betaFeedback

WorksPhysics and Astronomy

Local Measures in Odd de Sitter

Abstract

We show that massless theories in odd-dimensional de Sitter space have very simple measures—the modulus squared of the late-time wavefunctional—at tree level. They are local functionals; in particular, all wavefunction coefficients vanish at separated positions. All the usual singularities of nonlocal wavefunctions in even dimensions only affect the phase in the odd dimensional case. Hence a closed formula can be written for any number of points. We focus on three examples: the non-linear sigma model, Yang-Mills theory and General Relativity, from dimensions three to seven.

Introduction

The probability measure for massless field fluctuations in de Sitter space, given by the absolute value squared of the vacuum wavefunction, is typically a complicated functional, with wavefunction coefficients having support at separated field insertions. The wavefunction coefficients are the closest analogues of scattering amplitudes in cosmology,1 and a systematic development of this analogy has been very fruitful in recent years (see e.g. [5] and references therein). Due to redshift, conjugate momenta tend to have vanishingly small amplitude, meaning that the de Sitter vacuum is a heavily squeezed state. For that reason, typically we are interested in computing only the absolute value squared of the wavefunction, or equivalently the real part of wavefunction coefficients. The probability measure, i.e. the real part of the wavefunction coefficients, is the object from which we can compute quantum expectation values of fields at late times [6]. In this paper we point out that massless interacting gauge theories in odd-dimensional de Sitter space have local probability measures,2 which we write explicitly. This implies that they are analytic functions of the momenta.

In four-dimensional gauge theories, the organization of amplitudes on helicity sectors has proven to be particularly useful. At generic kinematics, tree-level all-plus and single-minus Yang-Mills amplitudes vanish, the first essentially because of supersymmetry, the second due to classical integrability of the self-dual sector of Yang-Mills theory3 (similar statements apply to gravity as well). However, the MHV sector, described at arbitrary multiplicity by the Parke-Taylor formula [11], is non-zero at very special kinematics. This has been highlighted in a series of recent papers [12, 13, 14] which have moreover found a general form of the amplitude at nn points. The same phenomenon persists in gravity. Finally the MHV amplitude is meromorphic and has support for any value of the momenta. Of course, all of these statements hold at tree-level.4

Local measures like the ones presented in this work provide, in our opinion, the closest analogue to special amplitudes. This is because the measure is only nonzero when all field insertions at the boundary are at the same location. In other words, the singularities associated to propagators are absent, and yet there is a nontrivial measure. Despite the locality of the measure, the expectation values of fields are nonlocal, although simpler than one would naively expect.5

We make this quantitative in § 2. We outline the setup in full generality, and explain why the finite nonlocal contributions to the wavefunction enter only its phase in odd dimensions. This is followed by reviewing one example of this phenomenon, first shown in [16], namely 5d Einstein gravity in de Sitter (dS) space. There, the data of the late-time probability measure is a Euclidean four-metric in the boundary, which is taken to have flat topology, and reads

∣ΨdS[g^]∣2=exp⁡(−Mpl3RdS32π8∫g^ W^[g^])(1)\left|\Psi_{\mathrm{dS}}[\hat{g}]\right|^2 = \exp\left(-\frac{M_{\mathrm{pl}}^3 R_{\mathrm{dS}}^3}{2}\frac{\pi}{8}\int\sqrt{\hat{g}}\,\hat{W}[\hat{g}]\right) \tag*{(1)}

where W^\hat{W} is the Weyl tensor of the boundary metric g^\hat{g}. This is a remarkably simple contribution, predicting the tree-level measure at any number of points, while in e.g. 4d gravity in de Sitter we have only recently been able to compute its analogue beyond four-points! Of course in such a theory cosmological correlators are nontrivial, and computing them remains a difficult task. However, they all contribute to the phase of the wavefunction, leaving no imprint on the measure. The fact that this contribution stems from infrared divergences of the bulk theory, explains locality through the holographic principle [17].

We further push this observation and show that Einstein gravity, Yang-Mills theory, and non-linear sigma models (NLSM) in odd-dimensional de Sitter space have similarly simple, local measures. The results are summarized in Table 1. Besides deriving the measures, in § 3 we report checks of our formulae from perturbation theory and the cosmological CPT theorem [18]. Finally, in § 4, we comment on possible generalizations to semi-local measures, special cases in three dimensions, and potential avenues for a deeper understanding of the origin of locality. Since the present work is restricted to tree level, we also comment on possible one-loop contributions, leaving a full analysis for future work.

Theorylog⁡∣ΨdS∣2\log\lvert\Psi_{\mathrm{dS}}\rvert^2
5d gravity−Mpl3RdS32π8∫d4xg^ W^ijklW^ijkl-\frac{M_{\mathrm{pl}}^3R_{\mathrm{dS}}^3}{2}\frac{\pi}{8}\int d^4x\sqrt{\hat{g}}\,\hat{W}_{ijkl}\hat{W}^{ijkl}
7d gravity−πMpl5RdS5384∫d6xg^[12W^ikljW^imnjW^  mnkl+3W^ij  klW^kl  mnW^mn  ij+W^iklm(δji□^−4R^ji+65R^δji)W^jklm]-\frac{\pi M_{\mathrm{pl}}^5R_{\mathrm{dS}}^5}{384}\int d^6x\sqrt{\hat{g}}\left[12\hat{W}_{iklj}\hat{W}^{imnj}\hat{W}^{kl}_{\ \ mn}+3\hat{W}_{ij}^{\ \ kl}\hat{W}_{kl}^{\ \ mn}\hat{W}_{mn}^{\ \ ij}+\hat{W}_{iklm}\left(\delta^i_j\hat{\Box}-4\hat{R}^i_j+\frac{6}{5}\hat{R}\delta^i_j\right)\hat{W}^{jklm}\right]
5d Yang-Mills−RdS4g52π2∫d4xF^ijaF^a ij-\frac{R_{\mathrm{dS}}}{4g_5^2}\frac{\pi}{2}\int d^4x\hat{F}^a_{ij}\hat{F}^{a\,ij}
7d Yang-Mills−RdS34g72π2∫d6x12[(D^jF^jia)(D^kF^  ka i)]-\frac{R_{\mathrm{dS}}^3}{4g_7^2}\frac{\pi}{2}\int d^6x\frac{1}{2}\left[(\hat{D}^j\hat{F}^a_{ji})(\hat{D}^k\hat{F}^{a\,i}_{\ \ k})\right]
3d NLSM−πRdS2g32∫d2x GAB(ϕ)∂iϕA∂iϕB-\frac{\pi R_{\mathrm{dS}}}{2g_3^2}\int d^2x\,G_{AB}(\phi)\partial_i\phi^A\partial^i\phi^B
5d NLSM−πRdS38g52∫d4x GAB(ϕ)(Di∂iϕA)(Dj∂jϕB)-\frac{\pi R_{\mathrm{dS}}^3}{8g_5^2}\int d^4x\,G_{AB}(\phi)(D^i\partial_i\phi^A)(D^j\partial_j\phi^B)
7d NLSM−πRdS5128g72∫d6x[Di(Di∂jϕA)Di(Dk∂kϕB)−RABCD(Dj∂jϕA)∂iϕB(Dk∂kϕC)∂iϕD]-\frac{\pi R_{\mathrm{dS}}^5}{128g_7^2}\int d^6x\left[D_i(D^i\partial_j\phi^A)D^i(D^k\partial_k\phi^B)-R_{ABCD}(D^j\partial_j\phi^A)\partial_i\phi^B(D^k\partial_k\phi^C)\partial^i\phi^D\right]

Table 1. Logarithm of the dS measure for gravity, Yang-Mills theory, and NLSM in different dimensions. In 7d Yang-Mills, the factor of 12\frac{1}{2} in the integrand does not arise from a difference in normalization: the action is normalized according to the same convention as in 5d. In 7d NLSM, RABCDR_{ABCD} is the Riemann tensor of the target space metric GABG_{AB}.

EAdS Origin of the Local Measure

The primary objective of this work is to determine the probability measure of the late-time quantum wavefunction in de Sitter space for a broad class of theories, namely gravity, Yang-Mills theory, and non-linear sigma models. Computing the cosmological wavefunction is generally a technically demanding problem: each coefficient receives contributions from contact and exchange diagrams involving bulk time integrals, together with boundary counterterms required to regulate late-time divergences. The result typically has a complicated nonlocal dependence on the external momenta. In the odd-dimensional theories considered here, however, it will turn out that all of this complexity is confined to the phase of the wavefunction, while the probability measure is determined entirely by a local logarithmic contribution. To see this, we spell out in detail the analytic continuation from Euclidean Anti-de Sitter (EAdS) and its features.

The EAdS argument

We consider a quantum field in (d+1)(d+1)-dimensional EAdS, using the Fefferman-Graham coordinate system

dsEAdS2=Gμνdxμdxν=RAdS2z2(dz2+gij(z,x)dxidxj),z>0,(2)\mathrm{d}s_{\mathrm{EAdS}}^2 = G_{\mu\nu}\mathrm{d}x^\mu\mathrm{d}x^\nu= \frac{R_{\mathrm{AdS}}^2}{z^2}\left(\mathrm{d}z^2 + g_{ij}(z,x)\mathrm{d}x^i\mathrm{d}x^j\right), \qquad z > 0, \tag*{(2)}

with i,j=1,…,di,j=1,\ldots,d. Here gij(z,x)g_{ij}(z,x) denotes the bulk spatial metric that reduces to the boundary metric g^ij(x)\hat{g}_{ij}(x) at z→0z \to0.

The late-time de Sitter wavefunction ΨdS\Psi_{\mathrm{dS}} is related to the EAdS partition function ZAdSZ_{\mathrm{AdS}} via the double analytic continuation

RAdS→iRdS,z→−iη,η<0,(3)R_{\mathrm{AdS}} \to iR_{\mathrm{dS}}, \qquad z \to-i\eta, \qquad\eta< 0, \tag*{(3)}

which maps the EAdS metric to the flat-slicing Lorentzian de Sitter metric:

dsdS2=RdS2η2(−dη2+gij(η,x)dxidxj).(4)\mathrm{d}s_{\mathrm{dS}}^2 = \frac{R_{\mathrm{dS}}^2}{\eta^2}\left(-\mathrm{d}\eta^2 + g_{ij}(\eta,x)\mathrm{d}x^i\mathrm{d}x^j\right). \tag*{(4)}

At the semiclassical level, the Euclidean partition function is given by the exponential of the on-shell action as ZAdS≃exp⁡(−SosEAdS)Z_{\mathrm{AdS}} \simeq\exp\left(-S_{\mathrm{os}}^{\mathrm{EAdS}}\right), and its analytic continuation yields the Lorentzian de Sitter wavefunction

ZAdS≃exp⁡(−SosEAdS[A^])⟶ΨdS[A^]≃exp⁡(iSosdS[A^]),(5)Z_{\mathrm{AdS}} \simeq\exp\left(-S_{\mathrm{os}}^{\mathrm{EAdS}}[\hat{A}]\right) \longrightarrow\Psi_{\mathrm{dS}}[\hat{A}] \simeq\exp\left(iS_{\mathrm{os}}^{\mathrm{dS}}[\hat{A}]\right), \tag*{(5)}

where A^\hat{A} stands for the Dirichlet boundary condition imposed for the field AA. Note that unlike SosEAdSS_{\mathrm{os}}^{\mathrm{EAdS}} which is strictly real, SosdSS_{\mathrm{os}}^{\mathrm{dS}} is complex.

Our main focus is on odd-dimensional de Sitter spaces, corresponding to even-dimensional conformal boundaries. For the theories considered here, the bulk actions carry an odd power of the AdS radius:

Sgrav, SNLSM=RAdSd−1×(⋯ ),SYM=RAdSd−3×(⋯ ),(6)S_{\mathrm{grav}},\,S_{\mathrm{NLSM}} = R_{\mathrm{AdS}}^{d-1} \times(\cdots), \qquad S_{\mathrm{YM}} = R_{\mathrm{AdS}}^{d-3} \times(\cdots), \tag*{(6)}

where both exponents are odd when the bulk dimension (d+1)(d+1) is odd. Under analytic continuation (RAdS)odd→i(±RdS)odd(R_{\mathrm{AdS}})^{\mathrm{odd}} \to i(\pm R_{\mathrm{dS}})^{\mathrm{odd}}, meaning that real terms in the Euclidean on-shell action are in general mapped to phases of the Lorentzian wavefunction.

To compute the late-time probability measure,

log⁡∣ΨdS∣2=−2Im⁡SosdS,(7)\log\lvert\Psi_{\mathrm{dS}}\rvert^2=-2\operatorname{Im}S_{\mathrm{os}}^{\mathrm{dS}}, \tag*{(7)}

we follow similar steps as in [19, 20]. The regulated EAdS on-shell action takes the schematic form

SosEAdS=RAdSp[∑nS(2n)EAdSϵ2n+Slog⁡EAdSlog⁡ϵ+SfinEAdS],p odd,(8)S_{\mathrm{os}}^{\mathrm{EAdS}}=R_{\mathrm{AdS}}^p\left[\sum_n\frac{S_{(2n)}^{\mathrm{EAdS}}}{\epsilon^{2n}}+S_{\log}^{\mathrm{EAdS}}\log\epsilon+S_{\mathrm{fin}}^{\mathrm{EAdS}}\right],\qquad p\ \text{odd}, \tag*{(8)}

where ϵ\epsilon is the radial IR cutoff. Importantly, this structure relies on the analyticity of the bulk on-shell action in zz, as is the case for gravity, Yang–Mills theory, and the NLSM considered here. For a scalar field of generic mass, however, this analyticity is generally lost, and the on-shell action need not admit an expansion of the form above. Focusing on the analytic case described by Eq. (8), we now continue to de Sitter, under which the radial cutoff rotates as ϵ=−iϵη\epsilon=-i\epsilon_\eta (ϵη<0)(\epsilon_\eta<0), yielding

log⁡ϵ⟶log⁡(−ϵη)+iπ2.(9)\log\epsilon\longrightarrow\log(-\epsilon_\eta)+\frac{i\pi}{2}. \tag*{(9)}

Crucially, while power-law divergences and finite real terms only contribute to the phase (i.e., the imaginary part of the exponent) of the wavefunction, the logarithmic term produces a real contribution due to the iπ/2i\pi/2 shift, namely

SosEAdS→iSosdS=(−1)p+12iRdSp[∑n(−1)nS(2n)dSϵη2n+Slog⁡dSlog⁡(−ϵη)+SfindS]+(−1)p+32π2RdSpSlog⁡dS.(10)S_{\mathrm{os}}^{\mathrm{EAdS}}\to iS_{\mathrm{os}}^{\mathrm{dS}}=(-1)^{\frac{p+1}{2}}iR_{\mathrm{dS}}^p\left[\sum_n(-1)^n\frac{S_{(2n)}^{\mathrm{dS}}}{\epsilon_\eta^{2n}}+S_{\log}^{\mathrm{dS}}\log(-\epsilon_\eta)+S_{\mathrm{fin}}^{\mathrm{dS}}\right]+(-1)^{\frac{p+3}{2}}\frac{\pi}{2}R_{\mathrm{dS}}^pS_{\log}^{\mathrm{dS}}. \tag*{(10)}

Consequently, the probability measure is controlled entirely by the coefficient of the logarithmic divergence:

log⁡∣ΨdS∣2=(−1)p+32πRdSpSlog⁡dS.(11)\log\lvert\Psi_{\mathrm{dS}}\rvert^2=(-1)^{\frac{p+3}{2}}\pi R_{\mathrm{dS}}^pS_{\log}^{\mathrm{dS}}. \tag*{(11)}

Thus, the problem of determining the probability measure reduces to computing the regulated EAdS on-shell action for each theory, isolating its logarithmic divergence Slog⁡EAdSS_{\log}^{\mathrm{EAdS}}, and continuing it to de Sitter.

Now, we shall illustrate that the measure for odd-dimensional de Sitter space is local for gravity, Yang–Mills theory, and NLSM. An important distinction between these theories concerns the role of the boundary metric, which is itself dynamical in the gravitational case. For gravity, we retain the general metric g^ij\hat{g}_{ij} at the boundary, whereas for Yang–Mills and NLSM we restrict to a flat boundary for simplicity.6

Einstein gravity

We first recall the general argument applied to 5d Einstein gravity and subsequently state the result of the 7d case. The resulting probability measures, corresponding to 4d and 6d conformal gravity, are summarized in Table 1.

The setup is as follows. We use Dirichlet boundary condition to fix g^\hat{g} as the boundary metric. We also have a bulk Einstein metric obeying the usual d+1d+1-dimensional Einstein manifold condition, Rμν=−dgμνR_{\mu\nu}=-d g_{\mu\nu}. The regularized on-shell action of Eq. (8) for pure gravity takes the form [21]7

−S[g^]=Mpl3RAdS32[∫z≥ϵd5xG(R+12)+2∫d4xK]=Mpl3RAdS32[1ϵ4∫d4xg^ a0+1ϵ2∫d4xg^ a2−log⁡ϵ8∫g^ W^2−Sfin[g^]].(12)\begin{aligned} -S[\hat{g}]=\frac{M_{\mathrm{pl}}^{3}R_{\mathrm{AdS}}^{3}}{2}\left[\int_{z\geq\epsilon}d^{5}x\sqrt{G}(R+12)+2\int d^{4}xK\right] \\ &=\frac{M_{\mathrm{pl}}^{3}R_{\mathrm{AdS}}^{3}}{2}\left[\frac{1}{\epsilon^{4}}\int d^{4}x\sqrt{\hat{g}}\,a_{0}+\frac{1}{\epsilon^{2}}\int d^{4}x\sqrt{\hat{g}}\,a_{2}-\frac{\log\epsilon}{8}\int\sqrt{\hat{g}}\,\widehat{W}^{2}-S_{\mathrm{fin}}[\hat{g}]\right]. \tag*{(12)} \end{aligned}

Upon analytic continuation (Eq. (3)), it is only the logarithmic divergence that contributes to the measure of the theory in de Sitter space, which is why we only state such term explicitly. Substituting the above expansion into Eq. (11) gives the result shown in Table 1.

Yang-Mills theory

We next apply the same framework to Yang-Mills theory. We derive the late-time probability measure explicitly in 5d de Sitter space and summarize its 7d counterpart in Table 1.

The action of Yang-Mills theory on an asymptotically EAdS5_5 reads

SEYM=14g52∫z≥ϵdzz∫d5xG FμνaFaμν,(13)S_{\mathrm{E}}^{\mathrm{YM}}=\frac{1}{4g_{5}^{2}}\int_{z\geq\epsilon}\frac{dz}{z}\int d^{5}x\sqrt{G}\,F_{\mu\nu}^{a}F^{a\mu\nu}, \tag*{(13)}

where Fμνa=∂μAνa−∂νAμa+fabcAμbAνcF_{\mu\nu}^{a}=\partial_{\mu}A_{\nu}^{a}-\partial_{\nu}A_{\mu}^{a}+f^{abc}A_{\mu}^{b}A_{\nu}^{c}. In five dimensions g52g_{5}^{2} has dimensions of length, and hence RAdSg52\frac{R_{\mathrm{AdS}}}{g_{5}^{2}} is dimensionless. We impose radial gauge Az=0A_{z}=0 and Dirichlet boundary conditions

Ai(z,x)→z→0A^i(x).(14)A_{i}(z,x)\xrightarrow{z\to0}\hat{A}_{i}(x). \tag*{(14)}

The residual zz-independent gauge transformations act on A^i\hat{A}_{i} as an ordinary 4d gauge transformation. Denote the field strength of A^i\hat{A}_{i} as

F^ij=∂iA^j−∂jA^i+[A^i,A^j],(15)\widehat{F}_{ij}=\partial_{i}\hat{A}_{j}-\partial_{j}\hat{A}_{i}+[\hat{A}_{i},\hat{A}_{j}], \tag*{(15)}

and restricting to a flat boundary, we use

D^iX=∂^iX+[A^i,X](16)\hat{D}_{i}X=\hat{\partial}_{i}X+[\hat{A}_{i},X] \tag*{(16)}

for the derivative covariant with respect to both the boundary metric g^ij\hat{g}_{ij} and the gauge connection A^i\hat{A}_{i}. In radial gauge, the action can consequently be written as

SEYM=RAdS4g52∫z≥ϵdzz∫d4xg[FijaFa ij+2gij∂zAia∂zAja].(17)S_{\mathrm{E}}^{\mathrm{YM}}=\frac{R_{\mathrm{AdS}}}{4g_{5}^{2}}\int_{z\geq\epsilon}\frac{dz}{z}\int d^{4}x\sqrt{g}\left[F_{ij}^{a}F^{a\,ij}+2g^{ij}\partial_{z}A_{i}^{a}\partial_{z}A_{j}^{a}\right]. \tag*{(17)}

Thus the complete classical on-shell action carries the common dimensionless prefactor RAdSg52\frac{R_{\mathrm{AdS}}}{g_5^2}, which is the Yang-Mills analogue of the gravitational coefficient (MplL)3(M_{\mathrm{pl}}L)^3.

In order to obtain the near-boundary expansion of the gauge field, we perturbatively expand the metric and the gauge field in zz and solve the equation of motion order by order. The equations of motion read

Dμ(GFμν)=0.(18)D_\mu\left(\sqrt{G}F^{\mu\nu}\right)=0. \tag*{(18)}

Imposing the Dirichlet boundary condition we get

Ai(z,x)=A^i(x)+z2[A(2)i(x)−12log⁡(μz)D^jF^ji(x)]+…(19)A_i(z,x)=\widehat{A}_i(x)+z^2\left[A_{(2)i}(x)-\frac{1}{2}\log(\mu z)\widehat{D}^{j}\widehat{F}_{ji}(x)\right]+\ldots \tag*{(19)}

Here, the appearance of the logarithmic term is due to a resonance in the equation of motion. Note that the coefficient of the logarithmic term is purely local in the boundary data. By contrast, A(2)iA_{(2)i} is not completely fixed by the near-boundary analysis. It is fully determined only after imposing an appropriate condition in the interior, such as regularity in EAdS, and therefore becomes in general a nonlocal functional of the boundary source A^i\widehat{A}_i. Using Eq. (19), we see that the second term in Eq. (17) is IR integrable, whereas integrating the first term over the radial direction produces

SE,ϵYM=−RAdS4g52log⁡ϵ∫d4x F^ijaF^a ij+Sfin.(20)S^{\mathrm{YM}}_{E,\epsilon}=-\frac{R_{\mathrm{AdS}}}{4g_5^2}\log\epsilon\int d^4x\,\widehat{F}^{a}_{ij}\widehat{F}^{a\,ij}+S_{\mathrm{fin}}. \tag*{(20)}

We find that the coefficient of the logarithmically divergent part is the 4d Yang-Mills theory on the boundary. Consequently, after analytic continuation, the dS5_5 wavefunction of Yang-Mills theory reads

∣ΨdS[A^]∣YM2=exp⁡[−RdS4g52∫d4x F^ijaF^a ij].(21)\left|\Psi_{\mathrm{dS}}[\widehat{A}]\right|_{\mathrm{YM}}^2=\exp\left[-\frac{R_{\mathrm{dS}}}{4g_5^2}\int d^4x\,\widehat{F}^{a}_{ij}\widehat{F}^{a\,ij}\right]. \tag*{(21)}

Non-Linear Sigma Model

Finally, we turn to applying this framework to the non-linear sigma model. Let ΦA(X)\Phi^A(X) be the coordinates for a map from the bulk spacetime to a target manifold M\mathcal{M} with target metric GAB(Φ)G_{AB}(\Phi). The 5d Euclidean NLSM action on the fixed EAdS background in Eq. (2) reads

SNLSM=RAdS32g52∫z≥ϵdzz3∫d4x g GAB(Φ)(∂zΦA∂zΦB+gij∂iΦA∂jΦB).(22)S_{\mathrm{NLSM}}=\frac{R_{\mathrm{AdS}}^3}{2g_5^2}\int_{z\geq\epsilon}\frac{dz}{z^3}\int d^4x\,\sqrt{g}\,G_{AB}(\Phi)\left(\partial_z\Phi^A\partial_z\Phi^B+g^{ij}\partial_i\Phi^A\partial_j\Phi^B\right). \tag*{(22)}

The equations of motion are

∂z2ΦA+ΓBCA(Φ)∂zΦB∂zΦC+(∂zgg−3z)∂zΦA+Di∂iΦA=0,(23)\partial_z^2\Phi^A+\Gamma^A_{BC}(\Phi)\partial_z\Phi^B\partial_z\Phi^C+\left(\frac{\partial_z\sqrt{g}}{\sqrt{g}}-\frac{3}{z}\right)\partial_z\Phi^A+D^i\partial_i\Phi^A=0, \tag*{(23)}

where ΓBCA(Φ)\Gamma^A_{BC}(\Phi) is the Christoffel symbol with respect to GAB(Φ)G_{AB}(\Phi) and we have introduced the covariant derivative

Di∂iΦA=1g∂i(ggij∂jΦA)+ΓBCA(Φ)gij∂iΦB∂jΦC.(24)D^i\partial_i\Phi^A=\frac{1}{\sqrt{g}}\partial_i\left(\sqrt{g}g^{ij}\partial_j\Phi^A\right)+\Gamma^A_{BC}(\Phi)g^{ij}\partial_i\Phi^B\partial_j\Phi^C. \tag*{(24)}

To solve them, we expand ΦA\Phi^{A} near the boundary, ΦA≃ϕA(x)+z2ϕ(2)A+⋯\Phi^{A} \simeq\phi^{A}(x)+z^{2}\phi^{A}_{(2)}+\cdots, along with the target space metric, and choose Dirichlet boundary condition for the field ΦA\Phi^{A}. Restricting to a flat boundary, the asymptotic solution is

ΦA≃ϕA+z24Di∂iϕA+O(z4log⁡z),(25)\Phi^{A} \simeq\phi^{A}+\frac{z^{2}}{4}D^{i}\partial_{i}\phi^{A}+O(z^{4}\log z), \tag*{(25)}

where the Laplacian is now with respect to flat metric and the logarithmic term is again due to resonances in the equations of motion. Because of the radial powers in Eq. (22), using Eq. (25) we can expand the kinetic terms as

∂zΦA∂zΦB+gij∂iΦA∂jΦB≃∂iϕA∂iϕB+z24(Di∂iϕA)(Dj∂jϕB)+z24∂i(Dj∂jϕA)∂iϕB+z24∂iϕA∂i(Dj∂jϕB)+O(z4log⁡z).(26)\begin{aligned} \partial_{z}\Phi^{A}\partial_{z}\Phi^{B}+g^{ij}\partial_{i}\Phi^{A}\partial_{j}\Phi^{B} &\simeq\partial_{i}\phi^{A}\partial^{i}\phi^{B}+\frac{z^{2}}{4}(D^{i}\partial_{i}\phi^{A})(D^{j}\partial_{j}\phi^{B}) \\ &\quad+\frac{z^{2}}{4}\partial_{i}(D^{j}\partial_{j}\phi^{A})\partial^{i}\phi^{B} +\frac{z^{2}}{4}\partial_{i}\phi^{A}\partial^{i}(D^{j}\partial_{j}\phi^{B}) \\ &\quad+O(z^{4}\log z). \tag*{(26)} \end{aligned}

Plugging back in Eq. (22), using near boundary expansion of the target space metric, integration by parts, and integrating out the radial direction, we get the on-shell action

SNLSM=RAdS3g52∫d4x GAB(14ϵ2∂iϕA∂iϕB+log⁡ϵ8(Di∂iϕA)(Dj∂jϕB)+O(ϵ2)).(27)S_{\mathrm{NLSM}}=\frac{R_{\mathrm{AdS}}^{3}}{g_{5}^{2}}\int d^{4}x\,G_{AB}\left(\frac{1}{4\epsilon^{2}}\partial_{i}\phi^{A}\partial^{i}\phi^{B}+\frac{\log\epsilon}{8}(D^{i}\partial_{i}\phi^{A})(D^{j}\partial_{j}\phi^{B})+O(\epsilon^{2})\right). \tag*{(27)}

Thus, the logarithmic divergence comes from the z2z^{2} field correction in Eq. (25). Now we can analytically continue the partition function. Once again, the real part of the wavefunction is generated by the phase of the logarithm as explained in Section 2 and reads

log⁡∣ΨdS∣2=−πRdS38g52∫d4x GAB(Di∂iϕA)(Dj∂jϕB).(28)\log\lvert\Psi_{\mathrm{dS}}\rvert^{2}=-\frac{\pi R_{\mathrm{dS}}^{3}}{8g_{5}^{2}}\int d^{4}x\,G_{AB}(D^{i}\partial_{i}\phi^{A})(D^{j}\partial_{j}\phi^{B}). \tag*{(28)}

We further include the contributions from a 3d and 7d bulk in Table 1.

Lorentzian Checks and Cosmological CPT

To illustrate these mechanisms purely on the dS side, we begin with an independent, explicit check of the local Yang–Mills probability measure in dS5_{5} using the Lorentzian wavefunction formalism. Specifically, we evaluate the two- and three-point wavefunction coefficients directly in the flat slicing,

ds2=RdS2η2(−dη2+dx2),−∞<η<0,(29)ds^{2}=\frac{R_{\mathrm{dS}}^{2}}{\eta^{2}}\left(-d\eta^{2}+dx^{2}\right),\qquad-\infty<\eta<0, \tag*{(29)}

working in temporal gauge Aηa=0A^{a}_{\eta}=0. The probability measure is determined by the imaginary part of the on-shell action, Eq. (11), which we compute order-by-order in the boundary field A^\hat{A}. We further present a partial explanation of the results using the cosmological CPT theorem [18].

Quadratic Yang-Mills Coefficient

The quadratic part of the action for the transverse spatial components (kiAia=0k_iA^a_i=0) reads

SYM⊃RdS2g52∫dη−η d4x[Aia′Aia′−12fijafa ij],(30)S_{\mathrm{YM}}\supset\frac{R_{\mathrm{dS}}}{2g_5^2}\int\frac{d\eta}{-\eta}\,d^4x\left[A_i^{a\prime}A_i^{a\prime}-\frac{1}{2}f^a_{ij}f^{a\,ij}\right], \tag*{(30)}

where fij=∂iAj−∂jAif_{ij}=\partial_iA_j-\partial_jA_i. The linearized equation of motion, Aia′′−1ηAia′+k2Aia=0A_i^{a\prime\prime}-\frac{1}{\eta}A_i^{a\prime}+k^2A_i^a=0, is solved by the Bunch-Davies bulk-to-boundary propagator Aia(η,k)=Kk(η)A^ia(k)A_i^a(\eta,\mathbf{k})=K_k(\eta)\hat{A}_i^a(\mathbf{k}), where

Kk(η)=−iπ2(−kη)H1(2)(−kη),(31)K_k(\eta)=-\frac{i\pi}{2}(-k\eta)H_1^{(2)}(-k\eta), \tag*{(31)}

where H1(2)H_1^{(2)} is the Hankel function of second kind of order 1. Using the small-argument expansion of the Hankel function,

Kk(η)=1−k2η22[log⁡(−kη2)+γE−12+iπ2]+O(η4log⁡η),(32)K_k(\eta)=1-\frac{k^2\eta^2}{2}\left[\log\left(\frac{-k\eta}{2}\right)+\gamma_E-\frac{1}{2}+\frac{i\pi}{2}\right]+O(\eta^4\log\eta), \tag*{(32)}

the quadratic part of the on-shell action reduces to the boundary term at late times (η∗→0−\eta_* \to0^-):

Sos⊃−RdS2g52∫d4k(2π)4Aia(η∗,k)Aia(η∗,−k)=RdS2g52∫d4k(2π)4k2A^ia(k)A^ia(−k)[log⁡(−kη∗2)+γE+iπ2].(33)S_{\rm os} \supset-\frac{R_{\rm dS}}{2g_5^2}\int\frac{d^4k}{(2\pi)^4}A_i^a(\eta_*,\mathbf{k})A_i^a(\eta_*,-\mathbf{k}) = \frac{R_{\rm dS}}{2g_5^2}\int\frac{d^4k}{(2\pi)^4}k^2\hat A_i^a(\mathbf{k})\hat A_i^a(-\mathbf{k})\left[\log\left(\frac{-k\eta_*}{2}\right)+\gamma_E+\frac{i\pi}{2}\right]. \tag*{(33)}

Extracting the imaginary part yields a purely local contribution:

Im⁡Sos⊃πRdS4g52∫d4k(2π)4k2A^ia(k)A^ia(−k)=πRdS8g52∫d4x f^ijaf^ija.(34)\operatorname{Im}S_{\rm os}\supset\frac{\pi R_{\rm dS}}{4g_5^2}\int\frac{d^4k}{(2\pi)^4}k^2\hat A_i^a(\mathbf{k})\hat A_i^a(-\mathbf{k})=\frac{\pi R_{\rm dS}}{8g_5^2}\int d^4x\,\hat f_{ij}^a\hat f_{ij}^a. \tag*{(34)}

reproducing the expected local quadratic measure of Eq. (21).

Cubic Yang-Mills Coefficient

At cubic order, the interaction is governed by

SYM⊃−RdS2g52∫dη(−η) d4x fabcfijaAibAjc.(35)S_{\rm YM}\supset-\frac{R_{\rm dS}}{2g_5^2}\int\frac{d\eta}{(-\eta)}\,d^4x\,f^{abc}f_{ij}^aA_i^bA_j^c. \tag*{(35)}

Evaluating this on the free solutions gives the cubic part of the on-shell action

Sos⊃−RdS2g52∫d4k1(2π)4d4k2(2π)4d4k3(2π)4(2π)4δ(4)(∑a=13ka)V3I3(k1,k2,k3),(36)S_{\rm os}\supset-\frac{R_{\rm dS}}{2g_5^2}\int\frac{d^4k_1}{(2\pi)^4}\frac{d^4k_2}{(2\pi)^4}\frac{d^4k_3}{(2\pi)^4}(2\pi)^4\delta^{(4)}\left(\sum_{a=1}^{3}\mathbf{k}_a\right)\mathcal{V}_3\mathcal{I}_3(k_1,k_2,k_3), \tag*{(36)}

where V3\mathcal{V}_3 is the momentum-space cubic tensor

V3=ifabc(k1iA^ja(k1)−k1jA^ia(k1))A^ib(k2)A^jc(k3),(37)\mathcal{V}_3=if^{abc}\left(k_{1i}\hat A_j^a(\mathbf{k}_1)-k_{1j}\hat A_i^a(\mathbf{k}_1)\right)\hat A_i^b(\mathbf{k}_2)\hat A_j^c(\mathbf{k}_3), \tag*{(37)}

and the time-dependent integral is defined over the Bunch-Davies contour (t=−η>0t=-\eta>0)

I3(k1,k2,k3)=∫0∞dtt∏a=13[−iπ2katH1(2)(kat)].(38)\mathcal{I}_3(k_1,k_2,k_3)=\int_0^\infty\frac{dt}{t}\prod_{a=1}^{3}\left[-\frac{i\pi}{2}k_atH_1^{(2)}(k_at)\right]. \tag*{(38)}

To compute the imaginary part without getting embroiled in fully nonlocal finite expressions, we map I3\mathcal{I}_3 to the Euclidean triple-K integral [22, 23] using the identity −iπxH1(2)(x)/2=ixK1(ix)-i\pi xH_1^{(2)}(x)/2=ixK_1(ix). This yields the analytic continuation of the triple-K integral in its momentum arguments, namely

I3(k1,k2,k3)=I2{1,1,1}(ik1,ik2,ik3).(39)\mathcal{I}_3(k_1,k_2,k_3)=I_{2\{1,1,1\}}(ik_1,ik_2,ik_3). \tag*{(39)}

The integral I2{1,1,1}I_{2\{1,1,1\}} possesses a logarithmic divergence at its lower endpoint (z→0z\to0), which we regulate by shifting the Bessel indices to 1+ε1+\varepsilon. Then the divergent part of the regulated integral exhibits a simple pole [23]:

I2{1+ε,1+ε,1+ε}div(ka)=−13ε+Freal(ka)+O(ε).(40)I_{2\{1+\varepsilon,1+\varepsilon,1+\varepsilon\}}^{\rm div}(k_a)=-\frac{1}{3\varepsilon}+F_{\rm real}(k_a)+O(\varepsilon). \tag*{(40)}

The crucial mechanism for generating the imaginary part lies in the homogeneity property of the triple-K integral under scaling, that is

I2{1+ε,1+ε,1+ε}(λka)=λ3εI2{1+ε,1+ε,1+ε}(ka).(41)I_{2\{1+\varepsilon,1+\varepsilon,1+\varepsilon\}}(\lambda k_a)=\lambda^{3\varepsilon}I_{2\{1+\varepsilon,1+\varepsilon,1+\varepsilon\}}(k_a). \tag*{(41)}

For the Lorentzian Bunch-Davies contour, we analytically continue the momenta by setting λ=i\lambda= i, giving

I2{1+ε,1+ε,1+ε}(ika)=e3iπε/2[−13ε+Freal(ka)+O(ε)].(42)I_{2\{1+\varepsilon,1+\varepsilon,1+\varepsilon\}}(ika)=e^{3i\pi\varepsilon/2}\left[-\frac{1}{3\varepsilon}+F_{\mathrm{real}}(ka)+\mathcal{O}(\varepsilon)\right]. \tag*{(42)}

Expanding the exponential factor for small ε\varepsilon, the cross-term between the phase expansion and the simple pole produces a finite, universal imaginary contribution:

(3iπ2ε)(−13ε)=−iπ2.(43)\left(\frac{3i\pi}{2}\varepsilon\right)\left(-\frac{1}{3\varepsilon}\right)=-\frac{i\pi}{2}. \tag*{(43)}

Thus, the imaginary part of the integral originates exclusively from the log divergence pole:

Im⁡I3(k1,k2,k3)=−π2.(44)\operatorname{Im}\mathcal{I}_3(k_1,k_2,k_3)=-\frac{\pi}{2}. \tag*{(44)}

Because the pole is strictly independent of external momenta kak_a, all non-trivial momentum-dependent pieces are relegated to the real part (the phase of the wavefunction), while the imaginary part remains entirely local. Substituting this back into the cubic on-shell action yields the exact local measure contribution

Im⁡Sos⊃πRdS4g52∫d4x fabcf^ijaA^ibA^jc,(45)\operatorname{Im} S_{\mathrm{os}}\supset\frac{\pi R_{\mathrm{dS}}}{4g_5^2}\int d^4x\,f^{abc}\hat f_{ij}^{a}\hat A_i^b\hat A_j^c, \tag*{(45)}

matching the prediction from the EAdS analytic continuation through cubic order in Eq. (21).

CPT Constraints and Locality

The origin of the local measure in low-point Yang-Mills theory is related to the cosmological CPT theorem [18]. Specifically, below we demonstrate that the wavefunction coefficients of Yang-Mills theory in dSd+1\mathrm{dS}_{d+1} do not contribute to the probability measure unless they exhibit logarithmic divergences. The same result goes through for NLSM and GR. This provides a purely de Sitter-based derivation of the phenomenon.

The wavefunction coefficients are defined through the perturbative expansion of the logarithm of the late-time wavefunction,

log⁡Ψ[A^]=∑L=0∞∑n=2∞Wn(L)[A^]=∑L=0∞∑n=2∞1n!∫[∏a=1nddka(2π)dA^(ka)](2π)dδ(d)(∑a=1nka)Tn(ka)ψ^n(L)(k1,…,kn),(46)\begin{aligned} \log\Psi[\hat A]=\sum_{L=0}^{\infty}\sum_{n=2}^{\infty}\mathcal{W}_{n}^{(L)}[\hat A] \\ &=\sum_{L=0}^{\infty}\sum_{n=2}^{\infty}\frac{1}{n!}\int\left[\prod_{a=1}^{n}\frac{d^{d}k_a}{(2\pi)^d}\hat A(k_a)\right](2\pi)^d\delta^{(d)}\left(\sum_{a=1}^{n}k_a\right)T_n(k_a)\hat\psi_n^{(L)}(k_1,\ldots,k_n), \tag*{(46)} \end{aligned}

where LL denotes the loop order, Tn(ka)T_n(k_a) is the momentum-space tensor structure, and field indices and their contractions are left implicit. The cosmological CPT theorem [18] dictates that finite wavefunction coefficients satisfy the reality-like condition

[ψ^n(L)]∗=eiπ((d+1)(L−1)+dn−∑a=1nΔa)ψ^n(L).(47)\left[\hat\psi_n^{(L)}\right]^*=e^{i\pi\left((d+1)(L-1)+dn-\sum_{a=1}^{n}\Delta_a\right)}\hat\psi_n^{(L)}. \tag*{(47)}

where Δa\Delta_a is the conformal dimension of the corresponding boundary operator. For a massless gauge field, the conformal dimension of the boundary operator is Δ^A=d−1\widehat{\Delta}_A = d - 1. Therefore, at tree level (L=0L = 0), it reduces to

[ψ^n(0)]∗=eiπ(−(d+1)+dn−n(d−1))ψ^n(0)=(−1)(−d−1+n)ψ^n(0).(48)\left[\widehat{\psi}_n^{(0)}\right]^* = e^{i\pi\left(-(d+1)+dn-n(d-1)\right)}\widehat{\psi}_n^{(0)} = (-1)^{(-d-1+n)}\widehat{\psi}_n^{(0)}. \tag*{(48)}

Note that this relation alone does not imply that the corresponding wavefunction terms are purely imaginary, as one must account for the parity-induced sign changes of the momentum-space tensor structures. To track these signs, consider the Yang-Mills action in temporal gauge, Aη=0A_\eta= 0. Pure Yang-Mills theory features two types of interaction vertices: the cubic vertex, which contains a single spatial derivative,

∫fabc(∂iAja−∂jAia)AibAjc,(49)\int f^{abc}(\partial_i A_j^a-\partial_j A_i^a)A_i^b A_j^c, \tag*{(49)}

and the quartic vertex, which contains no spatial derivatives,

∫fabcfadeAibAjcAidAje.(50)\int f^{abc}f^{ade}A_i^b A_j^c A_i^d A_j^e. \tag*{(50)}

Under simultaneous spatial parity inversion, ka→−ka\mathbf{k}_a \to-\mathbf{k}_a, every cubic vertex contributes a factor of −1-1, whereas the quartic vertex does not. The propagators and transverse projectors are even under this transformation.

For a connected tree diagram with V3V_3 cubic vertices, V4V_4 quartic vertices, II internal lines, and nn external legs, the tree topology implies

I=V3+V4−1.(51)I = V_3 + V_4 - 1. \tag*{(51)}

Counting the endpoints of all propagators yields

3V3+4V4=2I+n.(52)3V_3 + 4V_4 = 2I + n. \tag*{(52)}

Substituting II in the relation above gives the topological constraint

n=V3+2V4+2,(53)n = V_3 + 2V_4 + 2, \tag*{(53)}

which implies V3≡n(mod2)V_3 \equiv n \pmod{2}.

For a real gauge field in position space, the Fourier components satisfy A^ia(k)∗=A^ia(−k)\widehat{A}_i^a(\mathbf{k})^* = \widehat{A}_i^a(-\mathbf{k}). Taking the complex conjugate of Wn\mathcal{W}_n and changing integration variables via ka→−ka\mathbf{k}_a \to-\mathbf{k}_a leads to

Wn∗=(−1)n∫[∏a=1nddka(2π)dA^(ka)](2π)dδ(d)(∑a=1nka)Tn(ka)[ψ^n(0)]∗(k1,…,kn).(54)\mathcal{W}_n^* = (-1)^n \int\left[\prod_{a=1}^{n}\frac{\mathrm{d}^d k_a}{(2\pi)^d}\widehat{A}(\mathbf{k}_a)\right](2\pi)^d\delta^{(d)}\left(\sum_{a=1}^{n}\mathbf{k}_a\right)T_n(\mathbf{k}_a)\left[\widehat{\psi}_n^{(0)}\right]^*(\mathbf{k}_1,\ldots,\mathbf{k}_n). \tag*{(54)}

Applying the CPT constraint ψ^n∗=(−1)n−d−1ψ^n\widehat{\psi}_n^* = (-1)^{n-d-1}\widehat{\psi}_n then yields

Wn∗=(−1)n(−1)n−d−1Wn=(−1)−d−1Wn,(55)\mathcal{W}_n^* = (-1)^n(-1)^{n-d-1}\mathcal{W}_n = (-1)^{-d-1}\mathcal{W}_n, \tag*{(55)}

indicating that finite tree-level contributions are purely imaginary for odd $d+1.eriwa.

The non-triviality of the dSd+1_{d+1} probability measure arises from diagrams containing late-time logarithmic divergences, for which the simple finite-coefficient phase formula no longer applies. Consider a wavefunction coefficient with the following late-time asymptotic expansion

ψn(−η;ka)=Cn(ka)log⁡(−μη)+Bn(ka),(56)\psi_n(-\eta;k_a)=C_n(k_a)\log(-\mu\eta)+B_n(k_a), \tag*{(56)}

where CnC_n is the coefficient of the logarithmic divergence and BnB_n denotes its finite counterpart. The cosmological CPT theorem implies that

Cn∗=(−1)n+d+1Cn,(57)C_n^*=(-1)^{n+d+1}C_n, \tag*{(57)}

whereas the finite piece transforms as

Bn∗=(−1)n+d+1Bn−iπ(−1)n+d+1Cn.(58)B_n^*=(-1)^{n+d+1}B_n-i\pi(-1)^{n+d+1}C_n. \tag*{(58)}

Accounting for the parity of the momentum structures, it follows that the non-vanishing contributions to the probability measure stem solely from the logarithmically divergent parts of the wavefunction coefficients.

We can now relate the preceding argument to the explicit two- and three-point wavefunction calculations presented above. In both cases, the non-vanishing contributions to the probability measure are accompanied by logarithmic late-time divergences. Indeed, the quadratic on-shell action contains a term proportional to log⁡(−η∗)∫f^2\log(-\eta_*)\int\hat{f}^2, while the cubic time integral I3\mathcal{I}_3 develops a logarithmic divergence as η∗→0−\eta_* \to0^-. These observations are consistent with the cosmological CPT argument outlined above, according to which logarithmically divergent wavefunction coefficients evade the homogeneous reality constraint and can consequently give rise to nontrivial contributions to the probability measure.

To establish that the higher-point wavefunction coefficients do not contribute to the measure, however, the CPT argument alone is insufficient. One must additionally show that the coefficients with n≥5n \ge5 do not develop late-time logarithmic divergences, so that the ordinary CPT reality constraint applies without the logarithmic exception. Finiteness of these coefficients can be deduced from the asymptotic solution of the equation of motion that is the Lorentzian analogue of Eq. (19).

We have therefore obtained a purely Lorentzian explanation of the structure of the tree-level Yang-Mills measure in 5d dS: the CPT reality constraint eliminates the finite nonlocal contributions, while the only surviving contributions arise from logarithmic divergences whose coefficients are local functionals of the boundary gauge field.

Concluding Thoughts and Future Directions

In this paper, we have shown that a broad class of theories of interest in odd-dimensional de Sitter space have local and remarkably simple measures. More precisely, for Einstein gravity, Yang-Mills theory, and NLSM, nonlocal parts of the wavefunction reside entirely in its phase at the semiclassical level. This simplification follows from the logarithmic divergences of the corresponding Euclidean AdS on-shell actions, which, upon analytic continuation, entirely determine the de Sitter probability measure. We emphasize that the oddness of the bulk dimension is essential to our result. Indeed, massless fields in dS4\mathrm{dS}_4 exhibit nonlocal probability kernels already at quadratic order. It would also be interesting to understand this dimensional dependence directly from the Hankel expansions of bulk propagators, potentially identifying which higher-point interactions generate logarithmic divergences.

The cosmological CPT theorem offers a complementary perspective on the same mechanism. Suitably applied, through its reality-like conditions, it implies that finite tree-level wavefunction coefficients in the theories considered here contribute only to the phase of the wavefunction in odd-dimensional de Sitter space. Nevertheless, logarithmically divergent coefficients evade the corresponding reality condition. The main limitation of this approach is that it does not determine which coefficients develop such divergences, requiring them to be examined individually. In the EAdS derivation, this information is encoded in the local structure of Slog⁡EAdSS_{\log}^{\mathrm{EAdS}}. It is worth noting that the coefficient Slog⁡EAdSS_{\log}^{\mathrm{EAdS}} is closely related to the holographic conformal anomaly. Our result therefore relates the computation of late-time correlation functions in de Sitter space to that of correlation functions in certain classically conformal local Euclidean field theories. This discussion should, however, be understood within its perturbative regime of validity. In particular, more general gravitational saddles need not admit a real EAdS continuation, in which case the finite part of the continued on-shell action can acquire an imaginary component and give additional, potentially nonlocal, contributions to the probability measure [24, 25].

At the same time, the locality of the probability measure should not be confused with the locality of late-time correlation functions. Equal-time expectation values are obtained by integrating field insertions against ∣Ψ∣2|\Psi|^2,

⟨Φ^1⋯Φ^n⟩=∫DΦ^ Φ^1⋯Φ^n∣Ψ[Φ^]∣2∫DΦ^ ∣Ψ[Φ^]∣2.(59)\left\langle\widehat{\Phi}_1 \cdots\widehat{\Phi}_n \right\rangle= \frac{\int\mathcal{D}\widehat{\Phi}\,\widehat{\Phi}_1 \cdots\widehat{\Phi}_n |\Psi[\widehat{\Phi}]|^2} {\int\mathcal{D}\widehat{\Phi}\,|\Psi[\widehat{\Phi}]|^2}. \tag*{(59)}

Even for a local probability action, the propagator is typically nonlocal. Consequently, exchange diagrams built from local probability vertices give rise to nonlocal connected correlators.

The dS5\mathrm{dS}_5 Yang-Mills example illustrates an additional simplification. Since the logarithm of its probability measure is proportional to ∫F^2\int\widehat{F}^2, the elementary probability vertices terminate at quartic order. Yet, connected correlators of arbitrarily high multiplicity are generally nonzero and can be constructed recursively by joining the cubic and quartic probability vertices with propagators.

It is also natural to ask whether this simplicity survives the inclusion of additional interactions. The EAdS argument suggests that locality is not restricted to the two-derivative theories studied here. One can consider higher-curvature gravitational interactions, higher powers of the Yang-Mills field strength, shift-symmetric scalar interactions, and higher-derivative interactions of general pp-form gauge fields. The argument continues to hold, generically because contractions with the bulk inverse metric yield a factor of REAdSd+1−2mR_{\mathrm{EAdS}}^{d+1-2m}, where mm counts the inverse metrics. In odd bulk dimensions, this remains an odd power of the radius, so the same analytic continuation applies. Hence, provided the near-boundary expansions of the fields remain analytic in the radial direction zz, higher-derivative interactions may modify the local functional determining the probability measure without spoiling its locality.

Another possible generalization concerns semi-local measures, whose wavefunction coefficients have support only when subsets of the boundary insertion points coincide. For instance, a three-point kernel may contain a term of the form

K3(x1,x2,x3)⊃δ(d)(x1−x2) F(x1,x3),(60)\mathcal{K}_{3}(x_{1},x_{2},x_{3}) \supset\delta^{(d)}(x_{1}-x_{2})\,\mathcal{F}(x_{1},x_{3}), \tag*{(60)}

where F\mathcal{F} is generically nonlocal. Interacting scalar theories may provide examples of this structure. A generic ϕn\phi^{n} potential can modify the leading asymptotic behavior of a massless scalar and invalidate the standard expansion around arbitrary finite Dirichlet boundary conditions. It would be interesting to investigate semi-local contributions from the perspective of cosmological CPT, particularly whether it constrains their possible forms and enables a broader classification of late-time probability measures.

The case of three-dimensional de Sitter space deserves separate considerations. Pure gravity has no local propagating degrees of freedom, and its 2d2d logarithmic functional is topological on a closed boundary. The leading order dynamics of the gauge fields is controlled by Chern-Simons theory, which again is topological and produces a wavefunction which is a pure phase.8 Although the topological nature of gauge fields and gravity makes the three-dimensional case particularly simple, nontrivial results can still arise in scalar theories, as illustrated by the NLSM.

Our results further ask for a comparison with recent Grassmannian formulations of four-dimensional CFT correlators [27]. Since the five-dimensional measures studied here are obtained by analytic continuation from EAdS5\mathrm{EAdS}_{5}, these constructions may provide an independent way of examining the locality of the probability measure. In particular, it would be useful to understand how the logarithmic contributions associated with holographic anomalies are encoded in the Grassmannian framework.

An important open question is how locality is modified beyond tree level. Loops can introduce anomalous dimensions, additional logarithms, and nonlocal contributions that cosmological CPT alone does not exclude. More generally, understanding how far the simplification we have found extends, and whether it admits a more direct explanation in terms of the underlying structure of cosmological wavefunction coefficients, remains an open problem.

Acknowledgments

We are grateful to Carlos Duaso Pueyo, Emine Şeyma Kutluk, Manuel Loparco, Juan Maldacena, Charlotte Sleight, and Massimo Taroni for helpful comments and discussions. GLP, NR and BT are supported by the ERC (NOTIMEFORCOSMO, 101126304), by Scuola Normale, and by INFN (IS GSS-Pi). GLP is also supported by the Italian Ministry of Universities and Research (MUR) under contract 20223ANFHR (PRIN2022). NR is further supported by the European Union’s Horizon Europe research and innovation programme under the Marie Skłodowska-Curie Actions – Global Postdoctoral Fellowship (ACROSS, 101267248).

References

  1. [1]D. Marolf, I.A. Morrison and M. Srednicki, Perturbative S-matrix for massive scalar fields in global de Sitter space, Class. Quant. Grav. 30 (2013) 155023 [1209.6039].DOI
  2. [2]S. Melville and G.L. Pimentel, de Sitter S matrix for the masses, Phys. Rev. D 110 (2024) 103530 [2309.07092].arxiv.org/abs/2309.07092
  3. [3]S. Melville and G.L. Pimentel, A de Sitter S-matrix from amputated cosmological correlators, JHEP 08 (2024) 211 [2404.05712].arxiv.org/abs/2404.05712
  4. [4]T.R. Taylor and B. Zhu, Three-gluon scattering amplitude in de Sitter spacetime, JHEP 10 (2026) 008 [2604.24844].arxiv.org/abs/2604.24844
  5. [5]P. Benincasa, Amplitudes meet Cosmology: A (Scalar) Primer, 2203.15330.DOI
  6. [6]J.M. Maldacena, Non-Gaussian features of primordial fluctuations in single field inflationary models, JHEP 05 (2003) 013 [astro-ph/0210063].DOI
  7. [7]E. Skvortsov and R. Van Dongen, Dirichlet, Neumann, Mixed and self-dual holography: (self-dual) Yang–Mills theory II, 2606.26302.arxiv.org/abs/2606.26302
  8. [8]S. Heuveline, R. Ruzziconi, A. Sheta and A. Strominger, Self-Dual Yang-Mills in AdS₄, 2608.27731.DOI
  9. [9]C. Chowdhury and S. Nagy, All Order Self-Dual Correlators From Symmetries, 2610.32956.arxiv.org/abs/2610.32956
  10. [10]H. Gomez, R. Lipinski Jusinskas, A. Lipstein and C. Lopez-Arcos, All-multiplicity AdS₄ correlators in the self-dual sector, 2610.32957.arxiv.org/abs/2610.32957
  11. [11]S.J. Parke and T.R. Taylor, An Amplitude for n Gluon Scattering, Phys. Rev. Lett. 56 (1986) 2459.pubmed.ncbi.nlm.nih.gov/10032998
  12. [12]R. Monteiro and D. O’Connell, The Kinematic Algebra From the Self-Dual Sector, JHEP 07 (2011) 007 [1105.2565].DOI
  13. [13]A. Guevara, A. Lupsasca, D. Skinner, A. Strominger and K. Weil, Single-minus gluon tree amplitudes are nonzero, 2602.12176.arxiv.org/abs/2602.12176
  14. [14]A. Guevara, A. Lupsasca, D. Skinner, A. Strominger and K. Weil, Single-minus graviton tree amplitudes are nonzero, 2603.04330.arxiv.org/abs/2603.04330
  15. [15]Z. Bern, L.J. Dixon, D.C. Dunbar and D.A. Kosower, One loop n point gauge theory amplitudes, unitarity and collinear limits, Nucl. Phys. B 425 (1994) 217 [hep-ph/9403226].DOI
  16. [16]J. Maldacena, Einstein Gravity from Conformal Gravity, 1105.5632.
  17. [17]L. Susskind and E. Witten, The Holographic bound in anti-de Sitter space, hep-th/9805114.DOI
  18. [18]H. Goodhew, A. Thavanesan and A.C. Wall, The cosmological CPT theorem, SciPost Phys. 20 (2026) 049 [2408.17406].arxiv.org/abs/2408.17406
  19. [19]S. de Haro, S.N. Solodukhin and K. Skenderis, Holographic reconstruction of space-time and renormalization in the AdS / CFT correspondence, Commun. Math. Phys. 217 (2001) 595 [hep-th/0002230].
  20. [20]K. Skenderis, Lecture notes on holographic renormalization, Class. Quant. Grav. 19 (2002) 5849 [hep-th/0209067].DOI
  21. [21]M. Henningson and K. Skenderis, The Holographic Weyl anomaly, JHEP 07 (1998) 023 [hep-th/9806087].DOI
  22. [22]A. Bzowski, P. McFadden and K. Skenderis, Implications of conformal invariance in momentum space, JHEP 03 (2014) 111 [1304.7760].
  23. [23]A. Bzowski, P. McFadden and K. Skenderis, Renormalised CFT 3-point functions of scalars, currents and stress tensors, JHEP 11 (2018) 159 [1805.12100].
  24. [24]J. Cotler, V. Ivo and J. Maldacena, Logarithmic gravity from a very slow roll limit of inflation, 2609.38052.arxiv.org/abs/2609.38052
  25. [25]J. Cotler, P. Dhivakar, W. Harvey, V. Ivo and K. Jensen, Runaways in the no-boundary wavefunction, 2610.23743.arxiv.org/abs/2610.23743
  26. [26]G. Mathys, G.L. Pimentel and F. Rost, AdS₃ Recursion Relations, Double Copy and CFT₂ Ward Identities, 2609.12046.
  27. [27]A. Bala, S. Jain and D.K. S, The Conformal Grassmannian: A Symplectic Bi-Grassmannian for CFT₄ Correlators, 2605.06811.arxiv.org/abs/2605.06811

Paper details

vlnc:W_atrsjbg5
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 2

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents