Inflation with a constant rate of roll
aa r X i v : . [ a s t r o - ph . C O ] A p r RESCEU-51/14
Inflation with a constant rate of roll
Hayato Motohashi , Alexei A. Starobinsky , , and Jun’ichi Yokoyama , , Kavli Institute for Cosmological Physics, The University of Chicago, Chicago, Illinois 60637, U.S.A. Research Center for the Early Universe (RESCEU),Graduate School of Science, The University of Tokyo, Tokyo 113-0033, Japan L. D. Landau Institute for Theoretical Physics RAS, Moscow 119334, Russia Department of Physics, Graduate School of Science,The University of Tokyo, Tokyo 113-0033, Japan Kavli Institute for the Physics and Mathematics of the Universe (Kavli IPMU),WPI, UTIAS, The University of Tokyo, Kashiwa, Chiba 277-8568, Japan
We consider an inflationary scenario where the rate of inflaton roll defined by ¨ φ/H ˙ φ remainsconstant. The rate of roll is small for slow-roll inflation, while a generic rate of roll leads to theinteresting case of ‘constant-roll’ inflation. We find a general exact solution for the inflaton poten-tial required for such inflaton behaviour. In this model, due to non-slow evolution of background,the would-be decaying mode of linear scalar (curvature) perturbations may not be neglected. Itcan even grow for some values of the model parameter, while the other mode always remains con-stant. However, this always occurs for unstable solutions which are not attractors for the givenpotential. The most interesting particular cases of constant-roll inflation remaining viable with themost recent observational data are quadratic hilltop inflation (with cutoff) and natural inflation(with an additional negative cosmological constant). In these cases even-order slow-roll parametersapproach non-negligible constants while the odd ones are asymptotically vanishing in the quasi-deSitter regime. I. INTRODUCTION
The inflationary paradigm based on the assumption of the existence of a quasi-de Sitter stage in the early Universebefore the hot radiation dominated Big Bang [1–5] has now become a well established part of modern cosmology.It makes definite predictions about present properties of the Universe which have been confirmed by numerousobservations. The most common way to realize this quasi-de Sitter expansion is to employ a scalar field (inflaton)with an approximately flat potential and consider a slow-roll solution of the field equations. It is also possible torealize inflationary expansion without introducing such an inflaton field by upgrading the gravitational action from theEinstein-Hilbert one to a nontrivial function of the Ricci scalar, f ( R ). However, using the conformal transformation,equations for an f ( R ) type inflation can be transformed to the Einstein-Hilbert action with a canonical scalar fielddubbed scalaron with its potential defined by the form of f ( R ). In this approach, the R inflation [1], including itsgeneralization to accommodate present dark energy [6–8] or to consider a small deviation from it [9–11], can generateprimordial scalar and tensor metric fluctuations of the same form as scalar field inflation.Given that slow-roll inflationary models with an approximately flat inflaton potential yield a simple realization ofinflation with viable observational predictions, it is natural to ask what happens if we omit the assumption of theinflaton slow roll. The slow-roll approximation assumes that in the Klein-Gordon equation for the inflaton given by¨ φ + 3 H ˙ φ + ∂V∂φ = 0 , (1)the second derivative term ¨ φ is negligible compared to other terms. The necessity to go beyond this approximationand to use some exact solutions instead has been already occurred in a number of cases, in particular, when theinflaton potential V ( φ ) has some local non-analytic feature [12] or is near its local extremum [13], if ∂V /∂φ = 0 holdsfor an extended period [14–16] (the latter case dubbed the ‘ultra-slow-roll’ inflation) and if the period of fast-rollinflation preceded about 50 e-folds of slow-roll inflation [17–19]. An important new feature appearing in all thesecases is that the curvature power spectrum becomes evolving on super-Hubble scales temporarily. Furthermore, thenon-Gaussianity consistency relation for single field inflation can be violated [16].Motivated by these phenomenologically interesting features, the ultra-slow-roll inflation was generalized in [20].Starting from an assumption of a constant rate of roll with ¨ φ/H ˙ φ = − − α , where nonzero α implies deviationfrom the flat potential, they derived a potential that satisfies the assumption approximately. However, it is actuallypossible to construct an exact solution under this assumption, which is the main topic of the present paper. We referto this class of models as the ‘constant-roll’ inflation. We shall show that the general exact solution for these modelsincludes power-law inflation [21, 22] and other two solutions: One of them is equivalent to a solution previously foundin [23] in a different context, and the other one amounts to a particular case of hilltop inflation [24] with cutoffor natural inflation [25] with an additional negative cosmological constant. We shall also investigate the evolutionof scalar (curvature) perturbations in these models. In general, these constant-roll models may have super-Hubbleevolution of the scalar perturbation, which is the same situation as the ultra-slow-roll inflation. However, we shallshow that for the particular cases of hilltop inflation and natural inflation do not suffer from it and they actuallyremain observationally feasible with the most recent constraints for the tilt of the scalar power spectrum and thetensor-to-scalar ratio.The rest of the paper is organized as follows. In § II, we determine the inflaton potential required for constant-rollinflation. We show that it is possible to derive exact solution for inflationary dynamics without assumptions made inthe literature. We shall see that in order to make the inflationary regime an attractor, one needs the constant-rollparameter α < − /
2. In § III, we explore scalar and tensor perturbations generated during constant-roll inflation.In § III A, we calculate power spectrum of the curvature perturbation and show that α & α . − § III B, we confirm that thesuper-Hubble evolution of the curvature perturbations in non-attractor model α > − /
2, whereas α < − / α . − § III C, we addressed tensor perturbation. We then conclude in § IV. Throughoutthe paper, we will work in the natural unit where c = 1, and the metric signature is ( − + ++). II. CONSTANT-ROLL INFLATION
We consider inflation driven by a minimally coupled scalar field defined by the action S = Z d x √− g (cid:20) M R − g µν ∂ µ φ∂ ν φ − V ( φ ) (cid:21) , (2)where M Pl = (8 πG ) − / . Working in the flat Friedmann-Lemaˆıtre-Robertson-Walker (FLRW) metric, the Friedmannequation and the equation of motion for the scalar field are given by3 M H = ˙ φ V, (3) − M ˙ H = ˙ φ , (4)¨ φ + 3 H ˙ φ + ∂V∂φ = 0 , (5)where a is the scale factor, a dot denotes the derivative with respect to t , H ≡ ˙ a/a is the Hubble parameter.Inflationary evolution is characterized by the slow roll parameters: ǫ ≡ − ˙ HH , ǫ n +1 ≡ ˙ ǫ n Hǫ n . (6)In the slow-roll inflation, the slow-roll parameters are small, | ǫ n | ≪
1, and the first terms of the right-hand side of(3) and the left-hand side of (5) can be neglected. As a result, an approximately flat spectrum of scalar (curvature)perturbations can be obtained, whose exact form depends on the functional shape of the potential V ( φ ) during thequasi-de Sitter expansion regime. Here, instead, we are interested in a different regime when ¨ φ is not negligible in(5). Following [20], we adopt an ansatz of a constant rate of roll ¨ φ/H ˙ φ during inflation and parametrize it as¨ φ = − (3 + α ) H ˙ φ, (7)with an arbitrary constant value of α . The standard slow-roll inflation occurs if α ≃ − α = 0. Thus, the constant-roll inflation interpolates between these two regimes. By the way,this shows that the term ‘ultra-slow-roll inflation’ is rather misleading. In fact, this regime has to be considered as aspecific case of fast-roll inflation since the slow-roll approximation breaks down during it.In [20], by neglecting ˙ φ / H = H ( φ ) anduse the Hamiltonian-Jacobi-like formalism [26, 27]. This approach is applicable as long as t = t ( φ ) is a single-valuedfunction, i.e. ˙ φ = 0. Since ˙ H = ˙ φdH/dφ holds in this approach, we can rewrite (4) as˙ φ = − M dHdφ . (8)Plugging the time derivative of (8) to (7), we obtain the differential equation for the Hubble parameter with respectto φ , d Hdφ = 3 + α M H. (9)The general solution of this equation is H ( φ ) = C exp r α φM Pl ! + C exp − r α φM Pl ! . (10)For α < −
3, the exponents become imaginary, and then C = C ∗ . Using (3) and (8), we find the inflaton potentialrequired for the exact solution (10): V ( φ ) = M " H − M (cid:18) dHdφ (cid:19) = M (cid:20) − αC exp (cid:18)p α ) φM Pl (cid:19) + 2(6 + α ) C C − αC exp (cid:18) − p α ) φM Pl (cid:19)(cid:21) . (11)Then we can derive evolution of inflaton φ ( t ) by solving (8), and obtain H ( t ) by plugging φ ( t ) back into (10). However,it should be kept in mind that the Hamilton-Jacobi formalism produces one solution φ ( t ) for the derived potential V ( φ ) only which needs not be an attractor solution (even an intermediate one). So, its stability with respect to allFLRW solutions with the same potential has to be checked each time.The most interesting particular solutions for α > − H = M e ± √ α φM Pl , (12) H = M cosh r α φM Pl ! , (13) H = M sinh r α φM Pl ! , (14)where M is an integration constant which determines the amplitude of the power spectrum of the curvature pertur-bation. Note that the third solution describes a bounce. For α < −
3, only the last two solutions have physical sense,and the hyperbolic functions in them have to be substituted by the corresponding trigonometric ones. Consequently,for α < − φ → − φ , we focus on H = M e √ α φM Pl . From (11), we then obtain V ( φ ) = − αM M exp (cid:18)p α ) φM Pl (cid:19) , (15) φ = − r
23 + α M Pl ln[(3 + α ) M t ] , (16) H = 1(3 + α ) t , (17) a ∝ t α . (18)The positivity of V ( φ ) requires α <
0, and for − < α < − n s − n t = 2(3 + 2 α ) / (2 + α ) < φ . But from the observational pointof view, power-law inflation is certainly excluded because of the absence of the large amount of primordial tensorperturbations (gravitational waves) that would be r = 8(1 − n s ) ≈ .
28 in this case.Plugging the second solution (13) to (11), we obtain for α > − V ( φ ) = 3 M M (cid:20) α (cid:26) − cosh (cid:18)p α ) φM Pl (cid:19)(cid:27)(cid:21) , (19) φ = M Pl r
23 + α ln (cid:20) coth (cid:18) α M t (cid:19)(cid:21) , (20) H = M coth[(3 + α ) M t ] , (21) a ∝ sinh / (3+ α ) [(3 + α ) M t ] . (22)This solution is equivalent to a solution found in [23] in a different context. Note also that (22) with α = 3( w − / p = wρ with w = const. This case encompasses a more special case of mimicking the ΛCDM model with w = 0,which is found in [28]. For − < α < V ( φ ) has a minimum at φ = 0. Thus, to end inflation, a kind of phasetransition at this point has to be assumed additionally, e.g. similar to that occurs in hybrid inflation [29].Contrary, for α < − V ( φ ) = 3 M M (cid:20) α (cid:26) − cos (cid:18)p | α | φM Pl (cid:19)(cid:27)(cid:21) , (23) φ = 2 s | α | M Pl arctan( e | α | Mt ) , (24) H = − M tanh ( | α | M t ) , (25) a ∝ cosh − / | α | ( | α | M t ) , (26)which is a particular case of hilltop inflation [24] with the inverted quadratic potential near the origin. As usual inthis case, we assume that φ → +0 at M t → −∞ , so the Hubble parameter remains positive and approaches M . Theninflation takes place as the inflaton rolls down from the origin, and ends near the point where the potential crosseszero. In fact, the latter has to be changed somewhere before this point and the initial condition for φ may be madeless restrictive. We shall also see that this case includes natural inflation [25] with an additional negative cosmologicalconstant. We discuss the above points in § III B.Finally, the third solution (14) yields V ( φ ) = − M M (cid:20) α (cid:26) (cid:18)p α ) φM Pl (cid:19)(cid:27)(cid:21) , (27) φ = − M Pl r
23 + α arctanh (cid:20) tan (cid:18) α M t (cid:19)(cid:21) , (28) H = − M tan[(3 + α ) M t ] , (29) a ∝ cos / (3+ α ) [(3 + α ) M t ] . (30)Although this is a mathematically allowed solution, it has ¨ a ( t ) <
0. Therefore, it cannot describe an inflationarymodel in the usual sense.For the following, we investigate the second solution (13), i.e. (19)–(22) with α > − α < −
3, andclarify if they serve observationally feasible inflationary scenario. In particular, we shall check the potential and theanalytic solutions for the former case (19) below as the potential for the latter case (23) amounts to a particular caseof natural inflation or hilltop inflation with V ∝ const − φ around φ = 0. We present the potential (19) in Fig. 1 (a)for α = − . . α = 0 yields a flat potential and it reducesto the case of the original ultra-slow-roll inflation. The evolution of inflaton is depicted in Fig. 1 (b). The field valuemonotonically decreases and approaches to the origin. For the case of α = 0 .
5, the inflaton climbs up the potentialand approaches to the top of the potential at the origin. We note that the Hubble parameter in Fig. 1 (c) approachesconstant as time goes by, which leads to de Sitter expansion of the universe, i.e. a ( t ) ∝ e Mt for M t ≫
1, which isdisplayed in Fig. 1 (d). In this limit, the effective inflaton mass squared becomes m ( φ ) ≡ ∂ V /∂φ = − α (3 + α ) M .Then the generic solution for the inflaton is the sum of two exponents e αMt and e − (3+ α ) Mt . However, our constant-roll (a) Α = Α = -
Α = - - Φ (cid:144) M Pl V (cid:144) M P l M (b)0.0 0.5 1.0 1.5 2.0 - - M t Φ (cid:144) M P l (c)0.0 0.5 1.0 1.5 2.00.00.51.01.52.02.53.0 M t H (cid:144) M (d)0.0 0.5 1.0 1.5 2.0012345 M t a (e)0.0 0.5 1.0 1.5 2.001234 M t Ε (f)0.0 0.5 1.0 1.5 2.0 - - - - M t Ε FIG. 1. (a) Potential (19), and its exact solution for (b) inflaton φ , (c) Hubble parameter H , (d) scale factor a , and slow-rollparameters (e) ǫ ≡ − ˙ H/H , (f) ǫ ≡ ˙ ǫ /Hǫ , for α = − . . ǫ approaches to − α ), which is not necessarily negligible, the Hubble parameter approaches constant and the scale factorevolves as exponentially. The higher slow-roll parameters are given by ǫ n +1 = 2 ǫ and ǫ n = ǫ . solution (20) chooses the second exponent only due to the definition (7). The same situation happens for α < − M t → −∞ . This is an illustration of the remark above that the Hamilton-Jacobi method of finding exactsolutions produces only one solution for the potential V ( φ ) determined by its use at the same time.We can also express the conformal time τ = R dt/a analytically in terms of the Gauss’ hypergeometric function F . For (19) with α > −
3, we obtain τ = ( − α α ) M (3 + α ) (cid:20) cosh[(3 + α ) M t ] F (cid:18) , α α ) ,
32 ; cosh [(3 + α ) M t ] (cid:19) − ( − − / F (cid:18) , α α ) ,
32 ; 1 (cid:19)(cid:21) , (31)where we fixed the integration constant to make τ → − M t → ∞ . On the other hand, for (23) with α < −
3, weobtain τ = i (2 + α ) M cosh α α [(3 + α ) M t ] F (cid:18) , α α ) , α α ) ; cosh [(3 + α ) M t ] (cid:19) − √ π Γ (cid:16) α α ) (cid:17) Γ (cid:16) α α ) (cid:17) , (32)where the integration constant is determined by τ → − M t → −
0. Of course, to get a realistic model, we haveto cut the potential somewhere before it becomes negative. In that case the end of the inflation should be definedaccordingly and we would choose integration constant to set τ → − ǫ = 2 ǫ + 2 ¨ φ/ ( H ˙ φ ) and the condition (7). This fact has been pointed out from the beginning of this class of models[14]. Our exact solution allows us to see the violation in higher order slow-roll parameters. For (19) with α > −
3, thefirst slow-roll parameter ǫ is given by ǫ = 3 + α cosh [(3 + α ) M t ] = 3 + αa α ) + 1 , (33)and for higher order slow-roll parameters ǫ n with n ≥ ǫ n = − α ) tanh [(3 + α ) M t ] , ǫ n +1 = 2(3 + α )cosh [(3 + α ) M t ] . (34)We show their evolution in Fig. 1 (e) and (f). In particular, for M t ≫ ǫ = ǫ n +1 ≃ α ) a − α ) , ǫ n ≃ − α ) . (35)Thus, the odd-order slow-roll parameters asymptotically approach to zero, while the even ones approach to − α ),which is not necessarily negligible, rather, ǫ n can be of order unity. Thus the slow-roll approximation clearlybreaks down. Nevertheless, the Hubble parameter approaches to a constant H ≃ M and the scale factor grows asexponentially, so this is still inflation. Likewise, for (23) with α < −
3, the slow-roll parameters are given by2 ǫ = ǫ n +1 = − α )sinh [(3 + α ) M t ] , ǫ n = − α )tanh [(3 + α ) M t ] , (36)and their asymptotic behaviors are 2 ǫ = ǫ n +1 → ǫ n → − α ) as M t → −∞ , which are the same withthe case for α > − α . The top left panel of Fig. 2 for α = − .
02 shows a typical attractor behavior where the phasespace flow converges to ˙ φ = 0 and φ = 0. This implies that the inflaton approaches to the global minimum of thepotential at φ = 0. Thus the analytic solution with negative α is an attractor. The top right panel for α = 0 of Fig. 2shows phase space flow converges various points depending on initial conditions. This is not surprising because wenote that for α = 0 the potential is exactly constant V = 3 M M , so it is invariant under translation φ → φ + const.The scalar field moves on the flat potential with the Hubble friction, and it eventually stops at a point which dependson an initial condition. In this case, we need to recover the integration constant for our analytic solution (20) anddetermine it according to an initial condition. Finally, the bottom panel of Fig. 2 for α = 0 .
02 shows that the inflatongoes to ˙ φ → ±∞ and φ → ±∞ , respectively. Therefore the analytic solution for positive α is not an attractor. Thisis obvious because the potential is not bounded from below. The scalar field rolls down to φ → ±∞ since V → −∞ .Hence, we need to make fine-tuning of the initial condition for scalar field obeying the analytical solution (20) forpositive α .For a general α > −
3, this special constant-roll solution for which φ ( t ) ∝ exp [ − (3 + α ) M t ] in the limit
M t ≫ φ ( t ) ∝ exp ( αM t ). Therefore, the stability condition is − (3 + α ) > α , or α < − / α ≈ −
3. The same formally refers to the α < − φ φ . α=− φ . φ α= φ φ . α= FIG. 2. Phase space diagram for α = − .
02 (top left), α = 0 (top right) and 0 .
02 (bottom).
III. EVOLUTION OF SCALAR AND TENSOR PERTURBATIONA. Scalar perturbation
We consider the gauge invariant curvature perturbation ζ k in our models (19) and (23). It relates to the metricperturbation through δg ij = a (1 − ζ ) δ ij in a gauge δφ = 0. The evolution of the mode function v k ≡ √ M Pl zζ k with z ≡ a √ ǫ is governed by the Mukhanov-Sasaki equation [30, 31]: v ′′ k + (cid:18) k − z ′′ z (cid:19) v k = 0 , (37)where a prime denotes a derivative with respect to the conformal time τ . The potential term z ′′ /z is exactly expressedin terms of slow-roll parameters: z ′′ z = a H (cid:18) − ǫ + 32 ǫ + 14 ǫ − ǫ ǫ + 12 ǫ ǫ (cid:19) . (38)We can solve (37) by the standard treatment of the slow-roll inflation except that ǫ n is not negligible whereas ǫ n − is. Starting from the sub-Hubble regime where k ≫ z ′′ /z , (37) reduces to v ′′ k + k v k = 0. We choose the adiabaticvacuum initial condition, i.e. no particles and minimum energy at τ → −∞ : v k ( τ ) = e − ikτ √ k . (39)As the mode k approaches to the Hubble radius crossing, the potential term z ′′ /z becomes non-negligible. Sinceduring generation of the curvature perturbation 2 ǫ = ǫ n +1 → ǫ n → − α ) for both cases with α > − α < −
3, we can simplify the potential term as z ′′ z ≃ (1 + α )(2 + α ) τ = ν − / τ , (40)where ν ≡ r (1 + α )(2 + α ) + 14 = (cid:12)(cid:12)(cid:12)(cid:12) α + 32 (cid:12)(cid:12)(cid:12)(cid:12) . (41)With (39) as the boundary condition, the solution is given in terms of Hankel function as usual, v k ( τ ) = √− πτ H (1) ν ( − kτ ) . (42)The power spectrum of the curvature perturbation is then given by∆ s ( k ) ≡ k π | ζ k | = H π M ǫ (cid:18) kaH (cid:19) π | H (1) ν ( − kτ ) | . (43)Using the asymptotic formula lim x → H (1) ν ( x ) ≃ − iπ Γ( ν ) (cid:0) x (cid:1) − ν , we obtain∆ s ( k ) = H π M ǫ ν − [Γ( ν )] π (cid:18) kaH (cid:19) − ν , (44)and the spectral index is then n s − − ν . For given n s from observations, the constant-roll parameter α is givenby α = 1 − n s n s − . (45)For instance, for n s = 0 .
96, we can choose α = 0 .
02 or − .
02. However, as we discussed in the previous section,the analytic solution for α = 0 .
02 is not an attractor solution. In addition, we should be careful for a possibleevolution of the curvature perturbation on super-Hubble scales [20]. Indeed, from (43) we confirmed that ∆ s ( k ) ∝ H − ν ǫ − a − ν ∝ a α +3+ | α +3 | evolves for α = 0 .
02, but approaches to a constant for α = − .
02. Therefore α = − .
02 is viable, and the standard slow-roll approximation works in this case. However, our exact solution givesa possibility to sum all higher-order corrections to it, at least for the background evolution.
B. Super-Hubble evolution
Alternatively, we can examine the behavior of curvature perturbations in the super-Hubble regime k ≪ z ′′ /z bydirectly solving v ′′ k − ( z ′′ /z ) v k = 0. The formal solution for this equation is given by a linear combination of z and z R dτ /z . Since ζ k ∝ v k /z , we thus obtain ζ k = A k + B k Z dta ǫ , (46)where A k and B k are integration constants. The first term expresses the constant mode of the curvature perturbations.For slow-roll inflation, the second term yields decaying mode, thus only first term remains, and as a result the curvatureperturbation is conserved outside the Hubble radius. However, this is not necessarily the case for constant-roll inflation,as we shall see below.For the model (19) with α > −
3, we can perform the integral of the second term of (46) analytically using (22)and (33). The result is Z dta ǫ = 2 M Z H a ˙ φ dt ∝ Z dt cosh [(3 + α ) M t ] sinh − α [(3 + α ) M t ]3 + α = ( − α α ) M (3 + α ) cosh [(3 + α ) M t ] u (cid:18) , α α ) ,
52 ; cosh [(3 + α ) M t ] (cid:19) , (47)where u ( a, b, c ; x ) is a solution for the hypergeometric differential equation: x (1 − x ) d udx + [ c − ( a + b + 1) x ] dudx − abu = 0 . (48)It has two independent solutions and the solutions that converge for x ≫ a + b − c ) >
0, two solutions are u = ( − x ) b − c (1 − x ) c − a − b F (1 − b, c − b, a − b + 1; 1 /x ) ≃ x − a , (49) u = ( − x ) a − c (1 − x ) c − a − b F (1 − a, c − a, b − a + 1; 1 /x ) ≃ x − b , (50)and (ii) for Re( a + b − c ) < u = ( − x ) a F ( a, a − c + 1 , a − b + 1; 1 /x ) ≃ x − a , (51) u = ( − x ) b F ( b, b − c + 1 , b − a + 1; 1 /x ) ≃ x − b . (52)For the case of (47), − < α < x / u ≃ const and x / u ≃ x α α ) , where x = cosh [(3 + α ) M t ]. Therefore, the first solution gives a constant mode, which is absorbedinto the definition of A k , whereas the second solution is a decaying mode for − < α < − /
2, or a growing mode for − / < α <
0. The remaining region α > x / u ≃ const and x / u ≃ x α α ) . These asymptotic behavior of two solutions are the same with the case (i). The first solution givesa constant mode while the second one is a growing mode as α > α < −
3, we can prove that the second term of (46) yields sum ofconstant mode and decaying mode as
M t increases from −∞ . The integral is given by Z dta ǫ ∝ − iα (3 + α ) M cosh α/ (3+ α ) [(3 + α ) M t ] u (cid:18) − , α α ) , α )2(3 + α ) ; cosh [(3 + α ) M t ] (cid:19) . (53)As a + b − c = − / < u ( a, b, c ; x ), its two independent solutions are given by u and u .Thus the asymptotic behavior of the above integral is x b u ≃ x b − a = x α α ) = x | α | and x b u ≃ const. Thelatter mode is a constant mode, and the former mode is a decaying mode as M t increases from −∞ , namely, as x = cosh [(3 + α ) M t ] decreases from + ∞ .In summary, in the super-Hubble regime a curvature perturbation has two modes which asymptotically approachto const , and cosh α α [(3 + α ) M t ] , (54)respectively. The latter one is a decaying mode for α < − /
2, but it is a growing mode for α > − /
2. The conditionfor the decaying mode coincides with the attractor condition for background evolution derived in the previous section.In comparison with slow-roll inflation, where we always have a constant mode and a decaying mode outside theHubble radius, we may have a growing mode in constant-roll inflation. This is because ǫ decays as a − α ) , which isfaster than a − for α > − /
2. As a result, ζ ≃ R dt/ ( a ǫ ) grows in this case. Any model with background evolutionwith ǫ = d ln ǫ /d ln a < − α > − / not an attractor during expansion.Thus, in this case such solution can typically occur for no more than a few e-folds.To confirm our discussion above, we numerically integrated (37) with analytic solution for the background quantitiesderived in the previous section. The result is shown in Fig. 3. In the left and right panel, we present the result for(19) with α = 0 .
02, and the result for (23) with α = − .
02, respectively. The solid blue line represents the amplitudeof the curvature perturbation | ζ k | , the red dashed line expresses ( k/aH ) and yellow dotted line is ( z ′′ /z ) / ( aH ) . Wenormalized the e-folds N = ln a so that k = z ′′ /z at N = 0, which is about a half e-fold earlier than the Hubbleradius crossing k = aH . Two dotted-dashed lines are asymptotic analytic solution for the curvature perturbation.The pink dotted-dashed line is the one for the sub-Hubble evolution, which is proportional to ( a √ ǫ ) − ∝ sinh α [(3+ α ) M t ] / cosh[(3+ α ) M t ]. The green dotted-dashed line is the one for the super-Hubble evolution, which is proportionalto cosh α α [(3 + α ) M t ]. In the left panel, for α = 0 .
02 we see that the curvature perturbation switches its evolutionat the Hubble radius crossing, and continuously evolves even on the super-Hubble regime. On the other hand, asdepicted in the right panel, for α = − .
02 the curvature perturbation remains constant after the Hubble radius exit,as expected.0
Α = Ζ k H z '' (cid:144) z L ‘ H a H L H k (cid:144) a H L - - N È Ζ k È Α = - Ζ k H k (cid:144) a H L H z '' (cid:144) z L ‘ H a H L - - N È Ζ k È FIG. 3. Evolution of the amplitude of the curvature perturbation (blue solid) for α = 0 .
02 (left) and − .
02 (right) as a functionof e-folds N = ln a obtained by integrating (37) numerically. The analytic solutions for ( k/aH ) (red dashed), and ( z ′′ /z ) / ( aH ) (yellow dotted) are also presented. Two dotted-dashed lines are asymptotic analytic solution for the curvature perturbation,which are proportional to ( a √ ǫ ) − ∝ sinh α [(3 + α ) Mt ] / cosh[(3 + α ) Mt ] for the earlier phase (pink dotted-dashed) andcosh α α [(3 + α ) Mt ] for the latter phase (green dotted-dashed), respectively. The situation for α = 0 .
02 above is similar to the generalized ultra-slow-roll inflation [20], which also has a growingmode and results in an unwanted amplification of the curvature perturbation outside the Hubble radius. Combinedwith the observational value of the amplitude of the scalar power spectrum, this growth of the curvature perturbationshould be compensated by imposing extremely small energy scale of inflation, at least M ≃ − M Pl . This is muchsmaller than the BBN bound M > O (MeV). The same occurs for constant-roll inflation. Thus, the case α & α . − φ = φ which is less than the critical value φ c = πM Pl / p | α | where V ( φ ) < H changes thesign. The aim of this modification is to finish inflation at φ = φ . This situation is completely analogous to whathappens in the case of power-law inflation. So, then our solution can be applied not to the whole evolution of theUniverse but to its evolution during inflation only (apart from its very end) including the observable range of e-folds.The best-fit value of the constant-roll parameter α and the prediction for the tensor-to-scalar ratio r will depend on φ . We shall see below that | α | has to be small for any reasonable value of φ , so φ c ≫ M Pl anyway. Thus,depending on the value of φ , such a model can describe both small-field and large-field inflation.Let us first consider the case when φ ≪ φ c . Then V ( φ ) can be approximated as V ( φ ) = 3 M M (cid:18) − α (3 + α ) φ M (cid:19) , (55)during the whole inflation. In this case n s − − α (3 + α ) /
3. So, the best fit from the recentCMB data is α ≈ − .
02. As was shown above, the curvature perturbation remains constant on super-Hubble scalesin this case. The number of e-folds from the end of inflation is N = M − Z φ φ VdV /dφ dφ = 3 α (3 + α ) ln (cid:18) φ φ (cid:19) ≈
50 ln (cid:18) φ φ (cid:19) . (56)Thus, due to numerical coincidence between the measured value of 1 − n s and 2 /N H where N H = 50 −
60 correspondsto the present Hubble radius H − , the inflaton field value in the observable range of e-folds during inflation is ≈ φ e − .This provides a possibility to significantly enlarge possible range for an initial condition for the inflaton field at thelocal beginning of inflation: it is sufficient to have its any value less than ≈ φ e − .If now φ . φ c , but not specifically close to it, the potential (23) and the corresponding exact constant-roll solution(24)–(26) may be taken as they are up to φ = φ . Then the model looks like the so called ‘natural inflation’ [25]but with the additional negative cosmological constant Λ = M (3 + α ) <
0. Also, it is always large-field inflation.In this case, the measured value of n s − α unambiguously, it provides only an upper bound on | α | . Still | α | has to be small, so | Λ | ≪ M . As a result, this cosmological constant is subdominant duringinflation. However, it affects higher-order slow-roll corrections. Therefore, the model (23) represents a novel exampleof viable and exactly analytically solvable (for the evolution of background and perturbations in the super-Hubbleregime) model of inflation.1The limiting, although somewhat fine-tuned case, occurs if φ = φ c − O ( M Pl ). Then at the last stage of inflation,when φ ≫ M Pl and ˜ φ ≫ M Pl where ˜ φ = φ c − φ , we get H ≈ M r | α | φM Pl , V ≈ α (3 + α )2 M ˜ φ , (57)i.e. just quadratic chaotic inflation. Then n s − − /N , and we obtain its correct value independently of the valueof α . However, the upper limit on α follows from the condition that the approximation (57) still works for N = N H when ˜ φ ≈ M Pl . From the condition that the latter quantity should be ≪ φ c , it follows that | α | ≪ .
02, i.e.much less than in the hilltop case φ ≪ φ c considered above. C. Tensor perturbation
Finally, we consider the tensor perturbation δg ij = a h ij for our models. The evolution equation for the modefunction of the tensor perturbation u k,λ ≡ aM Pl h k,λ / u ′′ k,λ + (cid:18) k − a ′′ a (cid:19) u k,λ = 0 , (58)where λ = + , × denotes the two polarization modes of the gravitational waves. As a ′′ /a = ( aH ) (2 − ǫ ) ≃ (2 +3 ǫ ) /τ ≃ /τ , we obtain (in agreement with [33]) ∆ t = 2 H π M , (59)and thus the standard consistency relation for the tensor-to-scalar ratio r ≡ ∆ t ∆ s ≈ ǫ = 32 M (cid:18) d ln H ( φ ) dφ (cid:19) ≈ M (cid:18) d ln V ( φ ) dφ (cid:19) (60)holds with both ≈ signs becoming = in the leading order of the slow-roll approximation.For the quadratic hilltop case φ ≪ φ c we, therefore, get r = 8(3 + α ) φ /M . So, parametrically by powers of | α | , r is of the order of N − . However, actually r can be much less than the latter quantity if φ ≪ M Pl . To get r ∼ N − as, e.g. in the R + R inflationary model [1], one needs φ ∼ M Pl . Larger values of r in constant-roll inflation,of the order N − parametrically, can be obtained if M Pl ≪ φ ∼ φ c . Then the exact background solution (24)–(26)has to be used. Finally, in the limiting case φ = φ c − O ( M Pl ), r reaches the value 8 /N ( ∼ .
14 for N = N H ) as forthe quadratic chaotic inflation, but this model lies just beyond the 2 σ CL contour for the recent Planck data [34].
IV. CONCLUSION
We have investigated an inflationary scenario where the rate of roll defined by ¨ φ/H ˙ φ = − − α remains constant.This class includes slow-roll inflation with negligible rate of roll and fast-roll inflation, in particular, the so called‘ultra-slow-roll’ one. We find all exact solutions satisfying the constant-rate-of-roll ansatz. They include power-lawinflation, the solution found in [23] in a different context, and particular (and somewhat modified) cases of hilltopinflation and natural inflation. In this class of models, even-order slow-roll parameters can be order of unity whileodd-order slow-roll parameters are asymptotically negligible. It turns out that it is difficult for α > − α which yields a slightly red-tilted power spectrum of curvature perturbations, they growoutside the Hubble radius. In order to reproduce the observed amplitude of fluctuations in the presence of suchgrowth, we need an extremely low energy scale of inflation, which is much smaller than BBN bound. Therefore, thecase with α > − α < −
3, thecurvature perturbation has a constant mode and a decaying mode on super-Hubble scales, as the standard slow-rollinflation does. Therefore, the model (23) with α . − n s ≈ .
96. The prediction for the tensor-to-scalar ratio r depends on the potential cutoff location, and can be bothless than 1%, and in the range 1% . r . ACKNOWLEDGMENTS
The authors thank J. D. Barrow and T. Suyama for useful discussions. H.M. was partially supported by JapanSociety for the Promotion of Science (JSPS) Postdoctoral Fellowships for Research Abroad and J.Y. by JSPS Grant-in-Aid for Scientific Research No. 23340058. A.S. acknowledges RESCEU hospitality as a visiting professor. He wasalso partially supported by the grant RFBR 14-02-00894 and by the Scientific Programme “Astronomy” of the RussianAcademy of Sciences. [1] A. A. Starobinsky, Phys.Lett.
B91 , 99 (1980).[2] K. Sato, Mon.Not.Roy.Astron.Soc. , 467 (1981).[3] A. H. Guth, Phys.Rev.
D23 , 347 (1981).[4] A. D. Linde, Phys.Lett.
B108 , 389 (1982).[5] A. Albrecht and P. J. Steinhardt, Phys.Rev.Lett. , 1220 (1982).[6] S. A. Appleby, R. A. Battye, and A. A. Starobinsky, JCAP , 005 (2010), arXiv:0909.1737 [astro-ph.CO].[7] H. Motohashi and A. Nishizawa, Phys.Rev. D86 , 083514 (2012), arXiv:1204.1472 [astro-ph.CO].[8] A. Nishizawa and H. Motohashi, Phys.Rev.
D89 , 063541 (2014), arXiv:1401.1023 [astro-ph.CO].[9] A. De Felice and S. Tsujikawa, Living Rev.Rel. , 3 (2010), arXiv:1002.4928 [gr-qc].[10] J. Martin, C. Ringeval, and V. Vennin, Phys.Dark Univ. (2014), 10.1016/j.dark.2014.01.003,arXiv:1303.3787 [astro-ph.CO].[11] H. Motohashi, Phys.Rev. D91 , 064016 (2015), arXiv:1411.2972 [astro-ph.CO].[12] A. A. Starobinsky, JETP Lett. , 489 (1992).[13] S. Inoue and J. Yokoyama, Phys.Lett. B524 , 15 (2002), arXiv:hep-ph/0104083 [hep-ph].[14] N. Tsamis and R. P. Woodard, Phys.Rev.
D69 , 084005 (2004), arXiv:astro-ph/0307463 [astro-ph].[15] W. H. Kinney, Phys.Rev.
D72 , 023515 (2005), arXiv:gr-qc/0503017 [gr-qc].[16] M. H. Namjoo, H. Firouzjahi, and M. Sasaki, Europhys.Lett. , 39001 (2013), arXiv:1210.3692 [astro-ph.CO].[17] C. R. Contaldi, M. Peloso, L. Kofman, and A. D. Linde, JCAP , 002 (2003), arXiv:astro-ph/0303636 [astro-ph].[18] L. Lello and D. Boyanovsky, JCAP , 029 (2014), arXiv:1312.4251 [astro-ph.CO].[19] D. K. Hazra, A. Shafieloo, G. F. Smoot, and A. A. Starobinsky, Phys.Rev.Lett. , 071301 (2014),arXiv:1404.0360 [astro-ph.CO].[20] J. Martin, H. Motohashi, and T. Suyama, Phys.Rev.
D87 , 023514 (2013), arXiv:1211.0083 [astro-ph.CO].[21] L. Abbott and M. B. Wise, Nucl.Phys.
B244 , 541 (1984).[22] F. Lucchin and S. Matarrese, Phys.Rev.
D32 , 1316 (1985).[23] J. D. Barrow, Phys.Rev.
D49 , 3055 (1994).[24] L. Boubekeur and D. Lyth, JCAP , 010 (2005), arXiv:hep-ph/0502047 [hep-ph].[25] K. Freese, J. A. Frieman, and A. V. Olinto, Phys.Rev.Lett. , 3233 (1990).[26] A. Muslimov, Class.Quant.Grav. , 231 (1990).[27] D. Salopek and J. Bond, Phys.Rev. D42 , 3936 (1990).[28] H. Motohashi and W. Hu, Phys.Rev.
D90 , 104008 (2014), arXiv:1408.4813 [hep-th].[29] A. D. Linde, Phys.Rev.
D49 , 748 (1994), arXiv:astro-ph/9307002 [astro-ph].[30] V. F. Mukhanov, JETP Lett. , 493 (1985).[31] M. Sasaki, Prog.Theor.Phys. , 1036 (1986).[32] R. Saito, J. Yokoyama, and R. Nagata, JCAP , 024 (2008), arXiv:0804.3470 [astro-ph].[33] A. A. Starobinsky, JETP Lett. , 682 (1979).[34] P. Ade et al. (Planck), Astron.Astrophys.571