim a^{-4} ρ ˉ ∼ a − 4 , the resulting components of Π i j \Pi_{ij} Π ij are comoving with the Universe, i.e., they remain constant in time as the Universe expands, unless they dynamically evolve due to, e.g., decaying MHD turbulence [71 , 72 , 73 , 74 ] . On the other hand, the gradient energy density of the scalar field, ρ G ϕ = 1 2 ∂ i ϕ ∂ i ϕ \rho_G^\phi=\frac{1}{2}\partial_i\phi\partial_i\phi ρ G ϕ = 2 1 ∂ i ϕ ∂ i ϕ , does not, in general, evolve proportional to a − 4 a^{-4} a − 4 . Therefore, the dynamical evolution of ρ ϕ / ρ ˉ \rho_{\phi}/\bar{\rho} ρ ϕ / ρ ˉ depends on the equation of state of the scalar field.
For each source, the stresses T i j T_{ij} T ij are expressed as a bilinear form in the vector field φ i = { a X i , i E i , i B i , ∂ i ϕ } \varphi_i = \{aX_i, iE_i, iB_i, \partial_i\phi\} φ i = { a X i , i E i , i B i , ∂ i ϕ } , which we describe in next Sec. IV. In the following, we first provide the statistical description of stochastic velocity u u u (or X X X ) and magnetic B B B fields in Sec. III A, including both longitudinal and normal spectral components, and a generic fractional helicity, allowing us to describe a broad range of hydromagnetic sources. We omit the contribution from electric fields, as we expect the conductivity to be very large during the radiation-dominated era [71 , 74 , 75 , 76 , 77 , 78 ] , such that E ≃ − u × B E \simeq-u \times B E ≃ − u × B is suppressed for subrelativistic fluid velocities. However, their inclusion in Eq. (32) and the extension of the results of this work to describe electrical fields is straightforward. Then, we provide the statistical description of the gradients of scalar fields sourcing GWs in Sec. III B. We introduce in Sec. III C a generic smoothed broken power law model for the spectral components. Analytical calculations of integrals involving this generic model are provided in App. A.
Velocity and magnetic fields
The two-point correlation function of a statistically homogeneous and isotropic velocity field is expressed in Fourier space as
⟨ u i ( k ) u j ∗ ( k ′ ) ⟩ = ( 2 π ) 6 δ 3 ( k − k ′ ) F i j u ( k ) 4 π k 3 , (33) \left\langle u_i(\boldsymbol{k})u_j^*(\boldsymbol{k}')\right\rangle= (2\pi)^6\delta^3(\boldsymbol{k}-\boldsymbol{k}')\frac{\mathcal{F}_{ij}^u(k)}{4\pi k^3}, \tag*{(33)} ⟨ u i ( k ) u j ∗ ( k ′ ) ⟩ = ( 2 π ) 6 δ 3 ( k − k ′ ) 4 π k 3 F ij u ( k ) , (33)
where F i j u ( k ) \mathcal{F}_{ij}^u(\boldsymbol{k}) F ij u ( k ) presents the following unique decomposition [65 ]
F i j u ( k ) = 1 2 P i j ( k ^ ) E N u ( k ) + k ^ i k ^ j E L u ( k ) + i 2 ϵ i j l k ^ l H u ( k ) . (34) \mathcal{F}_{ij}^u(\boldsymbol{k}) = \frac{1}{2}P_{ij}(\hat{\boldsymbol{k}})E_N^u(k)+\hat{k}_i\hat{k}_jE_L^u(k)+\frac{i}{2}\epsilon_{ijl}\hat{k}_lH_u(k). \tag*{(34)} F ij u ( k ) = 2 1 P ij ( k ^ ) E N u ( k ) + k ^ i k ^ j E L u ( k ) + 2 i ϵ ij l k ^ l H u ( k ) . (34)
The spectral functions E N u ( k ) E_N^u(k) E N u ( k ) and E L u ( k ) E_L^u(k) E L u ( k ) correspond respectively to the normal and longitudinal components of the velocity field u u u . In the case of magnetic fields, since they are solenoidal (∇ ⋅ B = 0 \nabla\cdot B=0 ∇ ⋅ B = 0 ), the longitudinal component vanishes, E L B ( k ) = 0 E_L^B(k)=0 E L B ( k ) = 0 ,
F i j B ( k ) = 1 2 P i j ( k ^ ) E B ( k ) + i 2 ϵ i j l k ^ l H B ( k ) . (35) \mathcal{F}_{ij}^B(\boldsymbol{k}) = \frac{1}{2}P_{ij}(\hat{\boldsymbol{k}})E_B(k)+\frac{i}{2}\epsilon_{ijl}\hat{k}_lH_B(k). \tag*{(35)} F ij B ( k ) = 2 1 P ij ( k ^ ) E B ( k ) + 2 i ϵ ij l k ^ l H B ( k ) . (35)
As in Eq. (19) , the spectral function F i j ( k ) \mathcal{F}_{ij}(\boldsymbol{k}) F ij ( k ) is defined such that its trace is the velocity or the magnetic spectrum:
E u ( k ) ≡ F i i u ( k ) = E N u ( k ) + E L u ( k ) , (36) E_u(k)\equiv\mathcal{F}_{ii}^u(k)=E_N^u(k)+E_L^u(k), \tag*{(36)} E u ( k ) ≡ F ii u ( k ) = E N u ( k ) + E L u ( k ) , (36)
and E B ( k ) = F i i B ( k ) E_B(k)=\mathcal{F}_{ii}^B(k) E B ( k ) = F ii B ( k ) . Integrated over ln k \ln k ln k , these spectra correspond to the ensemble average of the squared field, i.e., u r m s 2 u_{\mathrm{rms}}^2 u rms 2 and B r m s 2 B_{\mathrm{rms}}^2 B rms 2 . Under the ergodic assumption, this is approximated as a volume average [65 ]
u r m s 2 = ⟨ u 2 ( x ) ⟩ V = ∫ − ∞ ∞ E u ( k ) d ln k , (37) u_{\mathrm{rms}}^2=\left\langle u^2(\boldsymbol{x})\right\rangle_V=\int_{-\infty}^{\infty}E_u(k)\,\mathrm{d}\ln k, \tag*{(37)} u rms 2 = ⟨ u 2 ( x ) ⟩ V = ∫ − ∞ ∞ E u ( k ) d ln k , (37)
B r m s 2 = ⟨ B 2 ( x ) ⟩ V = ∫ − ∞ ∞ E B ( k ) d ln k . (38) B_{\mathrm{rms}}^2=\left\langle B^2(\boldsymbol{x})\right\rangle_V=\int_{-\infty}^{\infty}E_B(k)\,\mathrm{d}\ln k. \tag*{(38)} B rms 2 = ⟨ B 2 ( x ) ⟩ V = ∫ − ∞ ∞ E B ( k ) d ln k . (38)
We also characterize the degree of compressibility in the velocity field as the ratio of the longitudinal component to the total spectrum,
q ( k ) ≡ E L u ( k ) E L u ( k ) + E N u ( k ) ∈ [ 0 , 1 ] . (39) q(k)\equiv\frac{E_L^u(k)}{E_L^u(k)+E_N^u(k)}\in[0,1]. \tag*{(39)} q ( k ) ≡ E L u ( k ) + E N u ( k ) E L u ( k ) ∈ [ 0 , 1 ] . (39)
Then, the longitudinal and normal components can be expressed as fractions of the velocity spectrum,
E L u ( k ) = q ( k ) E u ( k ) , (40) E_L^u(k)=q(k)E_u(k), \tag*{(40)} E L u ( k ) = q ( k ) E u ( k ) , (40)
E N u ( k ) = [ 1 − q ( k ) ] E u ( k ) . (41) E_N^u(k)=[1-q(k)]E_u(k). \tag*{(41)} E N u ( k ) = [ 1 − q ( k )] E u ( k ) . (41)
Note that for magnetic fields, we always have q ( k ) = 0 q(k)=0 q ( k ) = 0 .
Causality implies that the spectrum of a generic vector field becomes proportional to k a E k^{a_E} k a E with odd a E ≥ 3 a_E\geq3 a E ≥ 3 in the k → 0 k\to0 k → 0 limit [65 ] ,
E ( k ) = E N ( k ) + E L ( k ) ∼ k a E with a E ≥ 3. (42) E(k)=E_N(k)+E_L(k)\sim k^{a_E}\quad\text{with }a_E\geq3. \tag*{(42)} E ( k ) = E N ( k ) + E L ( k ) ∼ k a E with a E ≥ 3. (42)
However, analyticity of Eq. (30b) implies that the term proportional to k ^ i k ^ j = k i k j / k 2 \hat{k}_i\hat{k}_j=k_i k_j/k^2 k ^ i k ^ j = k i k j / k 2 should be proportional to k a E k^{a_E} k a E with odd a E ≥ 5 a_E\geq5 a E ≥ 5 when k → 0 k\to0 k → 0 [79 ] ,
∣ E L ( k ) − 1 2 E N ( k ) ∣ ∼ k a E with a E ≥ 5. (43) \left|E_L(k)-\frac{1}{2}E_N(k)\right|\sim k^{a_E}\quad\text{with }a_E\geq5. \tag*{(43)} E L ( k ) − 2 1 E N ( k ) ∼ k a E with a E ≥ 5. (43)
Therefore, the spectrum of the source, F i j ( k ) \mathcal{F}_{ij}(k) F ij ( k ) becomes proportional to k 5 k^5 k 5 or steeper when the field satisfies one of the following two conditions: i ) i) i ) the field is irrotational (ω = ∇ × u = 0 \boldsymbol{\omega}=\nabla\times u=0 ω = ∇ × u = 0 ), such that E N ( k ) = 0 E_N(k)=0 E N ( k ) = 0 , or i i ) ii) ii ) the field is solenoidal (∇ ⋅ u = 0 \nabla\cdot u=0 ∇ ⋅ u = 0 or ∇ ⋅ B = 0 \nabla\cdot B=0 ∇ ⋅ B = 0 ), such that E L ( k ) = 0 E_L(k)=0 E L ( k ) = 0 [65 , 79 ] (see also App. C of ref. [39] for details).
Helical spectra H ( k ) H(k) H ( k ) are associated to mirror symmetry breaking in the statistical distribution of the vector field and, hence, H ( k ) = 0 H(k)=0 H ( k ) = 0 when the statistics of a vector field satisfy parity symmetry. They characterize the kinetic helicity H u ≡ ⟨ ω ⋅ u ⟩ \mathcal{H}_u\equiv\langle\boldsymbol{\omega}\cdot u\rangle H u ≡ ⟨ ω ⋅ u ⟩ or the magnetic helicity H A ≡ ⟨ A ⋅ B ⟩ \mathcal{H}_A\equiv\langle A\cdot B\rangle H A ≡ ⟨ A ⋅ B ⟩ [80 ] ,
H u ≡ ⟨ ω ( x ) ⋅ u ( x ) ⟩ V = ∫ − ∞ ∞ k H u ( k ) d ln k , (44) \mathcal{H}_u\equiv\langle\boldsymbol{\omega}(\boldsymbol{x})\cdot u(\boldsymbol{x})\rangle_V=\int_{-\infty}^{\infty}kH_u(k)\,\mathrm{d}\ln k, \tag*{(44)} H u ≡ ⟨ ω ( x ) ⋅ u ( x ) ⟩ V = ∫ − ∞ ∞ k H u ( k ) d ln k , (44)
H A ≡ ⟨ A ( x ) ⋅ B ( x ) ⟩ V = ∫ − ∞ ∞ H B ( k ) k d ln k . (45) \mathcal{H}_A\equiv\langle A(\boldsymbol{x})\cdot B(\boldsymbol{x})\rangle_V=\int_{-\infty}^{\infty}\frac{H_B(k)}{k}\,\mathrm{d}\ln k. \tag*{(45)} H A ≡ ⟨ A ( x ) ⋅ B ( x ) ⟩ V = ∫ − ∞ ∞ k H B ( k ) d ln k . (45)
where A A A is the vector potential, such that B = ∇ × A \boldsymbol{B} = \nabla\times A B = ∇ × A . Note that the extra k k k and k − 1 k^{-1} k − 1 in the integrals compared to Eqs. (33) appear to match, respectively, the dimensions of ω ( k ) = − i k × u ( k ) \boldsymbol{\omega}(k) = -i\,\boldsymbol{k} \times\boldsymbol{u}(k) ω ( k ) = − i k × u ( k ) and B ( k ) = − i k × A ( k ) \boldsymbol{B}(k) = -i\boldsymbol{k} \times A(k) B ( k ) = − i k × A ( k ) . Again, analyticity in Eq. (30b) requires that the helical spectrum of a causal field becomes proportional to k a H k^{a_H} k a H with even a H ≥ 4 a_H \ge4 a H ≥ 4 when k → 0 k \to0 k → 0 [68 , 79 ] ,
H ( k ) ∼ k a H with a H ≥ 4. (46) H(k) \sim k^{a_H} \quad\text{with } a_H \ge4. \tag*{(46)} H ( k ) ∼ k a H with a H ≥ 4. (46)
In ideal MHD, the current helicity is defined as
H B ≡ ⟨ B ( x ) ⋅ J ( x ) ⟩ V = ∫ − ∞ ∞ k H B ( k ) d ln k , (47) \mathcal{H}_B \equiv\langle\boldsymbol{B}(x) \cdot\boldsymbol{J}(x) \rangle_V = \int_{-\infty}^{\infty} k H_B(k)\,\mathrm{d}\ln k, \tag*{(47)} H B ≡ ⟨ B ( x ) ⋅ J ( x ) ⟩ V = ∫ − ∞ ∞ k H B ( k ) d ln k , (47)
where J = ∇ × B \boldsymbol{J} = \nabla\times\boldsymbol{B} J = ∇ × B is the current density under the assumption of infinite conductivity. We note that we are using the generic definition of the helicity of a vector field φ i ( x ) \varphi_i(x) φ i ( x ) as
H φ ≡ ⟨ φ ( x ) ⋅ [ ∇ × φ ( x ) ] ⟩ = ∫ − ∞ ∞ k H φ ( k ) d ln k . (48) \begin{aligned}
\mathcal{H}_{\varphi} \equiv\langle\varphi(x) \cdot[\nabla\times\varphi(x)] \rangle\\
&= \int_{-\infty}^{\infty} k H_{\varphi}(k)\,\mathrm{d}\ln k. \tag*{(48)}
\end{aligned} H φ ≡ ⟨ φ ( x ) ⋅ [ ∇ × φ ( x )]⟩ = ∫ − ∞ ∞ k H φ ( k ) d ln k . (48)
Based on the kinetic and the magnetic helicities, H u \mathcal{H}_u H u and H A \mathcal{H}_A H A , we can define helical length scales ξ u H \xi_u^H ξ u H and ξ A H \xi_A^H ξ A H as
ξ u H = ∫ E u ( k ) d ln k ∫ k H u ( k ) d ln k = u r m s 2 H u , (49) \xi_u^H = \frac{\int E_u(k)\,\mathrm{d}\ln k}{\int k H_u(k)\,\mathrm{d}\ln k} = \frac{u_{\mathrm{rms}}^2}{\mathcal{H}_u}, \tag*{(49)} ξ u H = ∫ k H u ( k ) d ln k ∫ E u ( k ) d ln k = H u u rms 2 , (49)
ξ A H = ∫ k − 1 H B ( k ) d ln k ∫ E B ( k ) d ln k = H A B r m s 2 . (50) \xi_A^H = \frac{\int k^{-1}H_B(k)\,\mathrm{d}\ln k}{\int E_B(k)\,\mathrm{d}\ln k} = \frac{\mathcal{H}_A}{B_{\mathrm{rms}}^2}. \tag*{(50)} ξ A H = ∫ E B ( k ) d ln k ∫ k − 1 H B ( k ) d ln k = B rms 2 H A . (50)
The helicity of a vector field can always be expressed as H φ = φ r m s 2 / ξ φ H \mathcal{H}_{\varphi} = \varphi_{\mathrm{rms}}^2/\xi_{\varphi}^H H φ = φ rms 2 / ξ φ H with ξ φ H \xi_{\varphi}^H ξ φ H defined as in Eq. (42a). Note that Eq. (42b) should not be confused with the magnetic integral length scale used in MHD literature to describe the decay of primordial magnetic fields [44 , 65 , 73 , 81 , 82 , 83 ] ,
ξ B = ∫ k − 1 E B ( k ) d ln k ∫ E B ( k ) d ln k , (51) \xi_B = \frac{\int k^{-1}E_B(k)\,\mathrm{d}\ln k}{\int E_B(k)\,\mathrm{d}\ln k}, \tag*{(51)} ξ B = ∫ E B ( k ) d ln k ∫ k − 1 E B ( k ) d ln k , (51)
which is not directly related to magnetic helicity. The realizability condition bounds the value of the helical spectrum [46 , 49 , 66 , 67 , 69 , 80 ] ,
∣ H ( k ) ∣ ≤ E N ( k ) , (52) \lvert H(k) \rvert\le E_N(k), \tag*{(52)} ∣ H ( k )∣ ≤ E N ( k ) , (52)
and the fractional helicity determines the ratio to exhaust the inequality,
ϵ ( k ) = H ( k ) E N ( k ) ∈ [ − 1 , 1 ] , (53) \epsilon(k) = \frac{H(k)}{E_N(k)} \in[-1,1], \tag*{(53)} ϵ ( k ) = E N ( k ) H ( k ) ∈ [ − 1 , 1 ] , (53)
defined when E N ( k ) ≠ 0 E_N(k) \ne0 E N ( k ) = 0 .
defined when E N ( k ) ≠ 0 E_N(k) \ne0 E N ( k ) = 0 . Note that E N ( k ) = 0 E_N(k) = 0 E N ( k ) = 0 iff ω = ∇ × u = 0 \boldsymbol{\omega} = \nabla\times\boldsymbol{u} = 0 ω = ∇ × u = 0 [65 ] , then H = 0 \mathcal{H} = 0 H = 0 and, hence, H ( k ) = 0 H(k) = 0 H ( k ) = 0 . The realizability condition also applies for the helical and integral length scales defined in Eqs. (42b) and (43), ξ B ≥ ξ A H \xi_B \ge\xi_A^H ξ B ≥ ξ A H or, equivalently, H A ≤ ξ B B r m s 2 \mathcal{H}_A \le\xi_B B_{\mathrm{rms}}^2 H A ≤ ξ B B rms 2 . Using the fractional helicity, we can rewrite the helical spectra of velocity and magnetic fields as
H u ( k ) = ϵ u ( k ) [ 1 − q ( k ) ] E u ( k ) , (54) H_u(k) = \epsilon_u(k)[1-q(k)]E_u(k), \tag*{(54)} H u ( k ) = ϵ u ( k ) [ 1 − q ( k )] E u ( k ) , (54)
H B ( k ) = ϵ B ( k ) E B ( k ) . (55) H_B(k) = \epsilon_B(k) E_B(k). \tag*{(55)} H B ( k ) = ϵ B ( k ) E B ( k ) . (55)
The realizability condition in Eq. (52) also implies that a H ≥ a E a_H \ge a_E a H ≥ a E in the k → 0 k \to0 k → 0 limit, which, together with causality (or analyticity of the correlation function), requires that H ( k → 0 ) ∼ k a H H(k \to0) \sim k^{a_H} H ( k → 0 ) ∼ k a H with even a H ≥ 6 a_H \ge6 a H ≥ 6 when the field is solenoidal with E N ( k → 0 ) ∼ k a E E_N(k \to0) \sim k^{a_E} E N ( k → 0 ) ∼ k a E and odd a E ≥ 5 a_E \ge5 a E ≥ 5 [79 ] . Since this is always the case for magnetic fields, we have
H B ( k → 0 ) ∼ k a H with a H ≥ 6. (56) H_B(k \to0) \sim k^{a_H} \quad\text{with } a_H \ge6. \tag*{(56)} H B ( k → 0 ) ∼ k a H with a H ≥ 6. (56)
In general, we expect ϵ B ( k ) ∼ k \epsilon_B(k) \sim k ϵ B ( k ) ∼ k (or steeper) when k → 0 k \to0 k → 0 .
As mentioned above, when the enthalpy and velocity perturbations are small, the velocity stresses can be approximated as X i X j = w γ 2 u i u j ≃ w u ˉ i u ˉ j X_iX_j = w\gamma^2 u_i u_j \simeq w\bar{u}_i\bar{u}_j X i X j = w γ 2 u i u j ≃ w u ˉ i u ˉ j . Hence, the spectral function F i j X ( k ) \mathcal{F}_{ij}^X(k) F ij X ( k ) , defined as
⟨ X i ( k ) X j ∗ ( k ′ ) ⟩ = ( 2 π ) 6 δ 3 ( k − k ′ ) F i j X ( k ) 4 π k 3 , (57) \langle X_i(k)X_j^*(k')\rangle= (2\pi)^6\delta^3(k-k')\frac{\mathcal{F}_{ij}^X(k)}{4\pi k^3}, \tag*{(57)} ⟨ X i ( k ) X j ∗ ( k ′ )⟩ = ( 2 π ) 6 δ 3 ( k − k ′ ) 4 π k 3 F ij X ( k ) , (57)
relevant to determine the source of GWs, can be approximated as
F i j X ( k ) ≃ w ˉ F i j u ( k ) . (58) \mathcal{F}_{ij}^X(k) \simeq\bar{w}\,\mathcal{F}_{ij}^u(k). \tag*{(58)} F ij X ( k ) ≃ w ˉ F ij u ( k ) . (58)
Similarly, the mean fluid kinetic energy density is ρ ˉ k i n = ⟨ X 2 ⟩ = X r m s 2 ≃ w ˉ ⟨ u 2 ⟩ = w ˉ u r m s 2 \bar{\rho}_{\mathrm{kin}} = \langle X^2\rangle= X_{\mathrm{rms}}^2 \simeq\bar{w}\langle u^2\rangle= \bar{w}u_{\mathrm{rms}}^2 ρ ˉ kin = ⟨ X 2 ⟩ = X rms 2 ≃ w ˉ ⟨ u 2 ⟩ = w ˉ u rms 2 . In many parts of this work, we take this approximation to describe the source of GWs in terms of the velocity field as this has been commonly done in the literature. However, we note that our description can in general be applied to large enthalpy and velocity fluctuations using the spectral components of X i X_i X i instead of u i u_i u i [57 ] .
Scalar fields
Scalar fields source GWs through their kinetic term, proportional to their gradients, T i j ⊃ ∂ i ϕ ∂ j ϕ T_{ij} \supset\partial_i\phi\,\partial_j\phi T ij ⊃ ∂ i ϕ ∂ j ϕ . The two-point correlator of this source is then purely longitudinal,
⟨ ∂ i ϕ ( k ) ∂ j ϕ ∗ ( k ′ ) ⟩ = k k ′ k ^ i k ^ j ′ ⟨ ϕ ( k ) ϕ ∗ ( k ′ ) ⟩ = ( 2 π ) 6 δ 3 ( k − k ′ ) k ^ i k ^ j E ϕ ( k ) 4 π k (59) \left\langle\partial_i\phi(\boldsymbol{k})\,\partial_j\phi^*(\boldsymbol{k}')\right\rangle= kk'\hat{k}_i\hat{k}'_j\left\langle\phi(\boldsymbol{k})\phi^*(\boldsymbol{k}')\right\rangle
= (2\pi)^6\delta^3(\boldsymbol{k}-\boldsymbol{k}')\hat{k}_i\hat{k}_j\frac{E_\phi(k)}{4\pi k} \tag*{(59)} ⟨ ∂ i ϕ ( k ) ∂ j ϕ ∗ ( k ′ ) ⟩ = k k ′ k ^ i k ^ j ′ ⟨ ϕ ( k ) ϕ ∗ ( k ′ ) ⟩ = ( 2 π ) 6 δ 3 ( k − k ′ ) k ^ i k ^ j 4 π k E ϕ ( k ) (59)
such that E ϕ L ( k ) = k 2 E ϕ ( k ) E_\phi^{\mathrm{L}}(k)=k^2E_\phi(k) E ϕ L ( k ) = k 2 E ϕ ( k ) . Therefore, this source can be described with a single longitudinal spectral function, E ϕ L ( k ) E_\phi^{\mathrm{L}}(k) E ϕ L ( k ) . In this case, causality implies that E ϕ L ( k ) = k 2 E ϕ ( k ) ∼ k 2 + a E E_\phi^{\mathrm{L}}(k)=k^2E_\phi(k)\sim k^{2+a_E} E ϕ L ( k ) = k 2 E ϕ ( k ) ∼ k 2 + a E with odd a E ≥ 3 a_E\geq3 a E ≥ 3 .
The mean squared value of the scalar field and the mean gradient energy density are defined as
ϕ r m s 2 ≡ ⟨ ϕ 2 ( x ) ⟩ V = ∫ − ∞ ∞ E ϕ ( k ) d ln k (60) \phi_{\mathrm{rms}}^2\equiv\left\langle\phi^2(x)\right\rangle_V=\int_{-\infty}^{\infty}E_\phi(k)\,\mathrm{d}\ln k \tag*{(60)} ϕ rms 2 ≡ ⟨ ϕ 2 ( x ) ⟩ V = ∫ − ∞ ∞ E ϕ ( k ) d ln k (60)
ρ G ϕ ≡ 1 2 a 2 ⟨ ∣ ∇ ϕ ∣ 2 ( x ) ⟩ V = 1 2 a 2 ∫ − ∞ ∞ E ϕ L ( k ) d ln k . (61) \rho_G^\phi\equiv\frac{1}{2a^2}\left\langle\lvert\nabla\phi\rvert^2(x)\right\rangle_V=\frac{1}{2a^2}\int_{-\infty}^{\infty}E_\phi^{\mathrm{L}}(k)\,\mathrm{d}\ln k. \tag*{(61)} ρ G ϕ ≡ 2 a 2 1 ⟨ ∣ ∇ ϕ ∣ 2 ( x ) ⟩ V = 2 a 2 1 ∫ − ∞ ∞ E ϕ L ( k ) d ln k . (61)
We can define a squared length scale ξ ϕ 2 \xi_\phi^2 ξ ϕ 2 that characterizes their ratio as
ξ ϕ 2 = ∫ k − 2 E ϕ L ( k ) d ln k ∫ E ϕ L ( k ) d ln k = ϕ r m s 2 2 a 2 ρ G ϕ . (62) \xi_\phi^2=\frac{\int k^{-2}E_\phi^{\mathrm{L}}(k)\,\mathrm{d}\ln k}{\int E_\phi^{\mathrm{L}}(k)\,\mathrm{d}\ln k}=\frac{\phi_{\mathrm{rms}}^2}{2a^2\rho_G^\phi}. \tag*{(62)} ξ ϕ 2 = ∫ E ϕ L ( k ) d ln k ∫ k − 2 E ϕ L ( k ) d ln k = 2 a 2 ρ G ϕ ϕ rms 2 . (62)
Smoothed broken power law model
A common description for the spectral shape of a causal field sourcing GWs (e.g., velocity, electromagnetic, or scalar field) is a broken power law in k k k , presenting different scaling at small and large k k k .
As discussed in previous sections, due to causality, it is expected to find a Saffman spectrum at large scales [84 ] , i.e., E ( k → 0 ) ∼ k a E E(k\to0)\sim k^{a_E} E ( k → 0 ) ∼ k a E with odd a E ≥ 3 a_E\geq3 a E ≥ 3 , where a E = 3 a_E=3 a E = 3 corresponds to white noise. This condition applies to the spectra of all of the sourcing fields, as well as to the one of the anisotropic stresses, E Π ( k ) E_\Pi(k) E Π ( k ) , and of the GWs, Ω G W ( k ) \Omega_{\mathrm{GW}}(k) Ω GW ( k ) [41 , 65 ] . Since values of a E < 3 a_E<3 a E < 3 would violate causality at large scales, they cannot be found in fields generated by causal processes during the radiation-dominated era. However, a < 3 a<3 a < 3 can be found in fields generated during inflation as, for example, inflationary magnetic fields [66 , 68 , 85 , 86 ] . On the other hand, for the helical components of the spectra, H ( k ) H(k) H ( k ) , as well as the chiral H Π ( k ) H_\Pi(k) H Π ( k ) and Ξ G W ( k ) \Xi_{\mathrm{GW}}(k) Ξ GW ( k ) , the general causality condition requires them to be proportional to k a H k^{a_H} k a H , with even a H ≥ 4 a_H\geq4 a H ≥ 4 , in the k → 0 k\to0 k → 0 limit [79 ] .
In the particular case of vector fields spectra, this causality condition can be more restrictive. Indeed, velocity and magnetic fields spectra are required to follow a Batchelor or steeper spectrum (a E ≥ 5 a_E\geq5 a E ≥ 5 ) at large scales when the field is irrotational or solenoidal [see Eq. (37)]. Due to the realizability condition, ∣ H ( k ) ∣ ≤ E N ( k ) \lvert H(k)\rvert\leq E_N(k) ∣ H ( k )∣ ≤ E N ( k ) [see Eq. (44)], when the field is solenoidal, causality implies that H ( k → 0 ) ∼ k a H H(k\to0)\sim k^{a_H} H ( k → 0 ) ∼ k a H with even a H ≥ 6 a_H\geq6 a H ≥ 6 [79 ] .
At larger wave numbers, i.e., smaller scales, the field becomes correlated. This is commonly characterized by the integral length scale ξ \xi ξ , defined in Eq. (51) . For k ≫ k ∗ k\gg k_* k ≫ k ∗ , where k ∗ ∼ ξ − 1 k_*\sim\xi^{-1} k ∗ ∼ ξ − 1 is the spectral peak, the spectrum decays with a slope that depends on the properties of the source. For example, velocity fields in stationary Kolmogorov turbulence2 are expected to follow E u ( k ) ∼ k − 2 / 3 E_u(k)\sim k^{-2/3} E u ( k ) ∼ k − 2/3 at k ≫ k ∗ k\gg k_* k ≫ k ∗ , while the spectrum of acoustic perturbations (i.e., purely compressional velocity field with ω = 0 \omega=0 ω = 0 ) is expected to be E u ( k ) ∼ k − 1 E_u(k)\sim k^{-1} E u ( k ) ∼ k − 1 at k ≫ k ∗ k\gg k_* k ≫ k ∗ [26 , 27 , 28 , 29 , 30 , 32 , 33 , 34 , 36 , 37 , 39 , 43 ] .
Eventually, at larger wave numbers, the spectra are expected to follow an exponential damping due to diffusive processes, for example, viscosity and magnetic diffusivity dissipating respectively velocity and magnetic fields at small scales. This is a generic feature that is also expected for acoustic turbulence at wave numbers above the inverse scale that characterizes the thickness of the shocks [32 , 33 , 36 , 37 , 38 ] and for sound waves in first-order phase transitions for scales above the inverse bubble wall thickness [39 ] .
In the case of a scalar field in a first-order phase transition, its spectrum is less well understood as it has been less explored numerically. The simulations of refs. [24 , 25 ] show that E ϕ ( k ) E_\phi(k) E ϕ ( k ) becomes proportional to k 3 k^3 k 3 at small k ≪ k ∗ k\ll k_* k ≪ k ∗ , as expected from causality, and it becomes proportional to k − b ϕ k^{-b_\phi} k − b ϕ with 2.5 ≲ b ϕ ≲ 3 2.5\lesssim b_\phi\lesssim3 2.5 ≲ b ϕ ≲ 3 in the thin-wall limit, i.e., when the two vacua of the first-order phase transition are nearly degenerate (see, for example, Figs. 6–7 in ref. [24 ] ).
For generality, in the present work we provide the theoretical framework to study smoothed broken power law spectra of the fields sourcing GWs, φ = { X , B , ϕ } \varphi=\{X,B,\phi\} φ = { X , B , ϕ } , which transition from k a k^{a} k a at k ≪ k ∗ k\ll k_\ast k ≪ k ∗ to k − b k^{-b} k − b at k ≫ k ∗ k\gg k_\ast k ≫ k ∗ , for generic slopes a , b > 0 a,b>0 a , b > 0 ,
E ( K ) = E ∗ ( b + a ) 1 α K a [ b + a K α ( a + b ) ] 1 α = E ∗ ζ ( K ) , (63) E(K)=E_\ast\frac{(b+a)^{\frac{1}{\alpha}}K^a}{\left[b+aK^{\alpha(a+b)}\right]^{\frac{1}{\alpha}}}=E_\ast\zeta(K), \tag*{(63)} E ( K ) = E ∗ [ b + a K α ( a + b ) ] α 1 ( b + a ) α 1 K a = E ∗ ζ ( K ) , (63)
where K ≡ k / k ∗ K\equiv k/k_\ast K ≡ k / k ∗ is the normalized wave number and α \alpha α is the smoothness between slopes around k ∗ k_\ast k ∗ . The normalized spectral shape ζ ( K ) \zeta(K) ζ ( K ) is defined such that ζ ( 1 ) = 1 \zeta(1)=1 ζ ( 1 ) = 1 and, hence, E ( 1 ) = E ∗ E(1)=E_\ast E ( 1 ) = E ∗ . The prefactor ( b + a ) 1 α (b+a)^{\frac{1}{\alpha}} ( b + a ) α 1 and the values of a a a and b b b in the denominator have been included to ensure that the spectral maximum is reached at k ∗ k_\ast k ∗ with the normalized value ζ ( 1 ) = 1 \zeta(1)=1 ζ ( 1 ) = 1 .
The model function of Eq. (63) has already been considered in previous studies of GW production from turbulence [35 , 50 , 51 ] [1, 9, 18] and sound waves [34 ] . In previous work, it was sometimes alternatively expressed as in Eq. (230) , which is equivalent to Eq. (63) taking α 2 = α ( a + b ) \alpha_2=\alpha(a+b) α 2 = α ( a + b ) . The detailed analytical calculations in this section are presented in App. A. For numerical values presented in this section, unless noted otherwise, we consider as an example the von Kármán spectrum with a = 5 a=5 a = 5 and b = 2 / 3 b=2/3 b = 2/3 , describing vortical Kolmogorov turbulence.
The average value of the squared field is
φ r m s 2 = ∫ − ∞ ∞ E ( k ) d ln k = E ∗ A a b ( α ) , (64) \varphi_{\mathrm{rms}}^2=\int_{-\infty}^{\infty}E(k)\,\mathrm{d}\ln k=E_\ast A_{ab}(\alpha), \tag*{(64)} φ rms 2 = ∫ − ∞ ∞ E ( k ) d ln k = E ∗ A ab ( α ) , (64)
with φ r m s 2 / a 2 ≡ φ ~ r m s 2 ≡ { ρ ˉ k i n , 2 ρ ˉ B , 2 ρ ˉ G ϕ } \varphi_{\mathrm{rms}}^2/a^2\equiv\widetilde{\varphi}_{\mathrm{rms}}^2\equiv\{\bar{\rho}_{\mathrm{kin}},2\bar{\rho}_B,2\bar{\rho}_G^\phi\} φ rms 2 / a 2 ≡ φ rms 2 ≡ { ρ ˉ kin , 2 ρ ˉ B , 2 ρ ˉ G ϕ } , where ρ ˉ k i n = X r m s 2 \bar{\rho}_{\mathrm{kin}}=X_{\mathrm{rms}}^2 ρ ˉ kin = X rms 2 , ρ ˉ B = 1 2 B r m s 2 \bar{\rho}_B=\frac{1}{2}B_{\mathrm{rms}}^2 ρ ˉ B = 2 1 B rms 2 and ρ ˉ G ϕ \bar{\rho}_G^\phi ρ ˉ G ϕ are the mean kinetic, magnetic, and scalar gradient energy densities. In general, we assume that both a a a and b b b are positive, which ensures that φ r m s 2 \varphi_{\mathrm{rms}}^2 φ rms 2 is finite. The geometric factor A a b A_{ab} A ab is [see Eq. (235) ]
A a b ( α ) ≡ ∫ − ∞ ∞ ζ ( K ) d ln K = I a b ; − 1 ( α ) = ( a + b b ) 1 α ( a b ) − a α ( a + b ) Γ [ a α ( a + b ) ] Γ [ b α ( a + b ) ] α ( a + b ) Γ [ 1 α ] . (65) \begin{aligned}
A_{ab}(\alpha)&\equiv\int_{-\infty}^{\infty}\zeta(K)\,\mathrm{d}\ln K=I_{ab;-1}(\alpha)\\
&=\left(\frac{a+b}{b}\right)^{\frac{1}{\alpha}}\left(\frac{a}{b}\right)^{-\frac{a}{\alpha(a+b)}}\frac{\Gamma\left[\frac{a}{\alpha(a+b)}\right]\Gamma\left[\frac{b}{\alpha(a+b)}\right]}{\alpha(a+b)\Gamma\left[\frac{1}{\alpha}\right]}. \tag*{(65)}
\end{aligned} A ab ( α ) ≡ ∫ − ∞ ∞ ζ ( K ) d ln K = I ab ; − 1 ( α ) = ( b a + b ) α 1 ( b a ) − α ( a + b ) a α ( a + b ) Γ [ α 1 ] Γ [ α ( a + b ) a ] Γ [ α ( a + b ) b ] . (65)
shown in Figure 28 , where I a b ; n I_{ab;n} I ab ; n corresponds to the moments of ζ ( K ) \zeta(K) ζ ( K ) [see Eq. (232) ],
Figure 28. Left panel: Integrated amplitude A a b ( α ) ≡ φ r m s 2 / E ∗ A_{ab}(\alpha) \equiv\varphi_{\mathrm{rms}}^{2}/E_{*} A ab ( α ) ≡ φ rms 2 / E ∗ [see Eq. (228) ] for different values of α \alpha α and for the slopes a = 3 a=3 a = 3 (red) and 5 5 5 (blue), and b = 2 / 3 b=2/3 b = 2/3 (solid) and 1 1 1 (dashed). The models used in refs. [1] and refs. [35 , 50 ] [18] correspond to α = 2 \alpha=2 α = 2 and α 2 = 2 \alpha_{2}=2 α 2 = 2 (i.e., α = 6 / 17 \alpha=6/17 α = 6/17 ), respectively. Right panel: Geometric factor A ~ a b ( α ) \widetilde{A}_{ab}(\alpha) A ab ( α ) found for the normalization of the spectrum E − 1 ( K ) ≡ d φ r m s 2 / d K E_{-1}(K)\equiv\mathrm{d}\varphi_{\mathrm{rms}}^{2}/\mathrm{d}K E − 1 ( K ) ≡ d φ rms 2 / d K used in previous work [see Eq. (252) ] for different values of α \alpha α and the same slopes as in the left panel.
I a b ; n ( α ) = ∫ 0 ∞ ζ ( K ) K n d K . (66) I_{ab;n}(\alpha)=\int_{0}^{\infty}\zeta(K)K^n\,\mathrm{d}K. \tag*{(66)} I ab ; n ( α ) = ∫ 0 ∞ ζ ( K ) K n d K . (66)
The geometric factor A a b A_{ab} A ab allows us to define the spectral shape in terms of the integrated spectrum,
i.e., the mean squared φ r m s 2 \varphi_{\mathrm{rms}}^2 φ rms 2 , by substituting Eq. (64) into Eq. (63) ,
E ( K ) = φ r m s 2 ζ i n t ( K ) , E(K)=\varphi_{\mathrm{rms}}^2\zeta_{\mathrm{int}}(K), E ( K ) = φ rms 2 ζ int ( K ) ,
with ζ i n t ( K ) ≡ ζ ( K ) A a b ( α ) \zeta_{\mathrm{int}}(K)\equiv\frac{\zeta(K)}{A_{ab}(\alpha)} ζ int ( K ) ≡ A ab ( α ) ζ ( K ) .
∫ ζ i n t ( K ) d ln K = 1. \int\zeta_{\mathrm{int}}(K)\,\mathrm{d}\ln K=1. ∫ ζ int ( K ) d ln K = 1.
Unlike ζ ( K ) \zeta(K) ζ ( K ) , the spectrum ζ i n t ( K ) \zeta_{\mathrm{int}}(K) ζ int ( K ) is normalized by its integral, ∫ ζ i n t ( K ) d ln K = 1 \int\zeta_{\mathrm{int}}(K)\,\mathrm{d}\ln K=1 ∫ ζ int ( K ) d ln K = 1 . It is shown in Figure 1 for a = 5 a=5 a = 5 , b = 2 / 3 b=2/3 b = 2/3 , and different values of α \alpha α . Larger values of α \alpha α correspond to sharper transitions and, hence, the corresponding values at the peak increase with α \alpha α to keep the same integrated values.
Figure 1. Normalized smoothed broken power law ζ i n t ( K ) ≡ ζ ( K ) / A a b ( α ) \zeta_{\mathrm{int}}(K)\equiv\zeta(K)/A_{ab}(\alpha) ζ int ( K ) ≡ ζ ( K ) / A ab ( α ) describing a von Kármán spectrum (i.e., a = 5 a=5 a = 5 and b = 2 / 3 b=2/3 b = 2/3 ), for different values of the smoothing parameter α \alpha α . For reference, the models of refs. [1] and [35 , 43 , 50 ] [18] use α = 2 \alpha=2 α = 2 and α = 6 / 17 \alpha=6/17 α = 6/17 , respectively.
For velocity fields with longitudinal (L), normal (N), and helical (H) components, each component can be expressed in terms of ζ i n t ν ( K ) \zeta_{\mathrm{int}}^\nu(K) ζ int ν ( K ) as
E L ν ( K ) = u r m s 2 q ( K ) ζ i n t ν ( K ) , (67) E_L^\nu(K)=u_{\mathrm{rms}}^2q(K)\zeta_{\mathrm{int}}^\nu(K), \tag*{(67)} E L ν ( K ) = u rms 2 q ( K ) ζ int ν ( K ) , (67)
E N ν ( K ) = u r m s 2 [ 1 − q ( K ) ] ζ i n t ν ( K ) , (68) E_N^\nu(K)=u_{\mathrm{rms}}^2[1-q(K)]\zeta_{\mathrm{int}}^\nu(K), \tag*{(68)} E N ν ( K ) = u rms 2 [ 1 − q ( K )] ζ int ν ( K ) , (68)
H u ( K ) = u r m s 2 ϵ u ( K ) [ 1 − q ( K ) ] ζ i n t ν ( K ) , (69) H_u(K)=u_{\mathrm{rms}}^2\epsilon_u(K)[1-q(K)]\zeta_{\mathrm{int}}^\nu(K), \tag*{(69)} H u ( K ) = u rms 2 ϵ u ( K ) [ 1 − q ( K )] ζ int ν ( K ) , (69)
where q ( K ) q(K) q ( K ) and ϵ u ( K ) \epsilon_u(K) ϵ u ( K ) are the degree of compressibility and fractional helicity introduced in Sec. III A. Note that ζ i n t u ( K ) \zeta_{\mathrm{int}}^u(K) ζ int u ( K ) corresponds to the normalized velocity spectrum, such that
E u ( K ) = E N u ( K ) + E L u ( K ) = u r m s 2 ζ i n t u ( K ) . (70) E_u(K)=E_N^u(K)+E_L^u(K)=u_{\mathrm{rms}}^2\zeta_{\mathrm{int}}^u(K). \tag*{(70)} E u ( K ) = E N u ( K ) + E L u ( K ) = u rms 2 ζ int u ( K ) . (70)
The spectral components of the magnetic field are normalized as in Eqs. (58) taking q = 0 q=0 q = 0 , while the spectrum of the scalar field gradients is normalized taking q = 1 q=1 q = 1 , such that E N ( k ) = 0 = H ( k ) E_{\mathrm{N}}(k)=0=H(k) E N ( k ) = 0 = H ( k ) . Note that, since we are interested in sources of GWs, we denote as ζ ϕ ( K ) \zeta_{\phi}(K) ζ ϕ ( K ) the normalized spectrum of the scalar field gradients, instead of the normalized spectrum of the scalar field, E ϕ ( K ) E_{\phi}(K) E ϕ ( K ) .
If the helical spectrum in Eq. (58c), proportional to ϵ ( K ) [ 1 − q ( K ) ] ζ ( K ) \epsilon(K)[1-q(K)]\zeta(K) ϵ ( K ) [ 1 − q ( K )] ζ ( K ) , can also be expressed as a smoothed broken power law with slopes a H a_{\mathrm{H}} a H , b H b_{\mathrm{H}} b H , and smoothness α H \alpha_{\mathrm{H}} α H , and with peak amplitude H ∗ = E ∗ ( 1 − q ∗ ) ϵ ∗ H_{*}=E_{*}(1-q_{*})\epsilon_{*} H ∗ = E ∗ ( 1 − q ∗ ) ϵ ∗ , where ϵ ∗ \epsilon_{*} ϵ ∗ and q ∗ q_{*} q ∗ are respectively the fractional helicity and degree of compressibility at the peak, the total helicity of the field is
H = ∫ − ∞ ∞ k H ( k ) d ln k = H ∗ k ∗ I a H b H ; 0 ( α H ) = H ∗ E ∗ φ r m s 2 k ∗ I a H b H ; 0 ( α H ) A a b ( α ) . (71) \begin{aligned}
\mathcal{H}=\int_{-\infty}^{\infty}kH(k)\,\mathrm{d}\ln k=H_{*}k_{*}I_{a_{\mathrm{H}}b_{\mathrm{H}};0}(\alpha_{\mathrm{H}}) \\
&=\frac{H_{*}}{E_{*}}\varphi_{\mathrm{rms}}^{2}k_{*}\frac{I_{a_{\mathrm{H}}b_{\mathrm{H}};0}(\alpha_{\mathrm{H}})}{\mathcal{A}_{ab}(\alpha)} . \tag*{(71)}
\end{aligned} H = ∫ − ∞ ∞ k H ( k ) d ln k = H ∗ k ∗ I a H b H ; 0 ( α H ) = E ∗ H ∗ φ rms 2 k ∗ A ab ( α ) I a H b H ; 0 ( α H ) . (71)
Note that this expression directly applies for helicities defined as in Eq. (48) (i.e., kinetic and current helicities), while the magnetic helicity includes a k − 1 k^{-1} k − 1 term in the integral [see Eq. (38b)],
H A = H B k ∗ 2 I a H b H ; − 2 ( α H ) I a H b H ; 0 ( α H ) = H ∗ E ∗ B r m s 2 k ∗ I a H b H ; − 2 ( α H ) A a b ( α ) . (72) \begin{aligned}
\mathcal{H}_{\mathrm{A}}=\frac{\mathcal{H}_{\mathrm{B}}}{k_{*}^{2}}\frac{I_{a_{\mathrm{H}}b_{\mathrm{H}};-2}(\alpha_{\mathrm{H}})}{I_{a_{\mathrm{H}}b_{\mathrm{H}};0}(\alpha_{\mathrm{H}})} \\
&=\frac{H_{*}}{E_{*}}\frac{\mathcal{B}_{\mathrm{rms}}^{2}}{k_{*}}\frac{I_{a_{\mathrm{H}}b_{\mathrm{H}};-2}(\alpha_{\mathrm{H}})}{\mathcal{A}_{ab}(\alpha)} . \tag*{(72)}
\end{aligned} H A = k ∗ 2 H B I a H b H ; 0 ( α H ) I a H b H ; − 2 ( α H ) = E ∗ H ∗ k ∗ B rms 2 A ab ( α ) I a H b H ; − 2 ( α H ) . (72)
The integral length scale of the field, given in Eq. (51) , is
k ∗ ξ = I a b ; − 2 ( α ) A a b ( α ) = B a b ( α ) . (73) k_{*}\xi=\frac{I_{ab;-2}(\alpha)}{\mathcal{A}_{ab}(\alpha)}=\mathcal{B}_{ab}(\alpha) . \tag*{(73)} k ∗ ξ = A ab ( α ) I ab ; − 2 ( α ) = B ab ( α ) . (73)
where the geometric factor B a b ( α ) \mathcal{B}_{ab}(\alpha) B ab ( α ) can be expressed as [see Eq. (236) ]
B a b ( α ) = ( a b ) 1 α ( a + b ) Γ [ a − 1 α ( a + b ) ] Γ [ b + 1 α ( a + b ) ] Γ [ a α ( a + b ) ] Γ [ b α ( a + b ) ] . (74) \mathcal{B}_{ab}(\alpha)=\left(\frac{a}{b}\right)^{\frac{1}{\alpha(a+b)}}\frac{\Gamma\left[\frac{a-1}{\alpha(a+b)}\right]\Gamma\left[\frac{b+1}{\alpha(a+b)}\right]}{\Gamma\left[\frac{a}{\alpha(a+b)}\right]\Gamma\left[\frac{b}{\alpha(a+b)}\right]} . \tag*{(74)} B ab ( α ) = ( b a ) α ( a + b ) 1 Γ [ α ( a + b ) a ] Γ [ α ( a + b ) b ] Γ [ α ( a + b ) a − 1 ] Γ [ α ( a + b ) b + 1 ] . (74)
The factor B a b ( α ) \mathcal{B}_{ab}(\alpha) B ab ( α ) is shown in Figure 29 of App. A.
Figure 29. Normalized integral length scale k ∗ ξ = B a b ( α ) k_{*}\xi=B_{ab}(\alpha) k ∗ ξ = B ab ( α ) for different values of α \alpha α and the same slopes as in Figure 28 .
For scalar fields, the length scale ξ ϕ \xi_{\phi} ξ ϕ defined in Eq. (62) relates the mean gradient energy density to the mean squared value of ϕ \phi ϕ . It is given as
k ∗ ξ ϕ = [ I a b ; − 3 ( α ) A a b ( α ) ] 1 2 = B a b ; 2 ( α ) . (75) k_{*}\xi_{\phi}=\left[\frac{I_{ab;-3}(\alpha)}{\mathcal{A}_{ab}(\alpha)}\right]^{\frac{1}{2}}=\mathcal{B}_{ab;2}(\alpha) . \tag*{(75)} k ∗ ξ ϕ = [ A ab ( α ) I ab ; − 3 ( α ) ] 2 1 = B ab ; 2 ( α ) . (75)
The coefficient B a b ; n \mathcal{B}_{ab;n} B ab ; n is defined for a generalized length scale in App. A [see, in particular, Eq. (237) ]. Setting n = 2 n=2 n = 2 , we find
B a b ; 2 = ( a b ) 1 α ( a + b ) ( Γ [ a − 2 α ( a + b ) ] Γ [ b + 2 α ( a + b ) ] Γ [ a α ( a + b ) ] Γ [ b α ( a + b ) ] ) 1 2 . (76) \mathcal{B}_{ab;2}=\left(\frac{a}{b}\right)^{\frac{1}{\alpha(a+b)}}\left(\frac{\Gamma\left[\frac{a-2}{\alpha(a+b)}\right]\Gamma\left[\frac{b+2}{\alpha(a+b)}\right]}{\Gamma\left[\frac{a}{\alpha(a+b)}\right]\Gamma\left[\frac{b}{\alpha(a+b)}\right]}\right)^{\frac{1}{2}} . \tag*{(76)} B ab ; 2 = ( b a ) α ( a + b ) 1 Γ [ α ( a + b ) a ] Γ [ α ( a + b ) b ] Γ [ α ( a + b ) a − 2 ] Γ [ α ( a + b ) b + 2 ] 2 1 . (76)
Note that the slopes a a a , b b b , and smoothness α \alpha α refer to the spectrum of the scalar field gradients E L ϕ ( K ) E_{\mathrm{L}}^{\phi}(K) E L ϕ ( K ) in Eq. (58a), taking q ( K ) = 1 q(K)=1 q ( K ) = 1 .
Spectra of the anisotropic stresses
The GW spectrum and its circular polarization are determined by the integrals over τ 1 \tau_{1} τ 1 and τ 2 \tau_{2} τ 2 of the symmetric and antisymmetric spectral functions E Π ( τ 1 , τ 2 , k ) E_{\Pi}(\tau_{1},\tau_{2},k) E Π ( τ 1 , τ 2 , k ) and H Π ( τ 1 , τ 2 , k ) H_{\Pi}(\tau_{1},\tau_{2},k) H Π ( τ 1 , τ 2 , k ) , respectively, as shown in Eqs. (24) and (26) . These functions represent the two-point unequal time correlator (UETC) of the anisotropic stresses, Π i j \Pi_{ij} Π ij , as defined in Eqs. (20). In this section, we present the relation between the UETC of the different sources (velocity, electromagnetic, and scalar fields) with that of the anisotropic stresses that source GWs. We omit the τ 1 \tau_{1} τ 1 and τ 2 \tau_{2} τ 2 dependence in the following for brevity, as this dependence does not affect the results presented in this section.
We start by describing the two-point correlation function of the stress-energy tensor,
⟨ T i j ( k ) T l m ∗ ( k ′ ) ⟩ = ∫ d 3 p ( 2 π ) 3 ∫ d 3 q ( 2 π ) 3 ∑ a , b ⟨ φ i a ( p ~ ) φ j a ( p ) φ l b ∗ ( q ~ ) φ m b ∗ ( q ) ⟩ . (77) \left\langle T_{ij}(k)T_{lm}^{*}(k')\right\rangle=\int\frac{\mathrm{d}^{3}p}{(2\pi)^{3}}\int\frac{\mathrm{d}^{3}q}{(2\pi)^{3}}\sum_{a,b}\left\langle\varphi_{i}^{a}(\widetilde{p})\varphi_{j}^{a}(p)\varphi_{l}^{b*}(\widetilde{q})\varphi_{m}^{b*}(q)\right\rangle. \tag*{(77)} ⟨ T ij ( k ) T l m ∗ ( k ′ ) ⟩ = ∫ ( 2 π ) 3 d 3 p ∫ ( 2 π ) 3 d 3 q a , b ∑ ⟨ φ i a ( p ) φ j a ( p ) φ l b ∗ ( q ) φ m b ∗ ( q ) ⟩ . (77)
where φ i a , b ( k ) = { a X i , i B i , i E i , − i k i ϕ } \varphi_{i}^{a,b}(k)=\{aX_{i},iB_{i},iE_{i},-ik_{i}\phi\} φ i a , b ( k ) = { a X i , i B i , i E i , − i k i ϕ } corresponds to each of the sourcing fields, in Fourier space, that
contribute to the stress-energy tensor,
T i j ( x ) = ∑ a φ i a ( x ) φ j a ( x ) . (78) T_{ij}(x)=\sum_{a}\varphi_{i}^{a}(x)\varphi_{j}^{a}(x) . \tag*{(78)} T ij ( x ) = a ∑ φ i a ( x ) φ j a ( x ) . (78)
We have defined p ~ = k − p \widetilde{p}=k-p p = k − p and q ~ = k ′ − q \widetilde{q}=k'-q q = k ′ − q to describe the convolution in Eq. (77) . For simplicity, we assume that the cross-correlations between different sources are negligible and focus on GW production by each of the sources separately. In reality, however, when two or more sources are present with similar amplitudes, the cross-correlations between the sourcing fields should be taken into account as they are non-vanishing when the sourcing fields are dynamically coupled.
Under the assumption that the cross-correlations of different fields are vanishing, and omitting the contributions from electric fields, which is justified given the very large conductivity of the primordial plasma [76 ] (see the discussion at the beginning of Sec. III), we can express the symmetric and antisymmetric spectra of the anisotropic stresses as
E Π ( k ) = E Π X ( k ) + E Π B ( k ) + E Π ϕ ( k ) , (79) E_{\Pi}(k) = E_{\Pi}^{X}(k) + E_{\Pi}^{B}(k) + E_{\Pi}^{\phi}(k), \tag*{(79)} E Π ( k ) = E Π X ( k ) + E Π B ( k ) + E Π ϕ ( k ) , (79)
H Π ( k ) = H Π X ( k ) + H Π B ( k ) . (80) H_{\Pi}(k) = H_{\Pi}^{X}(k) + H_{\Pi}^{B}(k). \tag*{(80)} H Π ( k ) = H Π X ( k ) + H Π B ( k ) . (80)
We denote E Π ϕ ( k ) E_{\Pi}^{\phi}(k) E Π ϕ ( k ) and H Π ϕ ( k ) H_{\Pi}^{\phi}(k) H Π ϕ ( k ) with ϕ = { X , B , ϕ } \phi= \{X, B, \phi\} ϕ = { X , B , ϕ } as the component corresponding to the sourcing field ϕ i = { a X i , i B i , − i k i ϕ } \phi_i = \{aX_i, iB_i, -ik_i\phi\} ϕ i = { a X i , i B i , − i k i ϕ } , respectively.
When the statistical distribution of the sourcing field ϕ i \phi_i ϕ i is Gaussian, we can apply Isserlis’ (or Wick’s) theorem to the four-point correlation of Eq. (77) , and express it as a superposition of products of two-point correlation functions. Introducing the spectral decomposition of the velocity field of Eqs. (30), and applying the TT operators Λ i j l m \Lambda_{ijlm} Λ ij l m and A i j l m A_{ijlm} A ij l m , we find the spectral functions describing the statistics of the anisotropic stresses Π i j = a − 2 Λ i j l m T l m / ρ ˉ \Pi_{ij} = a^{-2}\Lambda_{ijlm}T_{lm}/\bar{\rho} Π ij = a − 2 Λ ij l m T l m / ρ ˉ , defined in Eqs. (20) (see details in App. B). In general, we note that even if the initial perturbations of a field are statistically Gaussian, non-Gaussianities can develop due to the non-linear dynamics of the source. When the field is non-Gaussian, then its four-point correlation function cannot be decomposed applying Isserlis’ theorem and numerical simulations describing the non-linear dynamics are, in general, required. We then restrict our theoretical analysis to Gaussian fields in the present work.
Symmetric spectrum of the anisotropic stresses E Π E_{\Pi} E Π
Let us first focus on the different components of the symmetric spectrum of the anisotropic stresses, E Π ϕ ( k ) E_{\Pi}^{\phi}(k) E Π ϕ ( k ) , computed in App. B [see Eq. (B4a)], which can be decomposed in up to four terms,
a 4 E Π v o r t ( k ) = k 2 8 ρ ˉ 2 ∫ 0 ∞ d p p 2 E N ( p ) a^{4}E_{\Pi}^{\mathrm{vort}}(k) = \frac{k^{2}}{8\bar{\rho}^{2}}\int_{0}^{\infty}\frac{\mathrm{d}p}{p^{2}}E_{N}(p) a 4 E Π vort ( k ) = 8 ρ ˉ 2 k 2 ∫ 0 ∞ p 2 d p E N ( p )
× ∫ ∣ k − p ∣ k + p d p ~ p ~ 2 E N ( p ~ ) ( 1 + z 2 ) ( 1 + z ~ 2 ) , (81) \qquad\times\int_{\lvert k-p\rvert}^{k+p}\frac{\mathrm{d}\tilde{p}}{\tilde{p}^{2}}E_{N}(\tilde{p})(1+z^{2})(1+\tilde{z}^{2}), \tag*{(81)} × ∫ ∣ k − p ∣ k + p p ~ 2 d p ~ E N ( p ~ ) ( 1 + z 2 ) ( 1 + z ~ 2 ) , (81)
where i = { v o r t , c o m p , m i x , h e l } i = \{\mathrm{vort}, \mathrm{comp}, \mathrm{mix}, \mathrm{hel}\} i = { vort , comp , mix , hel } accounts for the vortical (NN), compressional (LL), mixed vortical-compressional (NL), and helical (HH) components.
When the field is purely vortical (e.g., ∇ ⋅ u = 0 \nabla\cdot u = 0 ∇ ⋅ u = 0 or ∇ ⋅ B = 0 \nabla\cdot B = 0 ∇ ⋅ B = 0 ), such that E L ( k ) = 0 E_{L}(k) = 0 E L ( k ) = 0 , and non-helical with H ( k ) = 0 H(k) = 0 H ( k ) = 0 , then E Π ( k ) E_{\Pi}(k) E Π ( k ) only contains the “vort” component [94 ] [5], given in Eq. (B8a),
a 4 E Π v o r t ( k ) = k 2 8 ρ ˉ 2 ∫ 0 ∞ d p p 2 E N ( p ) a^{4}E_{\Pi}^{\mathrm{vort}}(k) = \frac{k^{2}}{8\bar{\rho}^{2}}\int_{0}^{\infty}\frac{\mathrm{d}p}{p^{2}}E_{N}(p) a 4 E Π vort ( k ) = 8 ρ ˉ 2 k 2 ∫ 0 ∞ p 2 d p E N ( p )
× ∫ ∣ k − p ∣ k + p d p ~ p ~ 2 E N ( p ~ ) ( 1 + z 2 ) ( 1 + z ~ 2 ) , (82) \qquad\times\int_{\lvert k-p\rvert}^{k+p}\frac{\mathrm{d}\tilde{p}}{\tilde{p}^{2}}E_{N}(\tilde{p})(1+z^{2})(1+\tilde{z}^{2}), \tag*{(82)} × ∫ ∣ k − p ∣ k + p p ~ 2 d p ~ E N ( p ~ ) ( 1 + z 2 ) ( 1 + z ~ 2 ) , (82)
where p ~ = ∣ p ~ ∣ = p 2 + k 2 − 2 p k z \tilde{p} = \lvert\tilde{\boldsymbol{p}}\rvert= \sqrt{p^{2}+k^{2}-2pkz} p ~ = ∣ p ~ ∣ = p 2 + k 2 − 2 p k z , and z z z and z ~ \tilde{z} z ~ are, respectively, the dot products k ^ ⋅ p ^ \hat{k}\cdot\hat{p} k ^ ⋅ p ^ and k ^ ⋅ p ~ ^ \hat{k}\cdot\hat{\tilde{p}} k ^ ⋅ p ~ ^ . If the purely vortical field has a non-zero helicity, then an additional “hel” term, proportional to the helical spectrum, contributes to E Π ( k ) E_{\Pi}(k) E Π ( k ) [66 ] , given in Eq. (B8d),
a 4 E Π h e l ( k ) = k 2 2 ρ ˉ 2 ∫ 0 ∞ d p p 2 H ( p ) a^{4}E_{\Pi}^{\mathrm{hel}}(k) = \frac{k^{2}}{2\bar{\rho}^{2}}\int_{0}^{\infty}\frac{\mathrm{d}p}{p^{2}}H(p) a 4 E Π hel ( k ) = 2 ρ ˉ 2 k 2 ∫ 0 ∞ p 2 d p H ( p )
× ∫ ∣ k − p ∣ k + p d p ~ p ~ 2 H ( p ~ ) z z ~ . (83) \qquad\times\int_{\lvert k-p\rvert}^{k+p}\frac{\mathrm{d}\tilde{p}}{\tilde{p}^{2}}H(\tilde{p})\,z\tilde{z}. \tag*{(83)} × ∫ ∣ k − p ∣ k + p p ~ 2 d p ~ H ( p ~ ) z z ~ . (83)
On the other hand, if the field is purely compressional (e.g., ∇ × u = 0 \nabla\times u = 0 ∇ × u = 0 or ∇ × ∇ ϕ = 0 \nabla\times\nabla\phi= 0 ∇ × ∇ ϕ = 0 ), such that E N ( k ) = 0 E_{N}(k) = 0 E N ( k ) = 0 , then E Π ( k ) E_{\Pi}(k) E Π ( k ) presents only the “comp” component [21], given in Eq. (B8b),
a 4 E Π c o m p ( k ) = k 2 2 ρ ˉ 2 ∫ 0 ∞ d p p 2 E L ( p ) a^{4}E_{\Pi}^{\mathrm{comp}}(k) = \frac{k^{2}}{2\bar{\rho}^{2}}\int_{0}^{\infty}\frac{\mathrm{d}p}{p^{2}}E_{L}(p) a 4 E Π comp ( k ) = 2 ρ ˉ 2 k 2 ∫ 0 ∞ p 2 d p E L ( p )
× ∫ ∣ k − p ∣ k + p d p ~ p ~ 2 E L ( p ~ ) ( 1 − z 2 ) ( 1 − z ~ 2 ) . (84) \qquad\times\int_{\lvert k-p\rvert}^{k+p}\frac{\mathrm{d}\tilde{p}}{\tilde{p}^{2}}E_{L}(\tilde{p})(1-z^{2})(1-\tilde{z}^{2}). \tag*{(84)} × ∫ ∣ k − p ∣ k + p p ~ 2 d p ~ E L ( p ~ ) ( 1 − z 2 ) ( 1 − z ~ 2 ) . (84)
More generally, for velocity fields with both non-vanishing curl and divergence, the spectrum of the anisotropic stresses becomes the sum of the “vort”, the “comp”, and the “hel” components, together with an additional mixed “mix” term. This mixed term is given in Eq. (B8c),
a 4 E Π m i x ( k ) = k 2 2 ρ ˉ 2 ∫ 0 ∞ d p p 2 E L ( p ) a^{4}E_{\Pi}^{\mathrm{mix}}(k) = \frac{k^{2}}{2\bar{\rho}^{2}}\int_{0}^{\infty}\frac{\mathrm{d}p}{p^{2}}E_{L}(p) a 4 E Π mix ( k ) = 2 ρ ˉ 2 k 2 ∫ 0 ∞ p 2 d p E L ( p )
× ∫ ∣ k − p ∣ k + p d p ~ p ~ 2 E N ( p ~ ) ( 1 − z 2 z ~ 2 ) . (85) \qquad\times\int_{\lvert k-p\rvert}^{k+p}\frac{\mathrm{d}\tilde{p}}{\tilde{p}^{2}}E_{N}(\tilde{p})(1-z^{2}\tilde{z}^{2}). \tag*{(85)} × ∫ ∣ k − p ∣ k + p p ~ 2 d p ~ E N ( p ~ ) ( 1 − z 2 z ~ 2 ) . (85)
With the exception of E Π h e l ( k ) E_{\Pi}^{\mathrm{hel}}(k) E Π hel ( k ) , all components E Π i E_{\Pi}^{i} E Π i are positive-definite, and this term satisfies ∣ E Π h e l ( k ) ∣ ≤ E Π v o r t ( k ) \lvert E_{\Pi}^{\mathrm{hel}}(k)\rvert\le E_{\Pi}^{\mathrm{vort}}(k) ∣ E Π hel ( k )∣ ≤ E Π vort ( k ) , ensuring that the resulting E Π ( k ) E_{\Pi}(k) E Π ( k ) is always positive. The integrands K ( p , p ~ ) \mathcal{K}(p,\tilde{p}) K ( p , p ~ ) that multiply the spectra in Eqs. (70–(85) are all symmetric functions of p p p and p ~ \tilde{p} p ~ , with the exception of the mixed term that is computed from the product E L ( p ) E N ( p ~ ) E_{\mathrm{L}}(p)E_{\mathrm{N}}(\tilde{p}) E L ( p ) E N ( p ~ ) . This product is, in general, not symmetric in ( p , p ~ ) (p,\tilde{p}) ( p , p ~ ) unless the degree of compressibility q q q is a constant. However, the integrals over E L ( p ~ ) E N ( p ) E_{\mathrm{L}}(\tilde{p})E_{\mathrm{N}}(p) E L ( p ~ ) E N ( p ) and E N ( p ~ ) E L ( p ) E_{\mathrm{N}}(\tilde{p})E_{\mathrm{L}}(p) E N ( p ~ ) E L ( p ) can be shown to yield the same result and, hence, they have been combined in Eq. (85) .
In general, velocity fields can contain contributions from all four components E Π i E_{\Pi}^{i} E Π i , while the anisotropic stresses of magnetic fields can only contain a vortical and a helical component. On the other hand, the gradients of scalar fields only present the compressional component.
Following the normalization of refs. [34 , 35 ] [1], we re-express each component i = { v o r t , c o m p , m i x , h e l } i = \{\mathrm{vort},\ \mathrm{comp},\ \mathrm{mix},\ \mathrm{hel}\} i = { vort , comp , mix , hel } of the anisotropic stresses spectrum in terms of normalized functions. In first place, we introduce the normalization of Eqs. (58) to each spectral component of the source, E L φ ( K ) E_{\mathrm{L}}^{\varphi}(K) E L φ ( K ) , E N φ ( K ) E_{\mathrm{N}}^{\varphi}(K) E N φ ( K ) , and H φ ( K ) H_{\varphi}(K) H φ ( K ) . Then, each of the components of E Π φ ( K ) E_{\Pi}^{\varphi}(K) E Π φ ( K ) is expressed as [see Eq. (274) ]
E Π φ , i ( K ) = φ ~ r m s 4 ρ ˉ 2 C Π φ , i A a b 2 ( α ) K 3 ζ Π φ , i ( K ) , (86) E_{\Pi}^{\varphi,i}(K)=\frac{\tilde{\varphi}_{\mathrm{rms}}^{4}}{\bar{\rho}^{2}}\frac{C_{\Pi}^{\varphi,i}}{\mathcal{A}_{ab}^{2}(\alpha)}K^{3}\zeta_{\Pi}^{\varphi,i}(K), \tag*{(86)} E Π φ , i ( K ) = ρ ˉ 2 φ ~ rms 4 A ab 2 ( α ) C Π φ , i K 3 ζ Π φ , i ( K ) , (86)
where φ ~ r m s 2 ≡ φ r m s 2 / a 2 ≡ { ρ ˉ k i n , 2 ρ ˉ B , 2 ρ ˉ G } \tilde{\varphi}_{\mathrm{rms}}^{2}\equiv\varphi_{\mathrm{rms}}^{2}/a^{2}\equiv\{\bar{\rho}_{\mathrm{kin}},2\bar{\rho}_{B},2\bar{\rho}_{G}\} φ ~ rms 2 ≡ φ rms 2 / a 2 ≡ { ρ ˉ kin , 2 ρ ˉ B , 2 ρ ˉ G } . For small fluid perturbations, one can approximate ρ ˉ k i n = X r m s 2 ≃ w u ˉ r m s 2 \bar{\rho}_{\mathrm{kin}}=X_{\mathrm{rms}}^{2}\simeq w\bar{u}_{\mathrm{rms}}^{2} ρ ˉ kin = X rms 2 ≃ w u ˉ rms 2 , such that φ ~ r m s 2 / ρ ˉ ≃ Γ u ˉ r m s 2 \tilde{\varphi}_{\mathrm{rms}}^{2}/\bar{\rho}\simeq\Gamma\bar{u}_{\mathrm{rms}}^{2} φ ~ rms 2 / ρ ˉ ≃ Γ u ˉ rms 2 , where Γ ≡ w ˉ / ρ ˉ \Gamma\equiv\bar{w}/\bar{\rho} Γ ≡ w ˉ / ρ ˉ is the adiabatic ratio, which takes the value Γ = 4 / 3 \Gamma=4/3 Γ = 4/3 for a pure radiation fluid.
The different components of the normalized anisotropic stresses spectra, ζ Π φ , i ( K ) \zeta_{\Pi}^{\varphi,i}(K) ζ Π φ , i ( K ) , are given in Eqs. (B14). The geometrical factors C i φ > 0 C_{i}^{\varphi}>0 C i φ > 0 are defined such that ∣ ζ Π φ , i ( K ) ∣ → 1 \lvert\zeta_{\Pi}^{\varphi,i}(K)\rvert\to1 ∣ ζ Π φ , i ( K )∣ → 1 in the K → 0 K\to0 K → 0 limit [see Eqs. (B15)],
C v o r t = 7 15 ∫ 0 ∞ [ 1 − q ( P ) ] 2 ζ 2 ( P ) P 4 d P , (87) C_{\mathrm{vort}}=\frac{7}{15}\int_{0}^{\infty}[1-q(P)]^{2}\frac{\zeta^{2}(P)}{P^{4}}\,\mathrm{d}P, \tag*{(87)} C vort = 15 7 ∫ 0 ∞ [ 1 − q ( P ) ] 2 P 4 ζ 2 ( P ) d P , (87)
C c o m p = 8 15 ∫ 0 ∞ q 2 ( P ) ζ 2 ( P ) P 4 d P , (88) C_{\mathrm{comp}}=\frac{8}{15}\int_{0}^{\infty}q^{2}(P)\frac{\zeta^{2}(P)}{P^{4}}\,\mathrm{d}P, \tag*{(88)} C comp = 15 8 ∫ 0 ∞ q 2 ( P ) P 4 ζ 2 ( P ) d P , (88)
C m i x = 4 5 ∫ 0 ∞ q ( P ) [ 1 − q ( P ) ] ζ 2 ( P ) P 4 d P , (89) C_{\mathrm{mix}}=\frac{4}{5}\int_{0}^{\infty}q(P)[1-q(P)]\frac{\zeta^{2}(P)}{P^{4}}\,\mathrm{d}P, \tag*{(89)} C mix = 5 4 ∫ 0 ∞ q ( P ) [ 1 − q ( P )] P 4 ζ 2 ( P ) d P , (89)
C h e l = 1 3 ∫ 0 ∞ ε 2 ( P ) [ 1 − q ( P ) ] 2 ζ 2 ( P ) P 4 d P . (90) C_{\mathrm{hel}}=\frac{1}{3}\int_{0}^{\infty}\varepsilon^{2}(P)[1-q(P)]^{2}\frac{\zeta^{2}(P)}{P^{4}}\,\mathrm{d}P. \tag*{(90)} C hel = 3 1 ∫ 0 ∞ ε 2 ( P ) [ 1 − q ( P ) ] 2 P 4 ζ 2 ( P ) d P . (90)
In the particular case when the degree of compressibility and the fractional helicity are constant, all these coefficients can be expressed as
C i = c i ∫ 0 ∞ ζ 2 ( P ) P 4 d P ≡ C a b i ( α ) ≡ c i I 2 a , 2 b ; − 4 ( α / 2 ) = c i ( a + b b ) 2 α ( a b ) − 2 a − 3 α ( a + b ) Γ [ 2 a − 3 α ( a + b ) ] Γ [ 2 b + 3 α ( a + b ) ] α ( a + b ) Γ [ 2 α ] , (91) \begin{aligned}
C_i=c_i\int_{0}^{\infty}\frac{\zeta^{2}(P)}{P^{4}}\,\mathrm{d}P\equiv C_{ab}^{i}(\alpha)\equiv c_i I_{2a,2b;-4}(\alpha/2) \\
&=c_i\left(\frac{a+b}{b}\right)^{\frac{2}{\alpha}}\left(\frac{a}{b}\right)^{-\frac{2a-3}{\alpha(a+b)}}\frac{\Gamma\left[\frac{2a-3}{\alpha(a+b)}\right]\Gamma\left[\frac{2b+3}{\alpha(a+b)}\right]}{\alpha(a+b)\Gamma\left[\frac{2}{\alpha}\right]}, \tag*{(91)}
\end{aligned} C i = c i ∫ 0 ∞ P 4 ζ 2 ( P ) d P ≡ C ab i ( α ) ≡ c i I 2 a , 2 b ; − 4 ( α /2 ) = c i ( b a + b ) α 2 ( b a ) − α ( a + b ) 2 a − 3 α ( a + b ) Γ [ α 2 ] Γ [ α ( a + b ) 2 a − 3 ] Γ [ α ( a + b ) 2 b + 3 ] , (91)
where we have assumed that ζ ( K ) \zeta(K) ζ ( K ) is the smoothed broken power law of Eq. (63) , and I a b ; n ( α ) I_{ab;n}(\alpha) I ab ; n ( α ) is defined in Eq. (66) . The prefactor c i c_i c i is
c i = { 7 15 ( 1 − q ) 2 , 8 15 q 2 , 2 5 2 q ( 1 − q ) , 1 3 ε 2 ( 1 − q ) 2 } (92) c_i=\left\{\frac{7}{15}(1-q)^2,\ \frac{8}{15}q^2,\ \frac{2}{5}2q(1-q),\ \frac{1}{3}\varepsilon^2(1-q)^2\right\} \tag*{(92)} c i = { 15 7 ( 1 − q ) 2 , 15 8 q 2 , 5 2 2 q ( 1 − q ) , 3 1 ε 2 ( 1 − q ) 2 } (92)
for i = { v o r t , c o m p , m i x , h e l } i=\{\mathrm{vort},\ \mathrm{comp},\ \mathrm{mix},\ \mathrm{hel}\} i = { vort , comp , mix , hel } . The dependence of C a b i ( α ) / c i C_{ab}^{i}(\alpha)/c_i C ab i ( α ) / c i with α \alpha α is shown in Figure 30 .
Figure 30. Geometrical factors C a b ( α ) / c ≡ I 2 a , 2 b ; − 4 ( α / 2 ) C_{ab}(\alpha)/c \equiv I_{2a,2b;-4}(\alpha/2) C ab ( α ) / c ≡ I 2 a , 2 b ; − 4 ( α /2 ) (left panel) and C ~ a b ( α ) / c ≡ I 2 ( a − 1 ) , 2 ( b + 1 ) ; − 2 ( α / 2 ) \widetilde{C}_{ab}(\alpha)/c \equiv I_{2(a-1),2(b+1);-2}(\alpha/2) C ab ( α ) / c ≡ I 2 ( a − 1 ) , 2 ( b + 1 ) ; − 2 ( α /2 ) (right panel) that determine the amplitude of the anisotropic stresses in the K → 0 K \to0 K → 0 limit, respectively when using ζ ( K ) \zeta(K) ζ ( K ) and ζ − 1 ( K ~ ) \zeta_{-1}(\widetilde{K}) ζ − 1 ( K ) , for different values of α \alpha α and the same slopes as in Figure 28 . Notice that the ratios C a b j / c j C_{ab}^{j}/c_{j} C ab j / c j and C ~ a b j / c j \widetilde{C}_{ab}^{j}/c_{j} C ab j / c j do not depend on j j j , hence we can omit the index here.
We show in Figure 2 the normalized anisotropic stresses K 3 C Π φ , i ζ Π φ , i ( K ) / A a b 2 ( α ) K^{3}C_{\Pi}^{\varphi,i}\zeta_{\Pi}^{\varphi,i}(K)/\mathcal{A}_{ab}^{2}(\alpha) K 3 C Π φ , i ζ Π φ , i ( K ) / A ab 2 ( α ) that characterize E Π φ , i ( K ) E_{\Pi}^{\varphi,i}(K) E Π φ , i ( K ) in Eq. (86) for each component using the smoothed broken power law of Eq. (63) , taking the von Kármán spectrum with a = 5 a=5 a = 5 and b = 2 / 3 b=2/3 b = 2/3 for all terms. In reality, we note that for the helical spectrum a = 5 a=5 a = 5 is not compatible with causality and the realizability condition, which require a a a to be even with a ≥ 4 a\ge4 a ≥ 4 , or a ≥ 6 a\ge6 a ≥ 6 when the source is fully vortical (e.g., magnetic fields) [79 ] . For simplicity, we assume that the fractional helicity ε ( K ) \varepsilon(K) ε ( K ) remains constant until sufficiently small K K K , when it eventually becomes linear, satisfying the causality condition. Therefore, we omit its contribution at small K K K (see also footnote 3 ) and take a = 5 a=5 a = 5 also for the helical spectrum, such that we can directly compare all terms. Note that for generality, we omit the prefactors that depend on the degree of compressibility and fractional helicity for each spectral component [see Eqs. (58)]. Therefore, to compute E Π φ ( K ) E_{\Pi}^{\varphi}(K) E Π φ ( K ) , the prefactors need to be reincorporated for particular choices of q q q and ε \varepsilon ε . We also note that the mixed term is always proportional to q ( 1 − q ) q(1-q) q ( 1 − q ) , which takes a maximum value of 1 4 \frac{1}{4} 4 1 for q = 1 2 q=\frac{1}{2} q = 2 1 . Therefore, this term is always suppressed with respect to the one in Figure 2 by a factor of 4 or larger, depending on the value of q q q .
Figure 2. Normalized components of the spectrum of the anisotropic and unprojected stresses, K 3 C Π i ( K ) K^3 C_{\Pi}^{i}(K) K 3 C Π i ( K ) and K 3 C T i ( K ) K^3 C_{T}^{i}(K) K 3 C T i ( K ) , respectively in solid and dashed lines, with i = v o r t i = \mathrm{vort} i = vort (red), c o m p \mathrm{comp} comp (blue), m i x \mathrm{mix} mix (purple), and h e l \mathrm{hel} hel (green), divided by A a b 2 ( α ) \mathcal{A}_{ab}^{2}(\alpha) A ab 2 ( α ) , for the von Kármán spectrum with a = 5 a = 5 a = 5 , b = 2 / 3 b = 2/3 b = 2/3 , and α = 6 / 17 \alpha= 6/17 α = 6/17 (left panel) and 2 2 2 (right panel). Since the helical component can become negative, we show separately negative values of the anisotropic stresses with a dash-dotted line and of the unprojected stresses with a dotted line. For the plot, we are omitting the prefactors multiplying each of the components of the spectrum of the source as in Eqs. (58), so the lines correspond to the maximum possible amplitudes of each of the components, with the exception of the mixed term, which is bounded by 1 4 \frac{1}{4} 4 1 times the values in the plots, arising from maximizing q ( 1 − q ) ≤ 1 4 q(1-q) \le\frac{1}{4} q ( 1 − q ) ≤ 4 1 . We show in dashed black lines the different scalings with K K K for each component: ∼ K 3 \sim K^3 ∼ K 3 at small K ≪ 1 K \ll1 K ≪ 1 for all components; K − b K^{-b} K − b at K ≫ 1 K \gg1 K ≫ 1 for the vortical and mixed components and the unprojected compressional component; K − 2 b K^{-2b} K − 2 b at K ≫ 1 K \gg1 K ≫ 1 for the projected compressional and for the helical components. The latter scaling K − 2 b K^{-2b} K − 2 b dominates at large K K K for b < 1 b < 1 b < 1 (helical components) and b < 2 b < 2 b < 2 (compressional anisotropic component), but it needs to be changed for larger values of b b b as we show in Sec. IV C. The locations of the scaling lines are the same in both plots for a direct comparison of the relative amplitude of ζ Π ( K ) \zeta_{\Pi}(K) ζ Π ( K ) for the two values of α \alpha α .
For comparison, we also show in Figure 2 the spectral components of the unprojected stresses,
E Π φ , i ( K ) = φ ~ r m s 4 ρ ˉ 2 C T φ , i A a b 2 K 3 ζ T φ , i ( K ) , (93) E_{\Pi}^{\varphi,i}(K)=\frac{\tilde{\varphi}_{\mathrm{rms}}^{4}}{\bar{\rho}^{2}}\frac{C_{T}^{\varphi,i}}{\mathcal{A}_{ab}^{2}}K^{3}\zeta_{T}^{\varphi,i}(K), \tag*{(93)} E Π φ , i ( K ) = ρ ˉ 2 φ ~ rms 4 A ab 2 C T φ , i K 3 ζ T φ , i ( K ) , (93)
which are computed in App. B [see Eqs. (298) –B31)]. Taking the integral of E Π φ , i ( k ) E_{\Pi}^{\varphi,i}(k) E Π φ , i ( k ) over ln k \ln k ln k , we can find the mean squared anisotropic stresses, quantifying the amount of energy density that is transferred from the source into anisotropic stresses and, hence, is available to produce GWs. Let us integrate Eq. (86) to find,
∫ − ∞ ∞ E Π φ , i ( K ) d ln K = φ ~ r m s 4 ρ ˉ 2 E Π φ , i , (94) \int_{-\infty}^{\infty} E_{\Pi}^{\varphi,i}(K)\,\mathrm{d}\ln K = \frac{\widetilde{\varphi}_{\mathrm{rms}}^{4}}{\bar{\rho}^{2}} \mathcal{E}_{\Pi}^{\varphi,i}, \tag*{(94)} ∫ − ∞ ∞ E Π φ , i ( K ) d ln K = ρ ˉ 2 φ rms 4 E Π φ , i , (94)
where we have defined the efficiency factor E Π φ , i \mathcal{E}_{\Pi}^{\varphi,i} E Π φ , i as
E Π φ , i = C Π φ , i A a b 2 ( α ) ∫ 0 ∞ K 3 ζ Π φ , i ( K ) d ln K . (95) \mathcal{E}_{\Pi}^{\varphi,i} = \frac{C_{\Pi}^{\varphi,i}}{\mathcal{A}_{ab}^{2}(\alpha)} \int_{0}^{\infty} K^3 \zeta_{\Pi}^{\varphi,i}(K)\,\mathrm{d}\ln K. \tag*{(95)} E Π φ , i = A ab 2 ( α ) C Π φ , i ∫ 0 ∞ K 3 ζ Π φ , i ( K ) d ln K . (95)
This expression allows us to define a spectrum of the anisotropic stresses normalized by its integral,
ζ Π , i n t φ , i ( K ) ≡ C Π φ , i A a b 2 ( α ) E Π φ , i K 3 ζ Π φ , i ( K ) , (96) \zeta_{\Pi,\mathrm{int}}^{\varphi,i}(K) \equiv\frac{C_{\Pi}^{\varphi,i}}{\mathcal{A}_{ab}^{2}(\alpha)\mathcal{E}_{\Pi}^{\varphi,i}} K^3 \zeta_{\Pi}^{\varphi,i}(K), \tag*{(96)} ζ Π , int φ , i ( K ) ≡ A ab 2 ( α ) E Π φ , i C Π φ , i K 3 ζ Π φ , i ( K ) , (96)
such that
E Π φ , i ( K ) = φ ~ r m s 4 ρ ˉ 2 E Π φ , i ζ Π , i n t φ , i ( K ) . (97) E_{\Pi}^{\varphi,i}(K) = \frac{\widetilde{\varphi}_{\mathrm{rms}}^{4}}{\bar{\rho}^{2}} \mathcal{E}_{\Pi}^{\varphi,i}\zeta_{\Pi,\mathrm{int}}^{\varphi,i}(K). \tag*{(97)} E Π φ , i ( K ) = ρ ˉ 2 φ rms 4 E Π φ , i ζ Π , int φ , i ( K ) . (97)
The efficiency factors, E Π φ , i \mathcal{E}_{\Pi}^{\varphi,i} E Π φ , i , are shown in Figure 3 for a range of α = { 6 / 17 , 0.75 , 1 , 2 , 4 } \alpha= \{6/17,0.75,1,2,4\} α = { 6/17 , 0.75 , 1 , 2 , 4 } , a = { 3 , 5 } a = \{3,5\} a = { 3 , 5 } , and b = { 2 / 3 , 1 , 1.5 } b=\{2/3,1,1.5\} b = { 2/3 , 1 , 1.5 } . Their mean values over the considered range of parameters are
Figure 3. Efficiency factor E Π φ , i \mathcal{E}_{\Pi}^{\varphi,i} E Π φ , i , defined in Eq. (94) , for each component i = { v o r t , c o m p , m i x , h e l } i = \{\mathrm{vort},\mathrm{comp},\mathrm{mix},\mathrm{hel}\} i = { vort , comp , mix , hel } contributing to E Π φ , i ( K ) E_{\Pi}^{\varphi,i}(K) E Π φ , i ( K ) , divided respectively by ( 1 − q ) 2 (1-q)^2 ( 1 − q ) 2 , q 2 q^2 q 2 , 2 q ( 1 − q ) 2q(1-q) 2 q ( 1 − q ) , and ε 2 ( 1 − q ) 2 \varepsilon^2(1-q)^2 ε 2 ( 1 − q ) 2 assuming constants q q q and ε \varepsilon ε . For comparison, we show the analogous ratio for the unprojected stresses, E T φ ≡ ρ ˉ T φ / ρ ˉ r m s φ = 5 / 3 \mathcal{E}_{T}^{\varphi} \equiv\bar{\rho}_{T}^{\varphi}/\bar{\rho}_{\mathrm{rms}}^{\varphi} = 5/3 E T φ ≡ ρ ˉ T φ / ρ ˉ rms φ = 5/3 [see Eq. (103) ]. Dashed (solid) lines correspond to a = 5 a = 5 a = 5 (3 3 3 ), and three values of b = { 2 / 3 , 1 , 1.5 } b = \{2/3,1,1.5\} b = { 2/3 , 1 , 1.5 } are overplotted with larger opacity for larger b b b .
E Π v o r t ≃ ( 1 − q ) 2 ( 0.67 ± 0.02 ) , (98) \mathcal{E}_{\Pi}^{\mathrm{vort}} \simeq(1-q)^2(0.67\pm0.02), \tag*{(98)} E Π vort ≃ ( 1 − q ) 2 ( 0.67 ± 0.02 ) , (98)
E Π c o m p ≃ q 2 ( 0.12 ± 0.045 ) , (99) \mathcal{E}_{\Pi}^{\mathrm{comp}} \simeq q^2(0.12\pm0.045), \tag*{(99)} E Π comp ≃ q 2 ( 0.12 ± 0.045 ) , (99)
E Π m i x ≃ 2 q ( 1 − q ) ( 0.35 ± 0.008 ) , (100) \mathcal{E}_{\Pi}^{\mathrm{mix}} \simeq2q(1-q)(0.35\pm0.008), \tag*{(100)} E Π mix ≃ 2 q ( 1 − q ) ( 0.35 ± 0.008 ) , (100)
E Π h e l ≃ ε 2 ( 1 − q ) 2 ( 0.27 ± 0.07 ) , (101) \mathcal{E}_{\Pi}^{\mathrm{hel}} \simeq\varepsilon^2(1-q)^2(0.27\pm0.07), \tag*{(101)} E Π hel ≃ ε 2 ( 1 − q ) 2 ( 0.27 ± 0.07 ) , (101)
where we have assumed constant q q q and ε \varepsilon ε . The errors indicate the variation of the efficiency factors over the parameters a a a , b b b , and α \alpha α .
For comparison, the mean squared value of the unprojected stresses is [see details in App. B, in particular Eq. (312) ]
ρ ˉ T φ = a − 4 ∫ E T φ ( K ) d ln K = 5 3 φ ~ r m s 4 , (102) \bar{\rho}_{T}^{\varphi}=a^{-4}\int E_{T}^{\varphi}(K)\,\mathrm{d}\ln K=\frac{5}{3}\tilde{\varphi}_{\mathrm{rms}}^{4}, \tag*{(102)} ρ ˉ T φ = a − 4 ∫ E T φ ( K ) d ln K = 3 5 φ ~ rms 4 , (102)
such that the equivalent split of ρ ˉ T φ \bar{\rho}_{T}^{\varphi} ρ ˉ T φ into components as in Eq. (94) is
E T v o r t ( 1 − q ) 2 = E T c o m p q 2 = E T m i x 2 q ( 1 − q ) = ρ ˉ T φ φ ~ r m s 4 = 5 3 , (103) \frac{\mathcal{E}_{T}^{\mathrm{vort}}}{(1-q)^2}=\frac{\mathcal{E}_{T}^{\mathrm{comp}}}{q^2}=\frac{\mathcal{E}_{T}^{\mathrm{mix}}}{2q(1-q)}=\frac{\bar{\rho}_{T}^{\varphi}}{\tilde{\varphi}_{\mathrm{rms}}^{4}}=\frac{5}{3}, \tag*{(103)} ( 1 − q ) 2 E T vort = q 2 E T comp = 2 q ( 1 − q ) E T mix = φ ~ rms 4 ρ ˉ T φ = 3 5 , (103)
and E T h e l = 0 \mathcal{E}_{T}^{\mathrm{hel}}=0 E T hel = 0 . Unlike the efficiencies for the unprojected stresses, the anisotropic efficiencies E Π φ , i \mathcal{E}_{\Pi}^{\varphi,i} E Π φ , i depend on the spectral shape due to the appearance of the dot products z = p ^ ⋅ k ^ z=\hat{p}\cdot\hat{k} z = p ^ ⋅ k ^ and z ~ = p ~ ^ ⋅ k ^ \tilde{z}=\hat{\tilde{p}}\cdot\hat{k} z ~ = p ~ ^ ⋅ k ^ , introduced by the Λ \boldsymbol{\Lambda} Λ projection tensor. Furthermore, the projection tensor breaks the symmetry that makes the contribution from helicity to the unprojected stresses to vanish, such that helicity contributes to the mean squared value of the anisotropic stresses. Similarly, we see in Figure 2 that the helical contribution to the anisotropic stresses spectrum is also larger than its contribution to the unprojected stresses spectrum at wave numbers K ≳ 1 K\gtrsim1 K ≳ 1 . For the vortical and mixed4 components, we find that the efficiency is larger and much less sensitive to the spectral shape, with variations up to ∼ 30 % \sim30\% ∼ 30% , while the compressional and helical components are much more sensitive, with variations up to ∼ 40 % \sim40\% ∼ 40% . We find that steeper (larger a a a or larger b b b ) and sharper (larger α \alpha α ) spectra lead to a larger value of E Π \mathcal{E}_{\Pi} E Π . On the other hand, even though the helical component is negative at small K K K , hence, reducing the amplitude of the total spectral branch proportional to K 3 K^3 K 3 , it becomes positive around the peak and its net contribution to the total mean squared stresses is always positive, hence yielding overall larger efficiencies for helical fields. This result is consistent with a larger amplitude in the GW spectrum found in numerical simulations for magnetic fields with a larger fractional helicity [45 , 46 , 48 , 49 ] .
Interestingly, as already shown in ref. [48 ] , vortical modes are significantly more efficient at producing anisotropic stresses than compressional ones by a factor of E Π v o r t ( q = 0 ) / E Π c o m p ( q = 1 ) ∼ 6 \mathcal{E}_{\Pi}^{\mathrm{vort}}(q=0)/\mathcal{E}_{\Pi}^{\mathrm{comp}}(q=1)\sim6 E Π vort ( q = 0 ) / E Π comp ( q = 1 ) ∼ 6 . This seems in contradiction with the results of numerical simulations of GW production, which show that compressional motion is more efficient in producing GWs [44 , 45 , 46 , 48 , 49 ] . The difference in the GW amplitude can then only be due to the dynamics of the UETC of the source and it needs to compensate for the different efficiency at producing anisotropic stresses (see Figure 3 ). We address this issue in Sec. VI, where the CiT model [1], described in § 0.5 , is compared to the GW production from sound waves and stationary turbulence. There, we show that the actual dynamics of the UETC of the source impacts the resulting GW production when either the speed of propagation or decorrelation in time is large, leading to an enhancement of the GW signal with respect to the predictions of the CiT model. Therefore, when we consider sound waves in a pure radiation fluid, i.e., with an ultrarelativistic speed of sound c s 2 = 1 / 3 c_s^2=1/3 c s 2 = 1/3 , they transfer kinetic energy into GWs more efficiently than decaying vortical turbulence with subrelativistic decorrelation velocity u d c o r r ≪ 1 u_{\mathrm{dcorr}}\ll1 u dcorr ≪ 1 . These results clarify the origin of the different efficiencies found when converting the energy of the source into GWs. Furthermore, the presented model shows how relativistic fluid velocities can lead to a significant enhancement, similar to the one found in sound waves, of the GW signal from vortical turbulence, as we show in Sec. VI.
The purely vortical or compressional components have been extensively considered in the literature [26 , 28 , 29 , 30 , 34 , 35 , 36 , 37 , 38 , 40 , 41 , 42 , 43 , 46 , 48 , 49 , 50 , 66 , 67 , 77 , 95 ] [1, 5, 21]. Let us now consider the additional contributions to the anisotropic stresses spectrum when the velocity field5 has both normal and longitudinal components. In this case, assuming small fluid perturbations,
E Π u ( K ) = Γ 2 u r m s 4 C u A a b 2 ( α ) K 3 ζ Π u ( K ) , (104) E_{\Pi}^{u}(K)=\Gamma^{2}u_{\mathrm{rms}}^{4}\frac{C_{u}}{A_{ab}^{2}(\alpha)}K^{3}\zeta_{\Pi}^{u}(K), \tag*{(104)} E Π u ( K ) = Γ 2 u rms 4 A ab 2 ( α ) C u K 3 ζ Π u ( K ) , (104)
where
C u ζ Π u ( K ) = ∑ i C i u ζ Π u , i ( K ) , (105) C_{u}\zeta_{\Pi}^{u}(K)=\sum_{i}C_{i}^{u}\zeta_{\Pi}^{u,i}(K), \tag*{(105)} C u ζ Π u ( K ) = i ∑ C i u ζ Π u , i ( K ) , (105)
with i = { v o r t , c o m p , m i x , h e l } i=\{\mathrm{vort},\mathrm{comp},\mathrm{mix},\mathrm{hel}\} i = { vort , comp , mix , hel } . As before, the parameter c u c_{u} c u determines the value of the velocity anisotropic stresses spectrum in the K → 0 K\to0 K → 0 limit, such that ∣ ζ Π u ( K → 0 ) ∣ = 1 |\zeta_{\Pi}^{u}(K\to0)|=1 ∣ ζ Π u ( K → 0 ) ∣ = 1 .
For simplicity, let us now assume that the degree of compressibility q q q in Eq. (39) is a constant and that the velocity field is non-helical (i.e., ε u = 0 \varepsilon_{u}=0 ε u = 0 ). Then, C u C_{u} C u becomes [see Eqs. (91) and (92) ]
C u n o h e l = c u n o h e l ∫ 0 ∞ c u 2 ( P ) P 4 d P , (106) C_{u}^{\mathrm{nohel}}=c_{u}^{\mathrm{nohel}}\int_{0}^{\infty}\frac{c_{u}^{2}(P)}{P^{4}}\,\mathrm{d}P, \tag*{(106)} C u nohel = c u nohel ∫ 0 ∞ P 4 c u 2 ( P ) d P , (106)
where
c u n o h e l = 4 9 + 1 5 ( q − 1 3 ) 2 . (107) c_{u}^{\mathrm{nohel}}=\frac{4}{9}+\frac{1}{5}\left(q-\frac{1}{3}\right)^{2}. \tag*{(107)} c u nohel = 9 4 + 5 1 ( q − 3 1 ) 2 . (107)
The value c u n o h e l c_{u}^{\mathrm{nohel}} c u nohel recovers c v o r t = 7 / 15 c_{\mathrm{vort}}=7/15 c vort = 7/15 and c c o m p = 8 / 15 c_{\mathrm{comp}}=8/15 c comp = 8/15 for q = 0 q=0 q = 0 and 1 1 1 , respectively, and has a minimum value 4 / 9 4/9 4/9 at q = 1 / 3 q=1/3 q = 1/3 .
Similarly, if the helicity fraction ε u \varepsilon_{u} ε u , defined in Eq. (53) , is a non-zero constant, the resulting c u c_{u} c u , defined as in Eqs. (88), becomes3
c u = c u n o h e l − 1 3 ε u 2 ( 1 − q ) 2 , (108) c_{u}=c_{u}^{\mathrm{nohel}}-\frac{1}{3}\varepsilon_{u}^{2}(1-q)^{2}, \tag*{(108)} c u = c u nohel − 3 1 ε u 2 ( 1 − q ) 2 , (108)
where the negative sign comes from ζ Π h e l ( K ) \zeta_{\Pi}^{\mathrm{hel}}(K) ζ Π hel ( K ) being negative at small K K K (see Figure 3 ). As the helicity fraction increases, c u c_{u} c u decreases until it reaches its minimum value for ε u 2 = 1 \varepsilon_{u}^{2}=1 ε u 2 = 1 ,
c u ( ε u 2 = 1 ) = 2 3 − 2 15 ( 2 − q ) 2 , (109) c_{u}(\varepsilon_{u}^{2}=1)=\frac{2}{3}-\frac{2}{15}(2-q)^{2}, \tag*{(109)} c u ( ε u 2 = 1 ) = 3 2 − 15 2 ( 2 − q ) 2 , (109)
which varies from 2 / 15 2/15 2/15 for q = 0 q=0 q = 0 to 8 / 15 8/15 8/15 for q = 1 q=1 q = 1 . Note that the helical spectrum for the velocity field is proportional to ε u ( 1 − q ) \varepsilon_{u}(1-q) ε u ( 1 − q ) [see Eq. (58c)], hence for q = 1 q=1 q = 1 there is no helicity, and the values of c u c_{u} c u and c u n o h e l c_{u}^{\mathrm{nohel}} c u nohel coincide and are both given by c c o m p c_{\mathrm{comp}} c comp .
q = 1 q=1 q = 1 there is no helicity, and the values of c u c_{u} c u and c u n o h e l c_{u}^{\mathrm{nohel}} c u nohel coincide and are both given by c c o m p c_{\mathrm{comp}} c comp .
The normalized anisotropic stress spectra K 3 c u ζ Π u ( K ) K^{3}c_{u}\zeta_{\Pi}^{u}(K) K 3 c u ζ Π u ( K ) , divided by A a b 2 ( α ) A_{ab}^{2}(\alpha) A ab 2 ( α ) , for velocity fields with different degrees of compressibility q = { 0 , 0.5 , 0.8 , 0.95 , 1 } q=\{0,0.5,0.8,0.95,1\} q = { 0 , 0.5 , 0.8 , 0.95 , 1 } and different fractional kinetic helicities ∣ ε u ∣ ∈ [ 0 , 1 ] |\varepsilon_{u}|\in[0,1] ∣ ε u ∣ ∈ [ 0 , 1 ] are shown in Figure 4 for the von Kármán spectrum with a = 5 a=5 a = 5 , b = 2 / 3 b=2/3 b = 2/3 , and α = 6 / 17 \alpha=6/17 α = 6/17 and α = 2 \alpha=2 α = 2 . We note that even when the vorticity is subdominant in the system, i.e., the degree of compressibility is very close to one but not exactly one (e.g., 0.95 0.95 0.95 ), then the spectrum of the stresses is modified at large K K K , since E Π c o m p ( K ) ∼ K − 2 b E_{\Pi}^{\mathrm{comp}}(K)\sim K^{-2b} E Π comp ( K ) ∼ K − 2 b decays faster than E Π v o r t ( K ) ∼ K − b E_{\Pi}^{\mathrm{vort}}(K)\sim K^{-b} E Π vort ( K ) ∼ K − b at large K K K . The shaded regions represent variation of ∣ ε u ∣ |\varepsilon_{u}| ∣ ε u ∣ from 0 to 1. For low K K K , the helical contribution is negative and, hence, smaller amplitudes correspond to larger fractional helicities. At wave numbers at or above the peak, K ≳ 1 K\gtrsim1 K ≳ 1 , the helicity contribution becomes positive. In this range, the maximum amplitude corresponds to ε u = ± 1 \varepsilon_{u}=\pm1 ε u = ± 1 . In Figure 5 , we show the integrated efficiency H Π u \mathcal{H}_{\Pi}^{u} H Π u [see Eq. (95) ].
Figure 4. Normalized spectrum of the anisotropic stresses for a velocity von Kármán spectrum ζ u ( K ) \zeta_u(K) ζ u ( K ) with a = 5 a=5 a = 5 , b = 2 / 3 b=2/3 b = 2/3 , and α = 6 / 17 \alpha=6/17 α = 6/17 (upper panel) and α = 2 \alpha=2 α = 2 (lower panel) for different constant values of the degree of compressibility q q q (different colors) and fractional helicity ∣ ϵ u ∣ ∈ [ 0 , 1 ] |\epsilon_u| \in[0,1] ∣ ϵ u ∣ ∈ [ 0 , 1 ] (shaded regions). Lower panel: Efficiency factor E Π u \mathcal{E}_{\Pi}^{u} E Π u characterizing the transfer of energy into anisotropic stresses, as a function of q q q for different values of ∣ ϵ u ∣ ∈ [ 0 , 1 ] |\epsilon_u| \in[0,1] ∣ ϵ u ∣ ∈ [ 0 , 1 ] (shaded red region), indicating the largest efficiency for q = 0 q=0 q = 0 (vortical) and lowest one for q = 1 q=1 q = 1 (compressional) as well as a generic trend of larger efficiency for larger ∣ ϵ u ∣ |\epsilon_u| ∣ ϵ u ∣ . The value of 5 / 3 5/3 5/3 found for the unprojected stresses is shown for comparison. All panels apply to magnetic fields with different fractional helicities ϵ B ∈ [ 0 , 1 ] \epsilon_B \in[0,1] ϵ B ∈ [ 0 , 1 ] when q = 0 q=0 q = 0 .
Figure 5. Efficiency factors E Π u \mathcal{E}_{\Pi}^{u} E Π u (solid lines) and H Π u \mathcal{H}_{\Pi}^{u} H Π u (dashed lines) characterizing the transfer of energy from unprojected into anisotropic stresses for the von Kármán spectrum with a = 5 a=5 a = 5 , b = 2 / 3 b=2/3 b = 2/3 and two values of α = 6 / 17 \alpha=6/17 α = 6/17 and 2, as a function of q q q for different values of ∣ ϵ u ∣ ∈ ( 0 , 1 ] |\epsilon_u| \in(0,1] ∣ ϵ u ∣ ∈ ( 0 , 1 ] , indicating the largest efficiency for q = 0 q=0 q = 0 (vortical) and lowest one for q = 1 q=1 q = 1 (compressional) as well as a generic trend of larger efficiency for larger ∣ ϵ u ∣ |\epsilon_u| ∣ ϵ u ∣ . Note that the realizability condition requires ∣ H Π u ∣ < E Π u |\mathcal{H}_{\Pi}^{u}|<\mathcal{E}_{\Pi}^{u} ∣ H Π u ∣ < E Π u . Stresses from magnetic fields correspond to q = 0 q=0 q = 0 .
Antisymmetric spectrum of the anisotropic stresses H Π H_{\Pi} H Π
In the previous section, the symmetric UETC of the anisotropic stresses, E Π ( k ) E_{\Pi}(k) E Π ( k ) , has been computed. In this section, we compute the antisymmetric UETC of the anisotropic stresses, H Π ( τ 1 , τ 2 , k ) H_{\Pi}(\tau_{1},\tau_{2},k) H Π ( τ 1 , τ 2 , k ) , which sources the chiral spectrum of GWs [see description below Eq. (24) ]. For this purpose, we now project Eq. (77) with the tensor A i j l m A_{ijlm} A ij l m (see Appendix B for details).
We find that the antisymmetric spectrum of a vector source φ = { X , B } \varphi=\{X,B\} φ = { X , B } , H Π φ ( k ) H_{\Pi}^{\varphi}(k) H Π φ ( k ) , is only non-zero when the fractional helicity of the source is non-zero. For velocity fields, it can be decomposed into a compressional (i = c o m p i=\mathrm{comp} i = comp ) and a vortical (i = v o r t i=\mathrm{vort} i = vort ) component, while it has a unique vortical component for magnetic fields,
H Π u ( k ) = H Π u , v o r t ( k ) + H Π u , c o m p ( k ) , (110) H_{\Pi}^{u}(k)=H_{\Pi}^{u,\mathrm{vort}}(k)+H_{\Pi}^{u,\mathrm{comp}}(k), \tag*{(110)} H Π u ( k ) = H Π u , vort ( k ) + H Π u , comp ( k ) , (110)
H Π B ( k ) = H Π B , v o r t ( k ) . (111) H_{\Pi}^{B}(k)=H_{\Pi}^{B,\mathrm{vort}}(k). \tag*{(111)} H Π B ( k ) = H Π B , vort ( k ) . (111)
Even though H Π u ( k ) H_{\Pi}^{u}(k) H Π u ( k ) contains a compressional component, it vanishes for a purely compressional field, since then ε u → 0 \varepsilon_{u}\to0 ε u → 0 . Therefore, the presence of a compressional component always requires that the vortical component is non-zero.
The two components of the antisymmetric anisotropic stresses spectrum are given in Eqs. (B8e) and (B8f) (see also ref. [66 ] for H Π v o r t ( k ) H_{\Pi}^{\mathrm{vort}}(k) H Π vort ( k ) ),
a 4 H Π v o r t ( k ) = k 2 2 ρ ˉ 2 ∫ 0 ∞ d p p 2 E N ( p ) ∫ ∣ k − p ∣ k + p d p ~ p ~ 2 H ( p ~ ) z ( 1 + z ~ 2 ) , (112) a^{4}H_{\Pi}^{\mathrm{vort}}(k)=\frac{k^{2}}{2\bar{\rho}^{2}}\int_{0}^{\infty}\frac{\mathrm{d}p}{p^{2}}E_{N}(p)\int_{\lvert k-p\rvert}^{k+p}\frac{\mathrm{d}\tilde{p}}{\tilde{p}^{2}}H(\tilde{p})z(1+\tilde{z}^{2}), \tag*{(112)} a 4 H Π vort ( k ) = 2 ρ ˉ 2 k 2 ∫ 0 ∞ p 2 d p E N ( p ) ∫ ∣ k − p ∣ k + p p ~ 2 d p ~ H ( p ~ ) z ( 1 + z ~ 2 ) , (112)
a 4 H Π c o m p ( k ) = k 2 2 ρ ˉ 2 ∫ 0 ∞ d p p 2 E L ( p ) ∫ ∣ k − p ∣ k + p d p ~ p ~ 2 H ( p ~ ) z ( 1 − z ~ 2 ) . (113) a^{4}H_{\Pi}^{\mathrm{comp}}(k)=\frac{k^{2}}{2\bar{\rho}^{2}}\int_{0}^{\infty}\frac{\mathrm{d}p}{p^{2}}E_{L}(p)\int_{\lvert k-p\rvert}^{k+p}\frac{\mathrm{d}\tilde{p}}{\tilde{p}^{2}}H(\tilde{p})z(1-\tilde{z}^{2}). \tag*{(113)} a 4 H Π comp ( k ) = 2 ρ ˉ 2 k 2 ∫ 0 ∞ p 2 d p E L ( p ) ∫ ∣ k − p ∣ k + p p ~ 2 d p ~ H ( p ~ ) z ( 1 − z ~ 2 ) . (113)
The anisotropic stresses satisfying the TT properties can be decomposed into two polarization modes
Π i j ( k ) = ∑ λ = ± e i j λ Π λ ( k ) , (114) \Pi_{ij}(\boldsymbol{k})=\sum_{\lambda=\pm}e_{ij}^{\lambda}\Pi_{\lambda}(k), \tag*{(114)} Π ij ( k ) = λ = ± ∑ e ij λ Π λ ( k ) , (114)
in the basis [66 , 96 ]
e i j ± = ( e 1 ± i e 2 ) i × ( e 1 ± i e 2 ) j . (115) e_{ij}^{\pm}=(e_{1}\pm i e_{2})_{i}\times(e_{1}\pm i e_{2})_{j}. \tag*{(115)} e ij ± = ( e 1 ± i e 2 ) i × ( e 1 ± i e 2 ) j . (115)
Then, ref. [66 ] shows that the ratio of the antisymmetric to symmetric spectral functions of the anisotropic stresses is
P Π ( k ) ≡ H Π ( k ) E Π ( k ) = ∣ Π + ( k ) ∣ 2 − ∣ Π − ( k ) ∣ 2 ∣ Π + ( k ) ∣ 2 + ∣ Π − ( k ) ∣ 2 ∈ [ − 1 , 1 ] . (116) \mathcal{P}_{\Pi}(k)\equiv\frac{H_{\Pi}(k)}{E_{\Pi}(k)}
=\frac{|\Pi_{+}(k)|^{2}-|\Pi_{-}(k)|^{2}}{|\Pi_{+}(k)|^{2}+|\Pi_{-}(k)|^{2}}\in[-1,1]. \tag*{(116)} P Π ( k ) ≡ E Π ( k ) H Π ( k ) = ∣ Π + ( k ) ∣ 2 + ∣ Π − ( k ) ∣ 2 ∣ Π + ( k ) ∣ 2 − ∣ Π − ( k ) ∣ 2 ∈ [ − 1 , 1 ] . (116)
Similarly to the realizability condition for the sourcing fields, ∣ H ( k ) ∣ ≤ E ( k ) |H(k)|\leq E(k) ∣ H ( k ) ∣ ≤ E ( k ) , this implies that ∣ H Π ( k ) ∣ ≤ E Π ( k ) |H_{\Pi}(k)|\leq E_{\Pi}(k) ∣ H Π ( k ) ∣ ≤ E Π ( k ) . The polarization P Π ( k ) \mathcal{P}_{\Pi}(k) P Π ( k ) characterizes the degree of circular polarization of the GWs, P G W ( k ) \mathcal{P}_{\mathrm{GW}}(k) P GW ( k ) . In particular, both functions become identical in the CiT model, as we show in § 0.5 C.
Using the normalization of Eqs. (58) for the spectral components of the source, as we have done in Eq. (86) for the components of the symmetric spectrum of the anisotropic stresses, we find [see Eq. (275) ]
H Π φ , i ( K ) = ξ r m s 4 ρ ˉ 2 C a b φ , H , i A a b 2 K 4 ζ Π φ , H , i ( K ) , (117) H_{\Pi}^{\varphi,i}(K)=\frac{\xi_{\mathrm{rms}}^{4}}{\bar{\rho}^{2}}\frac{C_{ab}^{\varphi,H,i}}{\mathcal{A}_{ab}^{2}}K^{4}\zeta_{\Pi}^{\varphi,H,i}(K), \tag*{(117)} H Π φ , i ( K ) = ρ ˉ 2 ξ rms 4 A ab 2 C ab φ , H , i K 4 ζ Π φ , H , i ( K ) , (117)
with i = { v o r t , c o m p } i=\{\mathrm{vort},\mathrm{comp}\} i = { vort , comp } . We note that, unlike the symmetric stress tensor, the antisymmetric stress tensor components are proportional to K 4 K^{4} K 4 in the limit K → 0 K\to0 K → 0 , as expected from causality [79 ] . The normalized ζ Π φ , H , i ( K ) \zeta_{\Pi}^{\varphi,H,i}(K) ζ Π φ , H , i ( K ) are given in Eqs. (B14e) and (B14f), and the prefactors C a b φ , H , i C_{ab}^{\varphi,H,i} C ab φ , H , i are again chosen such that ∣ ζ Π φ , H , i ( K ) ∣ → 1 \left|\zeta_{\Pi}^{\varphi,H,i}(K)\right|\to1 ζ Π φ , H , i ( K ) → 1 when K → 0 K\to0 K → 0 , given in Eqs. (B22)
C v o r t H = 2 15 ∣ ∫ 0 ∞ [ 1 − q ( P ) ] 2 ε ( P ) ζ 2 ( P ) P 5 d P ∣ , (118) C_{\mathrm{vort}}^{H}=\frac{2}{15}\left|\int_{0}^{\infty}[1-q(P)]^{2}\varepsilon(P)\frac{\zeta^{2}(P)}{P^{5}}\,\mathrm{d}P\right|, \tag*{(118)} C vort H = 15 2 ∫ 0 ∞ [ 1 − q ( P ) ] 2 ε ( P ) P 5 ζ 2 ( P ) d P , (118)
C c o m p H = 2 5 ∣ ∫ 0 ∞ q ( P ) [ 1 − q ( P ) ] ε ( P ) ζ 2 ( P ) P 5 d P ∣ . (119) C_{\mathrm{comp}}^{H}=\frac{2}{5}\left|\int_{0}^{\infty}q(P)[1-q(P)]\varepsilon(P)\frac{\zeta^{2}(P)}{P^{5}}\,\mathrm{d}P\right|. \tag*{(119)} C comp H = 5 2 ∫ 0 ∞ q ( P ) [ 1 − q ( P )] ε ( P ) P 5 ζ 2 ( P ) d P . (119)
When q q q and ε \varepsilon ε are both constants, and assuming that ζ ( P ) \zeta(P) ζ ( P ) is a smoothed broken power law as given in Eq. (63) , these coefficients can be expressed as
C i H = c i H ∫ 0 ∞ ζ 2 ( P ) P 5 d P = c i H I 2 a , 2 b ; − 5 ( α / 2 ) = c i H ( a + b b ) 2 α ( a b ) − 2 a − 4 α ( a + b ) Γ [ 2 a − 4 α ( a + b ) ] Γ [ 2 b + 4 α ( a + b ) ] α ( a + b ) Γ [ 2 α ] , (120) \begin{aligned}
C_{i}^{H}=c_{i}^{H}\int_{0}^{\infty}\frac{\zeta^{2}(P)}{P^{5}}\,\mathrm{d}P=c_{i}^{H}I_{2a,2b;-5}(\alpha/2) \\
&=c_{i}^{H}\left(\frac{a+b}{b}\right)^{\frac{2}{\alpha}}\left(\frac{a}{b}\right)^{-\frac{2a-4}{\alpha(a+b)}}\frac{\Gamma\left[\frac{2a-4}{\alpha(a+b)}\right]\Gamma\left[\frac{2b+4}{\alpha(a+b)}\right]}{\alpha(a+b)\Gamma\left[\frac{2}{\alpha}\right]}, \tag*{(120)}
\end{aligned} C i H = c i H ∫ 0 ∞ P 5 ζ 2 ( P ) d P = c i H I 2 a , 2 b ; − 5 ( α /2 ) = c i H ( b a + b ) α 2 ( b a ) − α ( a + b ) 2 a − 4 α ( a + b ) Γ [ α 2 ] Γ [ α ( a + b ) 2 a − 4 ] Γ [ α ( a + b ) 2 b + 4 ] , (120)
with
c i H = 2 15 ∣ ε ∣ ( 1 − q ) { 1 − q , 3 q } (121) c_{i}^{H}=\frac{2}{15}|\varepsilon|(1-q)\{1-q,\,3q\} \tag*{(121)} c i H = 15 2 ∣ ε ∣ ( 1 − q ) { 1 − q , 3 q } (121)
for i = { v o r t , c o m p } i=\{\mathrm{vort},\mathrm{comp}\} i = { vort , comp } . The dependence of C a b i , H ( α ) / c i H C_{ab}^{i,H}(\alpha)/c_{i}^{H} C ab i , H ( α ) / c i H is shown in Figure 31 .
Figure 31. Geometrical factors C a b H ( α ) / c = I 2 a , 2 b ; − 5 ( α / 2 ) \mathcal{C}^{\mathrm{H}}_{ab}(\alpha)/c=I_{2a,2b;-5}(\alpha/2) C ab H ( α ) / c = I 2 a , 2 b ; − 5 ( α /2 ) that determine the amplitude of the helical anisotropic stresses in the K → 0 K\to0 K → 0 limit for different values of α \alpha α and the same slopes as in Figure 28 . Notice that the ratio C a b j , H / C j H \mathcal{C}^{j,\mathrm{H}}_{ab}/\mathcal{C}^{\mathrm{H}}_j C ab j , H / C j H does not depend on j j j , hence we can omit the index here.
In analogy to Eq. (95) , we define an efficiency factor characterizing the conversion of the mean stress of the source into an antisymmetric component of the anisotropic stresses,
H Π φ , i = C a b φ , H , i A a b 2 ( α ) ∫ 0 ∞ K 3 ζ Π φ , H , i ( K ) d ln K , (122) \mathcal{H}_{\Pi}^{\varphi,i}=\frac{C_{ab}^{\varphi,H,i}}{\mathcal{A}_{ab}^{2}(\alpha)}\int_{0}^{\infty}K^{3}\zeta_{\Pi}^{\varphi,H,i}(K)\,\mathrm{d}\ln K, \tag*{(122)} H Π φ , i = A ab 2 ( α ) C ab φ , H , i ∫ 0 ∞ K 3 ζ Π φ , H , i ( K ) d ln K , (122)
which also allows us to define a function ζ Π , i n t φ , H , i ( K ) \zeta_{\Pi,\mathrm{int}}^{\varphi,H,i}(K) ζ Π , int φ , H , i ( K ) normalized by its integral, as in Eq. (96) , such that ∫ ζ Π , i n t φ , H , i ( K ) d ln K = 1 \int\zeta_{\Pi,\mathrm{int}}^{\varphi,H,i}(K)\,\mathrm{d}\ln K=1 ∫ ζ Π , int φ , H , i ( K ) d ln K = 1 . Due to the property in Eq. (116) , ∣ H Π φ ∣ < E Π φ \left|\mathcal{H}_{\Pi}^{\varphi}\right|<\mathcal{E}_{\Pi}^{\varphi} ∣ H Π φ ∣ < E Π φ after summing over the corresponding components. We note that the equality can never be exhausted because causality requires ε ∼ K \varepsilon\sim K ε ∼ K when K → 0 K\to0 K → 0 .
The helical spectrum of the anisotropic stresses for a generic velocity field is
H Π u ( K ) = Γ 2 u r m s 4 C u H A a b 2 ( α ) K 4 ζ Π u , H ( K ) , (123) H_{\Pi}^{u}(K)=\Gamma^{2}u_{\mathrm{rms}}^{4}\frac{C_{u}^{H}}{\mathcal{A}_{ab}^{2}(\alpha)}K^{4}\zeta_{\Pi}^{u,H}(K), \tag*{(123)} H Π u ( K ) = Γ 2 u rms 4 A ab 2 ( α ) C u H K 4 ζ Π u , H ( K ) , (123)
where
C u H ζ Π u , H ( K ) = ∑ i C i H ζ Π u , H , i ( K ) , (124) C_{u}^{H}\zeta_{\Pi}^{u,H}(K)=\sum_{i}C_{i}^{H}\zeta_{\Pi}^{u,H,i}(K), \tag*{(124)} C u H ζ Π u , H ( K ) = i ∑ C i H ζ Π u , H , i ( K ) , (124)
with i = { v o r t , c o m p } i=\{\mathrm{vort},\mathrm{comp}\} i = { vort , comp } . Note that the helical spectrum of the anisotropic stresses is proportional to the helicity fraction and the vortical component of the source. Therefore, if the source has a constant positive or negative helicity at all modes k k k , it will induce positive or negative stresses H Π u ( K ) H_{\Pi}^{u}(K) H Π u ( K ) , and GW polarization [46 , 49 , 67 ] . For constant q q q and ε u \varepsilon_{u} ε u , then C u H C_{u}^{H} C u H is given by Eq. (120) with c u H = 2 15 ∣ ε u ∣ ( 1 − q ) ( 1 + 2 q ) c_{u}^{H}=\frac{2}{15}|\varepsilon_{u}|(1-q)(1+2q) c u H = 15 2 ∣ ε u ∣ ( 1 − q ) ( 1 + 2 q ) , obtained adding up the two components in Eq. (121) .
The resulting normalized spectra K 4 C u H ζ Π u , H ( K ) K^{4}C_{u}^{H}\zeta_{\Pi}^{u,H}(K) K 4 C u H ζ Π u , H ( K ) , divided by A 2 \mathcal{A}^{2} A 2 and ∣ ε ∣ |\varepsilon| ∣ ε ∣ , for velocity fields with different degrees of compressibility are shown in Figure 6 . The efficiency factor ∣ H Π u ∣ \left|\mathcal{H}_{\Pi}^{u}\right| ∣ H Π u ∣ for different values of q q q and ε u \varepsilon_{u} ε u for the von Kármán spectrum with a = 5 a=5 a = 5 , b = 2 / 3 b=2/3 b = 2/3 , and two values of α = 6 / 17 \alpha=6/17 α = 6/17 and 2 2 2 , is shown in Figure 5 . The spectra and efficiencies for magnetic fields correspond to the fully vortical case (q = 0 q=0 q = 0 ).
Figure 6. Normalized helical spectra of the anisotropic stresses, divided by a constant fractional helicity, K 4 ζ Π H ( K ) / ∣ ε u ∣ K^{4} \zeta_{\Pi}^{\mathrm{H}}(K)/|\varepsilon_u| K 4 ζ Π H ( K ) /∣ ε u ∣ and normalized with A a b 2 ( α ) A_{ab}^{2}(\alpha) A ab 2 ( α ) , corresponding to a velocity field with a constant degree of compressibility q ∈ [ 0 , 1 ) q \in[0,1) q ∈ [ 0 , 1 ) . Notice that for q = 1 q=1 q = 1 , since there is no helicity [see Eq. (58c)], we have ζ Π H ( K ) = 0 \zeta_{\Pi}^{\mathrm{H}}(K)=0 ζ Π H ( K ) = 0 . The velocity spectrum is a smoothed broken power law with a = 5 a=5 a = 5 , b = 2 / 3 b=2/3 b = 2/3 , and α = 6 / 17 \alpha=6/17 α = 6/17 (upper panel) and α = 2 \alpha=2 α = 2 (lower panel). The helicity spectrum for a magnetic field source, ζ Π H , B ( K ) \zeta_{\Pi}^{\mathrm{H,B}}(K) ζ Π H , B ( K ) , corresponds to the purely vortical case, i.e., q = 0 q=0 q = 0 . Dashed lines correspond to the symmetric spectra K 3 C Π H ( K ) / A a b 2 ( α ) K^{3}C_{\Pi}^{\mathrm{H}}(K)/A_{ab}^{2}(\alpha) K 3 C Π H ( K ) / A ab 2 ( α ) shown in Figure 4 , with ∣ ε u ∣ = 1 |\varepsilon_u|=1 ∣ ε u ∣ = 1 . Scaling lines ∼ K 4 \sim K^{4} ∼ K 4 and ∼ K − b \sim K^{-b} ∼ K − b are shown at the same positions in both panels to allow for a direct comparison.
Smoothed broken power law fits
As we have discussed, the spectrum of the anisotropic stresses sourcing GWs (energy density and polarization) can in general be decomposed as a superposition of terms reflecting the contributions from different components of the spectrum of the source (compressional, vortical, helical). Each of these terms can be described by normalized spectral functions ζ Π ( K ) \zeta_{\Pi}(K) ζ Π ( K ) and ζ Π H ( K ) \zeta_{\Pi}^{H}(K) ζ Π H ( K ) following the normalization in Eqs. (74) and (96), which take values of one when K → 0 K\to0 K → 0 .
As it can be seen respectively in Figure 2 and Figure 6 , all of the contributions to the symmetric and antisymmetric spectra, ζ Π ( K ) \zeta_{\Pi}(K) ζ Π ( K ) and ζ Π H ( K ) \zeta_{\Pi}^{H}(K) ζ Π H ( K ) , become constant at K ≪ 1 K\ll1 K ≪ 1 and then decay proportional to a power law at K ≫ 1 K\gg1 K ≫ 1 whose exponent only depends on the power law decay exponent of the spectrum of the source, b b b . This motivates us to consider the following broken power law fits, analogous to the model given in Eq. (63) , but adapted for a = 0 a=0 a = 0 , to describe each component separately,
∣ ζ Π i ( K ) ∣ = [ 1 + ( K / K Π i ) α Π i ] − b Π i / α Π i (125) \left|\zeta_{\Pi}^{i}(K)\right|=\left[1+\left(K/K_{\Pi}^{i}\right)^{\alpha_{\Pi}^{i}}\right]^{-b_{\Pi}^{i}/\alpha_{\Pi}^{i}} \tag*{(125)} ζ Π i ( K ) = [ 1 + ( K / K Π i ) α Π i ] − b Π i / α Π i (125)
∣ ζ Π H , i ( K ) ∣ = [ 1 + ( K / K Π H , i ) α Π H , i ] − b Π H , i / α Π H , i , (126) \left|\zeta_{\Pi}^{\mathrm{H},i}(K)\right|=\left[1+\left(K/K_{\Pi}^{\mathrm{H},i}\right)^{\alpha_{\Pi}^{\mathrm{H},i}}\right]^{-b_{\Pi}^{\mathrm{H},i}/\alpha_{\Pi}^{\mathrm{H},i}}, \tag*{(126)} ζ Π H , i ( K ) = [ 1 + ( K / K Π H , i ) α Π H , i ] − b Π H , i / α Π H , i , (126)
where the large-K K K slope of the stresses spectrum is
b Π v o r t , m i x = b + 3 , (127) b_{\Pi}^{\mathrm{vort,mix}}=b+3, \tag*{(127)} b Π vort , mix = b + 3 , (127)
b Π c o m p = min ( 2 b + 3 , b + 5 ) b ≠ 2 , (128) b_{\Pi}^{\mathrm{comp}}=\min(2b+3,b+5)\qquad b\ne2, \tag*{(128)} b Π comp = min ( 2 b + 3 , b + 5 ) b = 2 , (128)
b Π h e l = min ( 2 b + 3 , b + 4 ) b ≠ 1 , (129) b_{\Pi}^{\mathrm{hel}}=\min(2b+3,b+4)\qquad b\ne1, \tag*{(129)} b Π hel = min ( 2 b + 3 , b + 4 ) b = 1 , (129)
b Π H , c o m p , v o r t = b + 4. (130) b_{\Pi}^{\mathrm{H,comp,vort}}=b+4. \tag*{(130)} b Π H , comp , vort = b + 4. (130)
The resulting fits found for the various components of the stresses are shown in Figure 7 for the von Kármán spectrum with a = 5 a=5 a = 5 , b = 2 / 3 b=2/3 b = 2/3 , and different values of α \alpha α . Note that b Π i b_{\Pi}^{i} b Π i and b Π H , i b_{\Pi}^{\mathrm{H},i} b Π H , i are completely determined by the small-scale broken power law slope b b b , as in Eqs. (104), hence the only fitting parameters are K Π i , α Π i K_{\Pi}^{i},\alpha_{\Pi}^{i} K Π i , α Π i and K Π H , i , α Π H , i K_{\Pi}^{\mathrm{H},i},\alpha_{\Pi}^{\mathrm{H},i} K Π H , i , α Π H , i . The values of the fitting parameters α Π i \alpha_{\Pi}^{i} α Π i and K Π i K_{\Pi}^{i} K Π i for each component are provided in Table 1 for a = 5 a=5 a = 5 , b = 2 / 3 b=2/3 b = 2/3 and 1 1 1 , describing von Kármán turbulence and sound waves, respectively, and a range of values of α \alpha α . For all the fits, we are assuming that q q q and ε \varepsilon ε are constants, and each spectral component is divided by the prefactors in Eqs. (58), as we have done in Figure 2 . In particular, for α = 6 / 17 \alpha=6/17 α = 6/17 the values of α Π v o r t \alpha_{\Pi}^{\mathrm{vort}} α Π vort and K Π v o r t K_{\Pi}^{\mathrm{vort}} K Π vort to fit the vortical component are approximately 2 and 1.3, respectively. This fit6 has been used in refs. [35 ] [9, 18], where the value of K ~ Π ≡ K Π / K − 1 ∗ ≃ 2.2 \widetilde{K}_{\Pi}\equiv K_{\Pi}/K_{-1}^{*}\simeq2.2 K Π ≡ K Π / K − 1 ∗ ≃ 2.2 was used (see Eq. (19) in ref. [35 ] ). The helical component ζ Π h e l ( K ) \zeta_{\Pi}^{\mathrm{hel}}(K) ζ Π hel ( K ) takes negative values at K ≲ 1 K\lesssim1 K ≲ 1 and then changes sign around K ∼ 1 K\sim1 K ∼ 1 , hence we provide a fit for the envelope of its absolute value.
Figure 7. Compressional (top left panel), vortical (top middle panel), mixed (top right panel), helical (bottom left panel) components of the normalized spectra of the anisotropic stresses K 3 ∣ ζ Π c o m p ( K ) ∣ K^{3}|\zeta^{\mathrm{comp}}_{\Pi}(K)| K 3 ∣ ζ Π comp ( K ) ∣ , and vortical (bottom middle panel) and compressional (bottom right panel) components of the normalized helical spectra of the anisotropic stresses K 4 ∣ ζ Π i , H ( K ) ∣ K^{4}|\zeta^{i,\mathrm{H}}_{\Pi}(K)| K 4 ∣ ζ Π i , H ( K ) ∣ . We take a smoothed broken power law for the spectrum of the source with a = 5 a=5 a = 5 and b = 2 / 3 b=2/3 b = 2/3 , and different values of α \alpha α . Solid lines indicate the numerical values [computed from Eqs. (B14)] while dashed lines are the fits provided using Eqs. (103) and the values in Table I. For increased accuracy, a double power law fit is included in dotted lines.
Table 1. Numerical parameters of the fits in Eqs. (103) found to describe the various components of the anisotropic stresses spectrum as a broken power law for the von Kármán and sound waves spectra with a = 5 a = 5 a = 5 and b = 2 / 3 b = 2/3 b = 2/3 and 1 respectively. Figure 7 shows the numerical and the fitted ζ Π v o r t ( K ) \zeta_{\Pi}^{\mathrm{vort}}(K) ζ Π vort ( K ) and ζ Π H , v o r t ( K ) \zeta_{\Pi}^{H,\mathrm{vort}}(K) ζ Π H , vort ( K ) functions.
As mentioned in the caption of Figure 2 , the asymptotic scaling of ζ Π i ( K ) \zeta_{\Pi}^{i}(K) ζ Π i ( K ) for i = { c o m p , h e l } i=\{\mathrm{comp},\mathrm{hel}\} i = { comp , hel } at large K K K is not a power law when b c o m p = 2 b_{\mathrm{comp}}=2 b comp = 2 and b h e l = 1 b_{\mathrm{hel}}=1 b hel = 1 due to the appearance of logarithmic corrections ∝ ln K \propto\ln K ∝ ln K . The scaling at large K K K is then modified from an asymptotic power law with b Π i = 2 b + 3 b_{\Pi}^{i}=2b+3 b Π i = 2 b + 3 for smaller b b b to b Π c o m p = b + 5 b_{\Pi}^{\mathrm{comp}}=b+5 b Π comp = b + 5 and b Π h e l = b + 4 b_{\Pi}^{\mathrm{hel}}=b+4 b Π hel = b + 4 at larger b b b as we show numerically in Figure 8 . We show the compressional and helical components for different values of b = { 2 / 3 , 1 , 1.5 , 2 } b=\{2/3,1,1.5,2\} b = { 2/3 , 1 , 1.5 , 2 } .
Figure 8. Normalized anisotropic stresses compressional (upper panel) and helical (lower panel) components for different values of b b b . At large K K K , the compressional component shows a scaling K 3 Π c o m p ( K ) ∼ K − 2 b K^{3}\Pi_{\mathrm{comp}}(K) \sim K^{-2b} K 3 Π comp ( K ) ∼ K − 2 b for b < 2 b < 2 b < 2 , ∼ K − 4 ln K \sim K^{-4}\ln K ∼ K − 4 ln K for b = 2 b = 2 b = 2 , and ∼ K − b − 2 \sim K^{-b-2} ∼ K − b − 2 for b > 2 b > 2 b > 2 . The helical component scales as K 3 Π h e l ( K ) ∼ K − 2 b K^{3}\Pi_{\mathrm{hel}}(K) \sim K^{-2b} K 3 Π hel ( K ) ∼ K − 2 b for b < 1 b < 1 b < 1 , ∼ K − 2 ln K \sim K^{-2}\ln K ∼ K − 2 ln K for b = 1 b = 1 b = 1 , and ∼ K − b − 1 \sim K^{-b-1} ∼ K − b − 1 for b > 1 b > 1 b > 1 .
The fits with a single power law in Eqs. (104), although qualitatively accurate, can present large relative errors of up to order one at some values of K K K when we scan the fits over several values of a = 3 a=3 a = 3 and 5 5 5 , b = { 2 / 3 , 1 , 1.5 , 2 } b=\{2/3,1,1.5,2\} b = { 2/3 , 1 , 1.5 , 2 } and α ∈ [ 0.05 , ∞ ) \alpha\in[0.05,\infty) α ∈ [ 0.05 , ∞ ) . Extending this fit to a double broken power law, multiplied by the function 1 + ln ( 1 + K / K L ) 1+\ln(1+K/K_L) 1 + ln ( 1 + K / K L ) , allows us to add a ln K \ln K ln K correction at large K ≫ K L K\gg K_L K ≫ K L for the special cases b c o m p = 2 b_{\mathrm{comp}}=2 b comp = 2 and b h e l = 1 b_{\mathrm{hel}}=1 b hel = 1 , and reduces the relative error of the fit below 10% for all scanned parameters. The numerical computations for all components, the scan for all values of the fits, and values of the fitting parameters are included in the public Python package CosmoGW [55 , 56 ] , with a tutorial to re-produce these results that will be publicly available after the publication of this manuscript.
Constant-in-Time model for gravitational wave production
In this work, we propose that the production of gravitational waves (GWs) by slowly evolving sources is accurately described by the “Constant-in-Time” (CiT) model, proposed in ref. [1] and validated with numerical simulations of decaying MHD and hydrodynamical turbulence in refs. [50 ] [1]. This model of GW production from decaying turbulence has been used for parameter estimation with LISA and PTA in refs. [35 , 51 ] [9, 18]. In the present work, we give a detailed derivation of the CiT model of ref. [1] and argue that this model can be universally applicable to any source of GWs with evolution timescales (of the source) that are slower than the characteristic GW production time δ τ G W = 1 / k \delta\tau_{\mathrm{GW}}=1/k δ τ GW = 1/ k [51 ] [1].
When the source decays slowly, such that one can assume that it remains constant in time for an effective duration period τ ∈ [ τ ∗ , τ f i n ] \tau\in[\tau_{*},\tau_{\mathrm{fin}}] τ ∈ [ τ ∗ , τ fin ] , the GW production at a certain wave number k k k occurs on a characteristic time interval δ τ G W = τ G W − τ ∗ ≃ 1 / k \delta\tau_{\mathrm{GW}}=\tau_{\mathrm{GW}}-\tau_{*}\simeq1/k δ τ GW = τ GW − τ ∗ ≃ 1/ k , as can be inferred from the Green’s function of the wave equation [see Eq. (21)]. Therefore, if the source starts to decay at a time τ d c = τ ∗ + δ τ d c \tau_{\mathrm{dc}}=\tau_{*}+\delta\tau_{\mathrm{dc}} τ dc = τ ∗ + δ τ dc , all modes with k > 1 / δ τ f i n ∼ 1 / δ τ d c k>1/\delta\tau_{\mathrm{fin}}\sim1/\delta\tau_{\mathrm{dc}} k > 1/ δ τ fin ∼ 1/ δ τ dc have already reached saturation and just oscillate around their saturated amplitude, being almost unaffected by the decay of the source (see, e.g., the results of numerical simulations for MHD turbulence in refs. [45 , 46 , 47 , 48 , 49 ] [1]). Hence, slowly-evolving sources with an effective finite duration δ τ f i n ∼ O ( δ τ d c ) \delta\tau_{\mathrm{fin}}\sim\mathcal{O}(\delta\tau_{\mathrm{dc}}) δ τ fin ∼ O ( δ τ dc ) can be studied using the CiT model to accurately describe the GW spectrum at wave numbers k > 1 / δ τ f i n k>1/\delta\tau_{\mathrm{fin}} k > 1/ δ τ fin . Note that in this model, δ τ f i n \delta\tau_{\mathrm{fin}} δ τ fin is introduced as an effective parameter to separate the sourcing phase from the phase of GW free propagation, found to be δ τ f i n ≃ 2 δ τ d c \delta\tau_{\mathrm{fin}}\simeq2\delta\tau_{\mathrm{dc}} δ τ fin ≃ 2 δ τ dc for the decaying MHD turbulence simulations of ref. [1]. Indeed, τ f i n \tau_{\mathrm{fin}} τ fin does not really represent the time at which the source disappears, as the source is physically expected to continue to evolve at τ > τ d c \tau>\tau_{\mathrm{dc}} τ > τ dc . To take the source decay phase into account, we go beyond the CiT model in Sec. VII by introducing a coherent unequal time correlator (UETC) of the source, and we study the impact of the decay, finding modifications in the IR tail of the GW spectrum, while wave numbers around and above the peak remain well described by the CiT model for moderate decay rates and large enough decay timescales (see Figure 25 and Figure 26 ).
Figure 25. Normalized spectral shape of the GW spectrum multiplied by the efficiency Ω ~ G W c o h \widetilde{\Omega}_{\mathrm{GW}}^{\mathrm{coh}} Ω GW coh in the coherent case, for various δ τ d c / τ ∗ \delta\tau_{\mathrm{dc}}/\tau_{\ast} δ τ dc / τ ∗ and respectively m = − 1 / 3 m=-1/3 m = − 1/3 (upper panel) and m = − 11 / 9 m=-11/9 m = − 11/9 (lower panel), for a purely vortical source with α = 2 \alpha=2 α = 2 , a = 5 a=5 a = 5 , b = 2 / 3 b=2/3 b = 2/3 . We consider R ∗ H ∗ = 10 − 2 R_{\ast}H_{\ast}=10^{-2} R ∗ H ∗ = 1 0 − 2 and sourcing duration δ τ f i n / τ ∗ = 100 \delta\tau_{\mathrm{fin}}/\tau_{\ast}=100 δ τ fin / τ ∗ = 100 . For comparison, we show in black lines the result in the CiT model (see also Figure 13 ).
Figure 26. Normalized spectral shape of the GW spectrum multiplied by the efficiency Ω ~ G W c o h \widetilde{\Omega}_{\mathrm{GW}}^{\mathrm{coh}} Ω GW coh in the coherent case, for various δ τ d c / τ ∗ \delta\tau_{\mathrm{dc}}/\tau_{\ast} δ τ dc / τ ∗ and respectively m = − 7 / 5 m=-7/5 m = − 7/5 (upper panel) and m = − 13 / 7 m=-13/7 m = − 13/7 (lower panel), for a purely compressional source with α = 2 \alpha=2 α = 2 , a = 5 a=5 a = 5 , b = 1 b=1 b = 1 . We consider R ∗ H ∗ = 10 − 2 R_{\ast}H_{\ast}=10^{-2} R ∗ H ∗ = 1 0 − 2 and sourcing duration δ τ f i n / τ ∗ = 100 \delta\tau_{\mathrm{fin}}/\tau_{\ast}=100 δ τ fin / τ ∗ = 100 . For comparison, we show in black lines the result in the CiT model (see also Figure 13 ).
At later times, once the GWs enter the free-propagation regime at τ > τ f i n \tau> \tau_{\mathrm{fin}} τ > τ fin , the rescaled tensor-mode perturbations h i j h_{ij} h ij generated from τ ∗ \tau_{*} τ ∗ to τ f i n \tau_{\mathrm{fin}} τ fin just propagate proportionally to cos ( k τ ) \cos(k\tau) cos ( k τ ) and sin ( k τ ) \sin(k\tau) sin ( k τ ) as long as the Hubble friction can be neglected, i.e., for modes with k 2 ≫ a ′ ′ / a k^2 \gg a''/a k 2 ≫ a ′′ / a [see Eq. (5)]. Even when the Universe is no longer dominated by radiation, we can still safely ignore the friction as long as the relevant GW modes were produced deeply inside the radiation-dominated era.
Even though the validity of the CiT model has been previously considered for decaying MHD turbulence [35 , 50 ] [1, 18], it is a generic model that applies to all types of slowly-varying sources of GWs [51 ] . We focus on examples of such sources in Sec. VI.
In particular, in Sec. VI A we review its applicability to decaying MHD turbulence when the turbulent decay timescale, τ d c \tau_{\mathrm{dc}} τ dc , is large, and the characteristic velocity of decorrelation in time, u d c o r r u_{\mathrm{dcorr}} u dcorr , is small. We consider a Gaussian decay of the UETC in the time difference τ − = τ 2 − τ 1 \tau_{-} = \tau_{2} - \tau_{1} τ − = τ 2 − τ 1 , proportional to exp ( − 1 2 u d c o r r 2 k 2 τ − 2 ) \exp\left(-\frac{1}{2}u_{\mathrm{dcorr}}^{2}k^{2}\tau_{-}^{2}\right) exp ( − 2 1 u dcorr 2 k 2 τ − 2 ) , to describe stationary turbulence following Kraichnan decorrelation [54 ] . In reality, u d c o r r u_{\mathrm{dcorr}} u dcorr is a function of times τ 1 , τ 2 \tau_{1},\tau_{2} τ 1 , τ 2 and k k k [40 , 42 , 43 , 50 , 54 , 97 ] [5] but here we estimate it to be constant for a direct comparison with the CiT model. For small decorrelation velocities u d c o r r u_{\mathrm{dcorr}} u dcorr (i.e., long decorrelation times), the CiT model is found to accurately describe the GW spectrum, contrary to previous analytical estimates describing vortical turbulence (see, for example, refs. [40 ] [5]). However, as u d c o r r u_{\mathrm{dcorr}} u dcorr becomes large, we recover the GW spectrum found in previous theoretical work [40 ] [5], together with an additional low-k k k tail that is described by the CiT model. We then estimate how large u d c o r r u_{\mathrm{dcorr}} u dcorr is required to be to modify the GW spectrum with respect to the one found following the CiT model. The low-k k k tail is analogous to the term introduced in ref. [34 ] for sound waves, where the applicability to
Kraichnan decorrelation is also discussed (see their Sec. V B). This finding explains why the CiT model accurately describes the GW spectrum found in simulations of decaying MHD turbulence [45 , 49 , 50 ] [1], since the decorrelation velocity in these simulations is subrelativistic.
In Sec. V B, we study the applicability of the CiT model to sound waves. In this case, the UETC becomes proportional to cos ( c s k τ − ) \cos(c_{\mathrm{s}}k\tau_{-}) cos ( c s k τ − ) , where c s c_{\mathrm{s}} c s is the speed of sound and plays the role of u d c o r r u_{\mathrm{dcorr}} u dcorr in determining a decorrelation velocity. Therefore, when c s c_{\mathrm{s}} c s is relativistic (e.g., c s 2 = 1 / 3 c_{\mathrm{s}}^{2}=1/3 c s 2 = 1/3 ), an additional resonant component of the GW spectrum grows at k ∼ c s k ∗ k \sim c_{\mathrm{s}}k_{*} k ∼ c s k ∗ on top of the one predicted by the CiT model, and dominates the GW spectrum when the source duration is long c s δ τ f i n k ∗ ≫ 1 c_{\mathrm{s}}\delta\tau_{\mathrm{fin}}k_{*} \gg1 c s δ τ fin k ∗ ≫ 1 [28 , 30 , 34 , 52 ] . We show that the component that dominates the GW spectrum at small k k k , found in refs. [34 , 52 ] , is accurately described by the CiT model. This is the same low-k k k tail that we find for Kraichnan decorrelation, such that the CiT model can be used to predict this contribution for both types of sources.
We show that both the amplitude and spectral peak of the resonant component decrease as c s c_{\mathrm{s}} c s decreases (see also ref. [63 ] ) and eventually the GW spectrum from sound waves reduces to the one found in the CiT at all k k k in the c s → 0 c_{\mathrm{s}} \to0 c s → 0 limit.
These results manifest that both turbulence and sound waves are well described by the CiT model when the corresponding decorrelation velocity of the UETC is subrelativistic, u d c o r r k ∗ δ τ f i n ≪ 1 u_{\mathrm{dcorr}}k_{*}\delta\tau_{\mathrm{fin}} \ll1 u dcorr k ∗ δ τ fin ≪ 1 or c s k ∗ δ τ f i n ≪ 1 c_{\mathrm{s}}k_{*}\delta\tau_{\mathrm{fin}} \ll1 c s k ∗ δ τ fin ≪ 1 , while they both present a similar resonant growing mode around the peak as these velocities become larger, such that the UETC time variation becomes of a similar order to the time variation of the Green’s function [34 ] [1]. Since the speed of sound is relativistic in the primordial radiation plasma, even if the fluid velocities are subrelativistic, we expect sound waves to be more efficient than subrelativistic vortical turbulence in producing GWs [35 , 45 ] , even though the compressional anisotropic stresses are more suppressed than the vortical ones (see Sec. IV and ref. [48 ] ).
In Sec. VI C, we describe the potential applicability of the CiT model to sources of GWs arising from scalar fields: bubble collisions in the bulk flow model [23 , 32 , 60 ] or feebly interacting particles [98 ] . In general, we point out that a very peculiar feature of the CiT model for GW production is the cubic, Ω G W ( k ) ∼ k 3 \Omega_{\mathrm{GW}}(k) \sim k^{3} Ω GW ( k ) ∼ k 3 , and linear, Ω G W ( k ) ∼ k \Omega_{\mathrm{GW}}(k) \sim k Ω GW ( k ) ∼ k , scaling of the GW spectrum at low and intermediate frequencies, which are very commonly found in the literature. It remains to be separately explored which of these sources is accurately described by the CiT, and under which assumptions, by studying their UETCs.
Model
The GW spectrum at any time τ \tau τ is computed in Eq. (24) as the integral over τ 1 \tau_{1} τ 1 and τ 2 \tau_{2} τ 2 of the UETC of the anisotropic stresses and the product of Green’s functions. Under the assumption that the UETC of the source E Π ( τ 1 , τ 2 , k ) E_{\Pi}(\tau_{1},\tau_{2},k) E Π ( τ 1 , τ 2 , k ) can be approximated as constant (hence, the Constant-in-Time or CiT model) for an effective period of GW sourcing, τ ∈ [ τ ∗ , τ f i n ] \tau\in[\tau_{*},\tau_{\mathrm{fin}}] τ ∈ [ τ ∗ , τ fin ] , i.e., E Π ( τ 1 , τ 2 , k ) ≈ E Π ( τ ∗ , τ ∗ , k ) ≡ E Π ∗ ( k ) E_{\Pi}(\tau_{1},\tau_{2},k) \approx E_{\Pi}(\tau_{*},\tau_{*},k) \equiv E_{\Pi}^{*}(k) E Π ( τ 1 , τ 2 , k ) ≈ E Π ( τ ∗ , τ ∗ , k ) ≡ E Π ∗ ( k ) , the GW spectrum becomes [1]
Ω G W ( τ , k ) = 3 E Π ∗ ( k ) F G W 0 Δ 2 ( τ , k ) , (131) \Omega_{\mathrm{GW}}(\tau,k)=3E_{\Pi}^{*}(k)F_{\mathrm{GW}}^{0}\Delta^{2}(\tau,k), \tag*{(131)} Ω GW ( τ , k ) = 3 E Π ∗ ( k ) F GW 0 Δ 2 ( τ , k ) , (131)
where Δ 2 \Delta^{2} Δ 2 is the time kernel, corresponding to the integral over times τ 1 \tau_{1} τ 1 and τ 2 \tau_{2} τ 2 of the product of Green’s functions in Eq. (24) . Since the integral over τ 1 \tau_{1} τ 1 and τ 2 \tau_{2} τ 2 becomes separable for the CiT model, the result is expressed as the square of the kernel Δ \Delta Δ ,
Δ ( τ , k ) = ∫ τ ∗ τ ˉ d τ ′ τ ′ cos [ k ( τ − τ ′ ) ] = cos ( k τ ) Δ C i ( k τ ˉ ) + sin ( k τ ) Δ S i ( k τ ˉ ) , (132) \begin{aligned}
\Delta(\tau,k)=\int_{\tau_{*}}^{\bar{\tau}}\frac{\mathrm{d}\tau'}{\tau'}\cos[k(\tau-\tau')] \\
&=\cos(k\tau)\Delta\mathrm{Ci}(k\bar{\tau})+\sin(k\tau)\Delta\mathrm{Si}(k\bar{\tau}), \tag*{(132)}
\end{aligned} Δ ( τ , k ) = ∫ τ ∗ τ ˉ τ ′ d τ ′ cos [ k ( τ − τ ′ )] = cos ( k τ ) Δ Ci ( k τ ˉ ) + sin ( k τ ) Δ Si ( k τ ˉ ) , (132)
where τ ˉ = min ( τ , τ f i n ) \bar{\tau}=\min(\tau,\tau_{\mathrm{fin}}) τ ˉ = min ( τ , τ fin ) , and
Δ C i ( k τ ′ ) ≡ C i ( k τ ′ ) − C i ( k τ ∗ ) , (133) \Delta\mathrm{Ci}(k\tau')\equiv\mathrm{Ci}(k\tau')-\mathrm{Ci}(k\tau_{*}), \tag*{(133)} Δ Ci ( k τ ′ ) ≡ Ci ( k τ ′ ) − Ci ( k τ ∗ ) , (133)
Δ S i ( k τ ′ ) ≡ S i ( k τ ′ ) − S i ( k τ ∗ ) , (134) \Delta\mathrm{Si}(k\tau')\equiv\mathrm{Si}(k\tau')-\mathrm{Si}(k\tau_{*}), \tag*{(134)} Δ Si ( k τ ′ ) ≡ Si ( k τ ′ ) − Si ( k τ ∗ ) , (134)
being C i ( x ) \mathrm{Ci}(x) Ci ( x ) and S i ( x ) \mathrm{Si}(x) Si ( x ) the cosine and sine integral functions. This kernel corresponds to a source produced during the radiation-domination era of the Universe. It can be extended to different cosmological backgrounds using the GW solution given in Eq. (12) , i.e., substituting 1 / τ 1/\tau 1/ τ by a ( τ ) H 2 ( τ ) / H ∗ a(\tau)\mathcal{H}^{2}(\tau)/\mathcal{H}_{*} a ( τ ) H 2 ( τ ) / H ∗ , and by using the corresponding Green’s function,
Δ ( τ , k ) = H ∗ ∫ τ ∗ τ ˉ d τ ′ H 2 ( τ ′ ) H ∗ 2 a ( τ ′ ) ∂ τ G ( τ , τ ′ , k ) . (135) \Delta(\tau,k)=\mathcal{H}_{*}\int_{\tau_{*}}^{\bar{\tau}}\mathrm{d}\tau'\frac{\mathcal{H}^{2}(\tau')}{\mathcal{H}_{*}^{2}}a(\tau')\partial_{\tau}G(\tau,\tau',k). \tag*{(135)} Δ ( τ , k ) = H ∗ ∫ τ ∗ τ ˉ d τ ′ H ∗ 2 H 2 ( τ ′ ) a ( τ ′ ) ∂ τ G ( τ , τ ′ , k ) . (135)
Note that, in general, we can normalize all times with the initial time τ ∗ \tau_{*} τ ∗ , and wave numbers and the Hubble rate with 1 / τ ∗ 1/\tau_{*} 1/ τ ∗ . We recall that for radiation domination τ ∗ H ∗ = 1 \tau_{*}\mathcal{H}_{*}=1 τ ∗ H ∗ = 1 .
When the source duration is short compared to the Hubble time, the expansion of the Universe can be ignored, and the kernel in Eq. (132) becomes
Δ f l a t ( τ , k ) = 1 τ ∗ ∫ τ ∗ τ ˉ d τ ′ cos [ k ( τ − τ ′ ) ] , (136) \Delta_{\mathrm{flat}}(\tau,k)=\frac{1}{\tau_{*}}\int_{\tau_{*}}^{\bar{\tau}}\mathrm{d}\tau'\cos[k(\tau-\tau')], \tag*{(136)} Δ flat ( τ , k ) = τ ∗ 1 ∫ τ ∗ τ ˉ d τ ′ cos [ k ( τ − τ ′ )] , (136)
which has the two following branches:
Δ f l a t ( τ < τ f i n , k ) = 1 k τ ∗ sin ( k δ τ ) , (137) \Delta_{\mathrm{flat}}(\tau<\tau_{\mathrm{fin}},k)=\frac{1}{k\tau_{*}}\sin(k\delta\tau), \tag*{(137)} Δ flat ( τ < τ fin , k ) = k τ ∗ 1 sin ( k δ τ ) , (137)
Δ f l a t ( τ > τ f i n , k ) = 1 k τ ∗ ( sin ( k δ τ ) − sin [ k ( τ − τ f i n ) ] ) , (138) \Delta_{\mathrm{flat}}(\tau>\tau_{\mathrm{fin}},k)=\frac{1}{k\tau_{*}}\left(\sin(k\delta\tau)-\sin[k(\tau-\tau_{\mathrm{fin}})]\right), \tag*{(138)} Δ flat ( τ > τ fin , k ) = k τ ∗ 1 ( sin ( k δ τ ) − sin [ k ( τ − τ fin )] ) , (138)
where δ τ = τ − τ ∗ \delta\tau=\tau-\tau_{*} δ τ = τ − τ ∗ .
The dynamical evolution of the squared kernel, Δ 2 ( τ ≤ τ f i n , k ) \Delta^{2}(\tau\leq\tau_{\mathrm{fin}},k) Δ 2 ( τ ≤ τ fin , k ) , for an expanding radiation-dominated Universe is shown in Figure 9 . This spectrum oscillates in time and wave numbers, and the
Figure 9. Dynamical evolution of the kernel Δ 2 ( τ , k ) \Delta^2(\tau,k) Δ 2 ( τ , k ) for times τ ≤ τ f i n \tau\leq\tau_{\rm fin} τ ≤ τ fin as a function of k k k for different values of δ τ \delta\tau δ τ (left panel), and as a function of δ τ \delta\tau δ τ for different values of k k k (right panel). Vertical lines correspond to the wave numbers and times when saturation is reached, k δ τ G W = 1 k\delta\tau_{\rm GW}=1 k δ τ GW = 1 . The thick black line in the right panel corresponds to the envelope of the kernel as time evolves, ln 2 ( 1 + δ τ / τ ∗ ) \ln^2(1+\delta\tau/\tau_\ast) ln 2 ( 1 + δ τ / τ ∗ ) [see Eq. (139) ], until saturation is reached at δ τ G W = 1 / k \delta\tau_{\rm GW}=1/k δ τ GW = 1/ k . The thick black line in the left panel shows the envelope of Δ 2 ( τ , k ) \Delta^2(\tau,k) Δ 2 ( τ , k ) at saturation [see Eq. (140) ], as a function of k k k .
frequency of oscillations in time is proportional to k k k , while the frequency of oscillations in wave numbers is proportional to δ τ \delta\tau δ τ . The amplitude of the oscillations is studied in App. C in its asymptotic limits and it allows us to characterize the resulting GW spectrum.
We observe two regimes in the dynamical evolution of Δ 2 ( τ ≤ τ f i n , k ) \Delta^{2}(\tau\leq\tau_{\mathrm{fin}},k) Δ 2 ( τ ≤ τ fin , k ) :
i ) i) i ) k δ τ ≡ k ( τ − τ ∗ ) ≤ 1 k\delta\tau\equiv k(\tau-\tau_{*})\leq1 k δ τ ≡ k ( τ − τ ∗ ) ≤ 1 : the GW modes grow with time,
Δ 2 ( τ , k ) = ln 2 ( 1 + δ τ / τ ∗ ) , (139) \Delta^{2}(\tau,k)=\ln^{2}(1+\delta\tau/\tau_{*}), \tag*{(139)} Δ 2 ( τ , k ) = ln 2 ( 1 + δ τ / τ ∗ ) , (139)
which does not depend on k k k and grows quadratically at early times δ τ ≪ τ ∗ \delta\tau\ll\tau_{*} δ τ ≪ τ ∗ .
i i ) ii) ii ) k δ τ > 1 k\delta\tau>1 k δ τ > 1 : the GW modes have reached saturated amplitudes and oscillate in time. The envelope of Δ 2 ( τ , k ) \Delta^{2}(\tau,k) Δ 2 ( τ , k ) becomes
Δ e n v 2 ( τ , k ) = ln 2 [ 1 + ( k τ ∗ ) − 1 ] , (140) \Delta_{\mathrm{env}}^{2}(\tau,k)=\ln^{2}\left[1+(k\tau_{*})^{-1}\right], \tag*{(140)} Δ env 2 ( τ , k ) = ln 2 [ 1 + ( k τ ∗ ) − 1 ] , (140)
which is constant in time and decreases with k k k , quadratically in the limit k ≫ 1 / τ ∗ k\gg1/\tau_{*} k ≫ 1/ τ ∗ .
Hence, at each mode k k k , we observe a growing phase of the GW spectrum given by ln 2 ( 1 + δ τ / τ ∗ ) \ln^{2}(1+\delta\tau/\tau_{*}) ln 2 ( 1 + δ τ / τ ∗ ) for a duration δ τ G W = 1 / k \delta\tau_{\mathrm{GW}}=1/k δ τ GW = 1/ k , i.e., until the spectral mode reaches its saturation value: ln 2 [ 1 + ( k τ ∗ ) − 1 ] \ln^{2}[1+(k\tau_{*})^{-1}] ln 2 [ 1 + ( k τ ∗ ) − 1 ] (regime i i i ). At later times, but still τ < τ f i n \tau<\tau_{\mathrm{fin}} τ < τ fin , the kernel function Δ 2 \Delta^{2} Δ 2 oscillates around its saturated amplitude (regime i i ii ii ). Omitting the expansion of the Universe, these functions become ( δ τ / τ ∗ ) − 2 (\delta\tau/\tau_{*})^{-2} ( δ τ / τ ∗ ) − 2 and ( k τ ∗ ) − 2 (k\tau_{*})^{-2} ( k τ ∗ ) − 2 .
The decay of the kernel proportional to k − 2 k^{-2} k − 2 in the k → ∞ k\to\infty k → ∞ limit is a consequence of the discontinuity introduced in the sourcing, assuming that it starts instantaneously at τ ∗ \tau_{*} τ ∗ [41 ] . In reality, we expect that the GW spectrum is modified by the dynamics during the period of sourcing, Δ τ s \Delta\tau_{\mathrm{s}} Δ τ s , at wave numbers k ≳ 1 / Δ τ s k\gtrsim1/\Delta\tau_{\mathrm{s}} k ≳ 1/Δ τ s , and eventually decays exponentially if the sourcing process is smooth. Therefore, the model we present is valid for k ≪ 1 / Δ τ s k\ll1/\Delta\tau_{\mathrm{s}} k ≪ 1/Δ τ s , and it needs to include the dynamics of the sourcing process at k ≳ 1 / Δ τ s k\gtrsim1/\Delta\tau_{\mathrm{s}} k ≳ 1/Δ τ s .
To show this explicitly, let us rewrite Eq. (132) using the function g ( τ ) = Θ ( τ − τ ∗ ) − Θ ( τ − τ f i n ) g(\tau)=\Theta(\tau-\tau_{*})-\Theta(\tau-\tau_{\mathrm{fin}}) g ( τ ) = Θ ( τ − τ ∗ ) − Θ ( τ − τ fin ) that takes into account the switching on and off of the source at τ ∗ \tau_{*} τ ∗ and τ f i n \tau_{\mathrm{fin}} τ fin ,
Δ ( τ , k ) = ∫ − ∞ τ d τ ′ τ ′ cos [ k ( τ − τ ′ ) ] g ( τ ′ ) . (141) \Delta(\tau,k)=\int_{-\infty}^{\tau}\frac{\mathrm{d}\tau'}{\tau'}\cos[k(\tau-\tau')]g(\tau'). \tag*{(141)} Δ ( τ , k ) = ∫ − ∞ τ τ ′ d τ ′ cos [ k ( τ − τ ′ )] g ( τ ′ ) . (141)
Integrating by parts, we find
Δ ( τ , k ) = ∑ τ s = τ ∗ , τ ˉ Δ s ( g ( τ ′ ) k τ ′ sin [ k ( τ − τ ′ ) ] ) + ∫ τ ∗ τ ˉ [ g ( τ ′ ) / τ ′ ] ′ k sin [ k ( τ − τ ′ ) ] d τ ′ (142) \Delta(\tau,k)=\sum_{\tau_s=\tau_\ast,\bar{\tau}}\Delta_s\left(\frac{g(\tau')}{k\tau'}\sin[k(\tau-\tau')]\right)+\int_{\tau_\ast}^{\bar{\tau}}\frac{[g(\tau')/\tau']'}{k}\sin[k(\tau-\tau')]\,\mathrm{d}\tau' \tag*{(142)} Δ ( τ , k ) = τ s = τ ∗ , τ ˉ ∑ Δ s ( k τ ′ g ( τ ′ ) sin [ k ( τ − τ ′ )] ) + ∫ τ ∗ τ ˉ k [ g ( τ ′ ) / τ ′ ] ′ sin [ k ( τ − τ ′ )] d τ ′ (142)
where Δ s \Delta_s Δ s corresponds to the jump of the function g ( τ ′ ) / ( k τ ′ ) sin [ k ( τ − τ ′ ) ] g(\tau')/(k\tau')\sin[k(\tau-\tau')] g ( τ ′ ) / ( k τ ′ ) sin [ k ( τ − τ ′ )] at τ s = τ ∗ , τ ˉ \tau_s=\tau_\ast,\bar{\tau} τ s = τ ∗ , τ ˉ . Since g ( τ ∗ + ) = g ( τ ∗ − ) = 1 g(\tau_\ast^+)=g(\tau_\ast^-)=1 g ( τ ∗ + ) = g ( τ ∗ − ) = 1 , g ( τ ˉ − ) = 0 g(\bar{\tau}^-)=0 g ( τ ˉ − ) = 0 , and g ( τ ˉ + ) = 1 − Θ ( τ ˉ − τ f i n ) g(\bar{\tau}^+)=1-\Theta(\bar{\tau}-\tau_{\rm fin}) g ( τ ˉ + ) = 1 − Θ ( τ ˉ − τ fin ) , the leading-order term of Δ \Delta Δ in the k → ∞ k\to\infty k → ∞ limit becomes
lim k → ∞ Δ ( τ , k ) = 1 k τ ∗ sin ( k δ τ ) − 1 k τ ˉ sin [ k ( τ − τ ˉ ) ] + O ( k − 2 ) . (143) \lim_{k\to\infty}\Delta(\tau,k)=\frac{1}{k\tau_\ast}\sin(k\delta\tau)-\frac{1}{k\bar{\tau}}\sin[k(\tau-\bar{\tau})]+\mathcal{O}(k^{-2}). \tag*{(143)} k → ∞ lim Δ ( τ , k ) = k τ ∗ 1 sin ( k δ τ ) − k τ ˉ 1 sin [ k ( τ − τ ˉ )] + O ( k − 2 ) . (143)
Notice that the second term only starts contributing to the envelope of Δ \Delta Δ at τ ≥ τ f i n \tau\geq\tau_{\rm fin} τ ≥ τ fin . Equation (143) shows that Δ ∼ k − 1 \Delta\sim k^{-1} Δ ∼ k − 1 in the k → ∞ k\to\infty k → ∞ limit and allows us to evaluate its amplitude in this limit.
If, on the other hand, we consider a C ∞ C^\infty C ∞ function g ( τ ) g(\tau) g ( τ ) describing a smooth sourcing process instead of an instantaneous one, and we assume for simplicity that the UETC of the source is a separable function proportional to g ( τ 1 ) g ( τ 2 ) g(\tau_1)g(\tau_2) g ( τ 1 ) g ( τ 2 ) , the kernel Δ \Delta Δ can still be expressed as in Eq. (141) . For example, let us consider the following g g g function,
g ( δ τ ) = 1 2 [ tanh ( 2 δ τ Δ τ s ) + 1 ] , (144) g(\delta\tau)=\frac{1}{2}\left[\tanh\left(\frac{2\delta\tau}{\Delta\tau_s}\right)+1\right], \tag*{(144)} g ( δ τ ) = 2 1 [ tanh ( Δ τ s 2 δ τ ) + 1 ] , (144)
where Δ τ s \Delta\tau_s Δ τ s determines the interval of time over which the sourcing function g g g changes from 0 to 1, which is approximately 2 Δ τ s 2\Delta\tau_s 2Δ τ s . Iteratively applying the same procedure as the one that led to Eq. (142) , and taking into account that g g g is C ∞ C^\infty C ∞ , one would find that Δ ( τ , k ) \Delta(\tau,k) Δ ( τ , k ) goes to zero faster than any power law when k → ∞ k\to\infty k → ∞ . We show the resulting kernel Δ 2 \Delta^2 Δ 2 in Figure 10 , compared to the one found using Eq. (132) , which assumes instantaneous growth of the source. Indeed, both follow the same dependence with k k k at all k < 1 / Δ τ s k < 1/\Delta\tau_s k < 1/Δ τ s .
Figure 10. Kernel Δ 2 ( τ , k ) \Delta^2(\tau,k) Δ 2 ( τ , k ) found in the CiT model when the source is assumed to start instantaneously at τ ∗ \tau_\ast τ ∗ (blue) compared to the one found when the growth is determined by a C ∞ C^\infty C ∞ function with a width 2 Δ τ s 2\Delta\tau_s 2Δ τ s around τ ∗ \tau_\ast τ ∗ (red). The kernels are evaluated at δ τ = 10 τ ∗ \delta\tau=10\tau_\ast δ τ = 10 τ ∗ and Δ τ s = 0.2 τ ∗ \Delta\tau_s=0.2\tau_\ast Δ τ s = 0.2 τ ∗ .
On the other hand, at times after τ f i n \tau_{\mathrm{fin}} τ fin , τ > τ f i n \tau> \tau_{\mathrm{fin}} τ > τ fin , the dynamical behavior of Δ 2 ( τ , k ) \Delta^2(\tau,k) Δ 2 ( τ , k ) presents a modification at large k k k , as it can already be seen in Eq. (143) , as a consequence of the discontinuity of the source at τ f i n \tau_{\mathrm{fin}} τ fin . The upper panel of Figure 11 shows the kernel Δ 2 \Delta^2 Δ 2 for δ τ f i n = τ ∗ \delta\tau_{\mathrm{fin}}=\tau_* δ τ fin = τ ∗ as a function of time δ τ = τ − τ ∗ \delta\tau=\tau-\tau_* δ τ = τ − τ ∗ at different wave numbers, while the lower panel shows the kernel evaluated at a late time, δ τ = 10 3 τ ∗ \delta\tau=10^3\tau_* δ τ = 1 0 3 τ ∗ , obtained if the sourcing is active for finite durations δ τ f i n = { 10 − 2 , 10 − 1 , 1 , 10 , 10 2 } τ ∗ < δ τ \delta\tau_{\mathrm{fin}}=\{10^{-2},10^{-1},1,10,10^2\}\tau_*<\delta\tau δ τ fin = { 1 0 − 2 , 1 0 − 1 , 1 , 10 , 1 0 2 } τ ∗ < δ τ . After τ \tau τ reaches τ f i n \tau_{\mathrm{fin}} τ fin , we find two additional regimes, depending on whether the modes have already saturated by this time, i.e., k > 1 / δ τ f i n k>1/\delta\tau_{\mathrm{fin}} k > 1/ δ τ fin , or not:
Figure 11. Upper panel: Kernel Δ 2 ( τ , k ) \Delta^2(\tau,k) Δ 2 ( τ , k ) evaluated for different values of k k k as a function of τ \tau τ for δ τ f i n = τ ∗ \delta\tau_{\mathrm{fin}}=\tau_* δ τ fin = τ ∗ . The vertical and horizontal lines are the same as in the right panel of Figure 9 for comparison. Lower panel: Kernel Δ 2 ( τ , k ) \Delta^2(\tau,k) Δ 2 ( τ , k ) evaluated using different source durations δ τ f i n / τ ∗ = { 10 − 2 , 10 − 1 , 1 , 10 , 10 2 } \delta\tau_{\mathrm{fin}}/\tau_*=\{10^{-2},10^{-1},1,10,10^2\} δ τ fin / τ ∗ = { 1 0 − 2 , 1 0 − 1 , 1 , 10 , 1 0 2 } at a time δ τ / τ ∗ = 10 3 ≫ τ f i n / τ ∗ \delta\tau/\tau_*=10^3\gg\tau_{\mathrm{fin}}/\tau_* δ τ / τ ∗ = 1 0 3 ≫ τ fin / τ ∗ . The vertical line indicates k = 1 / δ τ k=1/\delta\tau k = 1/ δ τ .
Δ e n v 2 ( τ , k ) = ln 2 ( 1 + δ τ f i n / τ ∗ ) . (145) \Delta^2_{\mathrm{env}}(\tau,k)=\ln^2(1+\delta\tau_{\mathrm{fin}}/\tau_*). \tag*{(145)} Δ env 2 ( τ , k ) = ln 2 ( 1 + δ τ fin / τ ∗ ) . (145)
Their amplitude remains constant. After they have been processed by the Green’s function at δ τ ≳ δ τ G W = 1 / k \delta\tau\gtrsim\delta\tau_{\mathrm{GW}}=1/k δ τ ≳ δ τ GW = 1/ k , these modes oscillate, keeping the same envelope amplitude.
iv) k δ τ f i n > 1 k\delta\tau_{\mathrm{fin}}>1 k δ τ fin > 1 : the GW modes had already saturated before τ f i n \tau_{\mathrm{fin}} τ fin is reached and were in the oscillatory regime ii) with an amplitude ln 2 [ 1 + ( k τ ∗ ) − 1 ] \ln^2[1+(k\tau_*)^{-1}] ln 2 [ 1 + ( k τ ∗ ) − 1 ] . At τ f i n \tau_{\mathrm{fin}} τ fin , the amplitude of the kernel Δ 2 \Delta^2 Δ 2 presents a discontinuity that depends on the wave number and the values of τ ∗ \tau_* τ ∗ and τ f i n \tau_{\mathrm{fin}} τ fin (see Eq. (143) and Figure 11 ), and then the function continues to oscillate in time with a saturated amplitude.
In the regime k ≥ 1 / δ τ f i n k\geq1/\delta\tau_{\mathrm{fin}} k ≥ 1/ δ τ fin , the spectrum shows the characteristic ln 2 [ 1 + 1 / ( k τ ∗ ) ] \ln^2[1+1/(k\tau_*)] ln 2 [ 1 + 1/ ( k τ ∗ )] that corresponds to all the saturated values reached at earlier times τ ≤ τ f i n \tau\leq\tau_{\mathrm{fin}} τ ≤ τ fin [1]. However, the abrupt switch off of the source enhances the envelope amplitude of Δ 2 \Delta^2 Δ 2 at τ ≥ τ f i n \tau\geq\tau_{\mathrm{fin}} τ ≥ τ fin and k > 1 / δ τ f i n k>1/\delta\tau_{\mathrm{fin}} k > 1/ δ τ fin , as we have explicitly shown in Eq. (143) . The exact modification of the spectrum depends on the specific values of k k k , τ \tau τ , and τ f i n \tau_{\mathrm{fin}} τ fin , with an upper bound over k k k (envelope) determined by the factor F ( τ f i n ) \mathcal{F}(\tau_{\mathrm{fin}}) F ( τ fin ) , computed in App. C [see, in particular, Eq. (328) ],
F ( τ f i n ) = 1 + τ ∗ τ f i n ∈ ( 1 , 2 ] . (146) \mathcal{F}(\tau_{\mathrm{fin}})=1+\frac{\tau_*}{\tau_{\mathrm{fin}}}\in(1,2]. \tag*{(146)} F ( τ fin ) = 1 + τ fin τ ∗ ∈ ( 1 , 2 ] . (146)
This factor is equivalent to the envelope of ∣ Δ ∣ |\Delta| ∣Δ∣ found in the k → ∞ k\to\infty k → ∞ in Eq. (143) , and asymptotically approaches one as τ f i n \tau_{\mathrm{fin}} τ fin increases, so it is mostly relevant for short source durations τ f i n \tau_{\mathrm{fin}} τ fin . Indeed, we see in Figure 11 (upper panel) a jump on Δ 2 \Delta^2 Δ 2 that can either be positive or negative depending on k k k at δ τ f i n \delta\tau_{\mathrm{fin}} δ τ fin , while we observe a resulting enhancement of the envelope amplitude in the lower panel at k > 1 / δ τ f i n k>1/\delta\tau_{\mathrm{fin}} k > 1/ δ τ fin , whose amplitude decreases as δ τ f i n \delta\tau_{\mathrm{fin}} δ τ fin increases. This factor becomes a constant in the limit of small source durations δ τ f i n ≪ τ ∗ \delta\tau_{\mathrm{fin}}\ll\tau_* δ τ fin ≪ τ ∗ . In this limit, F \mathcal{F} F takes its maximum value, F ( τ ∗ ) = 2 \mathcal{F}(\tau_*)=2 F ( τ ∗ ) = 2 . Therefore, this modification becomes maximal and independent of τ f i n \tau_{\mathrm{fin}} τ fin in flat space-time (i.e., when one omits the expansion of the Universe).
Taking into account the description discussed above for the CiT model, the envelope of the GW spectrum at times τ ≥ τ f i n \tau\geq\tau_{\mathrm{fin}} τ ≥ τ fin can be used to model the
GW spectrum at present time [35 ] [1, 9, 18]
Ω G W e n v ( δ τ f i n , k ) = 3 E Π ∗ ( k ) F G W 0 T G W ( δ τ f i n , k ) , (147) \Omega_{\mathrm{GW}}^{\mathrm{env}}(\delta\tau_{\mathrm{fin}},k)=3E_{\Pi}^{*}(k)F_{\mathrm{GW}}^{0}\mathcal{T}_{\mathrm{GW}}(\delta\tau_{\mathrm{fin}},k), \tag*{(147)} Ω GW env ( δ τ fin , k ) = 3 E Π ∗ ( k ) F GW 0 T GW ( δ τ fin , k ) , (147)
where the function T G W \mathcal{T}_{\mathrm{GW}} T GW , introduced in ref. [1], is
T G W ( δ τ f i n , k ) = { ln 2 ( 1 + δ τ f i n / τ ∗ ) , if k δ τ f i n ∗ < 1 , F 2 ln 2 [ 1 + ( k τ ∗ ) − 1 ] , if k δ τ f i n ∗ ≥ 1 , (148) \mathcal{T}_{\mathrm{GW}}(\delta\tau_{\mathrm{fin}},k)=
\begin{cases}
\ln^{2}(1+\delta\tau_{\mathrm{fin}}/\tau_{*}), & \text{if } k\delta\tau_{\mathrm{fin}}^{*}<1,\\
\mathcal{F}^{2}\ln^{2}[1+(k\tau_{*})^{-1}], & \text{if } k\delta\tau_{\mathrm{fin}}^{*}\ge1,
\end{cases} \tag*{(148)} T GW ( δ τ fin , k ) = { ln 2 ( 1 + δ τ fin / τ ∗ ) , F 2 ln 2 [ 1 + ( k τ ∗ ) − 1 ] , if k δ τ fin ∗ < 1 , if k δ τ fin ∗ ≥ 1 , (148)
with F \mathcal{F} F corresponding to the enhancement found in Eq. (146) . The separation between branches corresponds to the source duration δ τ f i n \delta\tau_{\mathrm{fin}} δ τ fin , modified to ensure continuity of the envelope T G W \mathcal{T}_{\mathrm{GW}} T GW [1],
δ τ f i n ∗ τ ∗ = ( 1 + δ τ f i n τ ∗ ) 1 / F − 1. (149) \frac{\delta\tau_{\mathrm{fin}}^{*}}{\tau_{*}}=\left(1+\frac{\delta\tau_{\mathrm{fin}}}{\tau_{*}}\right)^{1/\mathcal{F}}-1. \tag*{(149)} τ ∗ δ τ fin ∗ = ( 1 + τ ∗ δ τ fin ) 1/ F − 1. (149)
Its position reduces to δ τ f i n \delta\tau_{\mathrm{fin}} δ τ fin when F → 1 \mathcal{F}\to1 F → 1 . Hence, if the duration of the source is short and it is abruptly switched off, the envelope of the GW spectrum presents an enhancement by a factor F 2 \mathcal{F}^{2} F 2 at large k ≥ 1 / δ τ f i n ∗ k\ge1/\delta\tau_{\mathrm{fin}}^{*} k ≥ 1/ δ τ fin ∗ . In the limit of short durations δ τ f i n ≪ τ ∗ \delta\tau_{\mathrm{fin}}\ll\tau_{*} δ τ fin ≪ τ ∗ , we have F → 2 \mathcal{F}\to2 F → 2 and δ τ f i n ∗ → 1 2 δ τ f i n \delta\tau_{\mathrm{fin}}^{*}\to\frac{1}{2}\delta\tau_{\mathrm{fin}} δ τ fin ∗ → 2 1 δ τ fin .
At present time, the oscillations over k k k with frequency proportional to τ 0 \tau_{0} τ 0 can be averaged out at all k τ 0 ≫ 1 k\tau_{0}\gg1 k τ 0 ≫ 1 , as done to obtain the GW spectrum in Eq. (26) . Following the CiT model, the time kernel found when these fast oscillations are averaged out becomes
Δ a v 2 ( τ f i n , k ) = 1 2 [ Δ C i 2 ( k τ f i n ) + Δ S i 2 ( k τ f i n ) ] , (150) \Delta_{\mathrm{av}}^{2}(\tau_{\mathrm{fin}},k)=\frac{1}{2}\left[\Delta\mathrm{Ci}^{2}(k\tau_{\mathrm{fin}})+\Delta\mathrm{Si}^{2}(k\tau_{\mathrm{fin}})\right], \tag*{(150)} Δ av 2 ( τ fin , k ) = 2 1 [ Δ Ci 2 ( k τ fin ) + Δ Si 2 ( k τ fin ) ] , (150)
which is shown in Figure 12 . We note that the resulting kernel still presents some relic oscillations in k k k with frequency proportional to τ f i n \tau_{\mathrm{fin}} τ fin . The envelope of Eq. (150) is 1 2 T G W ( δ τ f i n , k ) \frac{1}{2}\mathcal{T}_{\mathrm{GW}}(\delta\tau_{\mathrm{fin}},k) 2 1 T GW ( δ τ fin , k ) , justifying the 1 2 \frac{1}{2} 2 1 factor introduced in refs. [35 ] [9, 18]. In flat space-time, the kernel averaged over fast oscillations becomes
Figure 12. Kernel Δ a v 2 \Delta_{\mathrm{av}}^{2} Δ av 2 averaged over fast frequency oscillations at late times τ ≫ τ f i n \tau\gg\tau_{\mathrm{fin}} τ ≫ τ fin , given in Eq. (150) for an expanding Universe, for different source durations δ τ f i n \delta\tau_{\mathrm{fin}} δ τ fin .
Δ a v , f l a t 2 ( τ f i n , k ) = 1 ( k τ ∗ ) 2 [ 1 − cos ( k δ τ f i n ) ] , (151) \Delta_{\mathrm{av,flat}}^{2}(\tau_{\mathrm{fin}},k)=\frac{1}{(k\tau_{*})^{2}}\left[1-\cos(k\delta\tau_{\mathrm{fin}})\right], \tag*{(151)} Δ av , flat 2 ( τ fin , k ) = ( k τ ∗ ) 2 1 [ 1 − cos ( k δ τ fin ) ] , (151)
whose envelope is also 1 2 T G W ( δ τ f i n , k ) \frac{1}{2}\mathcal{T}_{\mathrm{GW}}(\delta\tau_{\mathrm{fin}},k) 2 1 T GW ( δ τ fin , k ) taking the short-duration limit of the ln 2 \ln^{2} ln 2 functions, i.e.,
T G W , f l a t ( δ τ f i n , k ) = { ( δ τ f i n / τ ∗ ) 2 , if k δ τ f i n < 2 , 4 ( k τ ∗ ) − 2 , if k δ τ f i n ≥ 2. (152) \mathcal{T}_{\mathrm{GW,flat}}(\delta\tau_{\mathrm{fin}},k)=
\begin{cases}
(\delta\tau_{\mathrm{fin}}/\tau_{*})^{2}, & \text{if } k\delta\tau_{\mathrm{fin}}<2,\\
4(k\tau_{*})^{-2}, & \text{if } k\delta\tau_{\mathrm{fin}}\ge2.
\end{cases} \tag*{(152)} T GW , flat ( δ τ fin , k ) = { ( δ τ fin / τ ∗ ) 2 , 4 ( k τ ∗ ) − 2 , if k δ τ fin < 2 , if k δ τ fin ≥ 2. (152)
Equations (150) and (151) explicitly indicate that if one applies the procedure described in Sec. II C to compute the GW spectrum, where the GWs are already averaged over fast frequency oscillations at late times, it is crucial to make sure that the simulations are covering all times up to τ f i n \tau_{\mathrm{fin}} τ fin , i.e., to ensure that each mode k k k has reached saturation, to recover the correct results (see also discussion in App. A of ref. [33 ] ). Otherwise, the final GW spectrum is biased by the particular choice of the final time of the simulation.
In ref. [1], this model was used and compared with numerical simulations of decaying MHD turbulence. The numerical results were accurate at k δ τ d c ≳ 1 k\delta\tau_{\mathrm{dc}}\gtrsim1 k δ τ dc ≳ 1 with the model described above, with the exception of the factor F \mathcal{F} F , which does not appear in the numerical results. Since this is a consequence of the abrupt switching off of the source, which is not realistic for decaying turbulence (or any other time-decaying source), we introduce a coherent UETC in Sec. VII that allows us to describe the time decay of the source at late times. We find that the GW spectrum at wave numbers that have reached saturation before the source starts to decay, k δ τ d c ≳ 1 k\delta\tau_{\mathrm{dc}}\gtrsim1 k δ τ dc ≳ 1 , remains unmodified and described by the CiT model of Eq. (147) with F → 1 \mathcal{F}\to1 F → 1 , while smaller wave numbers are affected by the decay.
Gravitational wave template
Introducing the normalization of E Π ∗ ( k ) E_{\Pi}^{*}(k) E Π ∗ ( k ) described in Sec. IV A [see Eq. (86) ] into Eq. (131) , the GW spectrum at present time, averaged over fast frequency oscillations, becomes [35 ] [1, 9, 18]
Ω G W 0 ( K ) = 3 2 ρ ~ r m s 4 ρ ˉ 2 C A 2 F G W 0 K 3 S Π ∗ ( K ) T G W , (153) \Omega_{\mathrm{GW}}^{0}(K)=\frac{3}{2}\frac{\tilde{\rho}_{\mathrm{rms}}^{4}}{\bar{\rho}^{2}}\frac{\mathcal{C}}{\mathcal{A}^{2}}F_{\mathrm{GW}}^{0}K^{3}S_{\Pi}^{*}(K)\mathcal{T}_{\mathrm{GW}}, \tag*{(153)} Ω GW 0 ( K ) = 2 3 ρ ˉ 2 ρ ~ rms 4 A 2 C F GW 0 K 3 S Π ∗ ( K ) T GW , (153)
where K = k / k ∗ = k R ∗ K = k/k_* = k\mathcal{R}_* K = k / k ∗ = k R ∗ , ϕ ~ r m s 2 = { ρ ˉ k i n , 2 ρ ˉ B , 2 ρ ˉ G ϕ } \tilde{\phi}_{\rm rms}^2 = \{\bar{\rho}_{\rm kin}, 2\bar{\rho}_B, 2\bar{\rho}_G^\phi\} ϕ ~ rms 2 = { ρ ˉ kin , 2 ρ ˉ B , 2 ρ ˉ G ϕ } is the mean squared amplitude of the different type of sources considered ϕ ~ i = { X i , i B i / a , − i k i ϕ / a } \tilde{\phi}_i = \{X_i, iB_i/a, -ik_i\phi/a\} ϕ ~ i = { X i , i B i / a , − i k i ϕ / a } . Note that X r m s 2 / ρ ˉ ≃ Γ u r m s 2 X_{\rm rms}^2/\bar{\rho} \simeq\Gamma u_{\rm rms}^2 X rms 2 / ρ ˉ ≃ Γ u rms 2 in the limit of small perturbations with Γ = w ˉ / ρ ˉ \Gamma= \bar{w}/\bar{\rho} Γ = w ˉ / ρ ˉ . The envelope of the time kernel is described by the function T G W T_{\rm GW} T GW in Eq. (148) and the average over fast oscillations is taken care of by the extra 1 / 2 1/2 1/2 factor with respect to Eq. (131) [see Eq. (122)]. A main assumption of this model is that ϕ ~ r m s \tilde{\phi}_{\rm rms} ϕ ~ rms remains approximately constant over the sourcing period. If the dynamics of the source are slow, this assumption is always justified in flat space-time. Taking into account the expansion of the Universe, this assumption can be justified over cosmological times for radiation-like sources that dynamically evolve slowly over time, such that the ratio ϕ ~ r m s 2 / ρ ˉ ≈ const \tilde{\phi}_{\rm rms}^2/\bar{\rho} \approx\operatorname{const} ϕ ~ rms 2 / ρ ˉ ≈ const . For fluid or scalar fields with generic equation of state w w w , this ratio will be proportional to ρ G ϕ / ρ ˉ ∼ a 1 − 3 w \rho_G^\phi/\bar{\rho} \sim a^{1-3w} ρ G ϕ / ρ ˉ ∼ a 1 − 3 w for a radiation-dominated Universe and, hence, our model is generally applicable if the sourcing period is much shorter than one Hubble time. On the other hand, if the source of GWs corresponds to the dominant contribution to the energy density of the Universe, then both numerator and denominator scale with the same power law of a a a , and the ratio can be taken to be approximately constant.
As expected, whether choosing the normalization for the spectrum of the source, ζ ( K ) \zeta(K) ζ ( K ) , following the convention in this work, or following instead the convention of previous related work, ζ − 1 ( K ~ ) \zeta^{-1}(\tilde{K}) ζ − 1 ( K ~ ) [see discussion around Eq. (247) ], does not affect the resulting GW spectrum, which becomes
Ω G W 0 ( K ~ ) = 3 2 ϕ ~ r m s 4 ρ ˉ 2 C ~ A ~ 2 F G W 0 K ~ 3 ζ ~ Π ( K ~ ) T G W . (154) \Omega_{\rm GW}^0(\tilde{K}) = \frac{3}{2}\frac{\tilde{\phi}_{\rm rms}^4}{\bar{\rho}^2}\frac{\tilde{C}}{\tilde{A}^2}F_{\rm GW}^0\tilde{K}^3\tilde{\zeta}_{\Pi}(\tilde{K})T_{\rm GW}. \tag*{(154)} Ω GW 0 ( K ~ ) = 2 3 ρ ˉ 2 ϕ ~ rms 4 A ~ 2 C ~ F GW 0 K ~ 3 ζ ~ Π ( K ~ ) T GW . (154)
Indeed, using Eqs. (252) and (290) , and K ~ = K / K ∗ − 1 \tilde{K} = K/K_{*-1} K ~ = K / K ∗− 1 , one finds the equivalence between the normalization factors,
C ~ A ~ 2 K ~ 3 = C A 2 K 3 . (155) \frac{\tilde{C}}{\tilde{A}^2}\tilde{K}^3 = \frac{C}{A^2}K^3. \tag*{(155)} A ~ 2 C ~ K ~ 3 = A 2 C K 3 . (155)
The k k k dependence of the GW background in Eq. (153) has been previously described in ref. [1]. In summary, it presents:
Ω G W 0 ( K ) ∼ K 3 ln 2 [ 1 + ( R ∗ H ∗ ) / K ] , (156) \Omega_{\rm GW}^0(K) \sim K^3\ln^2\left[1+(\mathcal{R}_*\mathcal{H}_*)/K\right], \tag*{(156)} Ω GW 0 ( K ) ∼ K 3 ln 2 [ 1 + ( R ∗ H ∗ ) / K ] , (156)
in the range R ∗ / δ τ f i n ∗ < K < K G W \mathcal{R}_*/\delta\tau^*_{\rm fin} < K < K_{\rm GW} R ∗ / δ τ fin ∗ < K < K GW , where we have used 1 / τ ∗ = H ∗ 1/\tau_* = \mathcal{H}_* 1/ τ ∗ = H ∗ , and defined the comoving length scale R ∗ ≡ 1 / k ∗ \mathcal{R}_* \equiv1/k_* R ∗ ≡ 1/ k ∗ . K G W K_{\rm GW} K GW corresponds to the position of the GW spectral peak as we discuss below (see also discussion in ref. [1]). For subhorizon modes with k τ ∗ = K / ( R ∗ H ∗ ) ≫ 1 k\tau_* = K/(\mathcal{R}_*\mathcal{H}_*) \gg1 k τ ∗ = K / ( R ∗ H ∗ ) ≫ 1 , the logarithmic branch becomes the one in flat space-time, proportional to
lim k τ ∗ ≫ 1 Ω G W 0 ( K ) ∼ K ( R ∗ H ∗ ) 2 ; (157) \lim_{k\tau_*\gg1}\Omega_{\rm GW}^0(K) \sim K(\mathcal{R}_*\mathcal{H}_*)^2; \tag*{(157)} k τ ∗ ≫ 1 lim Ω GW 0 ( K ) ∼ K ( R ∗ H ∗ ) 2 ; (157)
Ω G W 0 ( K ) ∼ K 3 − b Π ln 2 [ 1 + ( R ∗ H ∗ ) / K ] , (158) \Omega_{\rm GW}^0(K) \sim K^{3-b_\Pi}\ln^2\left[1+(\mathcal{R}_*\mathcal{H}_*)/K\right], \tag*{(158)} Ω GW 0 ( K ) ∼ K 3 − b Π ln 2 [ 1 + ( R ∗ H ∗ ) / K ] , (158)
when K ≳ K G W K \gtrsim K_{\rm GW} K ≳ K GW , due to the decay of the normalized anisotropic stresses ζ Π ∗ ( K ) \zeta_\Pi^*(K) ζ Π ∗ ( K ) [see Eqs. (103)]. In Eq. (158) , − b Π -b_\Pi − b Π is the small-scale slope of ζ Π ( K ) \zeta_\Pi(K) ζ Π ( K ) , e.g., b Π = b + 3 b_\Pi= b+3 b Π = b + 3 for a purely vortical vector field (see Eqs. (104) for the slopes associated to the various components). As we generally consider causal sources with R ∗ H ∗ ≪ 1 \mathcal{R}_*\mathcal{H}_* \ll1 R ∗ H ∗ ≪ 1 , and K G W ≳ 1 K_{\rm GW} \gtrsim1 K GW ≳ 1 , then this branch satisfies K G W / ( R ∗ H ∗ ) ≫ 1 K_{\rm GW}/(\mathcal{R}_*\mathcal{H}_*) \gg1 K GW / ( R ∗ H ∗ ) ≫ 1 , and the GW spectrum becomes proportional to
lim k τ ∗ ≫ 1 Ω G W 0 ( K ) ∼ K 1 − b Π ( R ∗ H ∗ ) 2 . (159) \lim_{k\tau_*\gg1}\Omega_{\rm GW}^0(K) \sim K^{1-b_\Pi}(\mathcal{R}_*\mathcal{H}_*)^2. \tag*{(159)} k τ ∗ ≫ 1 lim Ω GW 0 ( K ) ∼ K 1 − b Π ( R ∗ H ∗ ) 2 . (159)
We are omitting the presence of an exponential decrease of the spectrum of the source below a certain dissipative scale, i.e., at K > K D K > K_D K > K D , for example due to viscosity, diffusivity, or shocks (see, e.g., refs. [25 , 42 , 51 , 82 ] [5]) and the dynamics of the production of the source over a period of time 2 Δ τ s 2\Delta\tau_s 2Δ τ s , which modifies the kernel at k > 1 / Δ τ s k > 1/\Delta\tau_s k > 1/Δ τ s (see Figure 10 ). In both cases, the GW spectrum presents an exponential decay at K > K e x p = min ( 1 / ( k ∗ Δ τ s ) , K D ) K > K_{\rm exp} = \min(1/(k_*\Delta\tau_s), K_D) K > K exp = min ( 1/ ( k ∗ Δ τ s ) , K D ) . As long as K e x p ≫ K G W K_{\rm exp} \gg K_{\rm GW} K exp ≫ K GW is satisfied, this correction can be safely omitted to describe the relevant modes of the GW spectrum.
Since GW spectra being proportional to K 3 K^3 K 3 and K K K at small and intermediate wave numbers and decaying at large wave numbers are commonly found in GW studies conducted in flat space-time and expanding backgrounds (see, e.g., refs. [23 , 31 , 32 , 34 , 35 , 45 , 49 , 52 , 60 , 98 , 99 ] [1]), the CiT model seems to be in general applicable to slowly-decaying sources of GWs [51 ] .
Following a common normalization for studies of GW backgrounds (see, e.g., refs. [30 , 33 , 34 , 35 , 45 , 49 , 51 , 100 ] [1, 7, 18]), we can write the GW spectrum of Eq. (153) as
Ω G W 0 = 3 F G W 0 Ω ~ G W ϕ ~ r m s 4 ρ ˉ 2 ( R ∗ H ∗ ) 2 S ( K ) . (160) \Omega_{\rm GW}^{0}=3F_{\rm GW}^{0}\widetilde{\Omega}_{\rm GW}\frac{\widetilde{\phi}_{\rm rms}^{4}}{\bar{\rho}^{2}}(R_{*}\mathcal{H}_{*})^{2}S(K). \tag*{(160)} Ω GW 0 = 3 F GW 0 Ω GW ρ ˉ 2 ϕ rms 4 ( R ∗ H ∗ ) 2 S ( K ) . (160)
The spectral shape is determined by the function S ( K ) S(K) S ( K ) ,
S ( K ) = N K 3 S Π ∗ ( K ) T G W ( K ) , (161) S(K)=\mathcal{N}K^{3}\mathcal{S}_{\Pi}^{*}(K)\mathcal{T}_{\rm GW}(K), \tag*{(161)} S ( K ) = N K 3 S Π ∗ ( K ) T GW ( K ) , (161)
where N \mathcal{N} N is a normalization constant that ensures that ∫ S ( K ) d ln K = 1 \int S(K)\,\mathrm{d}\ln K=1 ∫ S ( K ) d ln K = 1 , i.e.,
N − 1 = ∫ − ∞ ∞ K 3 S Π ∗ ( K ) T G W ( K ) d ln K . (162) \mathcal{N}^{-1}=\int_{-\infty}^{\infty}K^{3}\mathcal{S}_{\Pi}^{*}(K)\mathcal{T}_{\rm GW}(K)\,\mathrm{d}\ln K. \tag*{(162)} N − 1 = ∫ − ∞ ∞ K 3 S Π ∗ ( K ) T GW ( K ) d ln K . (162)
The efficiency factor Ω ~ G W \widetilde{\Omega}_{\rm GW} Ω GW determines the integrated amplitude of the GW spectrum [33 , 35 ] [7],
Ω ~ G W = C 2 A 2 N − 1 ( R ∗ H ∗ ) 2 . (163) \widetilde{\Omega}_{\rm GW}=\frac{C}{2A^{2}}\frac{\mathcal{N}^{-1}}{(R_{*}\mathcal{H}_{*})^{2}}. \tag*{(163)} Ω GW = 2 A 2 C ( R ∗ H ∗ ) 2 N − 1 . (163)
We note that the factor C / ( 2 A 2 ) C/(2A^{2}) C / ( 2 A 2 ) has been used in refs. [34 ] [18] to describe Ω ~ G W \widetilde{\Omega}_{\rm GW} Ω GW , based on the results of ref. [1].
We have introduced a quadratic scaling with the source length scale in Eq. (160) to take into account the dependence of the GW amplitude at its peak k G W k_{\rm GW} k GW [1] [see also Eq. (157) ]. To find the exact position of the peak of the GW spectrum, we thus need to find the value of K K K that maximizes K ζ Π ∗ S Π ( K ) K^{\zeta_{\Pi}^{*}}\mathcal{S}_{\Pi}(K) K ζ Π ∗ S Π ( K ) since the peak is in the regime where the kernel’s envelope, T G W ( K ) \mathcal{T}_{\rm GW}(K) T GW ( K ) , becomes proportional to K − 2 K^{-2} K − 2 . Using the fits of Eqs. (103), this implies that
K G W = max K [ K ζ Π ∗ S Π ( K ) ] = K Π ( b Π − 1 ) 1 / a Π . (164) K_{\rm GW}=\max_{K}\left[K^{\zeta_{\Pi}^{*}}\mathcal{S}_{\Pi}(K)\right]=\frac{K_{\Pi}}{(b_{\Pi}-1)^{1/a_{\Pi}}}. \tag*{(164)} K GW = K max [ K ζ Π ∗ S Π ( K ) ] = ( b Π − 1 ) 1/ a Π K Π . (164)
This value was used, e.g., in ref. [35 ] [see their Eq. (19a)] to characterize s t u r b = f t u r b R ∗ s_{\rm turb}=f_{\rm turb}R_{*} s turb = f turb R ∗ as the normalized peak frequency of the GW spectrum from MHD turbulence.
The normalized shape S ( K ) S(K) S ( K ) of the GW spectrum is shown in Figure 13 for a purely vortical source with a = 5 a=5 a = 5 , b = 2 / 3 b=2/3 b = 2/3 (upper panel) and for a compressional source with a = 5 a=5 a = 5 , b = 1 b=1 b = 1 (lower panel), for two values of α \alpha α : α = 6 / 17 \alpha=6/17 α = 6/17 and α = 2 \alpha=2 α = 2 . Their spectral peaks K G W K_{\rm GW} K GW and GW efficiency parameters Ω ~ G W \widetilde{\Omega}_{\rm GW} Ω GW are given in Table II for a range of values of α \alpha α . As anticipated in Eqs. (83), we find that Ω ~ G W \widetilde{\Omega}_{\rm GW} Ω GW is in general smaller for compressional sources than for vortical sources when we compare their values for the same slopes a a a and b b b , and smoothness α \alpha α . K G W K_{\rm GW} K GW are computed using Eq. (164) and the fitting values listed in Table I. For
Figure 13. Normalized spectral shape of the GW spectrum for a purely vortical source with a = 5 a=5 a = 5 , b = 2 / 3 b=2/3 b = 2/3 (upper panel) and for a purely compressional one with a = 5 a=5 a = 5 , b = 1 b=1 b = 1 (lower panel), for both α = 6 / 17 \alpha=6/17 α = 6/17 (solid lines) and 2 2 2 (dashed lines). We consider two values of R ∗ H ∗ = 10 − 2 R_{*}\mathcal{H}_{*}=10^{-2} R ∗ H ∗ = 1 0 − 2 (red lines) and 10 − 1 10^{-1} 1 0 − 1 (blue lines) with sourcing duration δ τ f i n / R ∗ = 100 \delta\tau_{\rm fin}/R_{*}=100 δ τ fin / R ∗ = 100 .
generic velocity, magnetic, and scalar fields sourcing GWs, all the components (vortical, compressional, mixed, and helical) need to be taken into account and included in Eqs. (161) and (163) to find the resulting GW spectrum following the CiT model.
Circular polarization of GWs
The degree of circular polarization P G W ( k ) \mathcal{P}_{\rm GW}(k) P GW ( k ) , as already defined above Eq. (20) , is given by the ratio of the chiral Ξ G W ( τ , k ) \Xi_{\rm GW}(\tau,k) Ξ GW ( τ , k ) to the symmetric Ω G W ( τ , k ) \Omega_{\rm GW}(\tau,k) Ω GW ( τ , k ) GW energy density spectrum. Applying the CiT model as in previous sections, Ξ G W ( τ , k ) \Xi_{\rm GW}(\tau,k) Ξ GW ( τ , k ) becomes
Table 2. Values of the GW spectral peak K G W K_{\mathrm{GW}} K GW [see Eq. (164) ] and of the GW efficiency in the CiT model Ω ~ G W \widetilde{\Omega}_{\mathrm{GW}} Ω GW for a purely vortical non-helical and a purely compressional source with slopes a = 5 a = 5 a = 5 and both b = 2 / 3 b = 2/3 b = 2/3 and 1 1 1 , and a range of values of the smoothness parameter α \alpha α . The values of Ω ~ G W \widetilde{\Omega}_{\mathrm{GW}} Ω GW correspond to the case R ∗ H ∗ = 10 − 2 \mathcal{R}_{*}\mathcal{H}_{*} = 10^{-2} R ∗ H ∗ = 1 0 − 2 and are computed at δ τ f i n / τ ∗ = 100 \delta\tau_{\mathrm{fin}}/\tau_{*} = 100 δ τ fin / τ ∗ = 100 . The values K G W v o r t K_{\mathrm{GW}}^{\mathrm{vort}} K GW vort and K G W c o m p K_{\mathrm{GW}}^{\mathrm{comp}} K GW comp respectively correspond to b = 2 / 3 b = 2/3 b = 2/3 and b = 1 b = 1 b = 1 .
[see Eq. (131) ]
Ξ G W ( τ , k ) = 3 H ∗ Π ( k ) F G W 0 Δ 2 ( τ , k ) . (165) \Xi_{\mathrm{GW}}(\tau,k) = 3 H_{*}^{\Pi}(k) F_{\mathrm{GW}}^{0}\Delta^{2}(\tau,k). \tag*{(165)} Ξ GW ( τ , k ) = 3 H ∗ Π ( k ) F GW 0 Δ 2 ( τ , k ) . (165)
Then, the degree of circular polarization becomes independent of time τ \tau τ and it corresponds to the ratio of the antisymmetric to the symmetric spectra of the anisotropic stresses, P Π ( k ) P_{\Pi}(k) P Π ( k ) [see Eq. (116) ],
P G W ( k ) ≡ Ξ G W ( τ , k ) Ω G W ( τ , k ) = P Π ( k ) = H Π ∗ ( k ) E Π ∗ ( k ) . (166) \mathcal{P}_{\mathrm{GW}}(k) \equiv\frac{\Xi_{\mathrm{GW}}(\tau,k)}{\Omega_{\mathrm{GW}}(\tau,k)} = P_{\Pi}(k) = \frac{H_{\Pi}^{*}(k)}{E_{\Pi}^{*}(k)}. \tag*{(166)} P GW ( k ) ≡ Ω GW ( τ , k ) Ξ GW ( τ , k ) = P Π ( k ) = E Π ∗ ( k ) H Π ∗ ( k ) . (166)
For generic velocity field, using Eqs. (104) and (123) we find
P G W u ( K ) = H Π u ( K ) E Π u ( K ) = K C H Π u , H ( K ) C Π u ( K ) , (167) \mathcal{P}_{\mathrm{GW}}^{u}(K) = \frac{H_{\Pi}^{u}(K)}{E_{\Pi}^{u}(K)} = K\frac{C_{\mathrm{H}\Pi}^{u,\mathrm{H}}(K)}{C_{\Pi}^{u}(K)}, \tag*{(167)} P GW u ( K ) = E Π u ( K ) H Π u ( K ) = K C Π u ( K ) C H Π u , H ( K ) , (167)
where each spectral component contributing to the normalized anisotropic stresses can be determined using Eqs. (B14). Since P G W ( k ) \mathcal{P}_{\mathrm{GW}}(k) P GW ( k ) is constructed from each component, accounting compressional, vortical, and helical components of the velocity spectrum, the degree of circular polarization depends on the degree of compressibility q ( k ) q(k) q ( k ) and on the fractional helicity ε u ( k ) \varepsilon_{u}(k) ε u ( k ) , as shown in Figure 14 , taking constant values for both ε u \varepsilon_{u} ε u (upper panel) and q q q (lower panel).
Figure 14. Upper panel: Degree of circular polarization ∣ P G W ( K ) ∣ |\mathcal{P}_{\mathrm{GW}}(K)| ∣ P GW ( K ) ∣ for different values of the degree of compressibility q ∈ [ 0 , 1 ] q \in[0,1] q ∈ [ 0 , 1 ] of a fully helical velocity field with ε u = ± 1 \varepsilon_{u} = \pm1 ε u = ± 1 . Lower panel: Same as upper panel but for a fully vortical field (q = 0 q = 0 q = 0 ) with different values of fractional helicity. In both panels, we take a = 5 a = 5 a = 5 and b = 2 / 3 b = 2/3 b = 2/3 .
Similarly, the degree of circular polarization for a magnetic field can be expressed using Eqs. (96–99) for H Π H_{\Pi} H Π and Eqs. (86–89) for E Π E_{\Pi} E Π , assuming constant fractional helicity, as
P G W B ( K ) = H Π B ( K ) E Π B ( K ) = K C H Π B , H ( K ) C B S ~ B Π ( K ) = K 2 ε B 7 − 5 ε B 2 J ζ Π B , H ( K ) I ζ Π B ( K ) , (168) \mathcal{P}_{\mathrm{GW}}^{B}(K) = \frac{H_{\Pi}^{B}(K)}{E_{\Pi}^{B}(K)} = K\frac{C_{\mathrm{H}\Pi}^{B,\mathrm{H}}(K)}{C_{B}\widetilde{S}_{B}^{\Pi}(K)}
= K\frac{2\varepsilon_{B}}{7-5\varepsilon_{B}^{2}}\frac{\mathcal{J}\,\zeta_{\Pi}^{B,\mathrm{H}}(K)}{\mathcal{I}\,\zeta_{\Pi}^{B}(K)}, \tag*{(168)} P GW B ( K ) = E Π B ( K ) H Π B ( K ) = K C B S B Π ( K ) C H Π B , H ( K ) = K 7 − 5 ε B 2 2 ε B I ζ Π B ( K ) J ζ Π B , H ( K ) , (168)
where I ≡ ∫ P − 4 ζ B 2 ( P ) d P \mathcal{I} \equiv\int P^{-4}\zeta_{B}^{2}(P)\,dP I ≡ ∫ P − 4 ζ B 2 ( P ) d P and J ≡ ∫ P − 5 ζ B 2 ( P ) d P \mathcal{J} \equiv\int P^{-5}\zeta_{B}^{2}(P)\,dP J ≡ ∫ P − 5 ζ B 2 ( P ) d P are integrals that determine C B C_{B} C B and C B H C_{B}^{\mathrm{H}} C B H , respectively, up to constants that appear multiplying in Eq. (168) if we assume constant fractional helicity [see Eqs. (91) and (118) ].
In Figure 14 , we see that the degree of compressibility q q q impacts the scales at which the spectrum becomes maximally polarized, delaying it towards smaller scales (larger K K K ) as the compressibility is increased, reaching a plateau at P G W ( K ) = ( 1 − q ) ε \mathcal{P}_{\mathrm{GW}}(K) = (1-q)\varepsilon P GW ( K ) = ( 1 − q ) ε (see upper panel of Figure 14 ). Note that for magnetic fields, q = 0 q = 0 q = 0 and all wave numbers above K ≳ 1 K \gtrsim1 K ≳ 1 in the GW spectrum are fully polarized for ε B = ± 1 \varepsilon_{B} = \pm1 ε B = ± 1 .
The lower panel of Figure 14 shows the spectrum of circular polarization for a vortical field with q = 0 q=0 q = 0 and for different values of the helicity fraction. For a fully helical field, a plateau appears at large wave numbers K ≳ 1 K\gtrsim1 K ≳ 1 , consistent with numerical results of helical decaying MHD turbulence (see Figure 5 of ref. [49 ] ) and previous theoretical estimates [67 , 69 ] . This plateau reaches a maximum value of one for P G W ( K ≫ 1 ) \mathcal{P}_{\rm GW}(K\gg1) P GW ( K ≫ 1 ) when the source is fully helical.
The maximum value of ∣ P G W ( K ) ∣ \left|\mathcal{P}_{\rm GW}(K)\right| ∣ P GW ( K ) ∣ as a function of ∣ ϵ B ∣ \left|\epsilon_B\right| ∣ ϵ B ∣ is shown in Figure 15 for different values of q q q , together with some simple fits. The fitting function for q = 0 q=0 q = 0 has a similar functional dependence to the one found for a deterministic magnetic Beltrami field sourcing GWs (see App. A of ref. [49 ] ), P G W B e l = 2 ϵ B / ( 1 + ϵ B 2 ) \mathcal{P}_{\rm GW}^{\rm Bel}=2\epsilon_B/(1+\epsilon_B^2) P GW Bel = 2 ϵ B / ( 1 + ϵ B 2 ) , shown with a dashed black line in Figure 15 . Furthermore, Eq. (168) explains the approximately linear dependence of P G W ( K ) \mathcal{P}_{\rm GW}(K) P GW ( K ) with the fractional helicity of the source (see Figure 6 of ref. [49 ] ), since C B H ∝ ϵ B C_B^H\propto\epsilon_B C B H ∝ ϵ B , while C B C_B C B varies more slowly with ϵ B \epsilon_B ϵ B . However, the dependence of C B H ζ Π B C_B^H\zeta_\Pi^B C B H ζ Π B with ϵ B \epsilon_B ϵ B (see also Figure 4 ) can lead to relevant corrections as ϵ B 2 \epsilon_B^2 ϵ B 2 approaches one, up to a factor of 3.5. Finally, we note that positive or negative helicity of the source only impacts the overall sign of the degree of circular polarization, as expected, but not its absolute value.
Figure 15. Maximum value of the circular polarization as a function of the helicity of the field (solid lines) for different values of q q q , and a = 5 a=5 a = 5 , b = 2 / 3 b=2/3 b = 2/3 , α = 6 / 17 \alpha=6/17 α = 6/17 . In dashed lines we show simple fits for each case with a relative error of less than 1%.
Applications of the Constant-in-Time model
In this section, we briefly discuss the applicability of the CiT model to a few example sources. As discussed above, this model had been developed [1] and validated with numerical simulations of decaying MHD turbulence [50 ] [1]. However, the most notable predictions of this model are the k 3 k^3 k 3 and k k k scalings of the GW spectrum at small and intermediate wave numbers, followed by a decay k 1 − b Π k^{1-b_\Pi} k 1 − b Π at large wave numbers that depends on the particular source. This type of spectrum is broadly found in the literature on GW production in the early Universe [23 , 34 , 45 , 52 , 60 , 98 ] [1].
It goes beyond the scope of this work to make an exhaustive list of these cases and check the applicability of the CiT model to each of them, so we just provide in the following a few examples relevant for first-order phase transitions: stationary turbulence, sound waves, bubble collisions, and feebly interacting particles. Our aim is to motivate the community to test their GW predictions against the CiT model when studying GW sources that vary slowly in time and explore in the future its applicability to other sources. Since this model allows us to make simple and robust predictions for the GW background, identifying the sources that can be accurately described by this model would allow preparing pipelines for GW searches in current and future GW experiments like Pulsar Timing Arrays or LISA.
MHD turbulence
Turbulent decay
The characteristic time of the decaying MHD turbulent cascade is the eddy turnover time δ τ e = τ e − τ ∗ = ( u ∗ k ∗ ) − 1 \delta\tau_e=\tau_e-\tau_*=(u_*k_*)^{-1} δ τ e = τ e − τ ∗ = ( u ∗ k ∗ ) − 1 , where u ∗ u_* u ∗ is a characteristic speed of the turbulent flow. We assume this speed to be the maximum, at the initial time, between the enthalpy-weighted rms velocity of the fluid, U r m s = ⟨ w u 2 ⟩ / w ˉ U_{\rm rms}=\langle w u^2\rangle/\bar w U rms = ⟨ w u 2 ⟩ / w ˉ and the Alfvén speed, which, during the radiation-domination era, is given by v A 2 = 2 ρ ˉ B / w ˉ = 2 3 ρ ˉ B / ρ ˉ v_A^2=2\bar\rho_B/\bar w=\frac{2}{3}\bar\rho_B/\bar\rho v A 2 = 2 ρ ˉ B / w ˉ = 3 2 ρ ˉ B / ρ ˉ [35 , 71 , 74 ] . The eddy turnover time corresponds to the time it takes for the eddies of size λ ∗ = 2 π / k ∗ \lambda_*=2\pi/k_* λ ∗ = 2 π / k ∗ to be processed by turbulence.
Comparing this timescale δ τ e \delta\tau_e δ τ e with the one found in the GW dynamics, δ τ G W = 1 / k \delta\tau_{\rm GW}=1/k δ τ GW = 1/ k , and taking into account that both v A v_A v A and u r m s u_{\rm rms} u rms are smaller than one due to causality, one finds that the GW dynamics is faster than the dynamical timescale of magnetic and velocity fields evolution for all wave numbers k > u ∗ k ∗ k > u_* k_* k > u ∗ k ∗ , as proposed in ref. [1]. For subrelativistic speeds, the range of wave numbers where this condition holds is larger and it only breaks down at the largest scales. There are, however, some discrepancies between the CiT model and the numerical results in the wave numbers below the spectral peak (i.e., for k < k ∗ k < k_* k < k ∗ ), which increase when the eddy turnover time becomes smaller. This is precisely because at these wave numbers, the timescale of GW production is larger than the eddy turnover time and the decay of the source has an impact on the GW production. In addition, the CiT model requires to introduce an effective finite duration after which the source is abruptly switched off in order to incorporate the effect of the turbulent decay. To go beyond these limitations, we introduce in Sec. VII a coherent-decay model for the UETC that accounts for the decay of the source. We find indeed that when the decay occurs at a characteristic timescale δ τ d c \delta\tau_{\rm dc} δ τ dc that is longer than that of the relevant GW dynamics, the impact on the GW background is restricted to k < δ τ d c − 1 k < \delta\tau_{\rm dc}^{-1} k < δ τ dc − 1 , allowing us to apply the CiT model at k > δ τ d c − 1 k > \delta\tau_{\rm dc}^{-1} k > δ τ dc − 1 , and incorporating the effects of the decay at k < δ τ d c − 1 k < \delta\tau_{\rm dc}^{-1} k < δ τ dc − 1 . Here, δ τ d c − 1 = u ∗ k ∗ \delta\tau_{\rm dc}^{-1}=u_*k_* δ τ dc − 1 = u ∗ k ∗ when the decay timescale is prescribed by the eddy turnover time δ τ e \delta\tau_e δ τ e .
Kraichnan decorrelation in stationary turbulence
Furthermore, turbulent motion is expected to decorrelate with time due to its non-linear nature [54 ] . This has been extensively studied and accounted for in turbulence theory and used for studies of GW production from stationary turbulence [42 , 43 , 50 , 82 ] [5]. In the following, we compute the impact of including a decorrelation velocity u d c o r r u_{\rm dcorr} u dcorr in the UETC of the velocity or magnetic fields and how the resulting GW spectrum compares to the one found in the CiT model. This regime was also explicitly studied in ref. [34 ] , where the usual assumption of linear growth with the source duration considered for stationary sources [40 ] is only found for k k k above a threshold value that depends on the decorrelation velocity (see Figure 8 in ref. [34 ] ). This indeed shows that our model allows us to compute the GW background when the velocities are relatively small, such that both the decay and the decorrelation timescales are larger than that of GW production. Since our focus is to test the universality of the CiT model and its extension to include decay of the source, we restrict our analysis in this section to stationary turbulence, and defer the study of decaying turbulence, including both the decay and the decorrelation of the UETC to a separate publication [101 ] .
An important caveat of our proposed model is that it omits the GW production during the process of turbulence development. As we pointed out in Sec. V A (see Figure 10 ), our model is applicable at all wave numbers below the inverse duration of the sourcing phase. If this sourcing phase occurs on a timescale similar to the dynamical timescale of the source, it could have an impact on the GW background at smaller wave numbers, even around the peak [41 , 42 ] . Therefore, the sourcing phase of turbulence needs to be studied in detail for accurate and fully reliable predictions for the GW background from realistic setups of MHD turbulence in the early Universe.
Omitting the decay of the source, the inertial range of the UETC of the velocity field in Kolmogorov stationary turbulence is expected to follow a Kraichnan decorrelation in time, [43 , 50 , 54 , 97 ]
E u ( τ 1 , τ 2 , k ) = E u ∗ ( k ) exp [ − 1 2 k 2 u d c o r r 2 ( k ) τ − 2 ] , (169) E_u(\tau_1,\tau_2,k)=E_u^*(k)\exp\left[-\frac{1}{2}k^2u_{\rm dcorr}^2(k)\tau_-^2\right], \tag*{(169)} E u ( τ 1 , τ 2 , k ) = E u ∗ ( k ) exp [ − 2 1 k 2 u dcorr 2 ( k ) τ − 2 ] , (169)
where E u ∗ = E u ( τ ∗ , k ) E_u^*=E_u(\tau_*,k) E u ∗ = E u ( τ ∗ , k ) and τ − ≡ τ 2 − τ 1 \tau_-\equiv\tau_2-\tau_1 τ − ≡ τ 2 − τ 1 .
Here we have introduced the decorrelation velocity u d c o r r ( k ) u_{\rm dcorr}(k) u dcorr ( k ) , which is close to a constant deep inside the inertial range
u d c o r r 2 ( k ) = 1 3 C u 2 u r m s 2 , with C u ∼ O ( 1 ) u_{\rm dcorr}^2(k)=\frac{1}{3}C_u^2u_{\rm rms}^2,\quad\text{with } C_u\sim\mathcal{O}(1) u dcorr 2 ( k ) = 3 1 C u 2 u rms 2 , with C u ∼ O ( 1 )
for k ∗ ≪ k ≪ k D k_*\ll k\ll k_D k ∗ ≪ k ≪ k D , {#eq:142}
where k ∗ k_* k ∗ and k D k_D k D are respectively the spectral peak and the dissipative scales. The decorrelation velocity can be generalized to a scale-dependent decorrelation velocity, as proposed in ref. [97 ] , already considered in the context of GW production from turbulence in refs. [43 , 50 ] ,
u d c o r r 2 ( k ) = ∫ 0 ∞ h ( p / k ) E u ( p ) d ln p , (170) u_{\rm dcorr}^2(k)=\int_0^\infty h(p/k)E_u(p)\,\mathrm{d}\ln p, \tag*{(170)} u dcorr 2 ( k ) = ∫ 0 ∞ h ( p / k ) E u ( p ) d ln p , (170)
where [97 ]
h ( x ) = 13 − 8 x 2 + 3 x 4 48 + ( 1 − x 2 ) 3 32 x ln ∣ 1 + x 1 − x ∣ . (171) h(x)=\frac{13-8x^2+3x^4}{48}
+\frac{(1-x^2)^3}{32x}\ln\left|\frac{1+x}{1-x}\right|. \tag*{(171)} h ( x ) = 48 13 − 8 x 2 + 3 x 4 + 32 x ( 1 − x 2 ) 3 ln 1 − x 1 + x . (171)
For the von Kármán spectrum considered in ref. [50 ] , the dependence of the decorrelation velocity with k k k can be approximated by the following fit
u d c o r r 2 ( K ) ≃ u r m s 2 3 ( 1 + 0.2 K 5 / 2 + 0.2 K ) 2 . (172) u_{\mathrm{dcorr}}^{2}(K) \simeq\frac{u_{\mathrm{rms}}^{2}}{3}\left(\frac{1+0.2K}{\sqrt{5/2}+0.2K}\right)^{2}. \tag*{(172)} u dcorr 2 ( K ) ≃ 3 u rms 2 ( 5/2 + 0.2 K 1 + 0.2 K ) 2 . (172)
This function takes values from u d c o r r 2 = 2 15 u r m s 2 u_{\mathrm{dcorr}}^{2}=\frac{2}{15}u_{\mathrm{rms}}^{2} u dcorr 2 = 15 2 u rms 2 when K → 0 K\to0 K → 0 , to 1 3 u r m s 2 \frac{1}{3}u_{\mathrm{rms}}^{2} 3 1 u rms 2 for large K K K , deep in the inertial range.
In the following, we want to show in which limit the Kraichnan decorrelation can be neglected for the computation of the GW spectrum from stationary turbulence and, hence, understand the regime of validity of the CiT model for this type of sourcing process. We will assume that the decorrelation velocity is a constant, such that the kernel Δ 2 \Delta^{2} Δ 2 does not depend on the spectrum of the source and we can provide general results for this type of sources.
The GW spectrum at present time, found after averaging over fast frequency oscillations, then becomes, for a purely vortical velocity field [cf. Eq. (B14a)],
Ω G W 0 ( k ) = 3 8 Γ 2 u r m s 4 K 2 F G W 0 C A 2 × ∫ 0 ∞ d P P 2 ∫ 0 ∞ d P ~ P ~ 2 ( 1 + z 2 ) ( 1 + z ~ 2 ) × ζ u ( P ) ζ u ( P ~ ) Δ t u r b , a v 2 ( K , P , P ~ ) , (173) \Omega_{\mathrm{GW}}^{0}(k)=\frac{3}{8}\Gamma^{2}u_{\mathrm{rms}}^{4}K^{2}F_{\mathrm{GW}}^{0}\frac{\mathcal{C}}{\mathcal{A}^{2}}
\times\int_{0}^{\infty}\frac{\mathrm{d}P}{P^{2}}\int_{0}^{\infty}\frac{\mathrm{d}\widetilde{P}}{\widetilde{P}^{2}}(1+z^{2})(1+\widetilde{z}^{2})
\times\zeta_{u}(P)\zeta_{u}(\widetilde{P})\Delta_{\mathrm{turb,av}}^{2}(K,P,\widetilde{P}), \tag*{(173)} Ω GW 0 ( k ) = 8 3 Γ 2 u rms 4 K 2 F GW 0 A 2 C × ∫ 0 ∞ P 2 d P ∫ 0 ∞ P 2 d P ( 1 + z 2 ) ( 1 + z 2 ) × ζ u ( P ) ζ u ( P ) Δ turb , av 2 ( K , P , P ) , (173)
where Δ t u r b , a v 2 \Delta_{\mathrm{turb,av}}^{2} Δ turb , av 2 is the kernel that includes the time decorrelation of the UETC and the averaged Green’s function,
Δ t u r b , a v 2 = 1 2 ∫ τ ∗ τ f i n d τ 1 τ 1 ∫ τ ∗ τ f i n d τ 2 τ 2 cos ( τ − k ∗ K ) × exp [ − 1 2 τ − 2 u d c o r r 2 k ∗ 2 ( P 2 + P ~ 2 ) ] . (174) \Delta_{\mathrm{turb,av}}^{2}=\frac{1}{2}\int_{\tau_{*}}^{\tau_{\mathrm{fin}}}\frac{\mathrm{d}\tau_{1}}{\tau_{1}}\int_{\tau_{*}}^{\tau_{\mathrm{fin}}}\frac{\mathrm{d}\tau_{2}}{\tau_{2}}\cos(\tau_{-}k_{*}K)
\times\exp\left[-\frac{1}{2}\tau_{-}^{2}u_{\mathrm{dcorr}}^{2}k_{*}^{2}(P^{2}+\widetilde{P}^{2})\right]. \tag*{(174)} Δ turb , av 2 = 2 1 ∫ τ ∗ τ fin τ 1 d τ 1 ∫ τ ∗ τ fin τ 2 d τ 2 cos ( τ − k ∗ K ) × exp [ − 2 1 τ − 2 u dcorr 2 k ∗ 2 ( P 2 + P 2 ) ] . (174)
We note that this kernel can be integrated analytically over τ − \tau_{-} τ − after a change of variable, reducing it to one integral over time τ 1 \tau_{1} τ 1 [101 ] .
Similarly to what we did in § 0.5 B for the CiT model [see Eq. (132)], we can define a normalized spectral shape S t u r b ( K ) \mathcal{S}_{\mathrm{turb}}(K) S turb ( K ) and a GW efficiency Ω ~ G W t u r b \widetilde{\Omega}_{\mathrm{GW}}^{\mathrm{turb}} Ω GW turb such that we can write the GW spectrum at present time as
Ω G W 0 ( k ) = 3 Γ 2 u r m s 4 F G W 0 Ω ~ G W t u r b × ( R ∗ H ∗ ) 2 S t u r b ( K ) , (175) \Omega_{\mathrm{GW}}^{0}(k)=3\Gamma^{2}u_{\mathrm{rms}}^{4}F_{\mathrm{GW}}^{0}\widetilde{\Omega}_{\mathrm{GW}}^{\mathrm{turb}}
\times(R_{*}\mathcal{H}_{*})^{2}\mathcal{S}_{\mathrm{turb}}(K), \tag*{(175)} Ω GW 0 ( k ) = 3 Γ 2 u rms 4 F GW 0 Ω GW turb × ( R ∗ H ∗ ) 2 S turb ( K ) , (175)
with
Ω ~ G W t u r b = C A 2 N t u r b − 1 ( R ∗ H ∗ ) 2 , (176) \widetilde{\Omega}_{\mathrm{GW}}^{\mathrm{turb}}=\frac{\mathcal{C}}{\mathcal{A}^{2}}\frac{\mathcal{N}_{\mathrm{turb}}^{-1}}{(R_{*}\mathcal{H}_{*})^{2}}, \tag*{(176)} Ω GW turb = A 2 C ( R ∗ H ∗ ) 2 N turb − 1 , (176)
S t u r b ( K ) = N t u r b K 2 8 × ∫ 0 ∞ d P P 2 ∫ 0 ∞ d P ~ P ~ 2 ( 1 + z 2 ) ( 1 + z ~ 2 ) × ζ u ( P ) ζ u ( P ~ ) Δ t u r b , a v 2 ( K , P , P ~ ) , (177) \mathcal{S}_{\mathrm{turb}}(K)=\mathcal{N}_{\mathrm{turb}}\frac{K^{2}}{8}
\times\int_{0}^{\infty}\frac{\mathrm{d}P}{P^{2}}\int_{0}^{\infty}\frac{\mathrm{d}\widetilde{P}}{\widetilde{P}^{2}}(1+z^{2})(1+\widetilde{z}^{2})
\times\zeta_{u}(P)\zeta_{u}(\widetilde{P})\Delta_{\mathrm{turb,av}}^{2}(K,P,\widetilde{P}), \tag*{(177)} S turb ( K ) = N turb 8 K 2 × ∫ 0 ∞ P 2 d P ∫ 0 ∞ P 2 d P ( 1 + z 2 ) ( 1 + z 2 ) × ζ u ( P ) ζ u ( P ) Δ turb , av 2 ( K , P , P ) , (177)
and where we have again defined N t u r b \mathcal{N}_{\mathrm{turb}} N turb such that ∫ S t u r b ( K ) d ln K = 1 \int\mathcal{S}_{\mathrm{turb}}(K)\,\mathrm{d}\ln K=1 ∫ S turb ( K ) d ln K = 1 .
In Figure 16 , we show the normalized GW spectrum Ω ~ G W t u r b S t u r b ( K ) \widetilde{\Omega}_{\mathrm{GW}}^{\mathrm{turb}}\mathcal{S}_{\mathrm{turb}}(K) Ω GW turb S turb ( K ) for different values of u d c o r r u_{\mathrm{dcorr}} u dcorr and two different source durations, δ τ f i n = τ ∗ \delta\tau_{\rm fin}=\tau_* δ τ fin = τ ∗ (lower panel) and δ τ f i n = 100 τ ∗ \delta\tau_{\rm fin}=100\tau_* δ τ fin = 100 τ ∗ (upper panel). We see that the GW spectrum of a field whose UETC includes Kraichnan decorrelation accurately recovers the CiT model, for R ∗ H ∗ = 10 − 2 \mathcal{R}_*\mathcal{H}_*=10^{-2} R ∗ H ∗ = 1 0 − 2 , when u d e c o r r 2 ≲ 10 − 4 u_{\rm decorr}^2\lesssim10^{-4} u decorr 2 ≲ 1 0 − 4 corresponding to subrelativistic speeds u r m s ≲ 10 − 2 u_{\rm rms}\lesssim10^{-2} u rms ≲ 1 0 − 2 . In general, we expect to recover the CiT model in the limit u d e c o r r k ∗ δ τ f i n ≪ 1 u_{\rm decorr}k_*\delta\tau_{\rm fin}\ll1 u decorr k ∗ δ τ fin ≪ 1 .
Figure 16. Normalized GW spectrum Ω ~ G W t u r b S t u r b ( K ) \widetilde{\Omega}_{\mathrm{GW}}^{\mathrm{turb}}\mathcal{S}_{\mathrm{turb}}(K) Ω GW turb S turb ( K ) in the case of purely vortical turbulence including Kraichnan decorrelation, for different constant decorrelation velocities u d c o r r u_{\mathrm{dcorr}} u dcorr . In both panels we consider a von Kármán spectrum with a = 5 a=5 a = 5 , b = 2 / 3 b=2/3 b = 2/3 , and α = 2 \alpha=2 α = 2 . The characteristic lengthscale of the source is R ∗ H ∗ = 10 − 2 R_{*}\mathcal{H}_{*}=10^{-2} R ∗ H ∗ = 1 0 − 2 and we consider two source durations: δ τ f i n H ∗ = 10 2 \delta\tau_{\mathrm{fin}}\mathcal{H}_{*}=10^{2} δ τ fin H ∗ = 1 0 2 (upper panel) and δ τ f i n H ∗ = 1 \delta\tau_{\mathrm{fin}}\mathcal{H}_{*}=1 δ τ fin H ∗ = 1 (lower panel). For comparison we plot in dashed black lines the spectrum in the CiT case, which is approached as u d c o r r u_{\mathrm{dcorr}} u dcorr decreases.
In Table III we show the values of Ω ~ G W t u r b \widetilde{\Omega}_{\rm GW}^{\rm turb} Ω GW turb for different source durations δ τ f i n / τ ∗ \delta\tau_{\rm fin}/\tau_* δ τ fin / τ ∗ and different values of the decorrelation velocity u d e c o r r u_{\rm decorr} u decorr , showing how the GW efficiency approaches the value found in the CiT model (i.e., Ω ~ G W ≃ 0.069 \widetilde{\Omega}_{\rm GW}\simeq0.069 Ω GW ≃ 0.069 ; see Table II) as u d e c o r r u_{\rm decorr} u decorr decreases, i.e., as the motion becomes subrelativistic. As the decorrelation velocity increases we see the growth of the peak of the resonant term at a wave number k k k proportional to k ∗ u d e c o r r k_*u_{\rm decorr} k ∗ u decorr . The resonant peak appears when u d e c o r r k ∗ τ ∗ Υ ≳ 1 u_{\rm decorr}k_*\tau_*\Upsilon\gtrsim1 u decorr k ∗ τ ∗ Υ ≳ 1 , and grows proportional to Υ = δ τ f i n H ∗ / ( 1 + δ τ f i n H ∗ ) \Upsilon=\delta\tau_{\rm fin}\mathcal{H}_*/(1+\delta\tau_{\rm fin}\mathcal{H}_*) Υ = δ τ fin H ∗ / ( 1 + δ τ fin H ∗ ) for radiation domination [34 ] , and linear in δ τ f i n \delta\tau_{\rm fin} δ τ fin in flat space-time. For smaller u d e c o r r k ∗ τ ∗ Υ u_{\rm decorr}k_*\tau_*\Upsilon u decorr k ∗ τ ∗ Υ , the resonant peak does not appear, the GW spectrum reduces to the one predicted by the CiT model, and the spectral peak becomes K G W K_{\rm GW} K GW in Eq. (164) . The amplitude of the resonant peak is proportional to u d e c o r r 2 u_{\rm decorr}^2 u decorr 2 . At large u d e c o r r u_{\rm decorr} u decorr the peak dominates and leads to a much larger GW efficiency than in the CiT model, as we can see in Table III . However, in the relativistic regime of velocities, we need to also take into account the decay of the source due to the non-linear dynamics of the fluid. Furthermore, we need to generalize the source of GWs to X r m s 2 = ρ ˉ k i n X_{\rm rms}^2=\bar{\rho}_{\rm kin} X rms 2 = ρ ˉ kin to also take into account large fluid enthalpy fluctuations. We leave the realistic prediction of the GW spectrum for turbulence with both Kraichnan decorrelation and decay of the source for future work [101 ] .
Table III. Values of Ω ~ G W t u r b \widetilde{\Omega}_{\rm GW}^{\rm turb} Ω GW turb for different u d e c o r r 2 u_{\rm decorr}^2 u decorr 2 and δ τ f i n / τ ∗ \delta\tau_{\rm fin}/\tau_* δ τ fin / τ ∗ computed from Eq. (148b) for a von Kármán vortical spectrum with a = 5 a=5 a = 5 , b = 2 / 3 b=2/3 b = 2/3 , and α = 2 \alpha=2 α = 2 . We take R ∗ H ∗ = 10 − 2 \mathcal{R}_*\mathcal{H}_*=10^{-2} R ∗ H ∗ = 1 0 − 2 as in Figure 17 . Note that for u d e c o r r 2 ≲ 10 − 4 u_{\rm decorr}^2\lesssim10^{-4} u decorr 2 ≲ 1 0 − 4 we recover the efficiency Ω ~ G W ≃ 0.069 \widetilde{\Omega}_{\rm GW}\simeq0.069 Ω GW ≃ 0.069 found in the CiT model (see Table II) for all values of δ τ f i n \delta\tau_{\rm fin} δ τ fin considered.
Sound waves
In studies of GW production from linearized sound waves, the UETC of the purely compressional velocity field is commonly approximated as E u ( τ 1 , τ 2 , k ) = E u ∗ ( k ) cos ( k c s τ − ) E_u(\tau_1,\tau_2,k)=E_u^*(k)\cos(kc_s\tau_-) E u ( τ 1 , τ 2 , k ) = E u ∗ ( k ) cos ( k c s τ − ) , with τ − = τ 2 − τ 1 \tau_-=\tau_2-\tau_1 τ − = τ 2 − τ 1 , E u ∗ ( k ) ≡ E u ( τ ∗ , τ ∗ , k ) E_u^*(k)\equiv E_u(\tau_*,\tau_*,k) E u ∗ ( k ) ≡ E u ( τ ∗ , τ ∗ , k ) , and c s c_s c s being the speed of sound [28 , 30 , 33 , 34 , 36 , 37 , 52 , 63 ] . Assuming that the compressional velocity field follows a Gaussian distribution and introducing the UETC of the anisotropic stresses [see Eqs. (104) and (B14b)] into the GW spectrum at present time [see Eq. (26) ], we find [34 ]
Ω G W 2 ( K ) = 3 Γ 2 u r m s 4 K 2 F G W 0 C A 2 \Omega_{\rm GW}^2(K)=3\Gamma^2 u_{\rm rms}^4 K^2 F_{\rm GW}^0 \frac{\mathcal{C}}{\mathcal{A}^2} Ω GW 2 ( K ) = 3 Γ 2 u rms 4 K 2 F GW 0 A 2 C
× ∫ d P P 2 ∫ d P ~ P ~ 2 ( 1 − z 2 ) ( 1 − z ~ 2 ) \qquad{}\times\int\frac{\mathrm{d}P}{P^2}\int\frac{\mathrm{d}\widetilde{P}}{\widetilde{P}^2}(1-z^2)(1-\widetilde{z}^2) × ∫ P 2 d P ∫ P 2 d P ( 1 − z 2 ) ( 1 − z 2 )
× ζ u ( P ) ζ u ( P ~ ) Δ s w , a v 2 ( K , P , P ~ ) , (178) \qquad{}\times\zeta_u(P)\zeta_u(\widetilde{P})\Delta_{\rm sw,av}^2(K,P,\widetilde{P}) \, , \tag*{(178)} × ζ u ( P ) ζ u ( P ) Δ sw , av 2 ( K , P , P ) , (178)
where ζ u ( K ) \zeta_u(K) ζ u ( K ) is the normalized spectrum of the velocity field [cf. Eq. (B11a)], as in Sec. IIIC (see also ref. [34 ] ). The time kernel for stationary sound waves, averaged over fast frequency oscillations at late times, Δ s w , a v 2 \Delta_{\rm sw,av}^2 Δ sw , av 2 , can be split into four components [34 , 52 ] ,
Δ s w , a v 2 = 1 4 ∑ m , n = ± 1 Δ s w , m n 2 , (179) \Delta_{\rm sw,av}^2=\frac{1}{4}\sum_{m,n=\pm1}\Delta_{\rm sw,mn}^2 \, , \tag*{(179)} Δ sw , av 2 = 4 1 m , n = ± 1 ∑ Δ sw , mn 2 , (179)
with
Δ s w , m n 2 ( p ^ m n ) = 1 2 [ Δ C i 2 ( p ^ m n τ f i n ) + Δ S i 2 ( p ^ m n τ f i n ) ] , (180) \Delta_{\rm sw,mn}^2(\widehat{p}_m n)=\frac{1}{2}\left[\Delta{\rm Ci}^2(\widehat{p}_m n\tau_{\rm fin})+\Delta{\rm Si}^2(\widehat{p}_m n\tau_{\rm fin})\right] \, , \tag*{(180)} Δ sw , mn 2 ( p m n ) = 2 1 [ Δ Ci 2 ( p m n τ fin ) + Δ Si 2 ( p m n τ fin ) ] , (180)
where p ^ m n = ( p + m p ~ ) c s + n k \widehat{p}_m n=(p+m\widetilde{p})c_s+nk p m n = ( p + m p ) c s + nk , and the Δ C i \Delta{\rm Ci} Δ Ci and Δ S i \Delta{\rm Si} Δ Si functions are defined as in Eqs. (107).
The dominant contribution initially considered in ref. [28 ] (see also refs. [34 , 52 ] ), corresponds to the resonant m = 1 m=1 m = 1 and n = − 1 n=-1 n = − 1 , and the resulting GW spectrum behaves very differently than the one found in the CiT model, as described in refs. [33 , 34 , 51 ] . On the other hand, the terms corresponding to m = − 1 m=-1 m = − 1 and n = ± 1 n=\pm1 n = ± 1 become identical after integration (see Figure 7 in ref. [52 ] ). In this case, we can observe that if the spectrum of the source is sharply peaked around k ∗ k_\ast k ∗ , such that ζ u ( K ) = δ ( K − 1 ) \zeta_u(K) = \delta(K - 1) ζ u ( K ) = δ ( K − 1 ) , then the only contribution is coming from P = P ~ = 1 P = \widetilde{P} = 1 P = P = 1 . Therefore, p ^ − 1 , ± 1 = ± k \hat{p}_{-1,\pm1} = \pm k p ^ − 1 , ± 1 = ± k , such that
Δ s w , − 1 , ± 1 2 = 1 2 [ Δ Ci 2 ( k τ f i n ) + Δ Si 2 ( k τ f i n ) ] . (181) \Delta^2_{\mathrm{sw},-1,\pm1} = \frac{1}{2}\left[\Delta\operatorname{Ci}^2(k\tau_{\mathrm{fin}}) + \Delta\operatorname{Si}^2(k\tau_{\mathrm{fin}})\right]. \tag*{(181)} Δ sw , − 1 , ± 1 2 = 2 1 [ Δ Ci 2 ( k τ fin ) + Δ Si 2 ( k τ fin ) ] . (181)
which is equivalent to the result found in the CiT model [see Eq. (150) ]. Hence, for sharply peaked velocity spectra, these contributions to the sound-wave production of GWs are expected to be well described by the CiT model. Indeed, this corresponds to the part of the GW spectrum found in ref. [34 ] below the spectral peak, which presents K 3 K^3 K 3 scaling at K < 1 / δ τ f i n K < 1/\delta\tau_{\mathrm{fin}} K < 1/ δ τ fin and becomes linear at K > 1 / δ τ f i n K > 1/\delta\tau_{\mathrm{fin}} K > 1/ δ τ fin . This scaling can be explained following the CiT model. Eventually, the dominant contribution (m = 1 , n = − 1 m = 1, n = -1 m = 1 , n = − 1 ) takes over, leading to a sharp bump around the spectral peak that grows with the source duration. This resonant term is analogous to the one found for stationary turbulence in previous section (see also refs. [40 , 82 ] ).
Following § 0.5 B, we again define the normalized spectral function S s w ( K ) S_{\mathrm{sw}}(K) S sw ( K ) , such that ∫ S s w ( K ) d ln K = 1 \int S_{\mathrm{sw}}(K)\,\mathrm{d}\ln K = 1 ∫ S sw ( K ) d ln K = 1 , as
S s w ( K ) = N s w K 2 ∫ d P P 2 ∫ d P ~ P ~ 2 ( 1 − z 2 ) ( 1 − z ~ 2 ) ζ u ( P ) ζ u ( P ~ ) Δ s w , a v 2 ( K , P , P ~ ) . (182) S_{\mathrm{sw}}(K) = \mathcal{N}_{\mathrm{sw}}K^2 \int\frac{\mathrm{d}P}{P^2}\int\frac{\mathrm{d}\widetilde{P}}{\widetilde{P}^2}(1-z^2)(1-\widetilde{z}^2)\zeta_u(P)\zeta_u(\widetilde{P})\Delta^2_{\mathrm{sw,av}}(K,P,\widetilde{P}). \tag*{(182)} S sw ( K ) = N sw K 2 ∫ P 2 d P ∫ P 2 d P ( 1 − z 2 ) ( 1 − z 2 ) ζ u ( P ) ζ u ( P ) Δ sw , av 2 ( K , P , P ) . (182)
By defining
Ω ~ G W s w = c 2 A 2 N s w − 1 ( R ∗ H ∗ ) 2 , (183) \widetilde{\Omega}^{\mathrm{sw}}_{\mathrm{GW}} = \frac{c}{2\mathcal{A}^2}\frac{\mathcal{N}_{\mathrm{sw}}^{-1}}{(\mathcal{R}_\ast\mathcal{H}_\ast)^2}, \tag*{(183)} Ω GW sw = 2 A 2 c ( R ∗ H ∗ ) 2 N sw − 1 , (183)
we can then rewrite Eq. (178) as
Ω G W 0 ( K ) = 3 Γ 2 u r m s 4 F G W 0 Ω ~ G W s w ( R ∗ H ∗ ) 2 S s w ( K ) . (184) \Omega^0_{\mathrm{GW}}(K) = 3\Gamma^2u_{\mathrm{rms}}^4 F^0_{\mathrm{GW}}\widetilde{\Omega}^{\mathrm{sw}}_{\mathrm{GW}}(\mathcal{R}_\ast\mathcal{H}_\ast)^2S_{\mathrm{sw}}(K). \tag*{(184)} Ω GW 0 ( K ) = 3 Γ 2 u rms 4 F GW 0 Ω GW sw ( R ∗ H ∗ ) 2 S sw ( K ) . (184)
We show in Figure 17 , a comparison of the normalized GW spectrum, Ω ~ G W s w S s w ( K ) \widetilde{\Omega}^{\mathrm{sw}}_{\mathrm{GW}}S_{\mathrm{sw}}(K) Ω GW sw S sw ( K ) , split into its different four contributions from Eq. (179) , with the normalized spectral shape S ( K ) S(K) S ( K ) in the CiT model [see Eqs. (160) –(163) ], for different values of the speed of sound, c s c_s c s . In Table IV we show the values of the efficiency factor Ω ~ G W \widetilde{\Omega}_{\mathrm{GW}} Ω GW for different final times and different values of c s 2 c_s^2 c s 2 . We see that for all values of c s 2 c_s^2 c s 2 the GW spectrum reaches saturation due to the expansion of the Universe with an amplitude proportional to Υ = δ τ f i n H ∗ / ( 1 + δ τ f i n H ∗ ) \Upsilon= \delta\tau_{\mathrm{fin}}\mathcal{H}_\ast/(1+\delta\tau_{\mathrm{fin}}\mathcal{H}_\ast) Υ = δ τ fin H ∗ / ( 1 + δ τ fin H ∗ ) when c s k ∗ δ τ f i n ≫ 1 c_sk_\ast\delta\tau_{\mathrm{fin}} \gg1 c s k ∗ δ τ fin ≫ 1 [34 ] . Moreover, as the sound speed decreases, also the efficiency quickly decreases, and
Figure 17. Upper panel: Four components of the normalized GW spectrum from sound waves with c s 2 = 1 / 3 c_s^2 = 1/3 c s 2 = 1/3 [34 , 52 ] , following the sound-shell model [28 , 30 ] , with m = 1 , n = − 1 m = 1, n = -1 m = 1 , n = − 1 [see Eq. (179) ] being the resonant contribution, and m = − 1 , n = ± 1 m = -1, n = \pm1 m = − 1 , n = ± 1 being described by the CiT model. Lower panel: Normalized GW spectrum for different values of c s c_s c s , showing how it converges to the GW spectrum predicted by the CiT model in the c s → 0 c_s \to0 c s → 0 limit. In both panels, the broken power law parameters for ζ u ( K ) \zeta_u(K) ζ u ( K ) are a = 5 a = 5 a = 5 , b = 1 b = 1 b = 1 , and α = 2 \alpha= 2 α = 2 . We use R ∗ H ∗ = 10 − 2 \mathcal{R}_\ast\mathcal{H}_\ast= 10^{-2} R ∗ H ∗ = 1 0 − 2 and δ τ f i n H ∗ = 10 2 \delta\tau_{\mathrm{fin}}\mathcal{H}_\ast= 10^2 δ τ fin H ∗ = 1 0 2 in both panels.
Table IV. Values of Ω ~ G W s w \widetilde{\Omega}_{\rm GW}^{\rm sw} Ω GW sw for different c s 2 c_s^2 c s 2 and δ τ f i n / τ ∗ \delta\tau_{\rm fin}/\tau_* δ τ fin / τ ∗ computed from Eq. (183) for a purely compressional source with a spectrum characterized by a = 5 a=5 a = 5 , b = 1 b=1 b = 1 , and α = 2 \alpha=2 α = 2 . We take R ∗ H ∗ = 10 − 2 R_*H_*=10^{-2} R ∗ H ∗ = 1 0 − 2 as in Figure 17 . Note that for c s 2 ≲ 10 − 4 c_s^2\lesssim10^{-4} c s 2 ≲ 1 0 − 4 we recover the efficiency Ω ~ G W ≃ 0.078 \widetilde{\Omega}_{\rm GW}\simeq0.078 Ω GW ≃ 0.078 found in the CiT model (see Table II) for all values of δ τ f i n \delta\tau_{\rm fin} δ τ fin considered.
for smaller values of c s 2 c_s^2 c s 2 saturation is reached earlier, as expected in the CiT model, for which saturation is predicted to occur by the time δ τ ∗ s w ≃ 1 / k ∗ \delta\tau_\ast^{\mathrm{sw}} \simeq1/k_\ast δ τ ∗ sw ≃ 1/ k ∗ . We note that as the sound speed increases, the resonant peak starts to dominate the total GW spectrum when c s k ∗ τ ≳ 1 c_sk_\ast\tau\gtrsim1 c s k ∗ τ ≳ 1 , and causes a large increase of the GW efficiency for the ultra-relativistic equation of state c s 2 = 1 / 3 c_s^2 = 1/3 c s 2 = 1/3 , compared to smaller values of c s c_s c s . This can explain why previous simulations of acoustic turbulence with c s 2 = 1 / 3 c_s^2 = 1/3 c s 2 = 1/3 always find a larger efficiency producing GWs compared to vortical turbulence with subrelativistic fluid and/or Alfven velocities [35 , 45 , 49 ] .
We note that when the fluid is the dominant component of the Universe, it is not consistent to vary c s c_s c s while keeping the evolution of the scale factor determined for a radiation-dominated Universe. In general, as c s c_s c s varies, the dynamics of the sound waves also vary [63 , 74 ] and the Green’s function of the GW equation is modified [63 ] . However, this simpler treatment allows us to understand what is the impact of c s c_s c s in the UETC, showing how c s c_s c s acts as a decorrelation velocity, such that the CiT model is recovered for long decorrelation times δ τ d c o r r = ( c s k ∗ ) − 1 ≫ τ ∗ \delta\tau_{\rm dcorr}=(c_sk_*)^{-1}\gg\tau_* δ τ dcorr = ( c s k ∗ ) − 1 ≫ τ ∗ , and fails to explain the full GW spectrum when c s k ∗ τ ∗ ≫ 1 c_sk_*\tau_*\gg1 c s k ∗ τ ∗ ≫ 1 , e.g., when c s 2 = 1 / 3 c_s^2=1/3 c s 2 = 1/3 .
We have shown that for a stationary UETC, approximated using a constant correlation velocity (u d c o r r u_{\rm dcorr} u dcorr for turbulence with Kraichnan decorrelation and c s c_s c s for sound waves), there are two distinguished regimes. For small u d c o r r k ∗ δ τ f i n ≪ 1 u_{\rm dcorr}k_*\delta\tau_{\rm fin}\ll1 u dcorr k ∗ δ τ fin ≪ 1 and c s k ∗ δ τ f i n ≪ 1 c_sk_*\delta\tau_{\rm fin}\ll1 c s k ∗ δ τ fin ≪ 1 , the GW spectrum sourced by both sound waves and stationary turbulence can be accurately described by the CiT model, which is consistent with the findings of numerical simulations for decaying MHD turbulence [45 , 49 , 50 ] [1]. However, as the correlation velocity increases, u d c o r r k ∗ τ ∗ Υ ≳ 1 u_{\rm dcorr}k_*\tau_*\Upsilon\gtrsim1 u dcorr k ∗ τ ∗ Υ ≳ 1 and c s k ∗ τ ∗ Υ ≳ 1 c_sk_*\tau_*\Upsilon\gtrsim1 c s k ∗ τ ∗ Υ ≳ 1 , with Υ = δ τ f i n H ∗ / ( 1 + δ τ f i n H ∗ ) \Upsilon=\delta\tau_{\rm fin}H_*/(1+\delta\tau_{\rm fin}H_*) Υ = δ τ fin H ∗ / ( 1 + δ τ fin H ∗ ) , the CiT model describes one component of the GW spectrum, which dominates at small k k k , but an additional resonant term grows proportional to the source duration, Ω G W ( K ) ∼ Υ = δ τ f i n H ∗ / ( 1 + δ τ f i n H ∗ ) \Omega_{\rm GW}(K)\sim\Upsilon=\delta\tau_{\rm fin}H_*/(1+\delta\tau_{\rm fin}H_*) Ω GW ( K ) ∼ Υ = δ τ fin H ∗ / ( 1 + δ τ fin H ∗ ) [34 ] , which dominates around the peak and at larger K K K . In this case, the location of the peak moves proportional to u d c o r r u_{\rm dcorr} u dcorr [40 ] or c s c_s c s [30 , 63 ] , with an amplitude proportional to u d c o r r 2 u_{\rm dcorr}^2 u dcorr 2 or c s 2 c_s^2 c s 2 . Therefore, we in general find that the presence of decorrelation in the UETC enhances the efficiency in producing GWs [41 ] [1], such that the CiT model provides a conservative estimate of the GW spectrum of these sources [1, 9, 18].
Bubble collisions
The GW production from scalar bubble collisions in first-order phase transitions was first studied following the envelope approximation [22 , 59 , 102 , 103 ] [4], where the scalar bubbles are assumed to vanish instantaneously after collisions. The GW production occurs when the thin shells of scalar field gradients collide. This model predicts a GW background proportional to k 3 k^3 k 3 at k ≪ k ∗ k\ll k_* k ≪ k ∗ and to k − 1 k^{-1} k − 1 at k ≫ k ∗ k\gg k_* k ≫ k ∗ , respectively.
Going beyond the envelope approximation, the bulk flow model was proposed for the GW production [23 , 60 ] , where the shells of gradient energy density are assumed to continue producing GWs even after collisions, instead of instantaneously vanishing, and their decay is taken into account. Simulations and analytical estimates following the bulk flow model predict a linear scaling ∼ k \sim k ∼ k at low wave numbers, and a power-law decay ∼ k − 3 \sim k^{-3} ∼ k − 3 at large wave numbers [23 , 60 ] . A similar GW spectrum (with k k k and k − 2 k^{-2} k − 2 scalings) has been found from feebly interacting particles during a phase transition in the dark sector [98 ] .
The linear scaling with k k k strongly resembles the predictions of the CiT model. In particular, in the bulk flow model, the scalar field gradients sourcing GWs are assumed to decay over a timescale δ τ d c \delta\tau_{\rm dc} δ τ dc . This timescale can be roughly taken to be the source duration, δ τ d c ∼ δ τ f i n \delta\tau_{\rm dc}\sim\delta\tau_{\rm fin} δ τ dc ∼ δ τ fin . Then, the linear-in-k k k scaling is expected for k ≥ δ τ f i n k\geq\delta\tau_{\rm fin} k ≥ δ τ fin , recovering the k 3 k^3 k 3 scaling at k < 1 / δ τ f i n k<1/\delta\tau_{\rm fin} k < 1/ δ τ fin , as predicted by the CiT model. Applying the CiT model, the GW spectrum decays at k ≳ k ∗ k\gtrsim k_* k ≳ k ∗ proportionally to k − b Π c o m p k^{-b_{\Pi}^{\rm comp}} k − b Π comp , where b Π c o m p = 2 b + 3 b_{\Pi}^{\rm comp}=2b+3 b Π comp = 2 b + 3 (if b < 2 b<2 b < 2 ) [cf. Eqs. (104)] depends on the UV-slope − b -b − b of the spectrum of the scalar field gradients E L ϕ ( k ) = k 2 E ϕ ( k ) E_L^\phi(k)=k^2E_\phi(k) E L ϕ ( k ) = k 2 E ϕ ( k ) , and k ∗ k_* k ∗ is its characteristic wave number.
As discussed in Sec. III C, the power spectrum E ϕ ( k ) E_\phi(k) E ϕ ( k ) for the scalar fields in a vacuum first-order phase transition found in the simulations of refs. [24 , 25 ] follows a broken power law with E ϕ ( k ) ∼ k − 2.5 E_\phi(k)\sim k^{-2.5} E ϕ ( k ) ∼ k − 2.5 at k ≫ k ∗ k \gg k_\ast k ≫ k ∗ , before falling exponentially at scales around the wall thickness. Since E L ϕ ( k ) = k 2 E ϕ ( k ) ∼ k − 0.5 E_L^\phi(k)=k^2 E_\phi(k)\sim k^{-0.5} E L ϕ ( k ) = k 2 E ϕ ( k ) ∼ k − 0.5 , then we find that the GW spectrum should decay as Ω G W ∼ k − 2 ( 1 + b ) = k − 3 \Omega_{\rm GW}\sim k^{-2(1+b)}=k^{-3} Ω GW ∼ k − 2 ( 1 + b ) = k − 3 , according to the CiT model, which exactly coincides with the bulk flow model. We show the resulting GW spectrum applying the CiT model to this case, for different values of α \alpha α and R ∗ H ∗ = 1 / ( k ∗ τ ∗ ) \mathcal{R}_\ast\mathcal{H}_\ast=1/(k_\ast\tau_\ast) R ∗ H ∗ = 1/ ( k ∗ τ ∗ ) in Figure 18 , where the normalized S c o m p ϕ ( K ) S^\phi_{\rm comp}(K) S comp ϕ ( K ) is defined as in § 0.5 B [cf. Eq. (161) ] and refers to a purely compressional spectrum, as expected for the scalar field gradients. Applying a similar analysis to the results from feebly interacting particles [98 ] , we would find that their UV spectrum is compatible with the CiT model if the anisotropic stress spectrum of the feebly interacting particle has a small-scale slope b Π c o m p = 3 b_\Pi^{\rm comp}=3 b Π comp = 3 . This corresponds to a flat scalar field gradients spectrum E L ϕ ( k ) ∼ k 0 E_L^\phi(k)\sim k^0 E L ϕ ( k ) ∼ k 0 at large k k k . Hence, the scalar field spectrum decays as E ϕ ( k ) ∼ k − 2 E_\phi(k)\sim k^{-2} E ϕ ( k ) ∼ k − 2 . Note that E L ϕ ( k ) E_L^\phi(k) E L ϕ ( k ) remains flat at large k k k until the damping scale is reached, i.e., for k ≪ k D k\ll k_D k ≪ k D . At larger k k k , the spectrum would decay exponentially, ensuring that the scalar field gradient energy converges.
Figure 18. Normalized spectral shape of the GW spectrum produced by a scalar field, whose gradients are described by a spectrum with slopes a = 5 a=5 a = 5 and b = 0.5 b=0.5 b = 0.5 , and smoothness parameters α = 6 / 17 \alpha=6/17 α = 6/17 (solid lines) and 2 2 2 (dashed lines). We use two values of the characteristic lengthscale R ∗ H ∗ = 10 − 2 \mathcal{R}_\ast\mathcal{H}_\ast=10^{-2} R ∗ H ∗ = 1 0 − 2 (red lines) and 10 − 1 10^{-1} 1 0 − 1 (blue lines) as in Figure 13 , and a sourcing duration δ τ f i n / τ ∗ = 100 \delta\tau_{\rm fin}/\tau_\ast=100 δ τ fin / τ ∗ = 100 .
The power spectrum of the scalar field in these studies of GW production should be computed for a direct comparison with the predictions of the CiT model before being able to conclude that the CiT model applies to these (or other) sources. This remains to be checked in future work.
Beyond the Constant-in-Time model: Coherent decay
In this section, we go beyond the CiT model and include the decay of the source. To take into account the decay, we restrict our treatment to sources for which the UETC of the anisotropic stresses is coherent in time [41 ] ,
E Π ( τ 1 , τ 2 , k ) = E Π ( τ 1 , k ) E Π ( τ 2 , k ) . (185) E_\Pi(\tau_1,\tau_2,k)=\sqrt{E_\Pi(\tau_1,k)E_\Pi(\tau_2,k)}. \tag*{(185)} E Π ( τ 1 , τ 2 , k ) = E Π ( τ 1 , k ) E Π ( τ 2 , k ) . (185)
This assumption allows us to introduce the time decay of the source while keeping the UETC as a separable function of τ 1 \tau_1 τ 1 and τ 2 \tau_2 τ 2 , simplifying the computation of the resulting GW spectrum. This model of the UETC holds for sources of GWs that have a long decorrelation timescale, as discussed in Sec. VI for sound waves or decaying MHD turbulence. Indeed, the authors in ref. [50 ] found that the velocity field followed Kraichnan decorrelation and the resulting GW spectrum was accurately described by the CiT model, such that only at relativistic speeds and large wave numbers the decorrelation has an impact on the resulting GW spectrum. In these cases, the decorrelation timescale is determined by the correlation velocity (c s c_s c s or u d c o r r u_{\rm dcorr} u dcorr ) and the characteristic wave number k ∗ k_\ast k ∗ [40 , 42 , 43 , 50 , 54 , 97 ] [5]. We show explicitly in a separate publication the comparison of the GW background obtained only accounting for decay with the one also including Kraichnan decorrelation on top of the decay of the source [101 ] .
Model
In the case where the UETC of the stress spectrum is coherent in time, as in Eq. (185) , the GW energy density spectrum, given in Eq. (24) , is
Ω G W ( τ , k ) = 3 E Π ∗ ( k ) F G W 0 Δ c o h 2 ( τ , k ) , (186) \Omega_{\rm GW}(\tau,k)=3E_\Pi^\ast(k)F_{\rm GW}^0\Delta_{\rm coh}^2(\tau,k), \tag*{(186)} Ω GW ( τ , k ) = 3 E Π ∗ ( k ) F GW 0 Δ coh 2 ( τ , k ) , (186)
where
Δ c o h ( τ , k ) = ∫ τ ∗ τ d τ ′ τ ′ E Π ( τ ′ , k ) E Π ∗ ( k ) cos k ( τ − τ ′ ) , (187) \Delta_{\rm coh}(\tau,k)
= \int_{\tau_\ast}^{\tau}\frac{\mathrm{d}\tau'}{\tau'}\sqrt{\frac{E_\Pi(\tau',k)}{E_\Pi^\ast(k)}}\cos k(\tau-\tau'), \tag*{(187)} Δ coh ( τ , k ) = ∫ τ ∗ τ τ ′ d τ ′ E Π ∗ ( k ) E Π ( τ ′ , k ) cos k ( τ − τ ′ ) , (187)
is the kernel found after incorporating the time dependence of the autocorrelator E Π ( τ , k ) E_\Pi(\tau,k) E Π ( τ , k ) , and where E Π ∗ ( k ) = E Π ( τ ∗ , k ) E_{\Pi}^{*}(k) = E_{\Pi}(\tau_{*}, k) E Π ∗ ( k ) = E Π ( τ ∗ , k ) is the initial spectrum of the anisotropic stresses.
The assumption of coherent correlation in time allows us to compute the GW spectrum from the time evolution of the stress autocorrelation spectrum, without the need to know the full form of the UETC at both times τ 1 \tau_{1} τ 1 and τ 2 \tau_{2} τ 2 . For non-linear decaying fields, e.g., decaying MHD turbulence, the decay of the peak wave number and the energy density of the sourcing field can be modeled with power laws in time [73 ] ,
k ∗ ( τ ) = k ∗ ( 1 + δ τ δ τ d c ) − s , (188) k_{*}(\tau) = k_{*}\left(1+\frac{\delta\tau}{\delta\tau_{\mathrm{dc}}}\right)^{-s}, \tag*{(188)} k ∗ ( τ ) = k ∗ ( 1 + δ τ dc δ τ ) − s , (188)
E ∗ ( τ ) = E ∗ ( 1 + δ τ δ τ d c ) − p , (189) E_{*}(\tau) = E_{*}\left(1+\frac{\delta\tau}{\delta\tau_{\mathrm{dc}}}\right)^{-p}, \tag*{(189)} E ∗ ( τ ) = E ∗ ( 1 + δ τ dc δ τ ) − p , (189)
where k ∗ k_{*} k ∗ and E ∗ E_{*} E ∗ refer to the peak wave number and amplitude of the source spectrum E ( k ) E(k) E ( k ) , δ τ = τ − τ ∗ \delta\tau=\tau-\tau_{*} δ τ = τ − τ ∗ , and δ τ d c = τ d c − τ ∗ \delta\tau_{\mathrm{dc}}=\tau_{\mathrm{dc}}-\tau_{*} δ τ dc = τ dc − τ ∗ is the typical decay timescale of the source. As an example, for MHD turbulence, taking the decay timescale as the eddy turnover time δ τ e \delta\tau_{\mathrm{e}} δ τ e (see also discussion in Sec. VI A) we have that, in the limit of infinite eddy turnover time, the turbulent fields do not decay, and we recover the CiT model.
The decay of the stress autocorrelation E Π ( τ , k ) E_{\Pi}(\tau,k) E Π ( τ , k ) , can be expressed as
E Π ( τ , k ) E Π ( τ ∗ , k ) = ζ Π r a t ( τ , k ) ( 1 + δ τ δ τ d c ) 2 m , (190) \frac{E_{\Pi}(\tau,k)}{E_{\Pi}(\tau_{*},k)}=\zeta_{\Pi}^{\mathrm{rat}}(\tau,k)\left(1+\frac{\delta\tau}{\delta\tau_{\mathrm{dc}}}\right)^{2m}, \tag*{(190)} E Π ( τ ∗ , k ) E Π ( τ , k ) = ζ Π rat ( τ , k ) ( 1 + δ τ dc δ τ ) 2 m , (190)
where m = 3 2 s − p m=\frac{3}{2}s-p m = 2 3 s − p is the power law decay of the stress spectrum, always taking negative values for typical decaying sources [37 , 104 ] , with the factors − 2 p -2p − 2 p and 3 s 3s 3 s respectively coming from the scaling of E Π E_{\Pi} E Π with E ∗ 2 E_{*}^{2} E ∗ 2 and K 3 K^{3} K 3 , where K ≡ k / k ∗ K\equiv k/k_{*} K ≡ k / k ∗ [see Eqs. (59) and (86)]. Here ζ Π r a t \zeta_{\Pi}^{\mathrm{rat}} ζ Π rat is the ratio of the normalized stress autocorrelation ζ Π ( K ) \zeta_{\Pi}(K) ζ Π ( K ) at time τ \tau τ to its value at τ ∗ \tau_{*} τ ∗ ,
ζ Π r a t ( τ , k ) ≡ ζ Π [ k / k ∗ ( τ ) ] ζ Π ( k / k ∗ ) . (191) \zeta_{\Pi}^{\mathrm{rat}}(\tau,k)\equiv\frac{\zeta_{\Pi}\left[k/k_{*}(\tau)\right]}{\zeta_{\Pi}(k/k_{*})}. \tag*{(191)} ζ Π rat ( τ , k ) ≡ ζ Π ( k / k ∗ ) ζ Π [ k / k ∗ ( τ ) ] . (191)
This yields the following expression for the GW spectrum,
Ω G W ( τ , k ) = 3 E Π ∗ ( k ) F G W 0 Δ p s 2 ( τ , k ; δ τ d c ) = 3 K 3 C A 2 Ω ~ r m s 4 ρ 2 ζ Π ∗ ( K ) F G W 0 Δ p s 2 ( τ , k ; δ τ d c ) , (192) \begin{aligned}
\Omega_{\mathrm{GW}}(\tau,k)=3E_{\Pi}^{*}(k)F_{\mathrm{GW}}^{0}\Delta_{\mathrm{ps}}^{2}(\tau,k;\delta\tau_{\mathrm{dc}}) \\
&=3K^{3}\frac{C}{A^{2}}\frac{\widetilde{\Omega}_{\mathrm{rms}}^{4}}{\rho^{2}}\zeta_{\Pi}^{*}(K)F_{\mathrm{GW}}^{0}\Delta_{\mathrm{ps}}^{2}(\tau,k;\delta\tau_{\mathrm{dc}}), \tag*{(192)}
\end{aligned} Ω GW ( τ , k ) = 3 E Π ∗ ( k ) F GW 0 Δ ps 2 ( τ , k ; δ τ dc ) = 3 K 3 A 2 C ρ 2 Ω rms 4 ζ Π ∗ ( K ) F GW 0 Δ ps 2 ( τ , k ; δ τ dc ) , (192)
where the function Δ p s ( τ , k ; δ τ d c ) \Delta_{\mathrm{ps}}(\tau,k;\delta\tau_{\mathrm{dc}}) Δ ps ( τ , k ; δ τ dc ) is defined as
Δ p s ( τ , k ; δ τ d c ) = ∫ τ ∗ τ d τ 1 τ 1 ( 1 + δ τ 1 δ τ d c ) m cos k ( τ − τ 1 ) ζ Π r a t ( τ 1 , k ) . (193) \Delta_{\mathrm{ps}}(\tau,k;\delta\tau_{\mathrm{dc}})=\int_{\tau_{*}}^{\tau}\frac{\mathrm{d}\tau_{1}}{\tau_{1}}\left(1+\frac{\delta\tau_{1}}{\delta\tau_{\mathrm{dc}}}\right)^{m}\cos k(\tau-\tau_{1})\sqrt{\zeta_{\Pi}^{\mathrm{rat}}(\tau_{1},k)}. \tag*{(193)} Δ ps ( τ , k ; δ τ dc ) = ∫ τ ∗ τ τ 1 d τ 1 ( 1 + δ τ dc δ τ 1 ) m cos k ( τ − τ 1 ) ζ Π rat ( τ 1 , k ) . (193)
Notice that, since we have included the decay of the source, it is not necessary in this case to define a final sourcing time τ f i n \tau_{\mathrm{fin}} τ fin , as we did for the CiT model (see § 0.5 ).
However, if we included a final sourcing time τ f i n \tau_{\mathrm{fin}} τ fin , then we would define a Δ p s \Delta_{\mathrm{ps}} Δ ps kernel where the upper limit of integration in Eq. (193) would be replaced by τ ˉ = min ( τ , τ f i n ) \bar{\tau}=\min(\tau,\tau_{\mathrm{fin}}) τ ˉ = min ( τ , τ fin ) and we would recover the CiT model result either in the case p = s = 0 p=s=0 p = s = 0 or in the limit of large decay timescale δ τ d c / τ ∗ → ∞ \delta\tau_{\mathrm{dc}}/\tau_{*}\to\infty δ τ dc / τ ∗ → ∞ , i.e.,
Δ ( τ , k ) = Δ 00 ( τ , k ; δ τ d c ) = lim δ τ d c / τ ∗ → ∞ Δ p s ( τ , k ; δ τ d c ) \Delta(\tau,k)=\Delta_{00}(\tau,k;\delta\tau_{\mathrm{dc}})=\lim_{\delta\tau_{\mathrm{dc}}/\tau_{*}\to\infty}\Delta_{\mathrm{ps}}(\tau,k;\delta\tau_{\mathrm{dc}}) Δ ( τ , k ) = Δ 00 ( τ , k ; δ τ dc ) = δ τ dc / τ ∗ → ∞ lim Δ ps ( τ , k ; δ τ dc )
[see Eq. (106)]. With such definition of Δ p s \Delta_{\mathrm{ps}} Δ ps , if we considered the present-day observable τ = τ 0 \tau=\tau_{0} τ = τ 0 and averaged over fast oscillations, we would find
Δ p s , a v 2 = 1 2 [ I sin 2 ( τ ˉ , k ) + I cos 2 ( τ ˉ , k ) ] , (194) \Delta_{\mathrm{ps,av}}^{2}=\frac{1}{2}\left[\mathcal{I}_{\sin}^{2}(\bar{\tau},k)+\mathcal{I}_{\cos}^{2}(\bar{\tau},k)\right], \tag*{(194)} Δ ps , av 2 = 2 1 [ I s i n 2 ( τ ˉ , k ) + I c o s 2 ( τ ˉ , k ) ] , (194)
where
I cos ( τ ˉ , k ) = ∫ τ ∗ τ ˉ d τ 1 τ 1 ( 1 + δ τ 1 δ τ d c ) m cos k τ 1 ζ Π r a t ( τ 1 , k ) , (195) \mathcal{I}_{\cos}(\bar{\tau},k)=\int_{\tau_{*}}^{\bar{\tau}}\frac{\mathrm{d}\tau_{1}}{\tau_{1}}\left(1+\frac{\delta\tau_{1}}{\delta\tau_{\mathrm{dc}}}\right)^{m}\cos k\tau_{1}\sqrt{\zeta_{\Pi}^{\mathrm{rat}}(\tau_{1},k)}, \tag*{(195)} I c o s ( τ ˉ , k ) = ∫ τ ∗ τ ˉ τ 1 d τ 1 ( 1 + δ τ dc δ τ 1 ) m cos k τ 1 ζ Π rat ( τ 1 , k ) , (195)
I sin ( τ ˉ , k ) = ∫ τ ∗ τ ˉ d τ 1 τ 1 ( 1 + δ τ 1 δ τ d c ) m sin k τ 1 ζ Π r a t ( τ 1 , k ) . (196) \mathcal{I}_{\sin}(\bar{\tau},k)=\int_{\tau_{*}}^{\bar{\tau}}\frac{\mathrm{d}\tau_{1}}{\tau_{1}}\left(1+\frac{\delta\tau_{1}}{\delta\tau_{\mathrm{dc}}}\right)^{m}\sin k\tau_{1}\sqrt{\zeta_{\Pi}^{\mathrm{rat}}(\tau_{1},k)}. \tag*{(196)} I s i n ( τ ˉ , k ) = ∫ τ ∗ τ ˉ τ 1 d τ 1 ( 1 + δ τ dc δ τ 1 ) m sin k τ 1 ζ Π rat ( τ 1 , k ) . (196)
In the following, except for App. § 1.4.1 , we will consider Δ p s \Delta_{\mathrm{ps}} Δ ps defined as in Eq. (193) . In general, we find that for small enough values of s s s the effect in Δ p s \Delta_{\mathrm{ps}} Δ ps from the ζ Π r a t \zeta_{\Pi}^{\mathrm{rat}} ζ Π rat term can be ignored [see App. E] and the resulting Δ p s ( τ , k ; δ τ d c ) \Delta_{\mathrm{ps}}(\tau,k;\delta\tau_{\mathrm{dc}}) Δ ps ( τ , k ; δ τ dc ) only depends on m m m and δ τ d c \delta\tau_{\mathrm{dc}} δ τ dc . This happens when the timescale of decay of ζ Π r a t \zeta_{\Pi}^{\mathrm{rat}} ζ Π rat is larger than the saturation timescale of the GW spectrum, which happens when [cf. Eq. (E1)]
s < 1 + ln ( k ∗ τ ∗ ) ln ( τ / τ ∗ ) for τ > τ ∗ . (197) s<1+\frac{\ln(k_{*}\tau_{*})}{\ln(\tau/\tau_{*})}\qquad\text{for }\tau>\tau_{*}. \tag*{(197)} s < 1 + ln ( τ / τ ∗ ) ln ( k ∗ τ ∗ ) for τ > τ ∗ . (197)
Hence, from now on we will safely restrict to values s ≤ 1 s\leq1 s ≤ 1 , which allow us to ignore this term and make the following approximation
Δ p s ( τ , k ; δ τ d c ) ≃ ∫ τ ∗ τ d τ 1 τ 1 ( 1 + δ τ 1 δ τ d c ) m cos k ( τ − τ 1 ) ≡ Δ m ( τ , k ; δ τ d c ) . (198) \begin{aligned}
\Delta_{\mathrm{ps}}(\tau,k;\delta\tau_{\mathrm{dc}})\simeq\int_{\tau_{*}}^{\tau}\frac{\mathrm{d}\tau_{1}}{\tau_{1}}\left(1+\frac{\delta\tau_{1}}{\delta\tau_{\mathrm{dc}}}\right)^{m}\cos k(\tau-\tau_{1}) \\
&\equiv\Delta_{m}(\tau,k;\delta\tau_{\mathrm{dc}}). \tag*{(198)}
\end{aligned} Δ ps ( τ , k ; δ τ dc ) ≃ ∫ τ ∗ τ τ 1 d τ 1 ( 1 + δ τ dc δ τ 1 ) m cos k ( τ − τ 1 ) ≡ Δ m ( τ , k ; δ τ dc ) . (198)
We compare the above defined function for different choices of δ τ d c / τ ∗ \delta\tau_{\rm dc}/\tau_\ast δ τ dc / τ ∗ in Figure 19 where we see that, as δ τ d c / τ ∗ \delta\tau_{\rm dc}/\tau_\ast δ τ dc / τ ∗ increases, Δ m ( τ , k ; δ τ d c ) \Delta_m(\tau,k;\delta\tau_{\rm dc}) Δ m ( τ , k ; δ τ dc ) approaches the function Δ 2 ( τ < τ f i n , k ) \Delta^2(\tau<\tau_{\rm fin},k) Δ 2 ( τ < τ fin , k ) in the CiT model [cf. Eq. (132) ].
Figure 19. Spectra Δ m 2 ( τ , k ; δ τ d c ) \Delta_m^2(\tau,k;\delta\tau_{\rm dc}) Δ m 2 ( τ , k ; δ τ dc ) [cf. Eq. (198) ] for different choices of δ τ d c / τ ∗ \delta\tau_{\rm dc}/\tau_\ast δ τ dc / τ ∗ for a negative value of m = − 1 3 m=-\frac{1}{3} m = − 3 1 . We see that as the decay timescale increases we recover, as expected, the result for Δ 2 ( τ < τ f i n , k ) \Delta^2(\tau<\tau_{\rm fin},k) Δ 2 ( τ < τ fin , k ) in the CiT model [cf. Eq. (132) ]. For shorter δ τ d c / τ ∗ \delta\tau_{\rm dc}/\tau_\ast δ τ dc / τ ∗ the spectra Δ m 2 ( τ , k ; δ τ d c ) \Delta_m^2(\tau,k;\delta\tau_{\rm dc}) Δ m 2 ( τ , k ; δ τ dc ) deviate from the CiT result for values of k < 1 / δ τ d c k<1/\delta\tau_{\rm dc} k < 1/ δ τ dc (vertical dashed lines), while the transition from the k → 0 k\to0 k → 0 to the k → ∞ k\to\infty k → ∞ regime starts at k ≃ 1 / δ τ k\simeq1/\delta\tau k ≃ 1/ δ τ (vertical dotted line) for all values of δ τ d c \delta\tau_{\rm dc} δ τ dc .
Moreover, in the upper panel of Figure 20 we show the large-scale limit of Δ m 2 \Delta_m^2 Δ m 2 for different choices of m m m and different values of δ τ d c / τ ∗ \delta\tau_{\rm dc}/\tau_\ast δ τ dc / τ ∗ , and we provide a fit (dashed black lines) of Δ m 2 / Δ 2 \Delta_m^2/\Delta^2 Δ m 2 / Δ 2 , tested for all values of δ τ d c / τ ∗ ∈ [ 10 − 5 , 10 5 ] \delta\tau_{\rm dc}/\tau_\ast\in[10^{-5},10^5] δ τ dc / τ ∗ ∈ [ 1 0 − 5 , 1 0 5 ] and − m ∈ [ 0.1 , 5 ] -m\in[0.1,5] − m ∈ [ 0.1 , 5 ] , in the form of the following broken power law
Figure 20. Upper panel: Amplitude in the k → 0 k\to0 k → 0 limit of Δ m 2 ( τ , k ; δ τ d c ) / Δ 2 ( τ , k ) \Delta_m^2(\tau,k;\delta\tau_{\rm dc})/\Delta^2(\tau,k) Δ m 2 ( τ , k ; δ τ dc ) / Δ 2 ( τ , k ) (solid lines) for τ / τ ∗ = 10 2 \tau/\tau_\ast=10^2 τ / τ ∗ = 1 0 2 , different values of δ τ d c \delta\tau_{\rm dc} δ τ dc and different choices of m m m . For all values of − m ∈ [ 0.1 , 5 ] -m\in[0.1,5] − m ∈ [ 0.1 , 5 ] we find the fit r 2 ( δ τ d c / τ ∗ ; x d c , α d c , β d c ) r^2(\delta\tau_{\rm dc}/\tau_\ast;x_{\rm dc},\alpha_{\rm dc},\beta_{\rm dc}) r 2 ( δ τ dc / τ ∗ ; x dc , α dc , β dc ) of Eq. (199) , which is overplotted in dashed black lines. Lower panel: Values of the fitting parameters for − m ∈ [ 0.1 , 5 ] -m\in[0.1,5] − m ∈ [ 0.1 , 5 ] .
r 2 ( δ τ d c / τ ∗ ; x d c , α d c , β d c ) = [ 1 + ( x d c δ τ d c / τ ∗ ) α d c ] − β d c α d c (199) r^2(\delta\tau_{\rm dc}/\tau_\ast;x_{\rm dc},\alpha_{\rm dc},\beta_{\rm dc})=\left[1+\left(\frac{x_{\rm dc}}{\delta\tau_{\rm dc}/\tau_\ast}\right)^{\alpha_{\rm dc}}\right]^{-\frac{\beta_{\rm dc}}{\alpha_{\rm dc}}} \tag*{(199)} r 2 ( δ τ dc / τ ∗ ; x dc , α dc , β dc ) = [ 1 + ( δ τ dc / τ ∗ x dc ) α dc ] − α dc β dc (199)
with the fitting parameters x d c x_{\rm dc} x dc , α d c \alpha_{\rm dc} α dc , β d c \beta_{\rm dc} β dc given in the lower panel of Figure 20 .
We can obtain an analytical expression for the resulting kernel Δ m 2 \Delta_m^2 Δ m 2 in the particular case in which δ τ d c = τ ∗ \delta\tau_{\rm dc}=\tau_\ast δ τ dc = τ ∗ which, in radiation-domination, is equivalent to δ τ d c = H ∗ − 1 \delta\tau_{\rm dc}=\mathcal{H}_\ast^{-1} δ τ dc = H ∗ − 1 . In this case Eq. (198) reduces to
Δ m ( τ , k ; δ τ d c = τ ∗ ) = τ ∗ − m ∫ τ ∗ τ τ 1 m − 1 cos k ( τ − τ 1 ) d τ 1 ≡ Δ m ( τ , k ) . (200) \begin{aligned}
\Delta_m(\tau,k;\delta\tau_{\rm dc}=\tau_\ast)=\tau_\ast^{-m}\int_{\tau_\ast}^{\tau}\tau_1^{m-1}\cos k(\tau-\tau_1)\,\mathrm{d}\tau_1 \\
&\equiv\Delta_m(\tau,k). \tag*{(200)}
\end{aligned} Δ m ( τ , k ; δ τ dc = τ ∗ ) = τ ∗ − m ∫ τ ∗ τ τ 1 m − 1 cos k ( τ − τ 1 ) d τ 1 ≡ Δ m ( τ , k ) . (200)
The function Δ m ( τ , k ) \Delta_m(\tau,k) Δ m ( τ , k ) is studied in App. § 1.4.1 ,7 where we show that it can be computed in terms of the upper incomplete gamma function $\Gamma(x,y)\equiv ∫ y ∞ t x − 1 e − t d t \int_y^\infty t^{x-1}e^{-t}\,\mathrm{d}t ∫ y ∞ t x − 1 e − t d t as [see Eq. (335) ]
Δ m ( τ , k ) = − 1 2 τ ∗ − m ( i k ) − m × { e i k τ [ Γ ( m , i k τ ) − Γ ( m , i k τ ∗ ) ] + e − i k τ e i π m [ Γ ( m , − i k τ ) − Γ ( m , − i k τ ∗ ) ] } . (201) \begin{aligned}
\Delta_m(\tau,k)={}&-\frac{1}{2}\tau_*^{-m}(ik)^{-m}\\
&\times\left\{e^{ik\tau}\left[\Gamma(m,ik\tau)-\Gamma(m,ik\tau_*)\right]\right.\\
&\left.\qquad+e^{-ik\tau}e^{i\pi m}\left[\Gamma(m,-ik\tau)-\Gamma(m,-ik\tau_*)\right]\right\}. \tag*{(201)}
\end{aligned} Δ m ( τ , k ) = − 2 1 τ ∗ − m ( ik ) − m × { e ik τ [ Γ ( m , ik τ ) − Γ ( m , ik τ ∗ ) ] + e − ik τ e iπ m [ Γ ( m , − ik τ ) − Γ ( m , − ik τ ∗ ) ] } . (201)
For the cases in which m m m is an integer, we can further simplify the analytical expression. We give in App. § 1.4.1 a method to find the analytical expression for any m m m and, in particular, the expressions for Δ − 1 \Delta_{-1} Δ − 1 and Δ − 2 \Delta_{-2} Δ − 2 are explicitly given in Eqs. (D22) and (D23),
Δ − 1 ( τ , k ) = − k τ ∗ cos k τ [ Si ( k τ ) − Si ( k τ ∗ ) ] + k τ ∗ sin k τ [ Ci ( k τ ) − Ci ( k τ ∗ ) ] − τ ∗ τ + cos k ( τ − τ ∗ ) , (202) \begin{aligned}
\Delta_{-1}(\tau,k)={}&-k\tau_*\cos k\tau\left[\operatorname{Si}(k\tau)-\operatorname{Si}(k\tau_*)\right]\\
&+k\tau_*\sin k\tau\left[\operatorname{Ci}(k\tau)-\operatorname{Ci}(k\tau_*)\right]\\
&-\frac{\tau_*}{\tau}+\cos k(\tau-\tau_*), \tag*{(202)}
\end{aligned} Δ − 1 ( τ , k ) = − k τ ∗ cos k τ [ Si ( k τ ) − Si ( k τ ∗ ) ] + k τ ∗ sin k τ [ Ci ( k τ ) − Ci ( k τ ∗ ) ] − τ τ ∗ + cos k ( τ − τ ∗ ) , (202)
Δ − 2 ( τ , k ) = − k 2 τ ∗ 2 2 cos k τ [ Ci ( k τ ) − Ci ( k τ ∗ ) ] − k 2 τ ∗ 2 2 sin k τ [ Si ( k τ ) − Si ( k τ ∗ ) ] + 1 2 [ cos k ( τ − τ ∗ ) + k τ ∗ sin k ( τ − τ ∗ ) ] − 1 2 τ ∗ 2 τ 2 . (203) \begin{aligned}
\Delta_{-2}(\tau,k)={}&-\frac{k^2\tau_*^2}{2}\cos k\tau\left[\operatorname{Ci}(k\tau)-\operatorname{Ci}(k\tau_*)\right]\\
&-\frac{k^2\tau_*^2}{2}\sin k\tau\left[\operatorname{Si}(k\tau)-\operatorname{Si}(k\tau_*)\right]\\
&+\frac{1}{2}\left[\cos k(\tau-\tau_*)+k\tau_*\sin k(\tau-\tau_*)\right]-\frac{1}{2}\frac{\tau_*^2}{\tau^2}. \tag*{(203)}
\end{aligned} Δ − 2 ( τ , k ) = − 2 k 2 τ ∗ 2 cos k τ [ Ci ( k τ ) − Ci ( k τ ∗ ) ] − 2 k 2 τ ∗ 2 sin k τ [ Si ( k τ ) − Si ( k τ ∗ ) ] + 2 1 [ cos k ( τ − τ ∗ ) + k τ ∗ sin k ( τ − τ ∗ ) ] − 2 1 τ 2 τ ∗ 2 . (203)
For comparison, the functions Δ 1 \Delta_1 Δ 1 and Δ 2 \Delta_2 Δ 2 are given in Eqs. (204) and (205) ,
Δ 1 ( τ , k ) = 1 k τ ∗ sin k ( τ − τ ∗ ) , (204) \Delta_1(\tau,k)=\frac{1}{k\tau_*}\sin k(\tau-\tau_*), \tag*{(204)} Δ 1 ( τ , k ) = k τ ∗ 1 sin k ( τ − τ ∗ ) , (204)
Δ 2 ( τ , k ) = 1 k 2 τ ∗ 2 [ 1 − cos k ( τ − τ ∗ ) ] + 1 k τ ∗ sin k ( τ − τ ∗ ) . (205) \Delta_2(\tau,k)=\frac{1}{k^2\tau_*^2}\left[1-\cos k(\tau-\tau_*)\right]+\frac{1}{k\tau_*}\sin k(\tau-\tau_*). \tag*{(205)} Δ 2 ( τ , k ) = k 2 τ ∗ 2 1 [ 1 − cos k ( τ − τ ∗ ) ] + k τ ∗ 1 sin k ( τ − τ ∗ ) . (205)
Note that for positive values of m m m , the autocorrelation stress spectrum increases with time, which is not physically compatible with a decaying source (e.g., in the case of decaying MHD turbulence). However, these cases might be relevant for the phase of turbulence production as long as the UETC follows the coherent approximation.
In App. § 1.4.1 we study the time evolution of the function Δ m ( τ , k ) \Delta_m(\tau,k) Δ m ( τ , k ) in the asymptotic limits k → 0 k\to0 k → 0 and k → ∞ k\to\infty k → ∞ .
Let us first focus on the limit k → 0 k\to0 k → 0 . For negative integers − n -n − n the function Δ − n 2 \Delta_{-n}^2 Δ − n 2 takes the asymptotic form [see Eq. (364) ]
lim k → 0 Δ − n 2 ( τ , k ) = 1 n 2 [ 1 − ( τ τ ∗ ) − n ] 2 . (206) \lim_{k\to0}\Delta_{-n}^2(\tau,k)=\frac{1}{n^2}\left[1-\left(\frac{\tau}{\tau_*}\right)^{-n}\right]^2. \tag*{(206)} k → 0 lim Δ − n 2 ( τ , k ) = n 2 1 [ 1 − ( τ ∗ τ ) − n ] 2 . (206)
The time evolution of Δ m ( τ , k ) \Delta_m(\tau,k) Δ m ( τ , k ) for different values of m m m in the limit k → 0 k\to0 k → 0 is shown in Figure 21 . For the values m = − n < 0 m=-n<0 m = − n < 0 , we find numerically that the relation Eq. (206) holds even for non-integer n n n .
Figure 21. Time evolution of the integrals Δ m 2 ( τ , k ) \Delta_m^2(\tau,k) Δ m 2 ( τ , k ) in the limit k → 0 k \to0 k → 0 for different values of m m m . The time evolution for m = − 1 , − 2 , m=-1,-2, m = − 1 , − 2 , and − 3 -3 − 3 can be analytically shown to follow the function in Eq. (206) . We find numerically that for non-integer values of m = − n < 0 m=-n<0 m = − n < 0 , this expression also holds (e.g., for m = − 0.25 , − 0.5 m=-0.25,-0.5 m = − 0.25 , − 0.5 ). The function Δ 1 2 \Delta_1^2 Δ 1 2 evolves as ( H ∗ δ τ ) 2 (\mathcal{H}_*\delta\tau)^2 ( H ∗ δ τ ) 2 , while for values of m ∈ ( 0 , 1 ) m\in(0,1) m ∈ ( 0 , 1 ) (e.g., 0.25 , 0.5 , 0.25,0.5, 0.25 , 0.5 , and 0.75 0.75 0.75 ) Δ m 2 \Delta_m^2 Δ m 2 diverges for large δ τ \delta\tau δ τ transitioning towards ln 2 ( H ∗ τ ) \ln^2(\mathcal{H}_*\tau) ln 2 ( H ∗ τ ) , which corresponds to the evolution of Δ 2 \Delta^2 Δ 2 in the CiT model [see Eq. (139) ]. For larger values of m > 1 m>1 m > 1 , the functions Δ m 2 \Delta_m^2 Δ m 2 diverge faster than δ τ 2 \delta\tau^2 δ τ 2 .
In the limit of large times τ ≫ τ ∗ \tau\gg\tau_* τ ≫ τ ∗ , Δ − n 2 ( τ , k → 0 ) \Delta_{-n}^2(\tau,k\to0) Δ − n 2 ( τ , k → 0 ) approaches the value [see Eq. (365) ]
lim k → 0 Δ − n 2 ( τ , k ) = 1 n 2 . (207) \lim_{k\to0}\Delta_{-n}^2(\tau,k)=\frac{1}{n^2}. \tag*{(207)} k → 0 lim Δ − n 2 ( τ , k ) = n 2 1 . (207)
This limit is shown in Figure 21 at large δ τ / τ ∗ \delta\tau/\tau_* δ τ / τ ∗ .
The values of the limit k → 0 k\to0 k → 0 for Δ m ( τ , k ) \Delta_m(\tau,k) Δ m ( τ , k ) with m = 1 , 2 m=1,2 m = 1 , 2 are instead [see Eqs. (D40) and (D41)]
lim k → 0 Δ 1 2 ( τ , k ) = ( δ τ / τ ∗ ) 2 , (208) \lim_{k\to0}\Delta_1^2(\tau,k)=(\delta\tau/\tau_*)^2, \tag*{(208)} k → 0 lim Δ 1 2 ( τ , k ) = ( δ τ / τ ∗ ) 2 , (208)
lim k → 0 Δ 2 2 ( τ , k ) = 1 4 ( δ τ / τ ∗ ) 4 + ( δ τ / τ ∗ ) 3 + ( δ τ / τ ∗ ) 2 . (209) \lim_{k\to0}\Delta_2^2(\tau,k)=\frac{1}{4}(\delta\tau/\tau_*)^4+(\delta\tau/\tau_*)^3+(\delta\tau/\tau_*)^2. \tag*{(209)} k → 0 lim Δ 2 2 ( τ , k ) = 4 1 ( δ τ / τ ∗ ) 4 + ( δ τ / τ ∗ ) 3 + ( δ τ / τ ∗ ) 2 . (209)
The function Δ 1 2 ( τ , k → 0 ) \Delta_1^2(\tau,k\to0) Δ 1 2 ( τ , k → 0 ) evolves quadratically with δ τ \delta\tau δ τ for all times, hence, it diverges faster than ln 2 ( H ∗ τ ) \ln^2(\mathcal{H}_*\tau) ln 2 ( H ∗ τ ) , as expected. On the other hand, the function Δ 2 2 ( τ , k → 0 ) \Delta_2^2(\tau,k\to0) Δ 2 2 ( τ , k → 0 ) evolves as δ τ 2 \delta\tau^2 δ τ 2 at small δ τ \delta\tau δ τ but as time increases, the term δ τ 4 \delta\tau^4 δ τ 4 dominates, making the function diverge even faster at large times. We find numerically that for values of m ∈ ( 0 , 1 ) m\in(0,1) m ∈ ( 0 , 1 ) the dynamical evolution of Δ m 2 ( τ , k → 0 ) \Delta_m^2(\tau,k\to0) Δ m 2 ( τ , k → 0 ) smoothly transitions from the CiT value ln 2 ( H ∗ δ τ ) \ln^2(\mathcal{H}_*\delta\tau) ln 2 ( H ∗ δ τ ) to ( H ∗ δ τ ) 2 (\mathcal{H}_*\delta\tau)^2 ( H ∗ δ τ ) 2 . In particular, the values m = 0.25 m=0.25 m = 0.25 , 0.5 0.5 0.5 , and 0.75 0.75 0.75 are shown in Figure 21 . Similarly, for values m ∈ ( 1 , 2 ) m\in(1,2) m ∈ ( 1 , 2 ) we find functions between Eqs. (208) and (209) , which diverge faster than δ τ 2 \delta\tau^2 δ τ 2 at late times.
In general, at wave numbers k k k , we find that each mode of the spectrum Δ m 2 ( τ , k ) \Delta_m^2(\tau,k) Δ m 2 ( τ , k ) evolves following the dynamical behavior shown in Figure 21 until the expected saturated value is reached at δ τ ∼ 1 / k \delta\tau\sim1/k δ τ ∼ 1/ k [1]. Hence, the expected saturated spectra Δ m 2 ( τ , k ) \Delta_m^2(\tau,k) Δ m 2 ( τ , k ) can be obtained substituting δ τ = 1 / k \delta\tau=1/k δ τ = 1/ k in Eq. (206) for Δ − n \Delta_{-n} Δ − n and in Eqs. (177) for Δ m \Delta_m Δ m with m = 1 , 2 m=1,2 m = 1 , 2 , thus giving
Δ − n 2 ( τ , k ) = 1 n 2 [ 1 − 1 ( 1 + H ∗ / k ) n ] 2 ∀ n > 0 , \Delta_{-n}^2(\tau,k)=\frac{1}{n^2}\left[1-\frac{1}{(1+\mathcal{H}_*/k)^n}\right]^2\quad\forall n>0, Δ − n 2 ( τ , k ) = n 2 1 [ 1 − ( 1 + H ∗ / k ) n 1 ] 2 ∀ n > 0 ,
Δ 1 2 ( τ , k ) = ( H ∗ / k ) 2 , \Delta_1^2(\tau,k)=(\mathcal{H}_*/k)^2, Δ 1 2 ( τ , k ) = ( H ∗ / k ) 2 ,
Δ 2 2 ( τ , k ) = 1 4 ( H ∗ / k ) 4 + H ∗ 3 / k 3 + H ∗ 2 / k 2 . (210) \Delta_2^2(\tau,k)=\frac{1}{4}(\mathcal{H}_*/k)^4+\mathcal{H}_*^3/k^3+\mathcal{H}_*^2/k^2. \tag*{(210)} Δ 2 2 ( τ , k ) = 4 1 ( H ∗ / k ) 4 + H ∗ 3 / k 3 + H ∗ 2 / k 2 . (210)
to be compared to the CiT result Δ 2 ( τ , k ) = ln 2 ( 1 + H ∗ / k ) \Delta^2(\tau,k)=\ln^2(1+\mathcal{H}_*/k) Δ 2 ( τ , k ) = ln 2 ( 1 + H ∗ / k ) , which was presented in ref. [1]. The spectra
Δ m 2 ( τ , k ) \Delta_m^2(\tau,k) Δ m 2 ( τ , k ) at large times, such that all modes k > 1 / δ τ k>1/\delta\tau k > 1/ δ τ have saturated, are shown in Figure 22 , compared to the envelope computed using Eq. (210) . From Eq. (210) , we find that in the limit k / H ∗ ≫ 1 k/\mathcal{H}_*\gg1 k / H ∗ ≫ 1 , the envelope tends to k − 2 k^{-2} k − 2 for all n n n , which is also seen in Figure 22 . We show in App. D that the behavior of Δ m 2 \Delta_m^2 Δ m 2 at large k → ∞ k\to\infty k → ∞ is, in fact, [cf. Eq. (370) ]
Figure 22. Spectra Δ m 2 ( τ , k ) \Delta_m^2(\tau,k) Δ m 2 ( τ , k ) for different values of m m m at large times (in the figure H ∗ τ = 10 3 \mathcal{H}_*\tau=10^3 H ∗ τ = 1 0 3 ) compared to their envelope, given in Eq. (210) for the cases m ≠ 0 m\ne0 m = 0 and in Eq. (140) for the CiT case m = 0 m=0 m = 0 .
Δ m ( τ , k ) ≃ k → ∞ 1 k τ ∗ sin k ( τ − τ ∗ ) , (211) \Delta_m(\tau,k) \underset{k\to\infty}{\simeq} \frac{1}{k\tau_*}\sin k(\tau-\tau_*), \tag*{(211)} Δ m ( τ , k ) k → ∞ ≃ k τ ∗ 1 sin k ( τ − τ ∗ ) , (211)
for any real value of m m m , which coincides with the result in the CiT model [cf. Eq. (322) ]. Hence we see that the envelope of Δ m 2 \Delta_m^2 Δ m 2 in the limit k → ∞ k\to\infty k → ∞ is ( H ∗ / k ) 2 (\mathcal{H}_*/k)^2 ( H ∗ / k ) 2 for any value of m m m .
In Figure 23 , we show the spectra Δ m 2 ( τ , k ) \Delta_m^2(\tau,k) Δ m 2 ( τ , k ) at different times for m = − 1 / 4 m=-1/4 m = − 1/4 , which can correspond, e.g., to the turbulent decay expected for nonhelical magnetic fields, with p = 1 p=1 p = 1 and s = 1 2 s=\frac{1}{2} s = 2 1 [83 ] . The spectra are compared with those obtained in the CiT model, given in Eq. (153) .
Figure 23. Spectra Δ m 2 ( τ , k ) \Delta_m^2(\tau,k) Δ m 2 ( τ , k ) , for m = − 1 / 4 m=-1/4 m = − 1/4 , at different times H ∗ δ τ = { 10 − 1 , 1 , 10 , 10 2 , 10 4 } \mathcal{H}_*\delta\tau=\{10^{-1},1,10,10^2,10^4\} H ∗ δ τ = { 1 0 − 1 , 1 , 10 , 1 0 2 , 1 0 4 } (colored solid lines). For comparison, we show the corresponding values of the envelope obtained using the CiT model Eq. (153) (dashed black line) and the envelope predicted using Eq. (210) (solid black line).
Gravitational wave template
In this section we have generalized the CiT model of § 0.5 to the case of coherent UETCs, described as in Eq. (185) . We have furthermore assumed that we can describe the decay of the source autocorrelation function Π ( τ , k ) \Pi(\tau,k) Π ( τ , k ) through the decay of its amplitude and peak wave number, as in Eqs. (159). This has led to the following expression for the Gravitational Wave spectrum [cf. Eq. (192) ]
Ω G W ( τ , k ) = 3 K 3 C A 2 Φ ~ r m s 4 ρ ˉ 2 S Π ⋆ ( K ) F G W 0 Δ p s 2 ( τ , k ; δ τ d c ) . (212) \Omega_{\rm GW}(\tau,k)=3K^3\frac{\mathcal{C}}{\mathcal{A}^2}\frac{\tilde{\Phi}_{\rm rms}^4}{\bar{\rho}^{2}}\mathcal{S}_{\Pi}^{\star}(K)F_{\rm GW}^{0}\Delta_{ps}^{2}(\tau,k;\delta\tau_{\rm dc}) . \tag*{(212)} Ω GW ( τ , k ) = 3 K 3 A 2 C ρ ˉ 2 Φ ~ rms 4 S Π ⋆ ( K ) F GW 0 Δ p s 2 ( τ , k ; δ τ dc ) . (212)
Here Δ p s 2 ( τ , k ; δ τ d c ) \Delta_{ps}^{2}(\tau,k;\delta\tau_{\rm dc}) Δ p s 2 ( τ , k ; δ τ dc ) , given in Eq. (193) , is a function which depends on both the decay exponents of the amplitude and peak wave number of E Π ( τ , k ) E_{\Pi}(\tau,k) E Π ( τ , k ) , respectively p p p and s s s , and on the decay timescale δ τ d c \delta\tau_{\rm dc} δ τ dc . In the case of moderate values of s < 1 s<1 s < 1 , as we show in App. E (see Eq. (372) and Figure 32 ) the dependence on ζ Π r a t \zeta_{\Pi}^{\rm rat} ζ Π rat in Eq. (193) can be neglected, which leads to the approximations Δ p s 2 ( τ , k ; δ τ d c ) ≃ Δ m 2 ( τ , k ; δ τ d c ) \Delta_{ps}^{2}(\tau,k;\delta\tau_{\rm dc})\simeq\Delta_{m}^{2}(\tau,k;\delta\tau_{\rm dc}) Δ p s 2 ( τ , k ; δ τ dc ) ≃ Δ m 2 ( τ , k ; δ τ dc ) [cf. Eq. (198) ], where m = 2 3 s − p m=\frac{2}{3}s-p m = 3 2 s − p . The function Δ m 2 ( τ , k ; δ τ d c ) \Delta_{m}^{2}(\tau,k;\delta\tau_{\rm dc}) Δ m 2 ( τ , k ; δ τ dc ) is shown in Figure 19 for different values of δ τ d c \delta\tau_{\rm dc} δ τ dc . In the case δ τ d c = τ ⋆ \delta\tau_{\rm dc}=\tau_{\star} δ τ dc = τ ⋆ , moreover, the function Δ m 2 \Delta_{m}^{2} Δ m 2 , given in Eq. (200) , can be computed analytically, and we derive in App. D its analytical expression, given in Eq. (201) .
Figure 32. Time evolution of the normalized stress spectrum ζ Π r a t ( τ , k ) = ζ Π ( τ , k ) / ζ Π ( τ ∗ , k ) \sqrt{\zeta_{\Pi}^{\mathrm{rat}}(\tau,k)}=\sqrt{\zeta_{\Pi}(\tau,k)/\zeta_{\Pi}(\tau_\ast,k)} ζ Π rat ( τ , k ) = ζ Π ( τ , k ) / ζ Π ( τ ∗ , k ) [see Eqs. (190) and (191) ] for α = 2 \alpha=2 α = 2 (upper panels) and α = 6 / 17 \alpha=6/17 α = 6/17 (lower panels), when the wave number of the peak evolves as k ∗ ( τ ) = k ∗ ( τ / τ ∗ ) − s k_\ast(\tau)=k_\ast(\tau/\tau_\ast)^{-s} k ∗ ( τ ) = k ∗ ( τ / τ ∗ ) − s (cf. Eq. (159a) with δ τ d c = τ ∗ \delta\tau_{\mathrm{dc}}=\tau_\ast δ τ dc = τ ∗ ), with s = 1 2 s=\frac{1}{2} s = 2 1 , and for a purely vortical non-helical field with a = 5 a=5 a = 5 and b = 2 / 3 b=2/3 b = 2/3 . Left panels show the stress spectra, normalized to their values at initial time τ ∗ \tau_\ast τ ∗ , at logarithmically spaced times between τ ∗ \tau_\ast τ ∗ (purple) and τ e n d = 10 3 τ ∗ \tau_{\mathrm{end}}=10^3\tau_\ast τ end = 1 0 3 τ ∗ (red). For each different time, the horizontal dashed line indicates the amplitude of ζ Π r a t \sqrt{\zeta_{\Pi}^{\mathrm{rat}}} ζ Π rat at large wave numbers, going as ( τ / τ ∗ ) − 11 s / 6 (\tau/\tau_\ast)^{-11s/6} ( τ / τ ∗ ) − 11 s /6 , while the vertical dotted line shows the position of the spectral peak, going as ( τ / τ ∗ ) − s (\tau/\tau_\ast)^{-s} ( τ / τ ∗ ) − s . The right panels show the time evolution at specific wave numbers. We see that at large times all wave numbers eventually evolve as ( τ / τ ∗ ) − 11 s / 6 (\tau/\tau_\ast)^{-11s/6} ( τ / τ ∗ ) − 11 s /6 .
As we have shown in this section (see Figure 22 ), the function Δ m 2 ( τ , k ) \Delta_{m}^{2}(\tau,k) Δ m 2 ( τ , k ) has the following envelope in the case of negative values of m m m [cf. Eq. (210) ]
Δ m 2 ( k ) = 1 m 2 [ 1 − ( 1 + H ⋆ k ) m ] 2 ∀ m < 0. (213) \Delta_m^2(k)=\frac{1}{m^2}\left[1-\left(1+\frac{\mathcal{H}_{\star}}{k}\right)^m\right]^2\qquad\forall m<0 . \tag*{(213)} Δ m 2 ( k ) = m 2 1 [ 1 − ( 1 + k H ⋆ ) m ] 2 ∀ m < 0. (213)
For this function we can then distinguish two regimes, respectively corresponding to large and small scales
lim k ≪ H ⋆ Δ m 2 ( τ , k ) ≃ 1 m 2 , (214) \lim_{k\ll\mathcal{H}_{\star}}\Delta_m^2(\tau,k)\simeq\frac{1}{m^2}, \tag*{(214)} k ≪ H ⋆ lim Δ m 2 ( τ , k ) ≃ m 2 1 , (214)
lim k ≫ H ⋆ Δ m 2 ( τ , k ) ≃ ( H ⋆ k ) 2 . (215) \lim_{k\gg\mathcal{H}_{\star}}\Delta_m^2(\tau,k)\simeq\left(\frac{\mathcal{H}_{\star}}{k}\right)^2 . \tag*{(215)} k ≫ H ⋆ lim Δ m 2 ( τ , k ) ≃ ( k H ⋆ ) 2 . (215)
Around the scale k ≃ H ⋆ k\simeq\mathcal{H}_{\star} k ≃ H ⋆ the envelope of Δ m 2 \Delta_m^2 Δ m 2 smoothly transitions from Eq. (182a) to Eq. (182b), following the exact form in Eq. (213) .
Hence, there are three different regimes for the Gravitational Wave spectrum.
iii) For large K > K G W K>K_{\rm GW} K > K GW , with K G W K_{\rm GW} K GW being the GW spectral peak, given in Eq. (164) as for the CiT model case, the GW spectrum goes as K 1 − b Π K^{1-b_{\Pi}} K 1 − b Π .
Hence the only changes with respect to the CiT model are at large scales, for K < R ⋆ H ⋆ K<\mathcal{R}_{\star}\mathcal{H}_{\star} K < R ⋆ H ⋆ , where
the suppression of Ω G W \Omega_{\rm GW} Ω GW is due to the decay of the source. At smaller scales the GW spectrum saturates with a timescale which is shorter than the decay timescale, hence we recover the same result as in the CiT model.
Similarly to Sec. V B we can define a normalized spectral shape S c o h ( K ) S_{\rm coh}(K) S coh ( K ) as
S c o h ( K ) = N c o h K 3 ζ Π ⋆ ( K ) Δ m 2 ( τ , k ) , (216) S_{\rm coh}(K)=\mathcal{N}_{\rm coh}K^3\zeta_{\Pi}^{\star}(K)\Delta_m^2(\tau,k), \tag*{(216)} S coh ( K ) = N coh K 3 ζ Π ⋆ ( K ) Δ m 2 ( τ , k ) , (216)
with
N c o h − 1 = ∫ − ∞ + ∞ K 3 ζ Π ⋆ ( K ) Δ m 2 ( τ , k ) d ln K , (217) \mathcal{N}_{\rm coh}^{-1}=\int_{-\infty}^{+\infty}K^3\zeta_{\Pi}^{\star}(K)\Delta_m^2(\tau,k)\,{\rm d}\ln K , \tag*{(217)} N coh − 1 = ∫ − ∞ + ∞ K 3 ζ Π ⋆ ( K ) Δ m 2 ( τ , k ) d ln K , (217)
and where we have restricted to the cases in which Δ p s 2 \Delta_{ps}^{2} Δ p s 2 [cf. Eq. (193) ] reduces to Δ m 2 ( τ , k ) \Delta_m^2(\tau,k) Δ m 2 ( τ , k ) [cf. Eq. (200) ]. Then we can define an efficiency factor
Ω ~ G W c o h = C 2 A 2 N c o h − 1 ( R ⋆ H ⋆ ) 2 , (218) \widetilde{\Omega}_{\rm GW}^{\rm coh}=\frac{\mathcal{C}}{2\mathcal{A}^2}\frac{\mathcal{N}_{\rm coh}^{-1}}{(\mathcal{R}_{\star}\mathcal{H}_{\star})^2}, \tag*{(218)} Ω GW coh = 2 A 2 C ( R ⋆ H ⋆ ) 2 N coh − 1 , (218)
such that the GW spectrum can be written as
Ω G W = 3 F G W 0 Ω ~ G W c o h Φ ~ r m s 4 ρ ˉ 2 ( R ⋆ H ⋆ ) 2 S c o h ( K ) . (219) \Omega_{\rm GW}=3F_{\rm GW}^{0}\widetilde{\Omega}_{\rm GW}^{\rm coh}\frac{\tilde{\Phi}_{\rm rms}^{4}}{\bar{\rho}^{2}}(\mathcal{R}_{\star}\mathcal{H}_{\star})^2S_{\rm coh}(K). \tag*{(219)} Ω GW = 3 F GW 0 Ω GW coh ρ ˉ 2 Φ ~ rms 4 ( R ⋆ H ⋆ ) 2 S coh ( K ) . (219)
We show in Figure 24 the normalized S c o h ( K ) S_{\rm coh}(K) S coh ( K ) multiplied by Ω ~ G W c o h \widetilde{\Omega}_{\rm GW}^{\rm coh} Ω GW coh for different values of m m m in the case δ τ d c = τ ⋆ \delta\tau_{\rm dc}=\tau_{\star} δ τ dc = τ ⋆ in which the function Δ m 2 \Delta_m^2 Δ m 2 can be computed analytically. Here we use for Δ m 2 \Delta_m^2 Δ m 2 its envelope which is given in Eq. (213) .
Figure 24. Normalized spectral shape of the GW spectrum multiplied by the efficiency Ω ~ G W \widetilde{\Omega}_{\rm GW} Ω GW , in the coherent case with δ τ d c = τ ∗ \delta\tau_{\rm dc}=\tau_\ast δ τ dc = τ ∗ and for different values of m m m , both for a purely vortical (upper panel) and purely compressional (lower panel) source with α = 2 \alpha=2 α = 2 , a = 5 a=5 a = 5 and respectively b = 2 / 3 b=2/3 b = 2/3 and b = 1 b=1 b = 1 for the vortical and compressional cases. We consider R ∗ H ∗ = 10 − 2 \mathcal{R}_\ast\mathcal{H}_\ast=10^{-2} R ∗ H ∗ = 1 0 − 2 and, for comparison, we show in black lines the result in the CiT model (see also Figure 13 ) with sourcing duration δ τ f i n / τ ∗ = 100 \delta\tau_{\rm fin}/\tau_\ast=100 δ τ fin / τ ∗ = 100 .
Moreover, in Table V we provide the GW efficiencies Ω ~ G W c o h \widetilde{\Omega}_{\rm GW}^{\rm coh} Ω GW coh for different values of α \alpha α and m m m . We consider physically motivated values of m m m . In particular, m = − 1 / 3 m=-1/3 m = − 1/3 and m = − 11 / 9 m=-11/9 m = − 11/9 are expected at late times for magnetically-dominated turbulence when the magnetic field is helical [73 ] or non-helical [104 ] , respectively, based on the conservation of helicity or the Hosking invariant. For kinetically-dominated turbulence, instead, we expect m = − 7 / 5 m=-7/5 m = − 7/5 or m = − 13 / 7 m=-13/7 m = − 13/7 for the IR slope of the velocity spectrum a = 3 a=3 a = 3 or a = 5 a=5 a = 5 , respectively [73 ] . When the velocity field is purely irrotational (i.e., acoustic turbulence), then a = 5 a=5 a = 5 and, hence, one expects m = − 7 / 5 m=-7/5 m = − 7/5 [36 , 37 ] .
In the case of δ τ d c ≠ τ ⋆ \delta\tau_{\rm dc}\ne\tau_{\star} δ τ dc = τ ⋆ the normalized S c o h ( K ) S_{\rm coh}(K) S coh ( K ) would be modified at k < 1 / δ τ d c k<1/\delta\tau_{\rm dc} k < 1/ δ τ dc with respect to the CiT model and the large-scale region of the GW spectrum would be suppressed by the factor r 2 ( δ τ d c ; x d c , α d c , β d c ) r^2(\delta\tau_{\rm dc};x_{\rm dc},\alpha_{\rm dc},\beta_{\rm dc}) r 2 ( δ τ dc ; x dc , α dc , β dc ) which we have introduced in Eq. (199) and shown in Figure 20 .
We show in Figure 25 and Figure 26 the normalized GW spectrum S c o h ( K ) S_{\rm coh}(K) S coh ( K ) multiplied by the GW efficiency Ω ~ G W c o h \widetilde{\Omega}_{\rm GW}^{\rm coh} Ω GW coh for different values of δ τ d c / τ ⋆ \delta\tau_{\rm dc}/\tau_{\star} δ τ dc / τ ⋆ both for purely vortical and purely compressional fields, and the same values of m m m as in Table V. Here we see that the effect of having a shorter decay timescale is more relevant for larger values of the decay parameter − m -m − m . Moreover, for large δ τ d c / τ ∗ \delta\tau_{\rm dc}/\tau_\ast δ τ dc / τ ∗ and small ∣ m ∣ |m| ∣ m ∣ , we see that the large-scale ∼ K 3 \sim K^3 ∼ K 3 regime of the GW spectrum is only reached below K ≲ 10 − 4 K\lesssim10^{-4} K ≲ 1 0 − 4 , corresponding to the prediction for the Constant-in-Time model, given that R ∗ H ∗ = 10 − 2 \mathcal{R}_\ast\mathcal{H}_\ast=10^{-2} R ∗ H ∗ = 1 0 − 2 and δ τ f i n = 10 2 τ ∗ \delta\tau_{\rm fin}=10^2\tau_\ast δ τ fin = 1 0 2 τ ∗ , which implies ( k ∗ δ τ f i n ) − 1 = 10 − 4 (k_\ast\delta\tau_{\rm fin})^{-1}=10^{-4} ( k ∗ δ τ fin ) − 1 = 1 0 − 4 . However, as ∣ m ∣ |m| ∣ m ∣ increases and as δ τ d c / τ ∗ \delta\tau_{\rm dc}/\tau_\ast δ τ dc / τ ∗ decreases, the ∼ K 3 \sim K^3 ∼ K 3 regime is reached earlier, and we define as the effective final time of the sourcing δ τ f i n e f f \delta\tau_{\rm fin}^{\rm eff} δ τ fin eff the inverse wave number
Table 5. Values of the GW efficiency in the coherent-decay model Ω ~ G W c o h \widetilde{\Omega}_{\rm GW}^{\rm coh} Ω GW coh for a purely vortical non-helical and a purely compressional source with slopes a = 5 a=5 a = 5 and b = 2 / 3 b=2/3 b = 2/3 for vortical (vort), while a = 5 a=5 a = 5 and b = 1 b=1 b = 1 for compressional (comp), and a range of values of the smoothness parameter α \alpha α . We use physically motivated values of m = − 1 / 3 , − 11 / 9 m=-1/3,-11/9 m = − 1/3 , − 11/9 for vortical [104 ] and m = − 7 / 5 , − 13 / 7 m=-7/5,-13/7 m = − 7/5 , − 13/7 for compressional [37 , 104 ] , as in Figure 24 . Here δ τ d c = τ ∗ \delta\tau_{\rm dc}=\tau_\ast δ τ dc = τ ∗ , R ∗ H ∗ = 10 − 2 \mathcal{R}_\ast\mathcal{H}_\ast=10^{-2} R ∗ H ∗ = 1 0 − 2 . Note that the values of Ω ~ G W c o h \widetilde{\Omega}_{\rm GW}^{\rm coh} Ω GW coh are always smaller than those in the CiT model (see Table II), which they approach as ∣ m ∣ |m| ∣ m ∣ decreases.
at which this regime is reached. We show in Figure 27 the values of δ τ f i n e f f / τ ∗ \delta\tau_{\rm fin}^{\rm eff}/\tau_\ast δ τ fin eff / τ ∗ for different choices of m m m , corresponding to those of Figure 25 and Figure 26 , together with some simple power law fits for each value of m m m .
Figure 27. Effective final time of the sourcing δ τ f i n e f f \delta\tau_{\rm fin}^{\rm eff} δ τ fin eff for the coherent-decay model (solid lines), as a function of the decay timescale, and for different values of m m m , corresponding to those of Figure 25 and Figure 26 . Dashed lines represent the fitting functions found for each case.
Conclusions
In this work, we have presented a general framework to describe the production of stochastic gravitational wave (GW) backgrounds from slowly-evolving sources in the early Universe, focusing in particular on the role of the temporal evolution and decorrelation of the source. The central quantity entering the computation of the GW spectrum is the unequal-time correlator (UETC) of the anisotropic stresses, E Π ( τ 1 , τ 2 , k ) E_\Pi(\tau_1,\tau_2,k) E Π ( τ 1 , τ 2 , k ) . Since the UETC is generally not known analytically, especially for nonlinear sources such as turbulent velocity and magnetic fields, a suitable modelling of its time dependence is essential to obtain reliable predictions for the resulting GW background.
Constant-in-Time (CiT) model. We have first studied the CiT model, which assumes that the source evolves on a timescale much longer than the characteristic timescale of GW production. Under this assumption, the UETC can be approximated as constant during an effective sourcing period δ τ f i n = τ f i n − τ ∗ \delta\tau_{\mathrm{fin}}=\tau_{\mathrm{fin}}-\tau_{\ast} δ τ fin = τ fin − τ ∗ ,
E Π ( τ 1 , τ 2 , k ) ≃ E Π ∗ ( k ) ≡ E Π ( τ ∗ , τ ∗ , k ) , (220) E_{\Pi}(\tau_1,\tau_2,k)\simeq E_{\Pi}^{\ast}(k)\equiv E_{\Pi}(\tau_{\ast},\tau_{\ast},k), \tag*{(220)} E Π ( τ 1 , τ 2 , k ) ≃ E Π ∗ ( k ) ≡ E Π ( τ ∗ , τ ∗ , k ) , (220)
and the GW spectrum can be written as
Ω G W ( τ , k ) = 3 E Π ∗ ( k ) F G W 0 Δ 2 ( τ , k ) , (221) \Omega_{\mathrm{GW}}(\tau,k)=3E_{\Pi}^{\ast}(k)F_{\mathrm{GW}}^{0}\Delta^{2}(\tau,k), \tag*{(221)} Ω GW ( τ , k ) = 3 E Π ∗ ( k ) F GW 0 Δ 2 ( τ , k ) , (221)
where Δ ( τ , k ) \Delta(\tau,k) Δ ( τ , k ) is the time kernel obtained by integrating the GW Green’s functions over the sourcing period [see Eq. (132) ], and F G W 0 F_{\mathrm{GW}}^{0} F GW 0 accounts for the redshift of the GW energy density from the moment of production to the present time. The main advantage of this approximation is that it separates the spectral properties of the source, encoded in E Π ∗ ( k ) E_{\Pi}^{\ast}(k) E Π ∗ ( k ) , from the time dependence of GW production, encoded in Δ ( τ , k ) \Delta(\tau,k) Δ ( τ , k ) .
A characteristic feature of the CiT model is the resulting GW spectral shape, which exhibits three different regimes. At sufficiently small wave numbers, the spectrum follows the causal infrared (IR) behaviour Ω G W ( k ) ∝ k 3 \Omega_{\mathrm{GW}}(k)\propto k^{3} Ω GW ( k ) ∝ k 3 . At intermediate wave numbers, between the inverse effective duration of the source and the peak scale of E Π ∗ ( k ) E_{\Pi}^{\ast}(k) E Π ∗ ( k ) , the spectrum develops an approximately linear dependence, Ω G W ( k ) ∝ k \Omega_{\mathrm{GW}}(k)\propto k Ω GW ( k ) ∝ k , which is, in the case of an expanding Universe, more precisely given by k 3 ln 2 ( 1 + H ∗ / k ) k^{3}\ln^{2}(1+\mathcal{H}_{\ast}/k) k 3 ln 2 ( 1 + H ∗ / k ) , with H ∗ \mathcal{H}_* H ∗ being the conformal Hubble rate at the time of GW production. Finally, at wave numbers above the peak of the UETC, the spectrum decreases with a power-law slope determined by the small-scale behaviour of the anisotropic stress spectrum. In particular, for a source whose anisotropic stress spectrum scales as E Π ( k ) ∝ k 3 − b Π E_{\Pi}(k) \propto k^{3-b_{\Pi}} E Π ( k ) ∝ k 3 − b Π , the ultraviolet (UV) GW spectrum behaves as Ω G W ( k ) ∝ k 1 − b Π \Omega_{\rm GW}(k) \propto k^{1-b_{\Pi}} Ω GW ( k ) ∝ k 1 − b Π . As an example, for Kolmogorov turbulence, which is characterized by b Π = 11 / 3 b_{\Pi}=11/3 b Π = 11/3 , this gives the well-known scaling Ω G W ( k ) ∝ k − 8 / 3 \Omega_{\rm GW}(k) \propto k^{-8/3} Ω GW ( k ) ∝ k − 8/3 . The UV behaviour is therefore not universal, but depends on the nature of the source and on the spectral properties of its anisotropic stresses, while the k 3 k^3 k 3 and linear regimes arise, respectively, from causality and from the Green’s function for GW production.
Applications of the CiT model. We have applied this description to velocity fields with generic compressibility and fractional helicity, magnetic fields with generic fractional helicity, and scalar fields. These cases encompass vortical and acoustic turbulence, sound waves, and scalar-field gradients. Although the underlying source spectra can be substantially different, the CiT framework provides a common description of their GW production in the regime in which both their decay and decorrelation timescales are longer than that of GW production. To make this description easy to use, we have modelled the source spectra with a smoothed broken power law, with slopes a a a and b b b at small and large wave numbers and smoothness parameter α \alpha α , and provided the corresponding analytical moments, as well as fits of the normalized anisotropic stresses and of the characteristic wave number k G W = K G W / R ∗ k_{\rm GW}=K_{\rm GW}/\mathcal{R}_* k GW = K GW / R ∗ of the GW peak, with R ∗ \mathcal{R}_* R ∗ being the correlation length of the source. This allows us to provide the following GW template,
Ω G W = 3 F G W 0 ϕ ~ r m s 4 ρ ˉ 2 ( R ∗ H ∗ ) 2 Ω ~ G W S ( K ) , (222) \Omega_{\rm GW}=3F_{\rm GW}^{0}\frac{\tilde{\phi}_{\rm rms}^{4}}{\bar{\rho}^{2}}(\mathcal{R}_*\mathcal{H}_*)^2\widetilde{\Omega}_{\rm GW}\,S(K), \tag*{(222)} Ω GW = 3 F GW 0 ρ ˉ 2 ϕ ~ rms 4 ( R ∗ H ∗ ) 2 Ω GW S ( K ) , (222)
where the amplitude is determined by the GW efficiency Ω ~ G W \widetilde{\Omega}_{\rm GW} Ω GW , the energy density fraction of the source φ i \varphi_i φ i , through the factor φ ~ r m s 2 / ρ ˉ 2 \tilde{\varphi}_{\rm rms}^{2}/\bar{\rho}^{2} φ ~ rms 2 / ρ ˉ 2 [see discussion below Eq. (86) ], and the characteristic length scale of the source in units of the Hubble radius, R ∗ H ∗ \mathcal{R}_*\mathcal{H}_* R ∗ H ∗ , while the spectral shape S ( K ) S(K) S ( K ) is normalized such that ∫ d ln K S ( K ) = 1 \int\mathrm{d}\ln K\,S(K)=1 ∫ d ln K S ( K ) = 1 . For helical sources, we have also computed, within the CiT model, the circular polarization of the GW background.
Decay of the source: coherent-decay model. An important limitation of the CiT approximation is that it does not describe explicitly the decay of the source. In order to overcome this limitation, we have introduced a coherent model for the UETC, in which this is factorized as
E Π ( τ 1 , τ 2 , k ) = E Π ( τ 1 , k ) E Π ( τ 2 , k ) . (223) E_{\Pi}(\tau_1,\tau_2,k)=\sqrt{E_{\Pi}(\tau_1,k)E_{\Pi}(\tau_2,k)}. \tag*{(223)} E Π ( τ 1 , τ 2 , k ) = E Π ( τ 1 , k ) E Π ( τ 2 , k ) . (223)
This assumption describes sources whose anisotropic stresses remain correlated over the relevant time interval of GW production. It allows us to account for the time evolution of the source while retaining a separable expression for the UETC, so that the GW spectrum can be computed analytically in certain regimes. To model this evolution, we have considered power-law decays of both the characteristic wave number and the amplitude of the source, k ∗ ( τ ) ∝ ( 1 + δ τ / δ τ d c ) − s k_*(\tau)\propto(1+\delta\tau/\delta\tau_{\rm dc})^{-s} k ∗ ( τ ) ∝ ( 1 + δ τ / δ τ dc ) − s and E ∗ ( τ ) ∝ ( 1 + δ τ / δ τ d c ) − p E_*(\tau)\propto(1+\delta\tau/\delta\tau_{\rm dc})^{-p} E ∗ ( τ ) ∝ ( 1 + δ τ / δ τ dc ) − p , where δ τ d c \delta\tau_{\rm dc} δ τ dc is the characteristic decay timescale. For moderate values of s ≲ 1 s\lesssim1 s ≲ 1 , we have shown that the dependence on the change in the UV spectral shape of E Π E_{\Pi} E Π can be approximately neglected, and the time kernel Δ c o h 2 \Delta_{\rm coh}^{2} Δ coh 2 depends only on the combination m = 2 3 s − p m=\frac{2}{3}s-p m = 3 2 s − p , hence we call it Δ m 2 \Delta_m^{2} Δ m 2 [see Eqs. (187) and (193) and, also, Eqs. (198) and (200) ]. This kernel can be computed analytically for the case δ τ d c = H ∗ − 1 \delta\tau_{\rm dc}=\mathcal{H}_*^{-1} δ τ dc = H ∗ − 1 , and presents large-scale deviations from this analytical expression for different values of δ τ d c \delta\tau_{\rm dc} δ τ dc and m m m , for which we have provided a fit.
An important consistency check of this model is that, in the limit of an infinitely long decay timescale, the source becomes effectively constant in time and the CiT result is recovered. For a finite decay timescale, instead, the source evolution modifies the GW spectrum at large scales. In particular, when the decay timescale is longer than the relevant GW production timescale, the impact of the decay is mainly restricted to sufficiently small wave numbers, k ≲ δ τ d c − 1 k \lesssim\delta\tau_{\rm dc}^{-1} k ≲ δ τ dc − 1 , where the amplitude of the k 3 k^3 k 3 regime is suppressed with respect to the CiT model. Moreover, in the coherent case, and for sufficiently short decay timescales, the intermediate linear regime disappears. At larger wave numbers, GW production occurs on a timescale short enough that the source can still be considered approximately constant, and the CiT model remains a good description. The coherent-decay model therefore provides a consistent way to describe the transition between the regime in which the source can be treated as constant and the regime in which its decay becomes relevant. In particular, it avoids the need to abruptly switch off the source after an effective finite duration δ τ f i n = τ f i n − τ ∗ \delta\tau_{\rm fin}=\tau_{\rm fin}-\tau_* δ τ fin = τ fin − τ ∗ , as is done in the CiT prescription.
The application to decaying magnetohydrodynamic (MHD) turbulence illustrates the relevance of this extension. In this case, the coherent-decay model reproduces the expected spectral regions, including the causal k 3 k^3 k 3 tail, the intermediate linear branch and the Kolmogorov ultraviolet scaling k − 8 / 3 k^{-8/3} k − 8/3 . However, the comparison between the two descriptions shows that the CiT model can overestimate GW production in the large-scale and intermediate regions. This difference is associated with the time evolution of the UETC: while the CiT approximation neglects its decay during the sourcing period, the coherent-decay model explicitly accounts for the decrease of the source amplitude and the evolution of its characteristic scale. The resulting GW spectrum is consequently sensitive to the decay rate and to the timescale over which the source evolves.
Temporal decorrelation of the source. We have also discussed the role of temporal decorrelation. In stationary turbulence, the UETC is expected to decorrelate in time because of the nonlinear dynamics. A commonly used description of this effect is the Kraichnan decorrelation model, which we have applied to hydrodynamical vortical turbulence. The CiT approximation is justified when the GW production timescale is much shorter than the source decorrelation timescale. In this regime, the source remains correlated during the relevant GW production interval and the CiT prediction is recovered. Conversely, when decorrelation becomes relevant, it can modify the resulting GW spectrum. We have found in particular that when the decorrelation velocity u d c o r r u_{\rm dcorr} u dcorr [see Eq. (141) and discussion above Eq. (145)] is
u d c o r r k ∗ τ ∗ R ≳ 1 (224) u_{\rm dcorr} k_* \tau_* \mathcal{R} \gtrsim1 \tag*{(224)} u dcorr k ∗ τ ∗ R ≳ 1 (224)
where R = δ τ f i n H ∗ / ( 1 + δ τ f i n H ∗ ) \mathcal{R}=\delta\tau_{\rm fin}\mathcal{H}_*/(1+\delta\tau_{\rm fin}\mathcal{H}_*) R = δ τ fin H ∗ / ( 1 + δ τ fin H ∗ ) and k ∗ = R ∗ − 1 k_*=\mathcal{R}_*^{-1} k ∗ = R ∗ − 1 , then the effect of Kraichnan decorrelation has to be taken into account in the resulting GW spectrum: a peak appears at a wave number proportional to k ∗ u d c o r r k_*u_{\rm dcorr} k ∗ u dcorr and leads to a larger GW amplitude than the CiT prediction. A similar condition holds for stationary sound waves, for which we find that a peak structure which makes the GW spectrum grow compared to the CiT prediction appears when the sound speed is
c s k ∗ τ ∗ R ≳ 1. (225) c_s k_* \tau_* \mathcal{R} \gtrsim1 . \tag*{(225)} c s k ∗ τ ∗ R ≳ 1. (225)
Hence, one can think of δ τ d c o r r K r a i c h n a n ≡ 1 / ( k ∗ u d c o r r ) \delta\tau_{\rm dcorr}^{\rm Kraichnan}\equiv1/(k_*u_{\rm dcorr}) δ τ dcorr Kraichnan ≡ 1/ ( k ∗ u dcorr ) and δ τ d c o r r s w ≡ 1 / ( k ∗ c s ) \delta\tau_{\rm dcorr}^{\rm sw}\equiv1/(k_*c_s) δ τ dcorr sw ≡ 1/ ( k ∗ c s ) as the typical timescales over which decorrelation becomes important, respectively for turbulence and sound waves. Indeed, the CiT model applies whenever all relevant sourcing times are much shorter than these typical decorrelation timescales, i.e., as long as
δ τ f i n ≪ δ τ d c o r r . (226) \delta\tau_{\rm fin} \ll\delta\tau_{\rm dcorr}. \tag*{(226)} δ τ fin ≪ δ τ dcorr . (226)
Outlook. The framework developed in this work provides a unified approach to the computation of GW spectra from a broad range of slowly-evolving sources in the radiation-dominated era, including velocity, magnetic and scalar fields. It also provides a systematic way to identify the regimes in which the CiT approximation is reliable and those in which the time evolution of the source must be taken into account. The resulting models have been implemented in the open-source CosmoGW [55 , 56 ] code, which can be used for numerical studies of different aspects of GW production from early Universe sources.
Several aspects deserve further investigation. In particular, a more detailed treatment of the growth phase of the source is needed, since the present description focuses on GW production during the source’s decay phase. Moreover, especially in the case of turbulence, in which the decorrelation of the source becomes particularly relevant in the ultrarelativistic nonlinear regime, a more complete description of realistic sources requires accounting simultaneously for both their decay and their temporal decorrelation. These improvements will be important for obtaining fully reliable predictions for the stochastic GW background produced by realistic sources in the early Universe.
Acknowledgments
ASM and ARP are supported through the Swiss National Science Foundation SNSF Ambizione grant 208807. The authors acknowledge the hospitality of CERN where part of this work was conducted. The authors are grateful for the support provided by Nordita during the program “Numerical Simulations of Early Universe Sources of Gravitational Waves,” by the Bernoulli Center during the program “Generation, evolution, and observations of cosmological magnetic fields,” by the Centro de Ciencias de Benasque Pedro Pascual during “The Dawn of Gravitational Wave Cosmology” conference, and by CERN during the workshop “Advancing gravitational wave predictions from cosmological first-order phase transitions,” where preliminary results of this work were presented. ASM and ARP acknowledge the support of CERN during the 2025 Pencil Code school and user meeting, where part of this work was conducted. ARP acknowledges the hospitality of the Yukawa Institute for Theoretical Physics during the Gravity and Cosmology 2024 workshop.
Moments of the smoothed broken power law model
The spectra of the velocity, magnetic, and scalar fields are modeled using a smoothed broken power law, introduced in Eq. (63) . In this section, we present analytical expressions to compute the different moments of the spectral shape that are relevant for our theoretical model description in Sec. III.
The normalized spectral shape ζ ( K ) ≡ E ( K ) / E ∗ \zeta(K) \equiv E(K)/E_\ast ζ ( K ) ≡ E ( K ) / E ∗ , being K ≡ k / k ∗ K \equiv k/k_\ast K ≡ k / k ∗ the normalized wave number, such that ζ ( 1 ) = 1 \zeta(1)=1 ζ ( 1 ) = 1 , is
ζ ( K ) = ( b + a ) 1 α K a [ b + a K α ( a + b ) ] 1 α . (227) \zeta(K)=\frac{(b+a)^{\frac{1}{\alpha}}K^a}{\left[b+aK^{\alpha(a+b)}\right]^{\frac{1}{\alpha}}}. \tag*{(227)} ζ ( K ) = [ b + a K α ( a + b ) ] α 1 ( b + a ) α 1 K a . (227)
with slopes a a a and − b -b − b at wave numbers K ≪ 1 K\ll1 K ≪ 1 and K ≫ 1 K\gg1 K ≫ 1 respectively. Both a a a and b b b are taken positive to have a converging value of the integral over ln k \ln k ln k ,
φ r m s 2 E ∗ = A a b ( α ) = ∫ − ∞ ∞ ζ ( K ) d ln K , (228) \frac{\varphi_{\mathrm{rms}}^2}{E_\ast}=\mathcal{A}_{ab}(\alpha)=\int_{-\infty}^{\infty}\zeta(K)\,\mathrm{d}\ln K, \tag*{(228)} E ∗ φ rms 2 = A ab ( α ) = ∫ − ∞ ∞ ζ ( K ) d ln K , (228)
as defined in Eq. (64) , where φ r m s 2 \varphi_{\mathrm{rms}}^2 φ rms 2 is the mean squared value of the field φ i \varphi_i φ i that is characterized by the spectral function E ( K ) E(K) E ( K ) . Therefore, we can define a spectrum with normalized integral,
ζ i n t ( K ) ≡ ζ ( K ) A a b ( α ) \zeta_{\mathrm{int}}(K)\equiv\frac{\zeta(K)}{\mathcal{A}_{ab}(\alpha)} ζ int ( K ) ≡ A ab ( α ) ζ ( K )
with
∫ − ∞ ∞ ζ i n t ( K ) d ln K = 1. (229) \int_{-\infty}^{\infty}\zeta_{\mathrm{int}}(K)\,\mathrm{d}\ln K=1. \tag*{(229)} ∫ − ∞ ∞ ζ int ( K ) d ln K = 1. (229)
Note that E ( K ) E(K) E ( K ) , E ∗ E_\ast E ∗ , and φ r m s 2 \varphi_{\mathrm{rms}}^2 φ rms 2 have the same units, while ζ ( K ) \zeta(K) ζ ( K ) and ζ i n t ( K ) \zeta_{\mathrm{int}}(K) ζ int ( K ) are dimensionless.
The parameter α \alpha α in Eq. (227) determines the smoothness of the transition between the two regimes around K ≃ 1 K\simeq1 K ≃ 1 . Alternatively, one can introduce the smoothness parameter such that
ζ ( K ) = ( b + a ) a + b α 2 K a ( b + a K α 2 ) a + b α 2 . (230) \zeta(K)=\frac{(b+a)^{\frac{a+b}{\alpha_2}}K^a}{\left(b+aK^{\alpha_2}\right)^{\frac{a+b}{\alpha_2}}}. \tag*{(230)} ζ ( K ) = ( b + a K α 2 ) α 2 a + b ( b + a ) α 2 a + b K a . (230)
as done, e.g., in refs. [35 , 50 ] [18]. We note that this spectral function is equivalent to the one in Eq. (227) if we use α = α 2 / ( a + b ) \alpha=\alpha_2/(a+b) α = α 2 / ( a + b ) , so the results presented in this section also apply to this definition of the generic broken power law. This motivates the use of α = 6 / 17 \alpha=6/17 α = 6/17 considered in refs. [35 ] [18] that reproduces the von Kármán spectrum with a = 5 a=5 a = 5 , b = 2 / 3 b=2/3 b = 2/3 , and α 2 = 2 \alpha_2=2 α 2 = 2 ,
ζ ( K ) = N K 5 ( 1 + 7.5 K 2 ) 17 6 . (231) \zeta(K)=\mathcal{N}\frac{K^5}{\left(1+7.5K^2\right)^{\frac{17}{6}}}. \tag*{(231)} ζ ( K ) = N ( 1 + 7.5 K 2 ) 6 17 K 5 . (231)
with N = ( 1 + a / b ) a + b α 2 = ( 1 + 7.5 ) 17 6 \mathcal{N}=(1+a/b)^{\frac{a+b}{\alpha_2}}=(1+7.5)^{\frac{17}{6}} N = ( 1 + a / b ) α 2 a + b = ( 1 + 7.5 ) 6 17 .
We now define the general integral that describes the moments of the spectral shape function ζ ( K ) \zeta(K) ζ ( K ) ,
I a b ; n ( α ) = ∫ 0 ∞ ζ ( K ) K n d K = ∫ − ∞ ∞ ζ ( K ) K n + 1 d ln K , (232) \begin{aligned}
\mathcal{I}_{ab;n}(\alpha)&=\int_0^\infty\zeta(K)K^n\,\mathrm{d}K\\
&=\int_{-\infty}^{\infty}\zeta(K)K^{n+1}\,\mathrm{d}\ln K, \tag*{(232)}
\end{aligned} I ab ; n ( α ) = ∫ 0 ∞ ζ ( K ) K n d K = ∫ − ∞ ∞ ζ ( K ) K n + 1 d ln K , (232)
which has the following analytical solution:
I a b ; n ( α ) = I a b ; n ( α ) Γ [ a + n + 1 α ( a + b ) ] Γ [ b − n − 1 α ( a + b ) ] Γ [ 1 α ] . (233) \mathcal{I}_{ab;n}(\alpha)=\mathcal{I}_{ab;n}(\alpha)\frac{\Gamma\left[\frac{a+n+1}{\alpha(a+b)}\right]\Gamma\left[\frac{b-n-1}{\alpha(a+b)}\right]}{\Gamma\left[\frac{1}{\alpha}\right]}. \tag*{(233)} I ab ; n ( α ) = I ab ; n ( α ) Γ [ α 1 ] Γ [ α ( a + b ) a + n + 1 ] Γ [ α ( a + b ) b − n − 1 ] . (233)
The first term is a normalization factor
I a b ; n ( α ) = 1 α ( a + b ) ( a + b b ) 1 α ( a b ) − a + n + 1 α ( a + b ) . (234) \mathcal{I}_{ab;n}(\alpha)=\frac{1}{\alpha(a+b)}\left(\frac{a+b}{b}\right)^{\frac{1}{\alpha}}\left(\frac{a}{b}\right)^{-\frac{a+n+1}{\alpha(a+b)}}. \tag*{(234)} I ab ; n ( α ) = α ( a + b ) 1 ( b a + b ) α 1 ( b a ) − α ( a + b ) a + n + 1 . (234)
and the second term in Eq. (233) is the complete beta function B ( a β , b β ) B(a_\beta,b_\beta) B ( a β , b β ) with a β = ( a + n + 1 ) / [ α ( a + b ) ] a_\beta=(a+n+1)/[\alpha(a+b)] a β = ( a + n + 1 ) / [ α ( a + b )] and b β = ( b − n − 1 ) / [ α ( a + b ) ] b_\beta=(b-n-1)/[\alpha(a+b)] b β = ( b − n − 1 ) / [ α ( a + b )] .
The integral I a b ; n \mathcal{I}_{ab;n} I ab ; n can be used to compute the different moments of the spectral functions. Let us start by considering the integrated spectrum of Eq. (228) ,
A a b ( α ) ≡ I a b ; − 1 ( α ) = 1 α ( a + b ) ( a + b b ) 1 α ( a b ) − a α ( a + b ) Γ [ a α ( a + b ) ] Γ [ b α ( a + b ) ] Γ [ 1 α ] . (235) \mathcal{A}_{ab}(\alpha)\equiv\mathcal{I}_{ab;-1}(\alpha)=\frac{1}{\alpha(a+b)}\left(\frac{a+b}{b}\right)^{\frac{1}{\alpha}}\left(\frac{a}{b}\right)^{-\frac{a}{\alpha(a+b)}}\frac{\Gamma\left[\frac{a}{\alpha(a+b)}\right]\Gamma\left[\frac{b}{\alpha(a+b)}\right]}{\Gamma\left[\frac{1}{\alpha}\right]}. \tag*{(235)} A ab ( α ) ≡ I ab ; − 1 ( α ) = α ( a + b ) 1 ( b a + b ) α 1 ( b a ) − α ( a + b ) a Γ [ α 1 ] Γ [ α ( a + b ) a ] Γ [ α ( a + b ) b ] . (235)
The factor A a b ( α ) \mathcal{A}_{ab}(\alpha) A ab ( α ) that relates the integrated spectrum and the peak spectral value is shown in Figure 28 for a few choices of a a a and b b b as a function of α \alpha α .
The integral length scale can be similarly computed as
k ∗ ξ ≡ B a b ( α ) ≡ 1 A a b ( α ) ∫ 0 ∞ K − 1 ζ ( K ) d ln K = I a b ; − 2 A a b ( α ) = ( a b ) 1 α 2 Γ [ a − 1 α ( a + b ) ] Γ [ b + 1 α ( a + b ) ] Γ [ a α ( a + b ) ] Γ [ b α ( a + b ) ] , (236) \begin{aligned}
k_\ast\xi\equiv\mathcal{B}_{ab}(\alpha)&\equiv\frac{1}{\mathcal{A}_{ab}(\alpha)}\int_0^\infty K^{-1}\zeta(K)\,\mathrm{d}\ln K\\
&=\frac{\mathcal{I}_{ab;-2}}{\mathcal{A}_{ab}(\alpha)}
=\left(\frac{a}{b}\right)^{\frac{1}{\alpha_2}}\frac{\Gamma\left[\frac{a-1}{\alpha(a+b)}\right]\Gamma\left[\frac{b+1}{\alpha(a+b)}\right]}{\Gamma\left[\frac{a}{\alpha(a+b)}\right]\Gamma\left[\frac{b}{\alpha(a+b)}\right]}, \tag*{(236)}
\end{aligned} k ∗ ξ ≡ B ab ( α ) ≡ A ab ( α ) 1 ∫ 0 ∞ K − 1 ζ ( K ) d ln K = A ab ( α ) I ab ; − 2 = ( b a ) α 2 1 Γ [ α ( a + b ) a ] Γ [ α ( a + b ) b ] Γ [ α ( a + b ) a − 1 ] Γ [ α ( a + b ) b + 1 ] , (236)
which is shown in Figure 29 .
In general, we can define different characteristic length scales from the n n n -th moment of the spectrum
k ∗ ξ ( n ) ≡ B a b ; n ( α ) ≡ [ 1 A a b ( α ) ∫ 0 ∞ ζ ( K ) K n d ln K ] 1 n = [ I a b ; − n − 1 ( α ) A a b ( α ) ] 1 n = ( a b ) 1 α ( a + b ) ( Γ [ a − n α ( a + b ) ] Γ [ b + n α ( a + b ) ] Γ [ a α ( a + b ) ] Γ [ b α ( a + b ) ] ) 1 n , (237) \begin{aligned}
k_\ast\xi^{(n)}\equiv\mathcal{B}_{ab;n}(\alpha)&\equiv\left[\frac{1}{\mathcal{A}_{ab}(\alpha)}\int_0^\infty\frac{\zeta(K)}{K^n}\,\mathrm{d}\ln K\right]^{\frac{1}{n}}\\
&=\left[\frac{\mathcal{I}_{ab;-n-1}(\alpha)}{\mathcal{A}_{ab}(\alpha)}\right]^{\frac{1}{n}}\\
&=\left(\frac{a}{b}\right)^{\frac{1}{\alpha(a+b)}}\left(\frac{\Gamma\left[\frac{a-n}{\alpha(a+b)}\right]\Gamma\left[\frac{b+n}{\alpha(a+b)}\right]}{\Gamma\left[\frac{a}{\alpha(a+b)}\right]\Gamma\left[\frac{b}{\alpha(a+b)}\right]}\right)^{\frac{1}{n}}, \tag*{(237)}
\end{aligned} k ∗ ξ ( n ) ≡ B ab ; n ( α ) ≡ [ A ab ( α ) 1 ∫ 0 ∞ K n ζ ( K ) d ln K ] n 1 = [ A ab ( α ) I ab ; − n − 1 ( α ) ] n 1 = ( b a ) α ( a + b ) 1 Γ [ α ( a + b ) a ] Γ [ α ( a + b ) b ] Γ [ α ( a + b ) a − n ] Γ [ α ( a + b ) b + n ] n 1 , (237)
such that B a b = B a b ; 1 \mathcal{B}_{ab}=\mathcal{B}_{ab;1} B ab = B ab ; 1 recovers Eq. (236) . For example, B a b ; − 1 \mathcal{B}_{ab;-1} B ab ; − 1 and B a b ; 1 \mathcal{B}_{ab;1} B ab ; 1 relate the kinetic and magnetic helicity, respectively, with the mean squared value of a vector field [see Eqs. (42)], and B a b ; 2 \mathcal{B}_{ab;2} B ab ; 2 relates the scalar field gradients energy density with the mean squared value of a scalar field in Eq. (62) .
At larger values of α \alpha α , the transition between the power laws of the model in Eq. (227) becomes sharper (see Figure 1 ). In the limit α → ∞ \alpha\to\infty α → ∞ (or α 2 → ∞ \alpha_{2}\to\infty α 2 → ∞ ), the model becomes a piece-wise broken power law,
ζ ( K ) = { K a for K ≤ 1 , K − b for K > 1. (238) \zeta(K)=
\begin{cases}
K^{a} & \text{for } K\le1,\\
K^{-b} & \text{for } K>1.
\end{cases} \tag*{(238)} ζ ( K ) = { K a K − b for K ≤ 1 , for K > 1. (238)
The integrated spectrum becomes
lim α → ∞ A a b ( α ) = a + b a b , (239) \lim_{\alpha\to\infty} A_{ab}(\alpha)=\frac{a+b}{ab}, \tag*{(239)} α → ∞ lim A ab ( α ) = ab a + b , (239)
which correspond to the constant values that are found at large α \alpha α in Figure 28 .
We can generalize this result to I a b ; n I_{ab;n} I ab ; n taking the limit α → ∞ \alpha\to\infty α → ∞ in Eq. (233) , by noting Γ ( a x ) → 1 / ( a x ) \Gamma(ax)\to1/(ax) Γ ( a x ) → 1/ ( a x ) when x → 0 x\to0 x → 0 ,
lim α → ∞ I a b ; n ( α ) = a + b ( a + n + 1 ) ( b − n − 1 ) . (240) \lim_{\alpha\to\infty} I_{ab;n}(\alpha)=\frac{a+b}{(a+n+1)(b-n-1)}. \tag*{(240)} α → ∞ lim I ab ; n ( α ) = ( a + n + 1 ) ( b − n − 1 ) a + b . (240)
Applying this result, the generic correlation length scale ξ ( n ) \xi^{(n)} ξ ( n ) for a piece-wise broken power-law is
lim α → ∞ B a b ; n = [ a b ( a − n ) ( b + n ) ] 1 n . (241) \lim_{\alpha\to\infty} B_{ab;n}=\left[\frac{ab}{(a-n)(b+n)}\right]^{\frac{1}{n}}. \tag*{(241)} α → ∞ lim B ab ; n = [ ( a − n ) ( b + n ) ab ] n 1 . (241)
For n = 1 n=1 n = 1 , k ∗ ξ = B a b ; 1 → a b / [ ( a − 1 ) ( b + 1 ) ] k_{*}\xi=B_{ab;1}\to ab/[(a-1)(b+1)] k ∗ ξ = B ab ; 1 → ab / [( a − 1 ) ( b + 1 )] when α → ∞ \alpha\to\infty α → ∞ , which corresponds to the asymptotic values seen at large α \alpha α in Figure 29 .
In some occasions, it is useful to rescale a function by a power of K K K . The resulting K m ζ ( K ) K^{m}\zeta(K) K m ζ ( K ) can be re-expressed as a new normalized function, ζ m ( K ~ ) \zeta_{m}(\widetilde{K}) ζ m ( K ) ,
K m ζ ( K ) = ζ m ∗ ζ m ( K ~ ) , (242) K^{m}\zeta(K)=\zeta_{m}^{*}\zeta_{m}(\widetilde{K}), \tag*{(242)} K m ζ ( K ) = ζ m ∗ ζ m ( K ) , (242)
defined such that ζ m ( 1 ) = 1 \zeta_{m}(1)=1 ζ m ( 1 ) = 1 as in Eq. (227) , with the new slopes m + a m+a m + a and m − b m-b m − b ,
ζ m ( K ~ ) = ( a + b ) 1 α K ~ a + m [ b − m + ( a + m ) K ~ α 2 ] 1 α , (243) \zeta_{m}(\widetilde{K})=
\frac{(a+b)^{\frac{1}{\alpha}}\widetilde{K}^{a+m}}
{\left[b-m+(a+m)\widetilde{K}^{\alpha_{2}}\right]^{\frac{1}{\alpha}}}, \tag*{(243)} ζ m ( K ) = [ b − m + ( a + m ) K α 2 ] α 1 ( a + b ) α 1 K a + m , (243)
where K ~ = K / K m ∗ \widetilde{K}=K/K_{m}^{*} K = K / K m ∗ is the new normalized wave number, with
K m ∗ ≡ k m ∗ k ∗ ∗ = ( b ( a + m ) a ( b − m ) ) 1 α ( a + b ) . (244) K_{m}^{*}\equiv\frac{k_{m}^{*}}{k_{*}^{*}}=
\left(\frac{b(a+m)}{a(b-m)}\right)^{\frac{1}{\alpha(a+b)}}. \tag*{(244)} K m ∗ ≡ k ∗ ∗ k m ∗ = ( a ( b − m ) b ( a + m ) ) α ( a + b ) 1 . (244)
indicating the location of the spectral peak after multiplying the original spectrum by K m K^m K m . The smoothness of the transition remains the same as that of the original ζ ( K ) \zeta(K) ζ ( K ) , characterized by α \alpha α . The amplitude of the function K m ζ ( K ) K^m\zeta(K) K m ζ ( K ) at K = K m ∗ K=K_m^* K = K m ∗ (i.e., K ~ = 1 \widetilde{K}=1 K = 1 ) is
ζ m ∗ = ( a + m a ) a + m α ( a + b ) ( b − m b ) b − m α ( a + b ) . (245) \zeta_m^*=\left(\frac{a+m}{a}\right)^{\frac{a+m}{\alpha(a+b)}}\left(\frac{b-m}{b}\right)^{\frac{b-m}{\alpha(a+b)}}. \tag*{(245)} ζ m ∗ = ( a a + m ) α ( a + b ) a + m ( b b − m ) α ( a + b ) b − m . (245)
Therefore, the moments of the rescaled function ζ m ( K ~ ) \zeta_m(\widetilde{K}) ζ m ( K ) [see Eq. (232) ], expressed with respect to the moments of the original ζ ( K ) \zeta(K) ζ ( K ) , are
I a + m , b − m ; n ( α ) = ∫ 0 ∞ ζ m ( K ~ ) K ~ n d K ~ = I a b ; n + m ( α ) [ K m ∗ ] n + 1 ζ m ∗ . (246) I_{a+m,b-m;n}(\alpha)=\int_0^\infty\zeta_m(\widetilde{K})\widetilde{K}^n\mathrm{d}\widetilde{K}
=\frac{I_{ab;n+m}(\alpha)}{[K_m^*]^{n+1}\zeta_m^*}. \tag*{(246)} I a + m , b − m ; n ( α ) = ∫ 0 ∞ ζ m ( K ) K n d K = [ K m ∗ ] n + 1 ζ m ∗ I ab ; n + m ( α ) . (246)
The difference between the two functions reduces as α \alpha α increases, vanishing in the α → ∞ \alpha\to\infty α → ∞ limit, where K m ∗ K_m^* K m ∗ and ζ m ∗ \zeta_m^* ζ m ∗ approach 1.
Equation (65) yields A 5 , 2 / 3 ( 2 ) ≃ 1.97 A_{5,2/3}(2)\simeq1.97 A 5 , 2/3 ( 2 ) ≃ 1.97 and A 5 , 2 / 3 ( 6 / 17 ) ≃ 2.88 A_{5,2/3}(6/17)\simeq2.88 A 5 , 2/3 ( 6/17 ) ≃ 2.88 for the values of α = 2 \alpha=2 α = 2 and 6 / 17 6/17 6/17 respectively considered in ref. [1] and refs. [35 , 50 ] [9, 18]. However, these works considered the following convention to define the velocity or magnetic spectra,
E − 1 ( K ) ≡ d ϕ r m s 2 d K = E ( K ) K = E ∗ ζ ( K ) K . (247) E_{-1}(K)\equiv\frac{\mathrm{d}\phi_{\mathrm{rms}}^2}{\mathrm{d}K}=\frac{E(K)}{K}=E_*\frac{\zeta(K)}{K}. \tag*{(247)} E − 1 ( K ) ≡ d K d ϕ rms 2 = K E ( K ) = E ∗ K ζ ( K ) . (247)
Since both normalizations are related by a factor of K K K , i.e., E − 1 ( K ) = E ( K ) / K E_{-1}(K)=E(K)/K E − 1 ( K ) = E ( K ) / K , we can apply the results described above for m = − 1 m=-1 m = − 1 to find [see Eqs. (242) –(245) ]
E − 1 ( K ) E ∗ = ζ ( K ) K = ζ − 1 ∗ ζ − 1 ( K ~ ) = ( a − 1 a ) a − 1 α ( a + b ) ( b + 1 b ) b + 1 α ( a + b ) ζ − 1 ( K ~ ) , (248) \frac{E_{-1}(K)}{E_*}=\frac{\zeta(K)}{K}=\zeta_{-1}^*\zeta_{-1}(\widetilde{K})
=\left(\frac{a-1}{a}\right)^{\frac{a-1}{\alpha(a+b)}}\left(\frac{b+1}{b}\right)^{\frac{b+1}{\alpha(a+b)}}\zeta_{-1}(\widetilde{K}), \tag*{(248)} E ∗ E − 1 ( K ) = K ζ ( K ) = ζ − 1 ∗ ζ − 1 ( K ) = ( a a − 1 ) α ( a + b ) a − 1 ( b b + 1 ) α ( a + b ) b + 1 ζ − 1 ( K ) , (248)
where
ζ − 1 ( K ~ ) = ( a + b ) 1 α K ~ a − 1 [ ( b + 1 ) + ( a − 1 ) K ~ α ( a + b ) ] 1 α , (249) \zeta_{-1}(\widetilde{K})=\frac{(a+b)^{\frac{1}{\alpha}}\widetilde{K}^{a-1}}{\left[(b+1)+(a-1)\widetilde{K}^{\alpha(a+b)}\right]^{\frac{1}{\alpha}}}, \tag*{(249)} ζ − 1 ( K ) = [ ( b + 1 ) + ( a − 1 ) K α ( a + b ) ] α 1 ( a + b ) α 1 K a − 1 , (249)
and K ~ = K / K − 1 ∗ \widetilde{K}=K/K_{-1}^* K = K / K − 1 ∗ with
K − 1 ∗ = ( b a a − 1 b + 1 ) 1 α ( a + b ) . (250) K_{-1}^*=\left(\frac{b}{a}\frac{a-1}{b+1}\right)^{\frac{1}{\alpha(a+b)}}. \tag*{(250)} K − 1 ∗ = ( a b b + 1 a − 1 ) α ( a + b ) 1 . (250)
The rescaling of ζ ( K ) \zeta(K) ζ ( K ) with 1 / K 1/K 1/ K leads to different moments of the new normalized spectral function ζ − 1 ( K ~ ) \zeta_{-1}(\widetilde{K}) ζ − 1 ( K ) [see Eq. (246) ],
I a − 1 , b + 1 ; n ( α ) = ∫ 0 ∞ ζ − 1 ( K ~ ) K ~ n d K ~ = I a b ; n − 1 ( α ) [ K − 1 ∗ ] n + 1 ζ − 1 ∗ . (251) I_{a-1,b+1;n}(\alpha)=\int_0^\infty\zeta_{-1}(\widetilde{K})\widetilde{K}^n\mathrm{d}\widetilde{K}
=\frac{I_{ab;n-1}(\alpha)}{[K_{-1}^*]^{n+1}\zeta_{-1}^*}. \tag*{(251)} I a − 1 , b + 1 ; n ( α ) = ∫ 0 ∞ ζ − 1 ( K ) K n d K = [ K − 1 ∗ ] n + 1 ζ − 1 ∗ I ab ; n − 1 ( α ) . (251)
In particular, the geometric factor A ~ a b \widetilde{A}_{ab} A ab found using the convention E − 1 ( K ) = d ϕ r m s 2 / d K E_{-1}(K)=\mathrm{d}\phi_{\mathrm{rms}}^2/\mathrm{d}K E − 1 ( K ) = d ϕ rms 2 / d K to define the normalized spectrum is
A ~ a b ( α ) = ∫ 0 ∞ ζ − 1 ( K ~ ) d K ~ = I a − 1 , b + 1 ; 0 ( α ) = A a b ( α ) ζ − 1 ∗ K − 1 ∗ . (252) \widetilde{A}_{ab}(\alpha)=\int_0^\infty\zeta_{-1}(\widetilde{K})\mathrm{d}\widetilde{K}
=I_{a-1,b+1;0}(\alpha)=\frac{A_{ab}(\alpha)}{\zeta_{-1}^*K_{-1}^*}. \tag*{(252)} A ab ( α ) = ∫ 0 ∞ ζ − 1 ( K ) d K = I a − 1 , b + 1 ; 0 ( α ) = ζ − 1 ∗ K − 1 ∗ A ab ( α ) . (252)
The resulting geometric factor is shown in Figure 28 (right panel) and allows us to recover the value of A ~ 5 , 2 / 3 ( 2 ) ≃ 2.064 \widetilde{A}_{5,2/3}(2)\simeq2.064 A 5 , 2/3 ( 2 ) ≃ 2.064 reported in ref. [1] [see their Eq. (9)], which slightly differs with respect to A 5 , 2 / 3 ( 2 ) ≃ 1.97 A_{5,2/3}(2)\simeq1.97 A 5 , 2/3 ( 2 ) ≃ 1.97 given the different convention. Of course, both conventions eventually yield the same results for physical observables like the GW spectrum, as we explicitly show when we present it in Eq. (153) .
Spectrum of the anisotropic stresses
In Sec. IV, we have described the spectrum of the anisotropic stresses, computed from the relevant terms in the stress-energy tensor, with these being expressed as T i j ( x ) = φ i ( x ) φ j ( x ) T_{ij}(x)=\varphi_i(x)\varphi_j(x) T ij ( x ) = φ i ( x ) φ j ( x ) where φ i = { a X i , i B i , ∂ i φ } \varphi_i=\{aX_i,iB_i,\partial_i\varphi\} φ i = { a X i , i B i , ∂ i φ } for velocity, magnetic, and scalar fields, respectively [see Eqs. (32) and (78) ]. Contracting Eqs. (20) with Λ i j l m \Lambda_{ijlm} Λ ij l m and A i j l m A_{ijlm} A ij l m , we find
1 ρ ˉ 2 ⟨ T i j ( k ) T l m ∗ ( k ′ ) ⟩ Λ i j l m ( k ^ ) = ( 2 π ) 6 δ 3 ( k − k ′ ) E Π ( k ) 4 π k 3 . (253) \frac{1}{\bar{\rho}^2}\langle T_{ij}(\boldsymbol{k})T_{lm}^*(\boldsymbol{k}')\rangle\Lambda_{ijlm}(\hat{\boldsymbol{k}})
=(2\pi)^6\delta^3(\boldsymbol{k}-\boldsymbol{k}')\frac{E_{\Pi}(k)}{4\pi k^3}. \tag*{(253)} ρ ˉ 2 1 ⟨ T ij ( k ) T l m ∗ ( k ′ )⟩ Λ ij l m ( k ^ ) = ( 2 π ) 6 δ 3 ( k − k ′ ) 4 π k 3 E Π ( k ) . (253)
1 ρ ˉ 2 ⟨ T i j ( k ) T l m ∗ ( k ′ ) ⟩ A i j l m ( k ^ ) = ( 2 π ) 6 δ 3 ( k − k ′ ) i H Π ( k ) 4 π k 3 , (254) \frac{1}{\bar{\rho}^2}\langle T_{ij}(\boldsymbol{k})T_{lm}^*(\boldsymbol{k}')\rangle A_{ijlm}(\hat{\boldsymbol{k}})
=(2\pi)^6\delta^3(\boldsymbol{k}-\boldsymbol{k}')i\frac{H_{\Pi}(k)}{4\pi k^3}, \tag*{(254)} ρ ˉ 2 1 ⟨ T ij ( k ) T l m ∗ ( k ′ )⟩ A ij l m ( k ^ ) = ( 2 π ) 6 δ 3 ( k − k ′ ) i 4 π k 3 H Π ( k ) , (254)
where we omit the arguments τ 1 \tau_1 τ 1 and τ 2 \tau_2 τ 2 for brevity as the results of this section are not affected by the times at which they are evaluated. We have used the orthogonality between Λ i j l m \Lambda_{ijlm} Λ ij l m and A i j l m A_{ijlm} A ij l m , such that Λ i j l m A i j l m = 0 \Lambda_{ijlm}A_{ijlm}=0 Λ ij l m A ij l m = 0 , and the norm of each tensor, A i j l m A i j l m = Λ i j l m Λ i j l m = 2 A_{ijlm}A_{ijlm}=\Lambda_{ijlm}\Lambda_{ijlm}=2 A ij l m A ij l m = Λ ij l m Λ ij l m = 2 . We have also used the following properties to relate the two-point correlation function of Π i j = a − 2 Λ i j l m T l m / ρ ˉ \Pi_{ij}=a^{-2}\Lambda_{ijlm}T_{lm}/\bar{\rho} Π ij = a − 2 Λ ij l m T l m / ρ ˉ with that of T i j T_{ij} T ij ,
Λ i j l m Λ i j a b Λ l m c d = Λ a b c d , (255) \Lambda_{ijlm}\Lambda_{ijab}\Lambda_{lmcd}=\Lambda_{abcd}, \tag*{(255)} Λ ij l m Λ ij ab Λ l m c d = Λ ab c d , (255)
A i j l m Λ i j a b Λ l m c d = A a b c d . (256) \mathcal{A}_{ijlm}\Lambda_{ijab}\Lambda_{lmcd}=\mathcal{A}_{abcd}. \tag*{(256)} A ij l m Λ ij ab Λ l m c d = A ab c d . (256)
The two-point correlation function of the stresses T i j T_{ij} T ij corresponding to the source φ i \varphi_i φ i is given in Eq. (77) ,
⟨ T i j ( k ) T l m ∗ ( k ′ ) ⟩ = ∫ d 3 p ( 2 π ) 3 ∫ d 3 q ( 2 π ) 3 × ⟨ φ i ( p ~ ) φ j ( p ) φ l ∗ ( q ~ ) φ m ∗ ( q ) ⟩ , (257) \langle T_{ij}(k)T^{*}_{lm}(k')\rangle=\int\frac{\mathrm{d}^{3}p}{(2\pi)^{3}}\int\frac{\mathrm{d}^{3}q}{(2\pi)^{3}}\times\langle\varphi_i(\tilde{p})\varphi_j(p)\varphi^{*}_l(\tilde{q})\varphi^{*}_m(q)\rangle, \tag*{(257)} ⟨ T ij ( k ) T l m ∗ ( k ′ )⟩ = ∫ ( 2 π ) 3 d 3 p ∫ ( 2 π ) 3 d 3 q × ⟨ φ i ( p ~ ) φ j ( p ) φ l ∗ ( q ~ ) φ m ∗ ( q )⟩ , (257)
where p ~ = k − p \tilde{p}=k-p p ~ = k − p and q ~ = k ′ − q \tilde{q}=k'-q q ~ = k ′ − q . Under the assumption that the sourcing fields φ i \varphi_i φ i are statistically Gaussian, Isserlis’ theorem is applied to decompose the four-point correlation function in Eq. (257) into a superposition of products of two-point correlation functions. Introducing the two-point correlation of the source, F i j φ ( k ) \mathcal{F}^{\varphi}_{ij}(k) F ij φ ( k ) , the resulting spectra of the anisotropic stresses for a source φ = { X , B , ϕ } \varphi=\{X,B,\phi\} φ = { X , B , ϕ } are
a 4 E Π φ ( k ) = k 3 2 π ρ ˉ 2 ∫ d 3 p p 3 p ~ 3 Λ i j l m ( k ^ ) × F i l φ ( p ) F j m φ ( p ~ ) , (258) a^{4}E^{\varphi}_{\Pi}(k)=\frac{k^{3}}{2\pi\bar{\rho}^{2}}\int\frac{\mathrm{d}^{3}p}{p^{3}\tilde{p}^{3}}\Lambda_{ijlm}(\hat{k})\times\mathcal{F}^{\varphi}_{il}(p)\mathcal{F}^{\varphi}_{jm}(\tilde{p}), \tag*{(258)} a 4 E Π φ ( k ) = 2 π ρ ˉ 2 k 3 ∫ p 3 p ~ 3 d 3 p Λ ij l m ( k ^ ) × F i l φ ( p ) F j m φ ( p ~ ) , (258)
a 4 H Π φ ( k ) = k 3 2 π ρ ˉ 2 ∫ d 3 p p 3 p ~ 3 A i j l m ( k ^ ) × F i l φ ( p ) F j m φ ( p ~ ) . (259) a^{4}H^{\varphi}_{\Pi}(k)=\frac{k^{3}}{2\pi\bar{\rho}^{2}}\int\frac{\mathrm{d}^{3}p}{p^{3}\tilde{p}^{3}}\mathcal{A}_{ijlm}(\hat{k})\times\mathcal{F}^{\varphi}_{il}(p)\mathcal{F}^{\varphi}_{jm}(\tilde{p}). \tag*{(259)} a 4 H Π φ ( k ) = 2 π ρ ˉ 2 k 3 ∫ p 3 p ~ 3 d 3 p A ij l m ( k ^ ) × F i l φ ( p ) F j m φ ( p ~ ) . (259)
where we have omitted the homogeneous contribution proportional to δ 3 ( k ) \delta^{3}(k) δ 3 ( k ) , and used the symmetries of the Λ i j l m \Lambda_{ijlm} Λ ij l m and A i j l m \mathcal{A}_{ijlm} A ij l m tensors under j ↔ m j\leftrightarrow m j ↔ m and i ↔ l i\leftrightarrow l i ↔ l transformations to combine the two remaining terms from Isserlis’ decomposition.
The tensor structure of the sourcing field φ \varphi φ can in general be decomposed into a longitudinal, a normal, and a helical component for any statistically homogeneous and isotropic field that violates mirror symmetry [see Eqs. (30)],
F i j φ ( k ) = 1 2 P i j ( k ^ ) E N φ ( k ) + k ^ i k ^ j E L φ ( k ) + i 2 ϵ i j l k ^ l H φ ( k ) . (260) \mathcal{F}^{\varphi}_{ij}(k)=\frac{1}{2}P_{ij}(\hat{k})E^{\varphi}_{N}(k)+\hat{k}_{i}\hat{k}_{j}E^{\varphi}_{L}(k)+\frac{i}{2}\epsilon_{ijl}\hat{k}_{l}H_{\varphi}(k). \tag*{(260)} F ij φ ( k ) = 2 1 P ij ( k ^ ) E N φ ( k ) + k ^ i k ^ j E L φ ( k ) + 2 i ϵ ij l k ^ l H φ ( k ) . (260)
For velocity fields, we consider all the components while magnetic fields are divergence-free and, hence, their longitudinal component always vanishes, E L B ( k ) = 0 E^{B}_{L}(k)=0 E L B ( k ) = 0 . The gradients of the scalar field can only present a longitudinal component, F L ϕ ( k ) = k 2 E ϕ ( k ) \mathcal{F}^{\phi}_{L}(k)=k^{2}E_{\phi}(k) F L ϕ ( k ) = k 2 E ϕ ( k ) , so both E N ϕ ( k ) = H ϕ ( k ) = 0 E^{\phi}_{N}(k)=H_{\phi}(k)=0 E N ϕ ( k ) = H ϕ ( k ) = 0 .
Introducing the general decomposition of Eq. (260) into the computation of the symmetric anisotropic stresses spectrum E Π ( k ) E_{\Pi}(k) E Π ( k ) of a single source [see Eq. (B4a)], we find that all cross-terms between N or L components and H vanish, so only NN (vortical), LL (compressional), NL (mixed), and HH (helical) products survive, yielding the following decomposition of E Π ( k ) E_{\Pi}(k) E Π ( k ) ,
E Π ( k ) = E Π v o r t ( k ) + E Π c o m p ( k ) + E Π m i x ( k ) + E Π h e l ( k ) . (261) E_{\Pi}(k)=E^{\mathrm{vort}}_{\Pi}(k)+E^{\mathrm{comp}}_{\Pi}(k)+E^{\mathrm{mix}}_{\Pi}(k)+E^{\mathrm{hel}}_{\Pi}(k). \tag*{(261)} E Π ( k ) = E Π vort ( k ) + E Π comp ( k ) + E Π mix ( k ) + E Π hel ( k ) . (261)
Hence, the vortical (vort) and compressional (comp) components are found when either E N ( k ) E_{N}(k) E N ( k ) or E L ( k ) E_{L}(k) E L ( k ) are non-zero respectively. The mixed (mix) component is found when both of them are non-zero, and the helical (hel) one when H ( k ) ≠ 0 H(k)\ne0 H ( k ) = 0 , which also requires E N ( k ) ≠ 0 E_{N}(k)\ne0 E N ( k ) = 0 due to the realizability condition ∣ H ( k ) ∣ ≤ E N ( k ) |H(k)|\le E_{N}(k) ∣ H ( k ) ∣ ≤ E N ( k ) .
Similarly, introducing the general decomposition of Eq. (260) into the computation of the antisymmetric H Π ( k ) H_{\Pi}(k) H Π ( k ) [see Eq. (B4b)], only cross-terms of H with N or L do not vanish, yielding the following split into vortical and compressional components,
H Π ( k ) = H Π v o r t ( k ) + H Π c o m p ( k ) . (262) H_{\Pi}(k)=H^{\mathrm{vort}}_{\Pi}(k)+H^{\mathrm{comp}}_{\Pi}(k). \tag*{(262)} H Π ( k ) = H Π vort ( k ) + H Π comp ( k ) . (262)
They are non-zero only when H ( k ) ≠ 0 H(k)\ne0 H ( k ) = 0 , which also implies E N ( k ) ≠ 0 E_{N}(k)\ne0 E N ( k ) = 0 . When E L ( k ) ≠ 0 E_{L}(k)\ne0 E L ( k ) = 0 , then we also find the compressional component.
Due to statistical isotropy, we convert the integrals over d 3 p \mathrm{d}^{3}p d 3 p in Eqs. (B4) into integrals over 2 π p 2 d p d z 2\pi p^{2}\mathrm{d}p\,\mathrm{d}z 2 π p 2 d p d z , where z = k ^ ⋅ p z=\hat{k}\cdot p z = k ^ ⋅ p . Then, since p ~ = p 2 + k 2 − 2 p k z \tilde{p}=\sqrt{p^{2}+k^{2}-2pkz} p ~ = p 2 + k 2 − 2 p k z , we can convert the integral over z z z as an integral over p ~ \tilde{p} p ~ using p ~ d p ~ = − p k d z \tilde{p}\,\mathrm{d}\tilde{p}=-pk\mathrm{d}z p ~ d p ~ = − p k d z and noting that the integration limits z ∈ [ − 1 , 1 ] z\in[-1,1] z ∈ [ − 1 , 1 ] correspond to p ~ ∈ [ k + p , ∣ k − p ∣ ] \tilde{p}\in[k+p,|k-p|] p ~ ∈ [ k + p , ∣ k − p ∣ ] . This allows us to preserve the symmetry over p , p ~ p,\tilde{p} p , p ~ in the integration. After projecting the two-point correlators with Λ i j l m \Lambda_{ijlm} Λ ij l m and A i j l m \mathcal{A}_{ijlm} A ij l m , respectively, we find the following components of E Π ( k ) E_{\Pi}(k) E Π ( k ) and H Π ( k ) H_{\Pi}(k) H Π ( k ) ,
a 4 E Π v o r t ( k ) = k 2 8 ρ ˉ 2 ∫ 0 ∞ d p p 2 E N ( p ) ∫ ∣ k − p ∣ k + p d p ~ p ~ 2 E N ( p ~ ) ( 1 + z 2 ) ( 1 + z ~ 2 ) , (263) a^{4}E_{\Pi}^{\mathrm{vort}}(k)=\frac{k^{2}}{8\bar{\rho}^{2}}\int_{0}^{\infty}\frac{\mathrm{d}p}{p^{2}}E_{N}(p)\int_{\lvert k-p\rvert}^{k+p}\frac{\mathrm{d}\tilde{p}}{\tilde{p}^{2}}E_{N}(\tilde{p})(1+z^{2})(1+\tilde{z}^{2}), \tag*{(263)} a 4 E Π vort ( k ) = 8 ρ ˉ 2 k 2 ∫ 0 ∞ p 2 d p E N ( p ) ∫ ∣ k − p ∣ k + p p ~ 2 d p ~ E N ( p ~ ) ( 1 + z 2 ) ( 1 + z ~ 2 ) , (263)
a 4 E Π c o m p ( k ) = k 2 2 ρ ˉ 2 ∫ 0 ∞ d p p 2 E L ( p ) ∫ ∣ k − p ∣ k + p d p ~ p ~ 2 E L ( p ~ ) ( 1 − z 2 ) ( 1 − z ~ 2 ) , (264) a^{4}E_{\Pi}^{\mathrm{comp}}(k)=\frac{k^{2}}{2\bar{\rho}^{2}}\int_{0}^{\infty}\frac{\mathrm{d}p}{p^{2}}E_{L}(p)\int_{\lvert k-p\rvert}^{k+p}\frac{\mathrm{d}\tilde{p}}{\tilde{p}^{2}}E_{L}(\tilde{p})(1-z^{2})(1-\tilde{z}^{2}), \tag*{(264)} a 4 E Π comp ( k ) = 2 ρ ˉ 2 k 2 ∫ 0 ∞ p 2 d p E L ( p ) ∫ ∣ k − p ∣ k + p p ~ 2 d p ~ E L ( p ~ ) ( 1 − z 2 ) ( 1 − z ~ 2 ) , (264)
a 4 E Π m i x ( k ) = k 2 2 ρ ˉ 2 ∫ 0 ∞ d p p 2 E L ( p ) ∫ ∣ k − p ∣ k + p d p ~ p ~ 2 E N ( p ~ ) ( 1 − z 2 z ~ 2 ) , (265) a^{4}E_{\Pi}^{\mathrm{mix}}(k)=\frac{k^{2}}{2\bar{\rho}^{2}}\int_{0}^{\infty}\frac{\mathrm{d}p}{p^{2}}E_{L}(p)\int_{\lvert k-p\rvert}^{k+p}\frac{\mathrm{d}\tilde{p}}{\tilde{p}^{2}}E_{N}(\tilde{p})(1-z^{2}\tilde{z}^{2}), \tag*{(265)} a 4 E Π mix ( k ) = 2 ρ ˉ 2 k 2 ∫ 0 ∞ p 2 d p E L ( p ) ∫ ∣ k − p ∣ k + p p ~ 2 d p ~ E N ( p ~ ) ( 1 − z 2 z ~ 2 ) , (265)
a 4 E Π h e l ( k ) = k 2 2 ρ ˉ 2 ∫ 0 ∞ d p p 2 H ( p ) ∫ ∣ k − p ∣ k + p d p ~ p ~ 2 H ( p ~ ) z z ~ , (266) a^{4}E_{\Pi}^{\mathrm{hel}}(k)=\frac{k^{2}}{2\bar{\rho}^{2}}\int_{0}^{\infty}\frac{\mathrm{d}p}{p^{2}}H(p)\int_{\lvert k-p\rvert}^{k+p}\frac{\mathrm{d}\tilde{p}}{\tilde{p}^{2}}H(\tilde{p})z\tilde{z}, \tag*{(266)} a 4 E Π hel ( k ) = 2 ρ ˉ 2 k 2 ∫ 0 ∞ p 2 d p H ( p ) ∫ ∣ k − p ∣ k + p p ~ 2 d p ~ H ( p ~ ) z z ~ , (266)
a 4 H Π v o r t ( k ) = k 2 4 ρ ˉ 2 ∫ 0 ∞ d p p 2 E N ( p ) ∫ ∣ k − p ∣ k + p d p ~ p ~ 2 H ( p ~ ) z ( 1 + z ~ 2 ) , (267) a^{4}H_{\Pi}^{\mathrm{vort}}(k)=\frac{k^{2}}{4\bar{\rho}^{2}}\int_{0}^{\infty}\frac{\mathrm{d}p}{p^{2}}E_{N}(p)\int_{\lvert k-p\rvert}^{k+p}\frac{\mathrm{d}\tilde{p}}{\tilde{p}^{2}}H(\tilde{p})z(1+\tilde{z}^{2}), \tag*{(267)} a 4 H Π vort ( k ) = 4 ρ ˉ 2 k 2 ∫ 0 ∞ p 2 d p E N ( p ) ∫ ∣ k − p ∣ k + p p ~ 2 d p ~ H ( p ~ ) z ( 1 + z ~ 2 ) , (267)
a 4 H Π c o m p ( k ) = k 2 2 ρ ˉ 2 ∫ 0 ∞ d p p 2 E L ( p ) ∫ ∣ k − p ∣ k + p d p ~ p ~ 2 H ( p ~ ) z ( 1 − z ~ 2 ) , (268) a^{4}H_{\Pi}^{\mathrm{comp}}(k)=\frac{k^{2}}{2\bar{\rho}^{2}}\int_{0}^{\infty}\frac{\mathrm{d}p}{p^{2}}E_{L}(p)\int_{\lvert k-p\rvert}^{k+p}\frac{\mathrm{d}\tilde{p}}{\tilde{p}^{2}}H(\tilde{p})z(1-\tilde{z}^{2}), \tag*{(268)} a 4 H Π comp ( k ) = 2 ρ ˉ 2 k 2 ∫ 0 ∞ p 2 d p E L ( p ) ∫ ∣ k − p ∣ k + p p ~ 2 d p ~ H ( p ~ ) z ( 1 − z ~ 2 ) , (268)
where z = k ^ ⋅ p ^ z=\hat{k}\cdot\hat{p} z = k ^ ⋅ p ^ and z ~ = k ^ ⋅ p ~ ^ \tilde{z}=\hat{k}\cdot\hat{\tilde{p}} z ~ = k ^ ⋅ p ~ ^ . Explicitly in terms of k k k , p p p , and p ~ \tilde{p} p ~ , these cosines are
z = k 2 + p 2 − p ~ 2 2 k p , z ~ = k 2 + p ~ 2 − p 2 2 k p ~ . (269) z=\frac{k^{2}+p^{2}-\tilde{p}^{2}}{2kp},\qquad\tilde{z}=\frac{k^{2}+\tilde{p}^{2}-p^{2}}{2k\tilde{p}}. \tag*{(269)} z = 2 k p k 2 + p 2 − p ~ 2 , z ~ = 2 k p ~ k 2 + p ~ 2 − p 2 . (269)
We note that the terms corresponding to a product of different spectral components, like E Π m i x E_{\Pi}^{\mathrm{mix}} E Π mix and H Π j H_{\Pi}^{j} H Π j with j = { v o r t , c o m p } j=\{\mathrm{vort},\mathrm{comp}\} j = { vort , comp } , are not fully symmetric as each function of the product is either evaluated at p p p or at p ~ \tilde{p} p ~ . However, since the integral is symmetric in p ↔ p ~ p\leftrightarrow\tilde{p} p ↔ p ~ , we have combined the two products into a single integral.
We consider the normalization of the spectral components as in Eqs. (58),
E i φ ( K ) = φ r m s 2 A a b ( α ) ζ i φ ( K ) , (270) E_{i}^{\varphi}(K)=\frac{\varphi_{\mathrm{rms}}^{2}}{A_{ab}(\alpha)}\zeta_{i}^{\varphi}(K), \tag*{(270)} E i φ ( K ) = A ab ( α ) φ rms 2 ζ i φ ( K ) , (270)
with i = L , N , H i=L,N,H i = L , N , H , and K = k / k ∗ K=k/k_{\ast} K = k / k ∗ , and where we have defined the components ζ i u ( K ) \zeta_{i}^{u}(K) ζ i u ( K ) as
ζ L u ( K ) = q ( K ) ζ u ( K ) , (271) \zeta_{L}^{u}(K)=q(K)\zeta_{u}(K), \tag*{(271)} ζ L u ( K ) = q ( K ) ζ u ( K ) , (271)
ζ N u ( K ) = [ 1 − q ( K ) ] ζ u ( K ) , (272) \zeta_{N}^{u}(K)=[1-q(K)]\zeta_{u}(K), \tag*{(272)} ζ N u ( K ) = [ 1 − q ( K )] ζ u ( K ) , (272)
ζ H u ( K ) = ϵ u ( K ) [ 1 − q ( K ) ] ζ u ( K ) . (273) \zeta_{H}^{u}(K)=\epsilon_{u}(K)[1-q(K)]\zeta_{u}(K). \tag*{(273)} ζ H u ( K ) = ϵ u ( K ) [ 1 − q ( K )] ζ u ( K ) . (273)
Then, the normalized stress spectra ζ Π ( K ) \zeta_{\Pi}(K) ζ Π ( K ) are defined such that ∣ ζ Π ( 0 ) ∣ = 1 \lvert\zeta_{\Pi}(0)\rvert=1 ∣ ζ Π ( 0 )∣ = 1 [34 ] [1] [see also Eq. (86) ],
E Π φ , j ( K ) = φ ~ r m s 4 ρ ˉ 2 1 A a b 2 K 3 C j φ ζ Π φ , j ( K ) , (274) E_{\Pi}^{\varphi,j}(K)=\frac{\tilde{\varphi}_{\mathrm{rms}}^{4}}{\bar{\rho}^{2}}\frac{1}{A_{ab}^{2}}K^{3}C_{j}^{\varphi}\zeta_{\Pi}^{\varphi,j}(K), \tag*{(274)} E Π φ , j ( K ) = ρ ˉ 2 φ ~ rms 4 A ab 2 1 K 3 C j φ ζ Π φ , j ( K ) , (274)
where j = { v o r t , c o m p , m i x , h e l } j=\{\mathrm{vort},\mathrm{comp},\mathrm{mix},\mathrm{hel}\} j = { vort , comp , mix , hel } and φ ~ r m s = φ r m s / a \tilde{\varphi}_{\mathrm{rms}}=\varphi_{\mathrm{rms}}/a φ ~ rms = φ rms / a . The geometrical factor C j C_{j} C j is chosen to ensure that ∣ ζ Π j ( 0 ) ∣ = 1 \lvert\zeta_{\Pi}^{j}(0)\rvert=1 ∣ ζ Π j ( 0 )∣ = 1 [34 ] [1]. Similarly, the normalized spectrum of the chiral stresses, ζ Π H , j ( K ) \zeta_{\Pi}^{H,j}(K) ζ Π H , j ( K ) [see Eq. (117) ], are defined such that
H Π φ , j ( K ) = φ ~ r m s 4 ρ ˉ 2 1 A a b 2 K 4 C j φ , H ζ Π φ , H , j ( K ) , (275) H_{\Pi}^{\varphi,j}(K)=\frac{\tilde{\varphi}_{\mathrm{rms}}^{4}}{\bar{\rho}^{2}}\frac{1}{A_{ab}^{2}}K^{4}C_{j}^{\varphi,H}\zeta_{\Pi}^{\varphi,H,j}(K), \tag*{(275)} H Π φ , j ( K ) = ρ ˉ 2 φ ~ rms 4 A ab 2 1 K 4 C j φ , H ζ Π φ , H , j ( K ) , (275)
with j = { v o r t , c o m p } j=\{\mathrm{vort},\mathrm{comp}\} j = { vort , comp } . Note that the components of the antisymmetric (chiral) stress spectrum are found to be proportional to K 4 K^{4} K 4 in the K → 0 K\to0 K → 0 limit, compatible with the expected causality condition [41 , 79 ] .
All the symmetric and antisymmetric normalized components of the anisotropic stress spectrum are
[ C ζ Π ( K ) ] v o r t = 1 8 K ∫ d P P 2 ζ N ( P ) ∫ d P ~ P ~ 2 ζ N ( P ~ ) ( 1 + z ~ 2 ) ( 1 + z 2 ) (276) [C\zeta_{\Pi}(K)]^{\mathrm{vort}}=\frac{1}{8K}\int\frac{\mathrm{d}P}{P^2}\zeta_N(P)\int\frac{\mathrm{d}\widetilde{P}}{\widetilde{P}^2}\zeta_N(\widetilde{P})(1+\widetilde{z}^{2})(1+z^{2}) \tag*{(276)} [ C ζ Π ( K ) ] vort = 8 K 1 ∫ P 2 d P ζ N ( P ) ∫ P 2 d P ζ N ( P ) ( 1 + z 2 ) ( 1 + z 2 ) (276)
[ C ζ Π ( K ) ] c o m p = 1 2 K ∫ d P P 2 ζ L ( P ) ∫ d P ~ P ~ 2 ζ L ( P ~ ) ( 1 − z 2 ) ( 1 − z ~ 2 ) (277) [C\zeta_{\Pi}(K)]^{\mathrm{comp}}=\frac{1}{2K}\int\frac{\mathrm{d}P}{P^2}\zeta_L(P)\int\frac{\mathrm{d}\widetilde{P}}{\widetilde{P}^2}\zeta_L(\widetilde{P})(1-z^{2})(1-\widetilde{z}^{2}) \tag*{(277)} [ C ζ Π ( K ) ] comp = 2 K 1 ∫ P 2 d P ζ L ( P ) ∫ P 2 d P ζ L ( P ) ( 1 − z 2 ) ( 1 − z 2 ) (277)
[ C ζ Π ( K ) ] m i x = 1 2 K ∫ d P P 2 ζ L ( P ) ∫ d P ~ P ~ 2 ζ N ( P ~ ) ( 1 − z 2 z ~ 2 ) (278) [C\zeta_{\Pi}(K)]^{\mathrm{mix}}=\frac{1}{2K}\int\frac{\mathrm{d}P}{P^2}\zeta_L(P)\int\frac{\mathrm{d}\widetilde{P}}{\widetilde{P}^2}\zeta_N(\widetilde{P})(1-z^{2}\widetilde{z}^{2}) \tag*{(278)} [ C ζ Π ( K ) ] mix = 2 K 1 ∫ P 2 d P ζ L ( P ) ∫ P 2 d P ζ N ( P ) ( 1 − z 2 z 2 ) (278)
[ C ζ Π ( K ) ] h e l = 1 2 K ∫ d P P 2 ζ H ( P ) ∫ d P ~ P ~ 2 ζ H ( P ~ ) z z ~ (279) [C\zeta_{\Pi}(K)]^{\mathrm{hel}}=\frac{1}{2K}\int\frac{\mathrm{d}P}{P^2}\zeta_H(P)\int\frac{\mathrm{d}\widetilde{P}}{\widetilde{P}^2}\zeta_H(\widetilde{P})z\widetilde{z} \tag*{(279)} [ C ζ Π ( K ) ] hel = 2 K 1 ∫ P 2 d P ζ H ( P ) ∫ P 2 d P ζ H ( P ) z z (279)
[ C ζ Π ( K ) ] H v o r t = 1 8 K 2 ∫ d P P 2 ζ N ( P ) ∫ d P ~ P ~ 2 ζ H ( P ~ ) [ z ( 1 + z ~ 2 ) + z ~ ( 1 + z 2 ) ] (280) [C\zeta_{\Pi}(K)]^{\mathrm{vort}}_H=\frac{1}{8K^2}\int\frac{\mathrm{d}P}{P^2}\zeta_N(P)\int\frac{\mathrm{d}\widetilde{P}}{\widetilde{P}^2}\zeta_H(\widetilde{P})\left[z(1+\widetilde{z}^{2})+\widetilde{z}(1+z^{2})\right] \tag*{(280)} [ C ζ Π ( K ) ] H vort = 8 K 2 1 ∫ P 2 d P ζ N ( P ) ∫ P 2 d P ζ H ( P ) [ z ( 1 + z 2 ) + z ( 1 + z 2 ) ] (280)
[ C ζ Π ( K ) ] H c o m p = 1 4 K 2 ∫ d P P 2 ζ L ( P ) ∫ d P ~ P ~ 2 ζ H ( P ~ ) [ z ( 1 − z ~ 2 ) + z ~ ( 1 − z 2 ) ] (281) [C\zeta_{\Pi}(K)]^{\mathrm{comp}}_H=\frac{1}{4K^2}\int\frac{\mathrm{d}P}{P^2}\zeta_L(P)\int\frac{\mathrm{d}\widetilde{P}}{\widetilde{P}^2}\zeta_H(\widetilde{P})\left[z(1-\widetilde{z}^{2})+\widetilde{z}(1-z^{2})\right] \tag*{(281)} [ C ζ Π ( K ) ] H comp = 4 K 2 1 ∫ P 2 d P ζ L ( P ) ∫ P 2 d P ζ H ( P ) [ z ( 1 − z 2 ) + z ( 1 − z 2 ) ] (281)
where we have also defined P = p / k ∗ P=p/k_\ast P = p / k ∗ and P ~ = p ~ / k ∗ \widetilde{P}=\widetilde{p}/k_\ast P = p / k ∗ . The normalized components ζ Π i j ( K ) \zeta_{\Pi}^{ij}(K) ζ Π ij ( K ) of the symmetric spectrum are shown in Figure 2 , while the sum of the components ζ Π H , i j ( K ) \zeta_{\Pi}^{H,ij}(K) ζ Π H , ij ( K ) of the antisymmetric spectrum is shown in Figure 6 .
The geometrical factors C j C_j C j correspond to the absolute values of the limits of the symmetric components C j ζ Π i j ( K ) C_j\zeta_{\Pi}^{ij}(K) C j ζ Π ij ( K ) when K → 0 K\to0 K → 0 . They can be directly found taking K → 0 K\to0 K → 0 , P ~ → P \widetilde{P}\to P P → P , and z ~ → − z \widetilde{z}\to-z z → − z in Eqs. (B14a) and (B14d), and transforming the integrals over P ~ \widetilde{P} P into integrals over z z z ,
C v o r t = 7 15 ∫ 0 ∞ ζ N 2 ( P ) P 4 d P , (282) C_{\mathrm{vort}}=\frac{7}{15}\int_0^\infty\frac{\zeta_N^2(P)}{P^4}\,\mathrm{d}P, \tag*{(282)} C vort = 15 7 ∫ 0 ∞ P 4 ζ N 2 ( P ) d P , (282)
C c o m p = 8 15 ∫ 0 ∞ ζ L 2 ( P ) P 4 d P , (283) C_{\mathrm{comp}}=\frac{8}{15}\int_0^\infty\frac{\zeta_L^2(P)}{P^4}\,\mathrm{d}P, \tag*{(283)} C comp = 15 8 ∫ 0 ∞ P 4 ζ L 2 ( P ) d P , (283)
C m i x = 4 5 ∫ 0 ∞ ζ L ( P ) ζ N ( P ) P 4 d P , (284) C_{\mathrm{mix}}=\frac{4}{5}\int_0^\infty\frac{\zeta_L(P)\zeta_N(P)}{P^4}\,\mathrm{d}P, \tag*{(284)} C mix = 5 4 ∫ 0 ∞ P 4 ζ L ( P ) ζ N ( P ) d P , (284)
C h e l = 1 3 ∫ 0 ∞ ζ H 2 ( P ) P 4 d P . (285) C_{\mathrm{hel}}=\frac{1}{3}\int_0^\infty\frac{\zeta_H^2(P)}{P^4}\,\mathrm{d}P. \tag*{(285)} C hel = 3 1 ∫ 0 ∞ P 4 ζ H 2 ( P ) d P . (285)
All of them have a different coefficient c j c_j c j [see also Eqs. (75)] multiplying the integral of ζ i 2 ( P ) / P 4 \zeta_i^2(P)/P^4 ζ i 2 ( P ) / P 4 over P P P . Note that, unlike the other contributions, ζ Π h e l ( K ) → − 1 \zeta_{\Pi}^{\mathrm{hel}}(K)\to-1 ζ Π hel ( K ) → − 1 in the K → 0 K\to0 K → 0 limit.
Using the smoothed broken-power law of App. A, we first note that any power of the normalized function ζ i n ( K ) \zeta_i^n(K) ζ i n ( K ) can be expressed using Eq. (227) with the mapping a → a n a\to an a → an , b → b n b\to bn b → bn , α → α / n \alpha\to\alpha/n α → α / n . Then, the integrals in Eqs. (B15) become
C a b j ( α ) c j = I 2 a , 2 b ; − 4 ( α / 2 ) = ( a + b b ) 2 α ( a b ) − 2 a − 3 α ( a + b ) × Γ [ 2 a − 3 α ( a + b ) ] Γ [ 2 b + 3 α ( a + b ) ] α ( a + b ) Γ [ 2 α ] , (286) \frac{C_{ab}^{j}(\alpha)}{c_j}=I_{2a,2b;-4}(\alpha/2)=\left(\frac{a+b}{b}\right)^{\frac{2}{\alpha}}\left(\frac{a}{b}\right)^{-\frac{2a-3}{\alpha(a+b)}}\times\frac{\Gamma\left[\frac{2a-3}{\alpha(a+b)}\right]\Gamma\left[\frac{2b+3}{\alpha(a+b)}\right]}{\alpha(a+b)\Gamma\left[\frac{2}{\alpha}\right]}, \tag*{(286)} c j C ab j ( α ) = I 2 a , 2 b ; − 4 ( α /2 ) = ( b a + b ) α 2 ( b a ) − α ( a + b ) 2 a − 3 × α ( a + b ) Γ [ α 2 ] Γ [ α ( a + b ) 2 a − 3 ] Γ [ α ( a + b ) 2 b + 3 ] , (286)
with c j = 7 15 ( 1 − q ) 2 c_j=\frac{7}{15}(1-q)^2 c j = 15 7 ( 1 − q ) 2 , 8 15 q 2 \frac{8}{15}q^2 15 8 q 2 , 4 5 q ( 1 − q ) \frac{4}{5}q(1-q) 5 4 q ( 1 − q ) , and 1 3 ε 2 ( 1 − q 2 ) \frac{1}{3}\varepsilon^2(1-q^2) 3 1 ε 2 ( 1 − q 2 ) , for the vortical, compressional, mixed, and helical components, respectively. We have assumed that the degree of compressibility q q q and the fractional helicity ε \varepsilon ε are constants, such that all components have the same spectral shape. The value of I 2 a , 2 b ; − 4 ( α / 2 ) I_{2a,2b;-4}(\alpha/2) I 2 a , 2 b ; − 4 ( α /2 ) is shown in Figure 30 as a function of α \alpha α for specific choices of the slopes a a a and b b b . In particular, when the broken power law becomes a piece-wise function [see Eq. (238) ], we find
lim α → ∞ C a b j ( α ) c j = 2 ( a + b ) ( 2 a − 3 ) ( 2 b + 3 ) . (287) \lim_{\alpha\to\infty}\frac{C_{ab}^{j}(\alpha)}{c_j}=\frac{2(a+b)}{(2a-3)(2b+3)}. \tag*{(287)} α → ∞ lim c j C ab j ( α ) = ( 2 a − 3 ) ( 2 b + 3 ) 2 ( a + b ) . (287)
In previous work [34 , 35 ] [1, 18], the normalized spectral function of the source was considered with a different normalization, such that their ζ ( K ) \zeta(K) ζ ( K ) corresponds to our ζ − 1 ( K ~ ) \zeta_{-1}(\widetilde{K}) ζ − 1 ( K ) [see Eq. (243) ], defined such that
φ r m s 2 E ∗ 2 = ∫ − ∞ ∞ ζ ( K ) d ln K = K ∗ − 1 ζ − 1 ∗ ∫ 0 ∞ ζ − 1 ( K ~ ) d K ~ . (288) \begin{aligned}
\frac{\varphi_{\mathrm{rms}}^2}{E_\ast^2}=\int_{-\infty}^{\infty}\zeta(K)\,\mathrm{d}\ln K \\
&=K_\ast^{-1}\zeta_{-1}^{\ast}\int_0^\infty\zeta_{-1}(\widetilde{K})\,\mathrm{d}\widetilde{K}. \tag*{(288)}
\end{aligned} E ∗ 2 φ rms 2 = ∫ − ∞ ∞ ζ ( K ) d ln K = K ∗ − 1 ζ − 1 ∗ ∫ 0 ∞ ζ − 1 ( K ) d K . (288)
Based on these normalizations, we can relate the moments of a power of the normalized function ζ l ( K ) \zeta^l(K) ζ l ( K ) with those of the rescaled function ζ m l ( K ~ ) \zeta_{m}^{l}(\widetilde{K}) ζ m l ( K ) ,
I l a , l b ; l m + n ( α / l ) = ∫ 0 ∞ ζ l ( K ) K l m + n d K = [ K m ∗ ] n + 1 ζ m ∗ l I l ( a + m ) , l ( b − m ) ; n ( α / l ) . (289) I_{la,lb;lm+n}(\alpha/l) = \int_{0}^{\infty} \zeta^{l}(K)K^{lm+n}\,\mathrm{d}K
= [K_{m}^{*}]^{n+1}\zeta_{m}^{*l}I_{l(a+m),l(b-m);n}(\alpha/l). \tag*{(289)} I l a , l b ; l m + n ( α / l ) = ∫ 0 ∞ ζ l ( K ) K l m + n d K = [ K m ∗ ] n + 1 ζ m ∗ l I l ( a + m ) , l ( b − m ) ; n ( α / l ) . (289)
Therefore, setting l = 2 l=2 l = 2 and m = − 1 m=-1 m = − 1 , the integrals in Eq. (286) can be expressed as
C a b j ( α ) = c j [ ζ − 1 ∗ ] 2 K − 1 ∗ 2 I 2 ( a − 1 ) , 2 ( b + 1 ) ; − 2 ( α / 2 ) ≡ [ ζ − 1 ∗ ] 2 K − 1 ∗ 2 C ~ a b j ( α ) , (290) C_{ab}^{j}(\alpha) = c_{j}\frac{[\zeta_{-1}^{*}]^{2}}{K_{-1}^{*2}}I_{2(a-1),2(b+1);-2}(\alpha/2)
\equiv\frac{[\zeta_{-1}^{*}]^{2}}{K_{-1}^{*2}}\widetilde{C}_{ab}^{j}(\alpha), \tag*{(290)} C ab j ( α ) = c j K − 1 ∗ 2 [ ζ − 1 ∗ ] 2 I 2 ( a − 1 ) , 2 ( b + 1 ) ; − 2 ( α /2 ) ≡ K − 1 ∗ 2 [ ζ − 1 ∗ ] 2 C ab j ( α ) , (290)
where C ~ a b ( α ) \widetilde{C}_{ab}(\alpha) C ab ( α ) is the geometrical factor that one finds when using ζ − 1 ( K ~ ) \zeta_{-1}(\widetilde{K}) ζ − 1 ( K ) as done in previous work [34 , 35 ] [1, 18]. C ~ a b ( α ) \widetilde{C}_{ab}(\alpha) C ab ( α ) shown in Figure 30 (right panel), for comparison with C a b ( α ) C_{ab}(\alpha) C ab ( α ) (left panel).
The geometric factor C C C was already introduced in ref. [1] and it becomes C 5 , 2 / 3 v o r t ( 2 ) ≃ 0.34 C_{5,2/3}^{\mathrm{vort}}(2) \simeq0.34 C 5 , 2/3 vort ( 2 ) ≃ 0.34 for a fully vortical non-helical von Kármán spectrum with a = 5 a=5 a = 5 , b = 2 / 3 b=2/3 b = 2/3 , and C 5 , 2 / 3 v o r t ( 6 / 17 ) ≃ 1.96 C_{5,2/3}^{\mathrm{vort}}(6/17) \simeq1.96 C 5 , 2/3 vort ( 6/17 ) ≃ 1.96 when α = 6 / 17 \alpha=6/17 α = 6/17 instead [35 , 50 ] [18]. For purely compressional motion, for example, sound waves studied in ref. [34 ] , with a = 5 a=5 a = 5 , b = 1 b=1 b = 1 , and α = 2 \alpha=2 α = 2 , we find C 5 , 1 c o m p ( 2 ) ≃ 0.34 C_{5,1}^{\mathrm{comp}}(2) \simeq0.34 C 5 , 1 comp ( 2 ) ≃ 0.34 . Since previous work used ζ − 1 ( K ~ ) \zeta_{-1}(\widetilde{K}) ζ − 1 ( K ) instead of our ζ ( K ) \zeta(K) ζ ( K ) , defined in Eq. (243) , as the normalized spectral shape [see discussion around Eq. (247) ], the modified prefactor is given in Eq. (290) ,
C ~ a b j ( α ) c j = ( a + b b + 1 ) 2 α ( a − 1 b + 1 ) − 2 a − 3 α ( a + b ) × Γ [ 2 a − 3 α ( a + b ) ] Γ [ 2 b + 3 α ( a + b ) ] α ( a + b ) Γ [ 2 α ] . (291) \frac{\widetilde{C}_{ab}^{j}(\alpha)}{c_{j}} =
\left(\frac{a+b}{b+1}\right)^{\frac{2}{\alpha}}
\left(\frac{a-1}{b+1}\right)^{-\frac{2a-3}{\alpha(a+b)}}
\times\frac{\Gamma\left[\frac{2a-3}{\alpha(a+b)}\right]\Gamma\left[\frac{2b+3}{\alpha(a+b)}\right]}{\alpha(a+b)\Gamma\left[\frac{2}{\alpha}\right]}. \tag*{(291)} c j C ab j ( α ) = ( b + 1 a + b ) α 2 ( b + 1 a − 1 ) − α ( a + b ) 2 a − 3 × α ( a + b ) Γ [ α 2 ] Γ [ α ( a + b ) 2 a − 3 ] Γ [ α ( a + b ) 2 b + 3 ] . (291)
Using this expression, we recover for vortical turbulence 4 C ~ 5 , 2 / 3 v o r t ( 2 ) ≃ 1.1 4\widetilde{C}_{5,2/3}^{\mathrm{vort}}(2) \simeq1.1 4 C 5 , 2/3 vort ( 2 ) ≃ 1.1 reported in ref. [1] (see their Eq. (22) and note that our velocity and magnetic spectra have a factor of 1 / 2 1/2 1/2 when compared to theirs, introducing a factor of 1 / 4 1/4 1/4 in C ~ \widetilde{C} C ), and 4 C ~ 5 , 2 / 3 v o r t ( 6 / 17 ) ≃ 2.35 4\widetilde{C}_{5,2/3}^{\mathrm{vort}}(6/17) \simeq2.35 4 C 5 , 2/3 vort ( 6/17 ) ≃ 2.35 . For comparison with previous work, this value of C ~ 5 , 2 / 3 v o r t \widetilde{C}_{5,2/3}^{\mathrm{vort}} C 5 , 2/3 vort when α = 6 / 17 \alpha=6/17 α = 6/17 recovers the values 2 C ~ / A ~ 2 ≃ 0.085 2\widetilde{C}/\widetilde{A}^{2} \simeq0.085 2 C / A 2 ≃ 0.085 (see footnote 1 in ref. [18]) and C ~ / ( 2 π 2 A ~ 2 ) ≃ 2 × 10 − 3 \widetilde{C}/(2\pi^{2}\widetilde{A}^{2}) \simeq2 \times10^{-3} C / ( 2 π 2 A 2 ) ≃ 2 × 1 0 − 3 (see footnote 3 in ref. [34 ] ).
The geometrical factors C j H C_{j}^{\mathrm{H}} C j H of the chiral components, with j = { v o r t , c o m p } j=\{\mathrm{vort},\mathrm{comp}\} j = { vort , comp } , correspond to the K → 0 K \to0 K → 0 limits of Eqs. (B14e) and (B14f). However, expanding P ~ → P \widetilde{P} \to P P → P and z ~ → − z \widetilde{z} \to-z z → − z , i.e., up to zero-th order in K K K , is not enough, as the resulting integrand exactly vanishes. This is expected from the causality condition for the spectrum of the anisotropic chiral stresses [41 , 79 ] , and motivated our choice to normalize H Π ( K ) H_{\Pi}(K) H Π ( K ) in Eq. (275) with K 4 K^{4} K 4 , introducing a 1 / K 2 1/K^{2} 1/ K 2 factor in Eqs. (B14e) and (B14f).
To find the leading-order values when K → 0 K \to0 K → 0 , we then need to expand z ~ \widetilde{z} z up to first order in K K K , z ~ → − z + K P ~ ( 1 − z 2 ) \widetilde{z} \to-z + \frac{K}{\widetilde{P}}(1-z^{2}) z → − z + P K ( 1 − z 2 ) and keep the second term in the expansion. We also expand P ~ → P − K z \widetilde{P} \to P-Kz P → P − K z and ζ ~ ( P ~ ) → ζ ( P ) − K ζ ′ ( P ) \tilde{\zeta}(\tilde{P}) \to\zeta(P) - K\zeta'(P) ζ ~ ( P ~ ) → ζ ( P ) − K ζ ′ ( P ) up to first order in K K K . Substituting the expansion of z ~ \tilde{z} z ~ up to first order in K K K , transforming integrals Eqs. (B14e) and (B14f) into integrals over z z z , and integrating by parts, we find
C H v o r t = 2 15 ∣ ∫ 0 ∞ d P P 5 ζ H ( P ) ζ N ( P ) [ 1 + P ζ N ( P ) ε ′ ( P ) ] ∣ (292) \mathcal{C}^{\mathrm{vort}}_{\mathrm{H}} = \frac{2}{15} \left\lvert\int_{0}^{\infty} \frac{\mathrm{d}P}{P^{5}} \zeta_{\mathrm{H}}(P)\zeta_{\mathrm{N}}(P) \left[1+P\zeta_{\mathrm{N}}(P)\varepsilon'(P)\right] \right\rvert \tag*{(292)} C H vort = 15 2 ∫ 0 ∞ P 5 d P ζ H ( P ) ζ N ( P ) [ 1 + P ζ N ( P ) ε ′ ( P ) ] (292)
C H c o m p = 2 5 ∣ ∫ 0 ∞ d P P 5 ζ H ( P ) ζ L ( P ) [ 1 + P ζ L ( P ) ε L ′ ( P ) ] ∣ (293) \mathcal{C}^{\mathrm{comp}}_{\mathrm{H}} = \frac{2}{5} \left\lvert\int_{0}^{\infty} \frac{\mathrm{d}P}{P^{5}} \zeta_{\mathrm{H}}(P)\zeta_{\mathrm{L}}(P) \left[1+P\zeta_{\mathrm{L}}(P)\varepsilon'_{\mathrm{L}}(P)\right] \right\rvert \tag*{(293)} C H comp = 5 2 ∫ 0 ∞ P 5 d P ζ H ( P ) ζ L ( P ) [ 1 + P ζ L ( P ) ε L ′ ( P ) ] (293)
where we have defined ε L ( K ) ≡ H ( K ) / E L ( K ) = ε ( K ) [ 1 − q ( K ) ] / q ( K ) \varepsilon_{\mathrm{L}}(K) \equiv H(K)/E_{\mathrm{L}}(K) = \varepsilon(K)[1-q(K)]/q(K) ε L ( K ) ≡ H ( K ) / E L ( K ) = ε ( K ) [ 1 − q ( K )] / q ( K ) . Note that for constant fractional helicity and degree of compressibility, ε ′ ( P ) = ε L ′ ( P ) = 0 \varepsilon'(P)=\varepsilon'_{\mathrm{L}}(P)=0 ε ′ ( P ) = ε L ′ ( P ) = 0 and the term in brackets for both terms reduces to one.
As before, assuming that ζ i ( P ) \zeta_i(P) ζ i ( P ) is given by the smoothed broken power law of App. A, and taking constant q q q and ε \varepsilon ε , we can express the geometrical factors C j H \mathcal{C}^{\mathrm{H}}_j C j H using Eq. (233) ,
C a b j , H ( α ) C j H = I 2 a , 2 b ; − 5 ( α / 2 ) = ( a + b b ) 2 α ( a b ) − 2 a − 4 α ( a + b ) Γ [ 2 a − 4 α ( a + b ) ] Γ [ 2 b + 4 α ( a + b ) ] α ( a + b ) Γ [ 2 α ] (294) \frac{\mathcal{C}^{j,\mathrm{H}}_{ab}(\alpha)}{\mathcal{C}^{\mathrm{H}}_j}=I_{2a,2b;-5}(\alpha/2)=\left(\frac{a+b}{b}\right)^{\frac{2}{\alpha}}\left(\frac{a}{b}\right)^{-\frac{2a-4}{\alpha(a+b)}}\frac{\Gamma\left[\frac{2a-4}{\alpha(a+b)}\right]\Gamma\left[\frac{2b+4}{\alpha(a+b)}\right]}{\alpha(a+b)\Gamma\left[\frac{2}{\alpha}\right]} \tag*{(294)} C j H C ab j , H ( α ) = I 2 a , 2 b ; − 5 ( α /2 ) = ( b a + b ) α 2 ( b a ) − α ( a + b ) 2 a − 4 α ( a + b ) Γ [ α 2 ] Γ [ α ( a + b ) 2 a − 4 ] Γ [ α ( a + b ) 2 b + 4 ] (294)
with C j H = 2 15 ( 1 − q ) 2 ∣ ε ∣ \mathcal{C}^{\mathrm{H}}_j=\frac{2}{15}(1-q)^2|\varepsilon| C j H = 15 2 ( 1 − q ) 2 ∣ ε ∣ , 2 5 q ( 1 − q ) ∣ ε ∣ \frac{2}{5}q(1-q)|\varepsilon| 5 2 q ( 1 − q ) ∣ ε ∣ for vortical and compressional components, respectively. The value of I 2 a , 2 b ; − 5 ( α / 2 ) I_{2a,2b;-5}(\alpha/2) I 2 a , 2 b ; − 5 ( α /2 ) is shown in Figure 31 . In the α → ∞ \alpha\to\infty α → ∞ limit,
lim α → ∞ C a b j , H ( α ) C j H = 2 ( a + b ) ( 2 a − 4 ) ( 2 b + 4 ) . (295) \lim_{\alpha\to\infty}\frac{\mathcal{C}^{j,\mathrm{H}}_{ab}(\alpha)}{\mathcal{C}^{\mathrm{H}}_j}=\frac{2(a+b)}{(2a-4)(2b+4)}. \tag*{(295)} α → ∞ lim C j H C ab j , H ( α ) = ( 2 a − 4 ) ( 2 b + 4 ) 2 ( a + b ) . (295)
For comparison, let us now consider the spectrum of the unprojected stresses for each source, E T φ ( k ) E^{\varphi}_{T}(k) E T φ ( k ) , obtained from the trace of F i j , l m T ( k ) F^{T}_{ij,lm}(k) F ij , l m T ( k ) ,
⟨ T i j ( k ) T i j ∗ ( k ′ ) ⟩ = ( 2 π ) 6 δ 3 ( k − k ′ ) E T ( k ) 4 π k 3 . (296) \left\langle T_{ij}(k)T^{*}_{ij}(k')\right\rangle=(2\pi)^6\delta^3(k-k')\frac{E_T(k)}{4\pi k^3}. \tag*{(296)} ⟨ T ij ( k ) T ij ∗ ( k ′ ) ⟩ = ( 2 π ) 6 δ 3 ( k − k ′ ) 4 π k 3 E T ( k ) . (296)
Introducing Isserlis’ decomposition into Eq. (257) , we find (see [68 ] for the case of helical magnetic fields)
E T φ ( k ) = k 3 4 π [ ∫ d 3 p p 3 p ~ 3 ( F i i φ ( p ) F j j φ ( p ~ ) + F i j φ ( p ) F j i φ ( p ~ ) ) + 1 3 δ 3 ( k ) ( ∫ d 3 p p 3 F i i φ ( p ) ) 2 ] . (297) E^{\varphi}_{T}(k)=\frac{k^3}{4\pi}\left[\int\frac{\mathrm{d}^3p}{p^3\tilde{p}^3}\left(F^{\varphi}_{ii}(p)F^{\varphi}_{jj}(\tilde{p})+F^{\varphi}_{ij}(p)F^{\varphi}_{ji}(\tilde{p})\right)+\frac{1}{3}\delta^3(k)\left(\int\frac{\mathrm{d}^3p}{p^3}F^{\varphi}_{ii}(p)\right)^2\right]. \tag*{(297)} E T φ ( k ) = 4 π k 3 [ ∫ p 3 p ~ 3 d 3 p ( F ii φ ( p ) F j j φ ( p ~ ) + F ij φ ( p ) F j i φ ( p ~ ) ) + 3 1 δ 3 ( k ) ( ∫ p 3 d 3 p F ii φ ( p ) ) 2 ] . (297)
The decomposition of E T φ ( k ) E^{\varphi}_{T}(k) E T φ ( k ) can be split into the same components as E Π φ ( k ) E^{\varphi}_{\Pi}(k) E Π φ ( k ) [see Eq. (261) ],
E T i ( k ) = k 2 ∫ 0 ∞ d p p 2 ∫ ∣ k − p ∣ k + p d p ~ p ~ 2 E j ( p ) E l ( p ~ ) K i ( μ ) + 4 π 3 c i k 3 δ 3 ( k ) ∫ 0 ∞ d p p E j ( p ) ∫ 0 ∞ d p ~ p ~ E l ( p ~ ) , (298) \begin{aligned}
E^i_T(k)={}&k^2\int_0^\infty\frac{\mathrm{d}p}{p^2}\int_{\lvert k-p\rvert}^{k+p}\frac{\mathrm{d}\tilde{p}}{\tilde{p}^2}E_j(p)E_l(\tilde{p})\mathcal{K}_i(\mu)\\
&+\frac{4\pi}{3}c_i k^3\delta^3(k)\int_0^\infty\frac{\mathrm{d}p}{p}E_j(p)\int_0^\infty\frac{\mathrm{d}\tilde{p}}{\tilde{p}}E_l(\tilde{p}), \tag*{(298)}
\end{aligned} E T i ( k ) = k 2 ∫ 0 ∞ p 2 d p ∫ ∣ k − p ∣ k + p p ~ 2 d p ~ E j ( p ) E l ( p ~ ) K i ( μ ) + 3 4 π c i k 3 δ 3 ( k ) ∫ 0 ∞ p d p E j ( p ) ∫ 0 ∞ p ~ d p ~ E l ( p ~ ) , (298)
where j = l = N j=l=\mathrm{N} j = l = N for i = v o r t i=\mathrm{vort} i = vort , j = l = L j=l=\mathrm{L} j = l = L for i = c o m p i=\mathrm{comp} i = comp , j = N j=\mathrm{N} j = N and l = L l=\mathrm{L} l = L for i = m i x i=\mathrm{mix} i = mix , and j = l = H j=l=\mathrm{H} j = l = H for i = h e l i=\mathrm{hel} i = hel . The last term of Eq. (298) is the contribution from the homogeneous (k = 0 k=0 k = 0 ) mode. The coefficients are c i = 1 , 1 , 2 , 0 c_i=1,1,2,0 c i = 1 , 1 , 2 , 0 for i = v o r t , c o m p , m i x , h e l i=\mathrm{vort},\mathrm{comp},\mathrm{mix},\mathrm{hel} i = vort , comp , mix , hel , respectively; the factor 2 in the mixed component accounts for the two orderings of N and L, which are combined in a single integral, and the helical component does not contribute due to antisymmetry. The kernels K i ( μ ) \mathcal{K}_i(\mu) K i ( μ ) are
K v o r t ( μ ) = 1 8 ( 5 + μ 2 ) , (299) \mathcal{K}_{\mathrm{vort}}(\mu)=\frac{1}{8}(5+\mu^2), \tag*{(299)} K vort ( μ ) = 8 1 ( 5 + μ 2 ) , (299)
K c o m p ( μ ) = 1 2 ( 1 + μ 2 ) , (300) \mathcal{K}_{\mathrm{comp}}(\mu)=\frac{1}{2}(1+\mu^2), \tag*{(300)} K comp ( μ ) = 2 1 ( 1 + μ 2 ) , (300)
K m i x ( μ ) = 1 2 ( 3 − μ 2 ) , (301) \mathcal{K}_{\mathrm{mix}}(\mu)=\frac{1}{2}(3-\mu^2), \tag*{(301)} K mix ( μ ) = 2 1 ( 3 − μ 2 ) , (301)
K h e l ( μ ) = 1 4 μ . (302) \mathcal{K}_{\mathrm{hel}}(\mu)=\frac{1}{4}\mu. \tag*{(302)} K hel ( μ ) = 4 1 μ . (302)
where μ = p ^ ⋅ p ~ ^ \mu=\hat{\boldsymbol{p}}\cdot\hat{\tilde{\boldsymbol{p}}} μ = p ^ ⋅ p ~ ^ , explicitly in terms of p p p , p ~ \tilde{p} p ~ , and k k k , is
μ = k 2 − p 2 − p ~ 2 2 p p ~ . (303) \mu=\frac{k^{2}-p^{2}-\tilde{p}^{2}}{2p\tilde{p}} . \tag*{(303)} μ = 2 p p ~ k 2 − p 2 − p ~ 2 . (303)
Again introducing the same normalization as in Eq. (274) , we obtain
E T φ , j ( K ) = φ ~ r m s 4 ρ 2 1 Λ a b 2 K 3 C T φ , j ζ T φ , j ( K ) . (304) E_{T}^{\varphi,j}(K)=\frac{\tilde{\varphi}_{\mathrm{rms}}^{4}}{\rho^{2}}\frac{1}{\Lambda_{ab}^{2}}K^{3}C_{T}^{\varphi,j}\zeta_{T}^{\varphi,j}(K). \tag*{(304)} E T φ , j ( K ) = ρ 2 φ ~ rms 4 Λ ab 2 1 K 3 C T φ , j ζ T φ , j ( K ) . (304)
The geometrical factors are computed taking the k → 0 k\to0 k → 0 limit of the expressions in Eqs. (298) , such that p ~ → p \tilde{p}\to p p ~ → p and μ → − 1 \mu\to-1 μ → − 1 ,
C T v o r t = 3 2 ∫ 0 ∞ ζ N 2 ( P ) P 4 d P , (305) C_{T}^{\mathrm{vort}}=\frac{3}{2}\int_{0}^{\infty}\frac{\zeta_{N}^{2}(P)}{P^{4}}\,\mathrm{d}P, \tag*{(305)} C T vort = 2 3 ∫ 0 ∞ P 4 ζ N 2 ( P ) d P , (305)
C T c o m p = 2 ∫ 0 ∞ ζ L 2 ( P ) P 4 d P , (306) C_{T}^{\mathrm{comp}}=2\int_{0}^{\infty}\frac{\zeta_{L}^{2}(P)}{P^{4}}\,\mathrm{d}P, \tag*{(306)} C T comp = 2 ∫ 0 ∞ P 4 ζ L 2 ( P ) d P , (306)
C T m i x = 2 ∫ 0 ∞ ζ L ( P ) ζ N ( P ) P 4 d P , (307) C_{T}^{\mathrm{mix}}=2\int_{0}^{\infty}\frac{\zeta_{L}(P)\zeta_{N}(P)}{P^{4}}\,\mathrm{d}P, \tag*{(307)} C T mix = 2 ∫ 0 ∞ P 4 ζ L ( P ) ζ N ( P ) d P , (307)
C T h e l = 1 2 ∫ 0 ∞ ζ H 2 ( P ) P 4 d P . (308) C_{T}^{\mathrm{hel}}=\frac{1}{2}\int_{0}^{\infty}\frac{\zeta_{H}^{2}(P)}{P^{4}}\,\mathrm{d}P. \tag*{(308)} C T hel = 2 1 ∫ 0 ∞ P 4 ζ H 2 ( P ) d P . (308)
The average squared value of the stress-energy tensor of a source φ \varphi φ can be directly computed from the two-point correlator function in physical space evaluated at r = 0 r=0 r = 0 ,
a 4 ρ T φ ≡ ⟨ T i j φ ( x ) T i j φ ( x ) ⟩ = ⟨ φ 4 ( x ) ⟩ = ∫ − ∞ ∞ E T φ ( K ) d ln K . (309) a^{4}\rho_{T}^{\varphi}\equiv\langle T_{ij}^{\varphi}(\boldsymbol{x})T_{ij}^{\varphi}(\boldsymbol{x})\rangle=\langle\varphi^{4}(\boldsymbol{x})\rangle
=\int_{-\infty}^{\infty}E_{T}^{\varphi}(K)\,\mathrm{d}\ln K. \tag*{(309)} a 4 ρ T φ ≡ ⟨ T ij φ ( x ) T ij φ ( x )⟩ = ⟨ φ 4 ( x )⟩ = ∫ − ∞ ∞ E T φ ( K ) d ln K . (309)
Decomposing the left-hand side using Isserlis’ theorem, and noting that, due to isotropy, ⟨ φ i φ j ⟩ = 1 3 δ i j φ r m s 2 \langle\varphi_{i}\varphi_{j}\rangle=\frac{1}{3}\delta_{ij}\varphi_{\mathrm{rms}}^{2} ⟨ φ i φ j ⟩ = 3 1 δ ij φ rms 2 , we find the mean squared value of T i j φ T_{ij}^{\varphi} T ij φ ,
ρ ˉ T φ = a − 4 ⟨ T i j φ ( x ) T i j φ ( x ) ⟩ = 5 3 φ ~ r m s 4 . (310) \bar{\rho}_{T}^{\varphi}=a^{-4}\langle T_{ij}^{\varphi}(\boldsymbol{x})T_{ij}^{\varphi}(\boldsymbol{x})\rangle=\frac{5}{3}\tilde{\varphi}_{\mathrm{rms}}^{4}. \tag*{(310)} ρ ˉ T φ = a − 4 ⟨ T ij φ ( x ) T ij φ ( x )⟩ = 3 5 φ ~ rms 4 . (310)
This result is also found integrating E T ( k ) E_{T}(k) E T ( k ) , as the sum over its components in Eq. (298) , over ln k \ln k ln k , as in Eq. (309) , and using the change of variables k d k → p p ~ d μ k\,\mathrm{d}k\to p\tilde{p}\,\mathrm{d}\mu k d k → p p ~ d μ ,
a 4 ρ T φ = ∫ 0 ∞ E T φ ( k ) d ln k = ∫ 0 ∞ d p p ∫ 0 ∞ d p ~ p ~ ∫ − 1 1 d μ ∑ j E j ( p ) E l ( p ~ ) K i ( μ ) + 1 3 ( ∫ 0 ∞ d p p [ E N ( p ) + E L ( p ) ] ) 2 . (311) a^{4}\rho_{T}^{\varphi}=\int_{0}^{\infty}E_{T}^{\varphi}(k)\,\mathrm{d}\ln k=\int_{0}^{\infty}\frac{\mathrm{d}p}{p}\int_{0}^{\infty}\frac{\mathrm{d}\tilde{p}}{\tilde{p}}\int_{-1}^{1}\mathrm{d}\mu\sum_{j}E_{j}(p)E_{l}(\tilde{p})K_{i}(\mu)
+\frac{1}{3}\left(\int_{0}^{\infty}\frac{\mathrm{d}p}{p}\left[E_{N}(p)+E_{L}(p)\right]\right)^{2}. \tag*{(311)} a 4 ρ T φ = ∫ 0 ∞ E T φ ( k ) d ln k = ∫ 0 ∞ p d p ∫ 0 ∞ p ~ d p ~ ∫ − 1 1 d μ j ∑ E j ( p ) E l ( p ~ ) K i ( μ ) + 3 1 ( ∫ 0 ∞ p d p [ E N ( p ) + E L ( p ) ] ) 2 . (311)
where j j j and l l l are the corresponding components of the source spectrum contributing to the stresses component i i i , as indicated below Eq. (298) . Using the kernels in Eqs. (B28), we find that the contribution to ρ T φ \rho_{T}^{\varphi} ρ T φ from the helical component vanishes, and the remaining ones become a separable product of p p p and p ~ \tilde{p} p ~ ,
ρ ˉ T φ = 5 3 a − 4 ( ∫ − ∞ ∞ [ E N ( P ) + E L ( P ) ] d ln P ) 2 = 5 3 φ ~ r m s 4 , (312) \bar{\rho}_{T}^{\varphi}=\frac{5}{3}a^{-4}\left(\int_{-\infty}^{\infty}\left[E_{N}(P)+E_{L}(P)\right]\mathrm{d}\ln P\right)^{2}
=\frac{5}{3}\tilde{\varphi}_{\mathrm{rms}}^{4}, \tag*{(312)} ρ ˉ T φ = 3 5 a − 4 ( ∫ − ∞ ∞ [ E N ( P ) + E L ( P ) ] d ln P ) 2 = 3 5 φ ~ rms 4 , (312)
as found in Eq. (310) .
Dynamical evolution of the Constant-in-Time kernel
In this section, we study the dynamical evolution of the kernel Δ 2 ( τ , k ) \Delta^{2}(\tau,k) Δ 2 ( τ , k ) that characterizes the GW spectrum in the Constant-in-Time (CiT) model described in § 0.5 [see Eq. (132) ].
We consider the two regimes described in § 0.5 A separately. Let us start with times τ ≤ τ f i n \tau\leq\tau_{\mathrm{fin}} τ ≤ τ fin , when the kernel takes the following form
Δ ( τ , k ) = cos ( k τ ) Δ C i ( k τ ) + sin ( k τ ) Δ S i ( k τ ) , (313) \Delta(\tau,k)=\cos(k\tau)\Delta\mathrm{Ci}(k\tau)+\sin(k\tau)\Delta\mathrm{Si}(k\tau), \tag*{(313)} Δ ( τ , k ) = cos ( k τ ) Δ Ci ( k τ ) + sin ( k τ ) Δ Si ( k τ ) , (313)
where Δ C i ( k τ ) = C i ( k τ ) − C i ( k τ ∗ ) \Delta\mathrm{Ci}(k\tau)=\mathrm{Ci}(k\tau)-\mathrm{Ci}(k\tau_{*}) Δ Ci ( k τ ) = Ci ( k τ ) − Ci ( k τ ∗ ) and Δ S i ( k τ ) = S i ( k τ ) − S i ( k τ ∗ ) \Delta\mathrm{Si}(k\tau)=\mathrm{Si}(k\tau)-\mathrm{Si}(k\tau_{*}) Δ Si ( k τ ) = Si ( k τ ) − Si ( k τ ∗ ) [see Eqs. (107)].
Let us consider the asymptotic limits k τ ≪ 1 k\tau\ll1 k τ ≪ 1 and k τ ≫ 1 k\tau\gg1 k τ ≫ 1 during the sourcing times τ ≤ τ f i n \tau\leq\tau_{\mathrm{fin}} τ ≤ τ fin . In the case when k τ = k τ ∗ + k δ τ ≪ 1 k\tau=k\tau_{*}+k\delta\tau\ll1 k τ = k τ ∗ + k δ τ ≪ 1 , which implies k τ ∗ ≪ 1 k\tau_{*}\ll1 k τ ∗ ≪ 1 and k δ τ ≪ 1 k\delta\tau\ll1 k δ τ ≪ 1 , the cosine and sine integral functions become
lim k τ ≪ 1 C i ( k τ ) = ln ( k τ ) + γ , (314) \lim_{k\tau\ll1}\mathrm{Ci}(k\tau)=\ln(k\tau)+\gamma, \tag*{(314)} k τ ≪ 1 lim Ci ( k τ ) = ln ( k τ ) + γ , (314)
lim k τ ≪ 1 S i ( k τ ) = k τ , (315) \lim_{k\tau\ll1}\mathrm{Si}(k\tau)=k\tau, \tag*{(315)} k τ ≪ 1 lim Si ( k τ ) = k τ , (315)
where γ \gamma γ is the Euler-Mascheroni constant. Hence, the kernel becomes
lim k τ ≪ 1 Δ 2 ( τ , k ) = ln 2 ( 1 + δ τ τ ∗ ) . (316) \lim_{k\tau\ll1}\Delta^{2}(\tau,k)=\ln^{2}\left(1+\frac{\delta\tau}{\tau_{*}}\right). \tag*{(316)} k τ ≪ 1 lim Δ 2 ( τ , k ) = ln 2 ( 1 + τ ∗ δ τ ) . (316)
Note that this result is valid only for a radiation-dominated Universe [1]. This limit of the kernel represents the regime i) listed in § 0.5 A, when the GW modes are growing in time (see right panel of Figure 9 ). For short durations, δ τ / τ ∗ ≪ 1 \delta\tau/\tau_{*}\ll1 δ τ / τ ∗ ≪ 1 , we recover the quadratic growth valid when the expansion of the Universe is neglected,
lim k τ ≪ 1 Δ f l a t 2 ( τ , k ) = ( δ τ τ ∗ ) 2 . (317) \lim_{k\tau\ll1}\Delta_{\mathrm{flat}}^{2}(\tau,k)=\left(\frac{\delta\tau}{\tau_{*}}\right)^{2}. \tag*{(317)} k τ ≪ 1 lim Δ flat 2 ( τ , k ) = ( τ ∗ δ τ ) 2 . (317)
In the opposite limit, k τ ≫ 1 k\tau\gg1 k τ ≫ 1 ,
lim k τ ≫ 1 Δ Ci ( k τ ) = sin k τ k τ − Ci ( k τ ∗ ) , (318) \lim_{k\tau\gg1}\Delta\operatorname{Ci}(k\tau)=\frac{\sin k\tau}{k\tau}-\operatorname{Ci}(k\tau_*), \tag*{(318)} k τ ≫ 1 lim Δ Ci ( k τ ) = k τ sin k τ − Ci ( k τ ∗ ) , (318)
lim k τ ≫ 1 Δ Si ( k τ ) = π 2 − cos k τ k τ − Si ( k τ ∗ ) , (319) \lim_{k\tau\gg1}\Delta\operatorname{Si}(k\tau)=\frac{\pi}{2}-\frac{\cos k\tau}{k\tau}-\operatorname{Si}(k\tau_*), \tag*{(319)} k τ ≫ 1 lim Δ Si ( k τ ) = 2 π − k τ cos k τ − Si ( k τ ∗ ) , (319)
where we have used Si ( x ) ≃ π 2 − cos x x \operatorname{Si}(x)\simeq\frac{\pi}{2}-\frac{\cos x}{x} Si ( x ) ≃ 2 π − x c o s x and Ci ( x ) ≃ sin x x \operatorname{Ci}(x)\simeq\frac{\sin x}{x} Ci ( x ) ≃ x s i n x , valid for large x x x . We can now further consider two limits, corresponding to k τ ∗ ≫ 1 k\tau_* \gg1 k τ ∗ ≫ 1 and k τ ∗ ≪ 1 k\tau_* \ll1 k τ ∗ ≪ 1 , which describe modes inside and outside the Hubble scale, respectively. When k τ ∗ ≫ 1 k\tau_* \gg1 k τ ∗ ≫ 1 , we find
lim τ , τ ∗ ≫ 1 / k Δ Ci ( k τ ) = sin k τ k τ − sin k τ ∗ k τ ∗ , (320) \lim_{\tau,\tau_*\gg1/k}\Delta\operatorname{Ci}(k\tau)=\frac{\sin k\tau}{k\tau}-\frac{\sin k\tau_*}{k\tau_*}, \tag*{(320)} τ , τ ∗ ≫ 1/ k lim Δ Ci ( k τ ) = k τ sin k τ − k τ ∗ sin k τ ∗ , (320)
lim τ , τ ∗ ≫ 1 / k Δ Si ( k τ ) = − cos k τ k τ + cos k τ ∗ k τ ∗ . (321) \lim_{\tau,\tau_*\gg1/k}\Delta\operatorname{Si}(k\tau)=-\frac{\cos k\tau}{k\tau}+\frac{\cos k\tau_*}{k\tau_*}. \tag*{(321)} τ , τ ∗ ≫ 1/ k lim Δ Si ( k τ ) = − k τ cos k τ + k τ ∗ cos k τ ∗ . (321)
Hence, the kernel in this limit becomes,
lim τ , τ ∗ ≫ 1 / k Δ 2 ( τ , k ) = sin 2 ( k δ τ ) ( k τ ∗ ) 2 , (322) \lim_{\tau,\tau_*\gg1/k}\Delta^2(\tau,k)=\frac{\sin^2(k\delta\tau)}{(k\tau_*)^2}, \tag*{(322)} τ , τ ∗ ≫ 1/ k lim Δ 2 ( τ , k ) = ( k τ ∗ ) 2 sin 2 ( k δ τ ) , (322)
which is an oscillatory function with an envelope amplitude 1 / k 2 1/k^2 1/ k 2 , describing the regime i i ) ii) ii ) of § 0.5 A. Since it describes sub-horizon modes k τ ∗ ≫ 1 k\tau_* \gg1 k τ ∗ ≫ 1 , this is the limit of the kernel that applies in flat space-time [see left panel of Figure 9 and Eq. (110a)]. However, when we consider super-horizon modes at the time of GW production, k τ ∗ ≪ 1 k\tau_* \ll1 k τ ∗ ≪ 1 , using Eq. (C2), we recover the k τ ≫ 1 k\tau\gg1 k τ ≫ 1 limit found in ref. [1],
lim k δ τ ≫ 1 Δ 2 ( τ , k ) = sin 2 ( k δ τ ) ln 2 [ 1 + ( k τ ∗ ) − 1 ] . (323) \lim_{k\delta\tau\gg1}\Delta^2(\tau,k)=\sin^2(k\delta\tau)\ln^2\left[1+(k\tau_*)^{-1}\right]. \tag*{(323)} k δ τ ≫ 1 lim Δ 2 ( τ , k ) = sin 2 ( k δ τ ) ln 2 [ 1 + ( k τ ∗ ) − 1 ] . (323)
We note that this expression also applies in the k τ ∗ ≫ 1 k\tau_* \gg1 k τ ∗ ≫ 1 limit since ln [ 1 + ( k τ ∗ ) − 1 ] \ln[1+(k\tau_*)^{-1}] ln [ 1 + ( k τ ∗ ) − 1 ] reduces to 1 / ( k τ ∗ ) 1/(k\tau_*) 1/ ( k τ ∗ ) in Eq. (322) . Actually, we find numerically that this is a good approximation at all k τ ∗ k\tau_* k τ ∗ . Let us now consider the case when τ ≥ τ f i n \tau\geq\tau_{\rm fin} τ ≥ τ fin . In this case, the kernel is
Δ ( τ , k ) = cos ( k τ ) Δ Ci ( k τ f i n ) + sin ( k τ ) Δ Si ( k τ f i n ) , (324) \Delta(\tau,k)=\cos(k\tau)\Delta\operatorname{Ci}(k\tau_{\rm fin})+\sin(k\tau)\Delta\operatorname{Si}(k\tau_{\rm fin}), \tag*{(324)} Δ ( τ , k ) = cos ( k τ ) Δ Ci ( k τ fin ) + sin ( k τ ) Δ Si ( k τ fin ) , (324)
and we see that a new timescale, τ f i n \tau_{\rm fin} τ fin , appears. Hence, let us again consider the limit k τ ∗ ≤ k τ f i n ≤ k τ ≪ 1 k\tau_*\leq k\tau_{\rm fin}\leq k\tau\ll1 k τ ∗ ≤ k τ fin ≤ k τ ≪ 1 . In this limit, the kernel becomes
lim k τ ≪ 1 Δ 2 ( τ , k ) = ln 2 ( 1 + δ τ f i n / τ ∗ ) , (325) \lim_{k\tau\ll1}\Delta^2(\tau,k)=\ln^2(1+\delta\tau_{\rm fin}/\tau_*), \tag*{(325)} k τ ≪ 1 lim Δ 2 ( τ , k ) = ln 2 ( 1 + δ τ fin / τ ∗ ) , (325)
where δ τ f i n = τ f i n − τ ∗ \delta\tau_{\rm fin}=\tau_{\rm fin}-\tau_* δ τ fin = τ fin − τ ∗ .
Integration of the Green's function in the coherent-decay case
In the limit k τ f i n ≥ k τ ∗ ≫ 1 k\tau_{\rm fin}\geq k\tau_*\gg1 k τ fin ≥ k τ ∗ ≫ 1 , the function becomes
lim k τ ∗ ≫ 1 Δ 2 ( τ , k ) = 1 ( k τ ∗ ) 2 [ cos k τ ( τ ∗ τ f i n sin k τ f i n − sin k τ ∗ ) − sin k τ ( τ ∗ τ f i n cos k τ f i n − cos k τ ∗ ) ] 2 . (326) \lim_{k\tau_*\gg1}\Delta^2(\tau,k)=\frac{1}{(k\tau_*)^2}\left[\cos k\tau\left(\frac{\tau_*}{\tau_{\rm fin}}\sin k\tau_{\rm fin}-\sin k\tau_*\right)-\sin k\tau\left(\frac{\tau_*}{\tau_{\rm fin}}\cos k\tau_{\rm fin}-\cos k\tau_*\right)\right]^2. \tag*{(326)} k τ ∗ ≫ 1 lim Δ 2 ( τ , k ) = ( k τ ∗ ) 2 1 [ cos k τ ( τ fin τ ∗ sin k τ fin − sin k τ ∗ ) − sin k τ ( τ fin τ ∗ cos k τ fin − cos k τ ∗ ) ] 2 . (326)
This shows that the k − 2 k^{-2} k − 2 branch is multiplied by the oscillatory term in brackets. The amplitude of these oscillations depends on the specific values of τ \tau τ , τ ∗ \tau_* τ ∗ , and τ f i n \tau_{\rm fin} τ fin . However, in general, the amplitude is bounded by the following function
lim k τ ∗ ≫ 1 Δ 2 ( τ , k ) ≤ F 2 ( τ f i n / τ ∗ ) ( k τ ∗ ) 2 , (327) \lim_{k\tau_*\gg1}\Delta^2(\tau,k)\leq\frac{\mathcal{F}^2(\tau_{\rm fin}/\tau_*)}{(k\tau_*)^2}, \tag*{(327)} k τ ∗ ≫ 1 lim Δ 2 ( τ , k ) ≤ ( k τ ∗ ) 2 F 2 ( τ fin / τ ∗ ) , (327)
where we have defined
F ( x ) = 1 + x − 1 ≥ 1. (328) \mathcal{F}(x)=1+x^{-1}\geq1. \tag*{(328)} F ( x ) = 1 + x − 1 ≥ 1. (328)
This function is always larger than one and it shows that for τ > τ f i n \tau>\tau_{\rm fin} τ > τ fin , the function Δ 2 ( τ , k ) \Delta^2(\tau,k) Δ 2 ( τ , k ) presents oscillations with an amplitude larger than that at earlier times, when τ ≤ τ f i n \tau\leq\tau_{\rm fin} τ ≤ τ fin , in the limit k τ ≥ k τ ∗ ≫ 1 k\tau\geq k\tau_*\gg1 k τ ≥ k τ ∗ ≫ 1 . As argued in § 0.5 A, this is a consequence of the discontinuity in the source at τ f i n \tau_{\rm fin} τ fin and it is not expected to be found if the source decays smoothly in time (see refs. [50 ] [1] and Sec. VII). As τ f i n \tau_{\rm fin} τ fin increases, this effect is less significant and in the limit τ f i n → ∞ \tau_{\rm fin}\to\infty τ fin → ∞ , we recover the case τ ≤ τ f i n \tau\leq\tau_{\rm fin} τ ≤ τ fin [see Eq. (322) ].
We present in this section some calculations that are used when we compute the power spectrum of the GW background produced by velocity or magnetic fields, under the assumption of coherent unequal time correlation of the stresses (see Sec. VII). We consider here that we can omit the factor ζ Π T T ( τ , k ) \zeta_{\Pi}^{\rm TT}(\tau,k) ζ Π TT ( τ , k ) appearing in Eq. (193) , which means that we restrict to cases with moderate s ≲ 1 s\lesssim1 s ≲ 1 (see App. E, in particular Eq. (372) and Figure 32 ). Under this assumption, and with the additional hypothesis that δ τ d c = τ ∗ \delta\tau_{\rm dc}=\tau_* δ τ dc = τ ∗ , the integral in Eq. (193) can be written as
Δ p s ( τ , k ; τ ∗ , τ f i n , δ τ d c = τ ∗ ) ≃ τ ∗ − m ∫ τ ∗ τ ˉ τ 1 m − 1 cos k ( τ − τ 1 ) d τ 1 ≡ Δ m ( τ , k ; τ ∗ , τ f i n ) , (329) \begin{aligned}
\Delta_{\rm ps}(\tau,k;\tau_*,\tau_{\rm fin},\delta\tau_{\rm dc}=\tau_*)\simeq\tau_*^{-m}\int_{\tau_*}^{\bar{\tau}}\tau_1^{m-1}\cos k(\tau-\tau_1)\,\mathrm{d}\tau_1 \\
&\equiv\Delta_m(\tau,k;\tau_*,\tau_{\rm fin}), \tag*{(329)}
\end{aligned} Δ ps ( τ , k ; τ ∗ , τ fin , δ τ dc = τ ∗ ) ≃ τ ∗ − m ∫ τ ∗ τ ˉ τ 1 m − 1 cos k ( τ − τ 1 ) d τ 1 ≡ Δ m ( τ , k ; τ ∗ , τ fin ) , (329)
where τ ˉ = min ( τ , τ f i n ) \bar{\tau}=\min(\tau,\tau_{\mathrm{fin}}) τ ˉ = min ( τ , τ fin ) , we use the subindex m = 3 2 s − p m=\frac{3}{2}s-p m = 2 3 s − p instead of the subindices p s ps p s [see discussion below Eq. (190) ], and we have defined the function
Δ m ( τ , k ; τ ∗ , τ f i n ) ≡ τ ∗ − m ∫ τ ∗ τ ˉ τ 1 m − 1 cos k ( τ − τ 1 ) d τ 1 . (330) \Delta_m(\tau,k;\tau_*,\tau_{\mathrm{fin}})\equiv\tau_*^{-m}\int_{\tau_*}^{\bar{\tau}}\tau_1^{m-1}\cos k(\tau-\tau_1)\,\mathrm{d}\tau_1. \tag*{(330)} Δ m ( τ , k ; τ ∗ , τ fin ) ≡ τ ∗ − m ∫ τ ∗ τ ˉ τ 1 m − 1 cos k ( τ − τ 1 ) d τ 1 . (330)
Taking into account that cos x = 1 2 ( e i x + e − i x ) \cos x=\frac{1}{2}\left(e^{ix}+e^{-ix}\right) cos x = 2 1 ( e i x + e − i x ) , Δ m \Delta_m Δ m can be expressed in the following form,
1 2 τ ∗ − m ∫ τ ∗ τ ˉ τ 1 m − 1 [ e i k ( τ − τ 1 ) + e − i k ( τ − τ 1 ) ] d τ 1 = 1 2 τ ∗ − m e i k τ ∫ τ ∗ τ ˉ τ 1 m − 1 e − i k τ 1 d τ 1 + 1 2 τ ∗ − m e − i k τ ∫ τ ∗ τ ˉ τ 1 m − 1 e i k τ 1 d τ 1 . (331) \begin{aligned}
\frac{1}{2}\tau_*^{-m}\int_{\tau_*}^{\bar{\tau}}\tau_1^{m-1}\left[e^{ik(\tau-\tau_1)}+e^{-ik(\tau-\tau_1)}\right]\mathrm{d}\tau_1 \\
&=\frac{1}{2}\tau_*^{-m}e^{ik\tau}\int_{\tau_*}^{\bar{\tau}}\tau_1^{m-1}e^{-ik\tau_1}\mathrm{d}\tau_1
+\frac{1}{2}\tau_*^{-m}e^{-ik\tau}\int_{\tau_*}^{\bar{\tau}}\tau_1^{m-1}e^{ik\tau_1}\mathrm{d}\tau_1. \tag*{(331)}
\end{aligned} 2 1 τ ∗ − m ∫ τ ∗ τ ˉ τ 1 m − 1 [ e ik ( τ − τ 1 ) + e − ik ( τ − τ 1 ) ] d τ 1 = 2 1 τ ∗ − m e ik τ ∫ τ ∗ τ ˉ τ 1 m − 1 e − ik τ 1 d τ 1 + 2 1 τ ∗ − m e − ik τ ∫ τ ∗ τ ˉ τ 1 m − 1 e ik τ 1 d τ 1 . (331)
To study these integrals, we use the upper incomplete gamma function Γ ( m , x ) \Gamma(m,x) Γ ( m , x ) ,
Γ ( m , x ) = ∫ x ∞ t m − 1 e − t d t . (332) \Gamma(m,x)=\int_x^\infty t^{m-1}e^{-t}\,\mathrm{d}t. \tag*{(332)} Γ ( m , x ) = ∫ x ∞ t m − 1 e − t d t . (332)
Using x = ± i k z x=\pm ikz x = ± ik z and the change of variable t = ± i k p t=\pm ikp t = ± ik p , we find
( i k ) − m Γ ( m , i k z ) = ∫ z − i ∞ p m − 1 e − i k p d p . (333) (ik)^{-m}\Gamma(m,ikz)=\int_z^{-i\infty}p^{m-1}e^{-ikp}\,\mathrm{d}p. \tag*{(333)} ( ik ) − m Γ ( m , ik z ) = ∫ z − i ∞ p m − 1 e − ik p d p . (333)
( − i k ) − m Γ ( m , − i k z ) = ∫ z + i ∞ p m − 1 e i k p d p . (334) (-ik)^{-m}\Gamma(m,-ikz)=\int_z^{+i\infty}p^{m-1}e^{ikp}\,\mathrm{d}p. \tag*{(334)} ( − ik ) − m Γ ( m , − ik z ) = ∫ z + i ∞ p m − 1 e ik p d p . (334)
By substituting these integrals in Eq. (331) , we find an expression for Δ m \Delta_m Δ m in terms of the incomplete Gamma function
Δ m ( τ , k ; τ ∗ , τ f i n ) = − 1 2 τ ∗ − m ( i k ) − m { e i k τ [ Γ ( m , i k τ ˉ ) − Γ ( m , i k τ ∗ ) ] + e − i k τ e i π m [ Γ ( m , − i k τ ˉ ) − Γ ( m , − i k τ ∗ ) ] } . (335) \Delta_m(\tau,k;\tau_*,\tau_{\mathrm{fin}})=-\frac{1}{2}\tau_*^{-m}(ik)^{-m}\left\{e^{ik\tau}\left[\Gamma(m,ik\bar{\tau})-\Gamma(m,ik\tau_*)\right]+e^{-ik\tau}e^{i\pi m}\left[\Gamma(m,-ik\bar{\tau})-\Gamma(m,-ik\tau_*)\right]\right\}. \tag*{(335)} Δ m ( τ , k ; τ ∗ , τ fin ) = − 2 1 τ ∗ − m ( ik ) − m { e ik τ [ Γ ( m , ik τ ˉ ) − Γ ( m , ik τ ∗ ) ] + e − ik τ e iπ m [ Γ ( m , − ik τ ˉ ) − Γ ( m , − ik τ ∗ ) ] } . (335)
where ( ± i k ) − m = k − m e ∓ i π m / 2 (\pm ik)^{-m}=k^{-m}e^{\mp i\pi m/2} ( ± ik ) − m = k − m e ∓ iπ m /2 .
Let us now show that for m = 0 m=0 m = 0 we recover Eq. (132) . In first place, we use the general expression Eq. (335) to get
Δ 0 ( τ , k ; τ ∗ , τ f i n ) = − 1 2 cos k τ [ Γ ( 0 , i k t ) + Γ ( 0 , − i k t ) ] τ ∗ τ ˉ − 1 2 i sin k τ [ Γ ( 0 , i k t ) − Γ ( 0 , − i k t ) ] τ ∗ τ ˉ . (336) \Delta_0(\tau,k;\tau_*,\tau_{\mathrm{fin}})=-\frac{1}{2}\cos k\tau\left[\Gamma(0,ikt)+\Gamma(0,-ikt)\right]_{\tau_*}^{\bar{\tau}}-\frac{1}{2}i\sin k\tau\left[\Gamma(0,ikt)-\Gamma(0,-ikt)\right]_{\tau_*}^{\bar{\tau}}. \tag*{(336)} Δ 0 ( τ , k ; τ ∗ , τ fin ) = − 2 1 cos k τ [ Γ ( 0 , ik t ) + Γ ( 0 , − ik t ) ] τ ∗ τ ˉ − 2 1 i sin k τ [ Γ ( 0 , ik t ) − Γ ( 0 , − ik t ) ] τ ∗ τ ˉ . (336)
where the notation f ( t ) ∣ τ ∗ τ ˉ = f ( τ ˉ ) − f ( τ ∗ ) f(t)|_{\tau_*}^{\bar{\tau}}=f(\bar{\tau})-f(\tau_*) f ( t ) ∣ τ ∗ τ ˉ = f ( τ ˉ ) − f ( τ ∗ ) is used for compactness. Using Eqs. (333) and (D6), we find
Γ ( 0 , ± i k τ ) = ∫ τ ∓ i ∞ e ∓ i k p p d p = ∫ τ ∓ i ∞ cos k p p d p ∓ i ∫ τ ∓ i ∞ sin k p p d p . (337) \begin{aligned}
\Gamma(0,\pm ik\tau)=\int_\tau^{\mp i\infty}\frac{e^{\mp ikp}}{p}\,\mathrm{d}p \\
&=\int_\tau^{\mp i\infty}\frac{\cos kp}{p}\,\mathrm{d}p\mp i\int_\tau^{\mp i\infty}\frac{\sin kp}{p}\,\mathrm{d}p. \tag*{(337)}
\end{aligned} Γ ( 0 , ± ik τ ) = ∫ τ ∓ i ∞ p e ∓ ik p d p = ∫ τ ∓ i ∞ p cos k p d p ∓ i ∫ τ ∓ i ∞ p sin k p d p . (337)
Hence, it follows that
[ Γ ( 0 , i k t ) + Γ ( 0 , − i k t ) ] τ ∗ τ ˉ = − 2 [ Ci ( k τ ˉ ) − Ci ( k τ ∗ ) ] . (338) \left[\Gamma(0,ikt)+\Gamma(0,-ikt)\right]_{\tau_*}^{\bar{\tau}}=-2\left[\operatorname{Ci}(k\bar{\tau})-\operatorname{Ci}(k\tau_*)\right]. \tag*{(338)} [ Γ ( 0 , ik t ) + Γ ( 0 , − ik t ) ] τ ∗ τ ˉ = − 2 [ Ci ( k τ ˉ ) − Ci ( k τ ∗ ) ] . (338)
[ Γ ( 0 , i k t ) − Γ ( 0 , − i k t ) ] τ ∗ τ ˉ = 2 i [ Si ( k τ ˉ ) − Si ( k τ ∗ ) ] . (339) \left[\Gamma(0,ikt)-\Gamma(0,-ikt)\right]_{\tau_*}^{\bar{\tau}}=2i\left[\operatorname{Si}(k\bar{\tau})-\operatorname{Si}(k\tau_*)\right]. \tag*{(339)} [ Γ ( 0 , ik t ) − Γ ( 0 , − ik t ) ] τ ∗ τ ˉ = 2 i [ Si ( k τ ˉ ) − Si ( k τ ∗ ) ] . (339)
Finally, substituting Eqs. (338) and (D11) into Eq. (336) , we find Eq. (132) , q.e.d.
We can now consider a few special cases of m = 3 2 s − p m=\frac{3}{2}s-p m = 2 3 s − p . In particular, if we consider negative integers m = − n m=-n m = − n , the incomplete Γ \Gamma Γ function at ± i k τ \pm ik\tau ± ik τ satisfies the following recurrence relation
Γ ( − n , i k τ ) = 1 n ! [ ( − 1 ) n Γ ( 0 , i k τ ) + e − i k τ ∑ q = 0 n − 1 ( − 1 ) q A n − q ( τ ) ] . (340) \Gamma(-n,ik\tau)=\frac{1}{n!}\left[(-1)^n\Gamma(0,ik\tau)+e^{-ik\tau}\sum_{q=0}^{n-1}(-1)^q A_{n-q}(\tau)\right]. \tag*{(340)} Γ ( − n , ik τ ) = n ! 1 [ ( − 1 ) n Γ ( 0 , ik τ ) + e − ik τ q = 0 ∑ n − 1 ( − 1 ) q A n − q ( τ ) ] . (340)
Γ ( − n , − i k τ ) = 1 n ! [ ( − 1 ) n Γ ( 0 , − i k τ ) + ( − 1 ) n e i k τ ∑ q = 0 n − 1 A n − q ( τ ) ] . (341) \Gamma(-n,-ik\tau)=\frac{1}{n!}\left[(-1)^n\Gamma(0,-ik\tau)+(-1)^ne^{ik\tau}\sum_{q=0}^{n-1}A_{n-q}(\tau)\right]. \tag*{(341)} Γ ( − n , − ik τ ) = n ! 1 [ ( − 1 ) n Γ ( 0 , − ik τ ) + ( − 1 ) n e ik τ q = 0 ∑ n − 1 A n − q ( τ ) ] . (341)
with
A a = ( a − 1 ) ! ( i k τ ) a . (342) A_a=\frac{(a-1)!}{(ik\tau)^a}. \tag*{(342)} A a = ( ik τ ) a ( a − 1 )! . (342)
In this case, we can substitute the values of the incomplete Γ \Gamma Γ into Eq. (335) ,
Δ − n ( τ , k ; τ ∗ , τ f i n ) = − 1 2 τ ∗ n i n k n n ! [ Δ − n , A + Δ − n , B ] . (343) \Delta_{-n}(\tau,k;\tau_*,\tau_{\mathrm{fin}})=-\frac{1}{2}\tau_*^n\frac{i^n k^n}{n!}\left[\Delta_{-n,A}+\Delta_{-n,B}\right]. \tag*{(343)} Δ − n ( τ , k ; τ ∗ , τ fin ) = − 2 1 τ ∗ n n ! i n k n [ Δ − n , A + Δ − n , B ] . (343)
where, for simplicity, we have split the integral in the terms proportional to Γ ( 0 , ± i k τ ) \Gamma(0,\pm ik\tau) Γ ( 0 , ± ik τ ) (A), and the rest (B), which can be computed separately. Hence,
Δ − n , A ( τ , k ; τ ∗ , τ f i n ) = ( − 1 ) n e i k τ [ Γ ( 0 , i k t ) ] τ ∗ τ ˉ + e − i k τ [ Γ ( 0 , − i k t ) ] τ ∗ τ ˉ = cos k τ [ ( − 1 ) n Γ ( 0 , i k t ) + Γ ( 0 , − i k t ) ] τ ∗ τ ˉ + i sin k τ [ ( − 1 ) n Γ ( 0 , i k t ) − Γ ( 0 , − i k t ) ] τ ∗ τ ˉ . (344) \begin{aligned}
\Delta_{-n,A}(\tau,k;\tau_\ast,\tau_{\rm fin})={}&(-1)^n e^{ik\tau}\left[\Gamma(0,ikt)\right]_{\tau_\ast}^{\bar{\tau}}+e^{-ik\tau}\left[\Gamma(0,-ikt)\right]_{\tau_\ast}^{\bar{\tau}}\\
={}&\cos k\tau\left[(-1)^n\Gamma(0,ikt)+\Gamma(0,-ikt)\right]_{\tau_\ast}^{\bar{\tau}}\\
&+i\sin k\tau\left[(-1)^n\Gamma(0,ikt)-\Gamma(0,-ikt)\right]_{\tau_\ast}^{\bar{\tau}}. \tag*{(344)}
\end{aligned} Δ − n , A ( τ , k ; τ ∗ , τ fin ) = = ( − 1 ) n e ik τ [ Γ ( 0 , ik t ) ] τ ∗ τ ˉ + e − ik τ [ Γ ( 0 , − ik t ) ] τ ∗ τ ˉ cos k τ [ ( − 1 ) n Γ ( 0 , ik t ) + Γ ( 0 , − ik t ) ] τ ∗ τ ˉ + i sin k τ [ ( − 1 ) n Γ ( 0 , ik t ) − Γ ( 0 , − ik t ) ] τ ∗ τ ˉ . (344)
Note that Δ − n , A \Delta_{-n,A} Δ − n , A will be a function of the sine and cosine integrals [see Eqs. (338) and (339) ], depending on the specific values of n n n . For different values of n n n , we have
Δ − 1 , A = − 2 i cos k τ [ Si ( k τ ˉ ) − Si ( k τ ∗ ) ] + 2 i sin k τ [ Ci ( k τ ˉ ) − Ci ( k τ ∗ ) ] , (345) \Delta_{-1,A}=-2i\cos k\tau\left[\operatorname{Si}(k\bar{\tau})-\operatorname{Si}(k\tau_\ast)\right]+2i\sin k\tau\left[\operatorname{Ci}(k\bar{\tau})-\operatorname{Ci}(k\tau_\ast)\right], \tag*{(345)} Δ − 1 , A = − 2 i cos k τ [ Si ( k τ ˉ ) − Si ( k τ ∗ ) ] + 2 i sin k τ [ Ci ( k τ ˉ ) − Ci ( k τ ∗ ) ] , (345)
Δ − 2 , A = − 2 cos k τ [ Ci ( k τ ˉ ) − Ci ( k τ ∗ ) ] − 2 sin k τ [ Si ( k τ ˉ ) − Si ( k τ ∗ ) ] . (346) \Delta_{-2,A}=-2\cos k\tau\left[\operatorname{Ci}(k\bar{\tau})-\operatorname{Ci}(k\tau_\ast)\right]-2\sin k\tau\left[\operatorname{Si}(k\bar{\tau})-\operatorname{Si}(k\tau_\ast)\right]. \tag*{(346)} Δ − 2 , A = − 2 cos k τ [ Ci ( k τ ˉ ) − Ci ( k τ ∗ ) ] − 2 sin k τ [ Si ( k τ ˉ ) − Si ( k τ ∗ ) ] . (346)
For other values of n n n , we note that Δ − n , A \Delta_{-n,A} Δ − n , A will be equal to Δ − 1 , A \Delta_{-1,A} Δ − 1 , A if n n n is odd and equal to Δ − 2 , A \Delta_{-2,A} Δ − 2 , A if n n n is even. The other term that appears in Eq. (343) is
Δ − n , B ( τ , k ; τ ∗ , τ f i n ) = [ e i k ( t − τ ) ∑ q = 0 n − 1 A n − q ( t ) ] τ ∗ τ ˉ + [ e − i k ( t − τ ) ∑ q = 0 n − 1 ( − 1 ) q A n − q ( t ) ] τ ∗ τ ˉ = [ 2 cos k ( t − τ ) ∑ q = 0 ⌊ 1 2 ( n − 1 ) ⌋ A n − 2 q ( t ) ] τ ∗ τ ˉ + [ 2 i sin k ( t − τ ) ∑ q = 0 ⌊ 1 2 ( n − 2 ) ⌋ A n − 2 q − 1 ( t ) ] τ ∗ τ ˉ . (347) \begin{aligned}
\Delta_{-n,B}(\tau,k;\tau_\ast,\tau_{\rm fin})={}&\left[e^{ik(t-\tau)}\sum_{q=0}^{n-1}A_{n-q}(t)\right]_{\tau_\ast}^{\bar{\tau}}+\left[e^{-ik(t-\tau)}\sum_{q=0}^{n-1}(-1)^qA_{n-q}(t)\right]_{\tau_\ast}^{\bar{\tau}}\\
={}&\left[2\cos k(t-\tau)\sum_{q=0}^{\left\lfloor\frac{1}{2}(n-1)\right\rfloor}A_{n-2q}(t)\right]_{\tau_\ast}^{\bar{\tau}}\\
&+\left[2i\sin k(t-\tau)\sum_{q=0}^{\left\lfloor\frac{1}{2}(n-2)\right\rfloor}A_{n-2q-1}(t)\right]_{\tau_\ast}^{\bar{\tau}}. \tag*{(347)}
\end{aligned} Δ − n , B ( τ , k ; τ ∗ , τ fin ) = = [ e ik ( t − τ ) q = 0 ∑ n − 1 A n − q ( t ) ] τ ∗ τ ˉ + [ e − ik ( t − τ ) q = 0 ∑ n − 1 ( − 1 ) q A n − q ( t ) ] τ ∗ τ ˉ 2 cos k ( t − τ ) q = 0 ∑ ⌊ 2 1 ( n − 1 ) ⌋ A n − 2 q ( t ) τ ∗ τ ˉ + 2 i sin k ( t − τ ) q = 0 ∑ ⌊ 2 1 ( n − 2 ) ⌋ A n − 2 q − 1 ( t ) τ ∗ τ ˉ . (347)
where ⌊ x ⌋ \lfloor x\rfloor ⌊ x ⌋ is the floor function. Hence, we will have terms proportional to inverse power laws of k τ k\tau k τ up to order − n -n − n . We can now compute the terms Δ − n , B \Delta_{-n,B} Δ − n , B for different values of n n n ,
Δ − 1 , B = − 2 i k [ cos k ( τ − τ ˉ ) τ ˉ − cos k ( τ − τ ∗ ) τ ∗ ] , (348) \Delta_{-1,B}=-\frac{2i}{k}\left[\frac{\cos k(\tau-\bar{\tau})}{\bar{\tau}}-\frac{\cos k(\tau-\tau_\ast)}{\tau_\ast}\right], \tag*{(348)} Δ − 1 , B = − k 2 i [ τ ˉ cos k ( τ − τ ˉ ) − τ ∗ cos k ( τ − τ ∗ ) ] , (348)
Δ − 2 , B = 2 k 2 [ cos k ( τ − τ ∗ ) τ ∗ 2 − cos k ( τ − τ ˉ ) τ ˉ 2 + k τ ∗ sin k ( τ − τ ∗ ) − k τ ˉ sin k ( τ − τ ˉ ) ] . (349) \Delta_{-2,B}=\frac{2}{k^2}\left[\frac{\cos k(\tau-\tau_\ast)}{\tau_\ast^2}-\frac{\cos k(\tau-\bar{\tau})}{\bar{\tau}^2}+\frac{k}{\tau_\ast}\sin k(\tau-\tau_\ast)-\frac{k}{\bar{\tau}}\sin k(\tau-\bar{\tau})\right]. \tag*{(349)} Δ − 2 , B = k 2 2 [ τ ∗ 2 cos k ( τ − τ ∗ ) − τ ˉ 2 cos k ( τ − τ ˉ ) + τ ∗ k sin k ( τ − τ ∗ ) − τ ˉ k sin k ( τ − τ ˉ ) ] . (349)
Finally, combining Δ − n , A \Delta_{-n,A} Δ − n , A and Δ − n , B \Delta_{-n,B} Δ − n , B allows us to compute the values of the integral Δ − n \Delta_{-n} Δ − n in Eq. (343) ,
Δ − 1 ( τ , k ; τ ∗ , τ f i n ) = − k τ ∗ cos k τ [ Si ( k τ ˉ ) − Si ( k τ ∗ ) ] + k τ ∗ sin k τ [ Ci ( k τ ˉ ) − Ci ( k τ ∗ ) ] − τ ∗ τ ˉ cos k ( τ − τ ˉ ) + cos k ( τ − τ ∗ ) , (350) \Delta_{-1}(\tau,k;\tau_\ast,\tau_{\rm fin})=-k\tau_\ast\cos k\tau\left[\operatorname{Si}(k\bar{\tau})-\operatorname{Si}(k\tau_\ast)\right]+k\tau_\ast\sin k\tau\left[\operatorname{Ci}(k\bar{\tau})-\operatorname{Ci}(k\tau_\ast)\right]-\frac{\tau_\ast}{\bar{\tau}}\cos k(\tau-\bar{\tau})+\cos k(\tau-\tau_\ast), \tag*{(350)} Δ − 1 ( τ , k ; τ ∗ , τ fin ) = − k τ ∗ cos k τ [ Si ( k τ ˉ ) − Si ( k τ ∗ ) ] + k τ ∗ sin k τ [ Ci ( k τ ˉ ) − Ci ( k τ ∗ ) ] − τ ˉ τ ∗ cos k ( τ − τ ˉ ) + cos k ( τ − τ ∗ ) , (350)
Δ − 2 ( τ , k ; τ ∗ , τ f i n ) = − k 2 τ ∗ 2 2 cos k τ [ Ci ( k τ ˉ ) − Ci ( k τ ∗ ) ] − k 2 τ ∗ 2 2 sin k τ [ Si ( k τ ˉ ) − Si ( k τ ∗ ) ] + 1 2 [ cos k ( τ − τ ∗ ) + k τ ∗ sin k ( τ − τ ∗ ) ] − 1 2 [ τ ∗ 2 τ ˉ 2 cos k ( τ − τ ˉ ) + τ ∗ τ ˉ k τ ∗ sin k ( τ − τ ˉ ) ] . (351) \begin{aligned}
\Delta_{-2}(\tau,k;\tau_\ast,\tau_{\rm fin})={}&-\frac{k^2\tau_\ast^2}{2}\cos k\tau\left[\operatorname{Ci}(k\bar{\tau})-\operatorname{Ci}(k\tau_\ast)\right]-\frac{k^2\tau_\ast^2}{2}\sin k\tau\left[\operatorname{Si}(k\bar{\tau})-\operatorname{Si}(k\tau_\ast)\right]\\
&+\frac{1}{2}\left[\cos k(\tau-\tau_\ast)+k\tau_\ast\sin k(\tau-\tau_\ast)\right]\\
&-\frac{1}{2}\left[\frac{\tau_\ast^2}{\bar{\tau}^2}\cos k(\tau-\bar{\tau})+\frac{\tau_\ast}{\bar{\tau}}k\tau_\ast\sin k(\tau-\bar{\tau})\right]. \tag*{(351)}
\end{aligned} Δ − 2 ( τ , k ; τ ∗ , τ fin ) = − 2 k 2 τ ∗ 2 cos k τ [ Ci ( k τ ˉ ) − Ci ( k τ ∗ ) ] − 2 k 2 τ ∗ 2 sin k τ [ Si ( k τ ˉ ) − Si ( k τ ∗ ) ] + 2 1 [ cos k ( τ − τ ∗ ) + k τ ∗ sin k ( τ − τ ∗ ) ] − 2 1 [ τ ˉ 2 τ ∗ 2 cos k ( τ − τ ˉ ) + τ ˉ τ ∗ k τ ∗ sin k ( τ − τ ˉ ) ] . (351)
Let us now consider the cases with m = n m=n m = n , being n n n positive integers. The case m = 1 m=1 m = 1 is straightforward
Δ 1 ( τ , k ; τ ∗ , τ f i n ) = τ ∗ ∫ τ ∗ τ ˉ cos k ( τ − τ 1 ) d τ 1 = 1 k τ ∗ [ sin k ( τ − τ ∗ ) − sin k ( τ − τ ˉ ) ] , (352) \Delta_1(\tau,k;\tau_\ast,\tau_{\rm fin})=\tau_\ast\int_{\tau_\ast}^{\bar{\tau}}\cos k(\tau-\tau_1)\,\mathrm{d}\tau_1=\frac{1}{k\tau_\ast}\left[\sin k(\tau-\tau_\ast)-\sin k(\tau-\bar{\tau})\right], \tag*{(352)} Δ 1 ( τ , k ; τ ∗ , τ fin ) = τ ∗ ∫ τ ∗ τ ˉ cos k ( τ − τ 1 ) d τ 1 = k τ ∗ 1 [ sin k ( τ − τ ∗ ) − sin k ( τ − τ ˉ ) ] , (352)
which can also be computed from Eq. (335) using
Γ ( 1 , ± i k τ ) = ± i k ∫ τ ∓ i ∞ e ∓ i k p d p = e ∓ i k τ . (353) \Gamma(1,\pm ik\tau)=\pm ik\int_{\tau}^{\mp i\infty}e^{\mp ikp}\,\mathrm{d}p=e^{\mp ik\tau}. \tag*{(353)} Γ ( 1 , ± ik τ ) = ± ik ∫ τ ∓ i ∞ e ∓ ik p d p = e ∓ ik τ . (353)
Larger values of n n n can be computed using the recursive formula for the upper incomplete gamma,
Γ ( n , ± i k τ ) = ( n − 1 ) Γ ( n − 1 , ± i k τ ) + ( ± i k τ ) n − 1 e ∓ i k τ . (354) \Gamma(n,\pm ik\tau)=(n-1)\Gamma(n-1,\pm ik\tau)+(\pm ik\tau)^{n-1}e^{\mp ik\tau}. \tag*{(354)} Γ ( n , ± ik τ ) = ( n − 1 ) Γ ( n − 1 , ± ik τ ) + ( ± ik τ ) n − 1 e ∓ ik τ . (354)
In particular, for n = 2 n=2 n = 2 , we have
Γ ( 2 , ± i k τ ) = e ∓ i k τ ( 1 ± i k τ ) , (355) \Gamma(2,\pm ik\tau)=e^{\mp ik\tau}(1\pm ik\tau), \tag*{(355)} Γ ( 2 , ± ik τ ) = e ∓ ik τ ( 1 ± ik τ ) , (355)
and hence, from Eq. (335) ,
Δ 2 ( τ , k ; τ ∗ , τ f i n ) = 1 2 ( k τ ∗ ) − 2 [ e − i k ( t − τ ) ( 1 + i k t ) + e i k ( t − τ ) ( 1 − i k t ) ] τ ∗ τ ˉ = 1 k 2 τ ∗ 2 [ cos k ( τ − τ ˉ ) − cos k ( τ − τ ∗ ) ] + 1 k τ ∗ [ sin k ( τ − τ ∗ ) − τ ˉ τ ∗ sin k ( τ − τ ˉ ) ] . (356) \begin{aligned}
\Delta_2(\tau,k;\tau_\ast,\tau_{\rm fin})={}&\frac{1}{2}(k\tau_\ast)^{-2}\left[e^{-ik(t-\tau)}(1+ikt)+e^{ik(t-\tau)}(1-ikt)\right]_{\tau_\ast}^{\bar{\tau}}\\
={}&\frac{1}{k^2\tau_\ast^2}\left[\cos k(\tau-\bar{\tau})-\cos k(\tau-\tau_\ast)\right]\\
&+\frac{1}{k\tau_\ast}\left[\sin k(\tau-\tau_\ast)-\frac{\bar{\tau}}{\tau_\ast}\sin k(\tau-\bar{\tau})\right]. \tag*{(356)}
\end{aligned} Δ 2 ( τ , k ; τ ∗ , τ fin ) = = 2 1 ( k τ ∗ ) − 2 [ e − ik ( t − τ ) ( 1 + ik t ) + e ik ( t − τ ) ( 1 − ik t ) ] τ ∗ τ ˉ k 2 τ ∗ 2 1 [ cos k ( τ − τ ˉ ) − cos k ( τ − τ ∗ ) ] + k τ ∗ 1 [ sin k ( τ − τ ∗ ) − τ ∗ τ ˉ sin k ( τ − τ ˉ ) ] . (356)
Asymptotic values of Δ m ( τ , k ; τ ∗ , τ f i n ) \Delta_m(\tau,k;\tau_\ast,\tau_{\mathrm{fin}}) Δ m ( τ , k ; τ ∗ , τ fin ) when k → 0 k \to0 k → 0
Let us now study the values of Δ m \Delta_m Δ m in the asymptotic limits k → 0 k \to0 k → 0 and k → ∞ k \to\infty k → ∞ .
We start by considering the case with negative integers m = − n m=-n m = − n , given in Eq. (343) as
Δ − n = − 1 2 τ ∗ i n k n n ! [ Δ − n , A + Δ − n , B ] , (357) \Delta_{-n}=-\frac{1}{2}\tau_\ast\frac{i^n k^n}{n!}\left[\Delta_{-n,A}+\Delta_{-n,B}\right], \tag*{(357)} Δ − n = − 2 1 τ ∗ n ! i n k n [ Δ − n , A + Δ − n , B ] , (357)
which shows that the asymptotic behavior of Δ − n \Delta_{-n} Δ − n will be determined by that of the functions inside the square brackets multiplied by k n k^n k n .
In the limit k → 0 k \to0 k → 0 , for even n = 2 l n=2l n = 2 l , the leading order term of the contribution from Δ − 2 l , A \Delta_{-2l,A} Δ − 2 l , A is
− 1 2 i 2 l k 2 l τ ∗ 2 l ( 2 l ) ! Δ − 2 l , A = ( − 1 ) l k 2 l τ ∗ 2 l ( 2 l ) ! { cos k τ [ Ci ( k τ ˉ ) − Ci ( k τ ∗ ) ] + sin k τ [ Si ( k τ ˉ ) − Si ( k τ ∗ ) ] } -\frac{1}{2}\frac{i^{2l}k^{2l}\tau_\ast^{2l}}{(2l)!}\Delta_{-2l,A}
=
\frac{(-1)^l k^{2l}\tau_\ast^{2l}}{(2l)!}
\left\{
\cos k\tau\left[\operatorname{Ci}(k\bar{\tau})-\operatorname{Ci}(k\tau_\ast)\right]
+\sin k\tau\left[\operatorname{Si}(k\bar{\tau})-\operatorname{Si}(k\tau_\ast)\right]
\right\} − 2 1 ( 2 l )! i 2 l k 2 l τ ∗ 2 l Δ − 2 l , A = ( 2 l )! ( − 1 ) l k 2 l τ ∗ 2 l { cos k τ [ Ci ( k τ ˉ ) − Ci ( k τ ∗ ) ] + sin k τ [ Si ( k τ ˉ ) − Si ( k τ ∗ ) ] }
→ k → 0 ( − 1 ) l ( k τ ∗ ) 2 l ( 2 l ) ! [ ln ( τ ˉ / τ ∗ ) + O ( ( k τ ∗ ) 2 ) ] , (358) \xrightarrow{k\to0}
\frac{(-1)^l(k\tau_\ast)^{2l}}{(2l)!}
\left[\ln(\bar{\tau}/\tau_\ast)+\mathcal{O}((k\tau_\ast)^2)\right], \tag*{(358)} k → 0 ( 2 l )! ( − 1 ) l ( k τ ∗ ) 2 l [ ln ( τ ˉ / τ ∗ ) + O (( k τ ∗ ) 2 ) ] , (358)
where the functions Si ( k x ) \operatorname{Si}(kx) Si ( k x ) and Ci ( k x ) \operatorname{Ci}(kx) Ci ( k x ) in the limit k x ≪ 1 kx\ll1 k x ≪ 1 are given in Eqs. (C2a) and (C2b). On the other hand, in the limit k → 0 k \to0 k → 0 , the leading order term of Δ − 2 l , B \Delta_{-2l,B} Δ − 2 l , B is given by [see Eq. (347) ],
− 1 2 i 2 l k 2 l τ ∗ 2 l ( 2 l ) ! Δ − 2 l , B → k → 0 ( − 1 ) l + 1 k 2 l τ ∗ 2 l ( 2 l ) ! { A 2 l ( τ ˉ ) − A 2 l ( τ ∗ ) + O ( ( k τ ∗ ) − 2 l + 1 ) } = 1 2 l [ 1 − ( τ ˉ / τ ∗ ) − 2 l ] + O ( k τ ∗ ) . (359) \begin{aligned}
-\frac{1}{2}\frac{i^{2l}k^{2l}\tau_\ast^{2l}}{(2l)!}\Delta_{-2l,B}
\xrightarrow{k\to0}
\frac{(-1)^{l+1}k^{2l}\tau_\ast^{2l}}{(2l)!}
\left\{A_{2l}(\bar{\tau})-A_{2l}(\tau_\ast)+\mathcal{O}((k\tau_\ast)^{-2l+1})\right\} \\
&=\frac{1}{2l}\left[1-(\bar{\tau}/\tau_\ast)^{-2l}\right]+\mathcal{O}(k\tau_\ast). \tag*{(359)}
\end{aligned} − 2 1 ( 2 l )! i 2 l k 2 l τ ∗ 2 l Δ − 2 l , B k → 0 ( 2 l )! ( − 1 ) l + 1 k 2 l τ ∗ 2 l { A 2 l ( τ ˉ ) − A 2 l ( τ ∗ ) + O (( k τ ∗ ) − 2 l + 1 ) } = 2 l 1 [ 1 − ( τ ˉ / τ ∗ ) − 2 l ] + O ( k τ ∗ ) . (359)
This shows that for even n n n , the value of Δ − n \Delta_{-n} Δ − n , which is the combination of the terms in Eqs. (D30) and (D31), is dominated by the latter term in the limit k → 0 k\to0 k → 0 , as long as l > 0 l>0 l > 0 , and its squared value is
lim k → 0 Δ − 2 l 2 ( τ , k ; τ ∗ , τ f i n ) = 1 ( 2 l ) 2 [ 1 − ( τ ˉ / τ ∗ ) − 2 l ] 2 . (360) \lim_{k\to0}\Delta_{-2l}^{2}(\tau,k;\tau_\ast,\tau_{\mathrm{fin}})
=
\frac{1}{(2l)^2}\left[1-(\bar{\tau}/\tau_\ast)^{-2l}\right]^2. \tag*{(360)} k → 0 lim Δ − 2 l 2 ( τ , k ; τ ∗ , τ fin ) = ( 2 l ) 2 1 [ 1 − ( τ ˉ / τ ∗ ) − 2 l ] 2 . (360)
At large τ ˉ / τ ∗ ≫ 1 \bar{\tau}/\tau_\ast\gg1 τ ˉ / τ ∗ ≫ 1 , given that l > 0 l>0 l > 0 , we find the following asymptotic value
lim k → 0 τ ˉ / τ ∗ → ∞ Δ − 2 l 2 ( τ , k ; τ ∗ , τ f i n ) = 1 ( 2 l ) 2 . (361) \lim_{\substack{k\to0\\ \bar{\tau}/\tau_\ast\to\infty}}
\Delta_{-2l}^{2}(\tau,k;\tau_\ast,\tau_{\mathrm{fin}})
=\frac{1}{(2l)^2}. \tag*{(361)} k → 0 τ ˉ / τ ∗ → ∞ lim Δ − 2 l 2 ( τ , k ; τ ∗ , τ fin ) = ( 2 l ) 2 1 . (361)
For odd n = 2 l − 1 n=2l-1 n = 2 l − 1 with l > 0 l>0 l > 0 , the leading order term in Δ − 2 l + 1 , A \Delta_{-2l+1,A} Δ − 2 l + 1 , A is, in the limit k → 0 k\to0 k → 0 ,
− 1 2 i 2 l − 1 k 2 l − 1 τ ∗ 2 l − 1 ( 2 l − 1 ) ! Δ − 2 l + 1 , A = ( − 1 ) l ( k τ ∗ ) 2 l − 1 ( 2 l − 1 ) ! { cos k τ [ Si ( k τ ˉ ) − Si ( k τ ∗ ) ] − sin k τ [ Ci ( k τ ˉ ) − Ci ( k τ ∗ ) ] } -\frac{1}{2}\frac{i^{2l-1}k^{2l-1}\tau_\ast^{2l-1}}{(2l-1)!}\Delta_{-2l+1,A}
=
\frac{(-1)^l(k\tau_\ast)^{2l-1}}{(2l-1)!}
\left\{
\cos k\tau\left[\operatorname{Si}(k\bar{\tau})-\operatorname{Si}(k\tau_\ast)\right]
-\sin k\tau\left[\operatorname{Ci}(k\bar{\tau})-\operatorname{Ci}(k\tau_\ast)\right]
\right\} − 2 1 ( 2 l − 1 )! i 2 l − 1 k 2 l − 1 τ ∗ 2 l − 1 Δ − 2 l + 1 , A = ( 2 l − 1 )! ( − 1 ) l ( k τ ∗ ) 2 l − 1 { cos k τ [ Si ( k τ ˉ ) − Si ( k τ ∗ ) ] − sin k τ [ Ci ( k τ ˉ ) − Ci ( k τ ∗ ) ] }
→ k → 0 ( − 1 ) l ( k τ ∗ ) 2 l ( 2 l − 1 ) ! [ τ ˉ τ ∗ − 1 − τ τ ∗ ln ( τ ˉ / τ ∗ ) + O ( k τ ∗ ) ] . (362) \xrightarrow{k\to0}
\frac{(-1)^l(k\tau_\ast)^{2l}}{(2l-1)!}
\left[\frac{\bar{\tau}}{\tau_\ast}-1-\frac{\tau}{\tau_\ast}\ln(\bar{\tau}/\tau_\ast)+\mathcal{O}(k\tau_\ast)\right]. \tag*{(362)} k → 0 ( 2 l − 1 )! ( − 1 ) l ( k τ ∗ ) 2 l [ τ ∗ τ ˉ − 1 − τ ∗ τ ln ( τ ˉ / τ ∗ ) + O ( k τ ∗ ) ] . (362)
The leading-order term in Δ − n , B \Delta_{-n,B} Δ − n , B for n = 2 l − 1 n=2l-1 n = 2 l − 1 with l > 0 l>0 l > 0 is, as before, proportional to A n A_n A n
− 1 2 i 2 l − 1 ( k τ ∗ ) 2 l − 1 ( 2 l − 1 ) ! Δ − 2 l + 1 , B → k → 0 1 2 l − 1 [ 1 − ( τ ˉ / τ ∗ ) − 2 l + 1 ] + O ( k τ ∗ ) , (363) -\frac{1}{2}\frac{i^{2l-1}(k\tau_\ast)^{2l-1}}{(2l-1)!}\Delta_{-2l+1,B}
\xrightarrow{k\to0}
\frac{1}{2l-1}\left[1-(\bar{\tau}/\tau_\ast)^{-2l+1}\right]+\mathcal{O}(k\tau_\ast), \tag*{(363)} − 2 1 ( 2 l − 1 )! i 2 l − 1 ( k τ ∗ ) 2 l − 1 Δ − 2 l + 1 , B k → 0 2 l − 1 1 [ 1 − ( τ ˉ / τ ∗ ) − 2 l + 1 ] + O ( k τ ∗ ) , (363)
which, as previously, dominates over the Δ − n , A \Delta_{-n,A} Δ − n , A contribution, which is of order k 2 l k^{2l} k 2 l for all l > 0 l>0 l > 0 .
Hence, for any m = − n m=-n m = − n , we find that the value of Δ m \Delta_m Δ m , in the limit k → 0 k\to0 k → 0 , is
lim k → 0 Δ − n 2 ( τ , k ; τ ∗ , τ f i n ) = 1 n 2 [ 1 − ( τ ˉ / τ ∗ ) − n ] 2 , (364) \lim_{k\to0}\Delta_{-n}^{2}(\tau,k;\tau_\ast,\tau_{\mathrm{fin}})
=
\frac{1}{n^2}\left[1-(\bar{\tau}/\tau_\ast)^{-n}\right]^2, \tag*{(364)} k → 0 lim Δ − n 2 ( τ , k ; τ ∗ , τ fin ) = n 2 1 [ 1 − ( τ ˉ / τ ∗ ) − n ] 2 , (364)
which, at large τ ˉ / τ ∗ ≫ 1 \bar{\tau}/\tau_\ast\gg1 τ ˉ / τ ∗ ≫ 1 , yields the following asymptotic value
lim k → 0 τ ˉ / τ ∗ → ∞ Δ − n 2 ( τ , k ; τ ∗ , τ f i n ) = 1 n 2 . (365) \lim_{\substack{k\to0\\ \bar{\tau}/\tau_\ast\to\infty}}
\Delta_{-n}^{2}(\tau,k;\tau_\ast,\tau_{\mathrm{fin}})
=\frac{1}{n^2}. \tag*{(365)} k → 0 τ ˉ / τ ∗ → ∞ lim Δ − n 2 ( τ , k ; τ ∗ , τ fin ) = n 2 1 . (365)
For positive integers m = n m=n m = n we find that, in the limit k → 0 k\to0 k → 0 , the functions Δ n \Delta_n Δ n for n = 1 , 2 n=1,2 n = 1 , 2 take the values [see Eqs. (352) and (356) ]
lim k → 0 Δ 1 2 ( τ , k ; τ ∗ , τ f i n ) = ( δ τ ˉ / τ ∗ ) 2 , (366) \lim_{k\to0}\Delta_1^2(\tau,k;\tau_\ast,\tau_{\mathrm{fin}})
=(\delta\bar{\tau}/\tau_\ast)^2, \tag*{(366)} k → 0 lim Δ 1 2 ( τ , k ; τ ∗ , τ fin ) = ( δ τ ˉ / τ ∗ ) 2 , (366)
lim k → 0 Δ 2 2 ( τ , k ; τ ∗ , τ f i n ) = ( δ τ ˉ τ ∗ ) 2 [ 1 + δ τ ˉ τ ∗ + 1 4 ( δ τ ˉ τ ∗ ) 2 ] , (367) \lim_{k\to0}\Delta_2^2(\tau,k;\tau_\ast,\tau_{\mathrm{fin}})
=\left(\frac{\delta\bar{\tau}}{\tau_\ast}\right)^2
\left[1+\frac{\delta\bar{\tau}}{\tau_\ast}+\frac{1}{4}\left(\frac{\delta\bar{\tau}}{\tau_\ast}\right)^2\right], \tag*{(367)} k → 0 lim Δ 2 2 ( τ , k ; τ ∗ , τ fin ) = ( τ ∗ δ τ ˉ ) 2 [ 1 + τ ∗ δ τ ˉ + 4 1 ( τ ∗ δ τ ˉ ) 2 ] , (367)
where δ τ ˉ ≡ τ ˉ − τ ∗ \delta\bar{\tau}\equiv\bar{\tau}-\tau_\ast δ τ ˉ ≡ τ ˉ − τ ∗ . For τ < τ f i n \tau<\tau_{\mathrm{fin}} τ < τ fin we have τ ˉ = τ \bar{\tau}=\tau τ ˉ = τ and these expressions reduce to
lim k → 0 τ < τ f i n Δ 1 2 ( τ , k ; τ ∗ , τ f i n ) = ( δ τ / τ ∗ ) 2 , (368) \lim_{\substack{k\to0\\ \tau<\tau_{\mathrm{fin}}}
}\Delta_1^2(\tau,k;\tau_\ast,\tau_{\mathrm{fin}})
=(\delta\tau/\tau_\ast)^2, \tag*{(368)} k → 0 τ < τ fin lim Δ 1 2 ( τ , k ; τ ∗ , τ fin ) = ( δ τ / τ ∗ ) 2 , (368)
lim k → 0 τ < τ f i n Δ 2 2 ( τ , k ; τ ∗ , τ f i n ) = 1 4 ( δ τ / τ ∗ ) 4 + ( δ τ / τ ∗ ) 3 + ( δ τ / τ ∗ ) 2 . (369) \lim_{\substack{k\to0\\ \tau<\tau_{\mathrm{fin}}}
}\Delta_2^2(\tau,k;\tau_\ast,\tau_{\mathrm{fin}})
=\frac{1}{4}(\delta\tau/\tau_\ast)^4+(\delta\tau/\tau_\ast)^3+(\delta\tau/\tau_\ast)^2. \tag*{(369)} k → 0 τ < τ fin lim Δ 2 2 ( τ , k ; τ ∗ , τ fin ) = 4 1 ( δ τ / τ ∗ ) 4 + ( δ τ / τ ∗ ) 3 + ( δ τ / τ ∗ ) 2 . (369)
Asymptotic values of Δ m ( τ , k ; τ ∗ , τ f i n ) \Delta_m(\tau,k;\tau_\ast,\tau_{\mathrm{fin}}) Δ m ( τ , k ; τ ∗ , τ fin ) when k → ∞ k \to\infty k → ∞
We now study the k → ∞ k \to\infty k → ∞ limit. To find the asymptotic form for k → ∞ k \to\infty k → ∞ of Δ m \Delta_m Δ m we can start directly from its definition in Eq. (330) and integrate by parts
Δ m ( τ , k ; τ ∗ , τ f i n ) = τ ∗ − m ∫ τ ∗ τ ˉ τ 1 m − 1 cos k ( τ − τ 1 ) d τ 1 = − τ ∗ − m k [ τ 1 m − 1 sin k ( τ − τ 1 ) ] τ ∗ τ ˉ + ( m − 1 ) τ ∗ − m k ∫ τ ∗ τ ˉ τ 1 m − 2 sin k ( τ − τ 1 ) d τ 1 ≃ k → ∞ 1 k τ ∗ m [ sin k ( τ − τ ∗ ) − ( τ ˉ τ ∗ ) m − 1 sin k ( τ − τ ˉ ) ] , (370) \begin{aligned}
\Delta_m(\tau,k;\tau_\ast,\tau_{\mathrm{fin}}) &= \tau_\ast^{-m}\int_{\tau_\ast}^{\bar{\tau}}\tau_1^{m-1}\cos k(\tau-\tau_1)\,\mathrm{d}\tau_1 \\
&= -\frac{\tau_\ast^{-m}}{k}\left[\tau_1^{m-1}\sin k(\tau-\tau_1)\right]_{\tau_\ast}^{\bar{\tau}}+\frac{(m-1)\tau_\ast^{-m}}{k}\int_{\tau_\ast}^{\bar{\tau}}\tau_1^{m-2}\sin k(\tau-\tau_1)\,\mathrm{d}\tau_1 \\
&\underset{k\to\infty}{\simeq}\frac{1}{k\tau_\ast^m}\left[\sin k(\tau-\tau_\ast)-\left(\frac{\bar{\tau}}{\tau_\ast}\right)^{m-1}\sin k(\tau-\bar{\tau})\right], \tag*{(370)}
\end{aligned} Δ m ( τ , k ; τ ∗ , τ fin ) = τ ∗ − m ∫ τ ∗ τ ˉ τ 1 m − 1 cos k ( τ − τ 1 ) d τ 1 = − k τ ∗ − m [ τ 1 m − 1 sin k ( τ − τ 1 ) ] τ ∗ τ ˉ + k ( m − 1 ) τ ∗ − m ∫ τ ∗ τ ˉ τ 1 m − 2 sin k ( τ − τ 1 ) d τ 1 k → ∞ ≃ k τ ∗ m 1 [ sin k ( τ − τ ∗ ) − ( τ ∗ τ ˉ ) m − 1 sin k ( τ − τ ˉ ) ] , (370)
where the boundary term from the integration by parts goes as k − 1 k^{-1} k − 1 , while the remaining integral can be shown to be, through an additional integration by parts, of O ( k − 2 ) \mathcal{O}(k^{-2}) O ( k − 2 ) .
We then see that, for m = 0 m=0 m = 0 , Eq. (370) recovers the result for the function Δ \Delta Δ in the CiT model [see Sec. C and in particular Eq. (C7)]. For m ≠ 0 m\ne0 m = 0 , instead, the envelope of the function for τ > τ f i n \tau>\tau_{\mathrm{fin}} τ > τ fin is modified and bounded by
[ Δ m 2 ] e n v ≤ k → ∞ 1 k 2 τ ∗ 2 [ 1 + ( τ f i n τ ∗ ) m − 1 ] 2 , (371) \left[\Delta_m^2\right]_{\mathrm{env}}\underset{k\to\infty}{\leq}\frac{1}{k^2\tau_\ast^2}\left[1+\left(\frac{\tau_{\mathrm{fin}}}{\tau_\ast}\right)^{m-1}\right]^2, \tag*{(371)} [ Δ m 2 ] env k → ∞ ≤ k 2 τ ∗ 2 1 [ 1 + ( τ ∗ τ fin ) m − 1 ] 2 , (371)
which reduces to Eqs. (327) and (C13) for m = 0 m=0 m = 0 . However, the additional contribution to the envelope for τ > τ f i n \tau>\tau_{\mathrm{fin}} τ > τ fin is an artifact of the abrupt switching off of the sourcing at the time τ f i n \tau_{\mathrm{fin}} τ fin . Hence, if we neglect this factor, we see that the large wave numbers are not affected by the decay of the source and we recover, in the k → ∞ k\to\infty k → ∞ limit, the same behavior as in the CiT model.
Fast evolution of the spectral peak in the coherent-decay model
Let us now study the time evolution of ζ Π r a t ( τ , k ) \zeta_{\Pi}^{\mathrm{rat}}(\tau,k) ζ Π rat ( τ , k ) , i.e., the function in Eq. (191) with k ∗ ( τ ) k_\ast(\tau) k ∗ ( τ ) evolving as in Eq. (159a). In the following, we will restrict ourselves, for simplicity, to the case δ τ d c = τ ∗ \delta\tau_{\mathrm{dc}}=\tau_\ast δ τ dc = τ ∗ , as in App. § 1.4.1 . Figure 32 shows the time evolution of this function for α = 2 \alpha=2 α = 2 and α = 6 / 17 \alpha=6/17 α = 6/17 , while the broken power law parameters a , b a,b a , b of ζ ( K ) \zeta(K) ζ ( K ) are always chosen to be 5 and 2 / 3 2/3 2/3 and we restrict to the case of a fully vortical non-helical vector field. We observe that, at time τ \tau τ , the wave numbers smaller than k ∗ ( τ / τ ∗ ) − s k_\ast(\tau/\tau_\ast)^{-s} k ∗ ( τ / τ ∗ ) − s are still not significantly affected by the time evolution of the peak and ζ Π r a t \zeta_{\Pi}^{\mathrm{rat}} ζ Π rat has an approximately constant value. In the large wave numbers region (i.e., k ≫ k ∗ k\gg k_\ast k ≫ k ∗ ), the asymptotic branch of ζ Π r a t \zeta_{\Pi}^{\mathrm{rat}} ζ Π rat , proportional to ( k τ ∗ s / k ∗ ) − 11 / 3 (k\tau_\ast^s/k_\ast)^{-11/3} ( k τ ∗ s / k ∗ ) − 11/3 , decays indeed as ( τ / τ ∗ ) − 11 s / 3 (\tau/\tau_\ast)^{-11s/3} ( τ / τ ∗ ) − 11 s /3 (c.f. the black dashed line in the right panels of Figure 32 ). The time evolution of ζ Π r a t \zeta_{\Pi}^{\mathrm{rat}} ζ Π rat at the intermediate wave numbers k / k ∗ ∈ ( τ ∗ − s τ s , 10 ) k/k_\ast\in(\tau_\ast^{-s}\tau^s,10) k / k ∗ ∈ ( τ ∗ − s τ s , 10 ) smoothly transitions from constant at τ ≲ τ ∗ ( k / k ∗ ) s \tau\lesssim\tau_\ast(k/k_\ast)^s τ ≲ τ ∗ ( k / k ∗ ) s to ( τ / τ ∗ ) − 11 s / 3 (\tau/\tau_\ast)^{-11s/3} ( τ / τ ∗ ) − 11 s /3 as k / k ∗ k/k_\ast k / k ∗ increases.
We find that, in the case with p = 1 p=1 p = 1 and s = 1 2 s=\frac{1}{2} s = 2 1 , shown in Figure 23 , the inclusion of the time evolution of ζ Π r a t \zeta_{\Pi}^{\mathrm{rat}} ζ Π rat does not lead to a significant difference in the resulting Δ p s 2 \Delta_{ps}^2 Δ p s 2 [cf. Eq. (193) ]. Note that, in fact, the evolution of ζ Π r a t \zeta_{\Pi}^{\mathrm{rat}} ζ Π rat is mostly relevant around the peak wave number k ∗ ( τ ) k_\ast(\tau) k ∗ ( τ ) , and that the time evolutions of k ∗ ( τ ) k_\ast(\tau) k ∗ ( τ ) and E ∗ ( τ ) E_\ast(\tau) E ∗ ( τ ) [cf. Eqs. (159)] are still being taken into account in the amplitude of E Π E_\Pi E Π via the τ 1 m \tau_1^m τ 1 m factor in Δ p s \Delta_{ps} Δ p s . Moreover, comparing the wave number range k > k ∗ ( τ / τ ∗ ) − s k>k_\ast(\tau/\tau_\ast)^{-s} k > k ∗ ( τ / τ ∗ ) − s affected by the decay, to the range in which the GW is already saturated k > 1 / τ k>1/\tau k > 1/ τ , we find that the evolution of ζ Π r a t \sqrt{\zeta_{\Pi}^{\mathrm{rat}}} ζ Π rat can be neglected when
k ∗ ( τ τ ∗ ) − s > 1 τ → s < 1 + ln ( k ∗ τ ∗ ) ln ( τ / τ ∗ ) for τ > τ ∗ . (372) \begin{aligned}
k_\ast\left(\frac{\tau}{\tau_\ast}\right)^{-s}>\frac{1}{\tau} \\
&\to s<1+\frac{\ln(k_\ast\tau_\ast)}{\ln(\tau/\tau_\ast)}\quad\text{for }\tau>\tau_\ast. \tag*{(372)}
\end{aligned} k ∗ ( τ ∗ τ ) − s > τ 1 → s < 1 + ln ( τ / τ ∗ ) ln ( k ∗ τ ∗ ) for τ > τ ∗ . (372)
Hence, for all subhorizon k ∗ > H k_\ast>\mathcal{H} k ∗ > H , the evolution of ζ Π r a t \zeta_{\Pi}^{\mathrm{rat}} ζ Π rat can be neglected for all values s < 1 s<1 s < 1 , i.e., when the evolution of the peak scale of ζ Π r a t \zeta_{\Pi}^{\mathrm{rat}} ζ Π rat is slower than the GW saturation.