Classical Preheating and Decoherence D. T. Son Department of Physics FM–15, University of Washington 6 9 Seattle, WA 98105, USA 9 1 January 1996 n a J 1 3 Abstract 1 We establish the equivalence between the quantum evolution of spatially v 7 homogeneous oscillations of a scalar field and that of an analogous classical 7 system with certain random initial condition. We argue that this observation 3 1 can be used for numerical simulation of the Universe in the preheating epoch. 0 6 We also explicitly demonstrate that the phenomenon of parametric resonance 9 thatleadstopreheatingissimultaneouslyaneffectivemechanismforgenerating / h quantum decoherence of the Universe. p - p e h : v i X r a UW/PT-96-01 1 1. Calculation of the reheating temperature is one the most important issue in inflationary cosmology. The temperature achieved soon after the inflation epoch is themaximaltemperatureinthewholehistoryoftheUniverse andmust besufficiently largeforthegenerationofbaryonnumbertobeeffective. Accordingtotheinflationary picture, the Universe is reheated by the process of transferring energy out of spatially homogeneous oscillations of the scalar inflaton field. The old theory of the reheating [1]describes thisenergy transfer asoccurring throughthedecay oftheinflatonquanta and predicts a relatively low speed for this process in theories with small couple constants. However, a recent revision [2] shows that in many case there exists a more rapidchannel ofenergytransfer, whichiscloselyrelatedtotheexistence ofparametric resonance in the spectra of the fields interacting with the inflaton and of the inflaton itself. These parametric resonance modes, typically forming continuous energetic bands, grow exponentially on the background of the inflaton field oscillations, and at some moment begin to take a large amount of energy from these oscillations. It has been argued in ref.[2] that due to the exponential character of the growth, the decay of the inflaton field oscillations is a rapid process of “explosion” rather than a slow damping. The amplitude of these oscillations drops to a small value after some time interval of the same order as of magnitude of the period of oscillations. This new mechanism, named in ref.[2] “preheating”, may change considerably the prediction of the reheating temperature. It has been argued that the phenomenon of parametric resonance may be important in other aspects as well [3, 4]. The complete theoretical description of preheating is still lacking right now, due to the complexity of interaction between the modes that appear via parametric reso- nance and the background, as well as between the modes themselves. However, some important steps have been made in this direction (see, for example [5, 6]). In one of the approaches [6] the problem is treated in the Hartree–Fock approximation, where one takes into account the back–reaction of the parametric–resonance modes to the background but neglects the scattering between these modes. The result is quite surprising: while it confirms that parametric resonance is the main mechanism of the damping of the inflaton field in the first time period, in the long run the amplitude of the oscillations typically remains constant (in some cases it is not much smaller than the initial amplitude before the explosion), or at least decays very slowly. This shows that it takes much longer time, of many order of magnitudes larger than the period of oscillations, to make the amplitude goes to 0. At late times, the created particles are mostly of very small energy, which means that the thermalization pro- 2 cess may also take a longer time interval than has been expected. The calculations in ref.[6], however, rely heavily on the Hartree–Fock approximation, so there remains the question whether these results reflect the real situation. In this paper we do not attempt to build a complete theory of preheating. How- ever, we will point out a possible way to perform reliable calculations in the nonlinear regime after inflation. Namely, we show that the behavior of the quantum system under consideration, during the preheating epoch, is equivalent to the evolution of the classical field, when one randomizes over a particular set of initial conditions. This result is established by comparing the perturbative series of the two theories. This fact makes possible numerical simulations of the Universe after inflation, since the problem is now completely classical. As a by–product, we demonstrate explicitly the development of decoherence of the Universe. We show that the phenomenon of parametric resonance provides an effective mechanism for the latter. 2. For simplicity we will work inthe model ofone scalar field only. Inother words, we neglect the interaction of the inflaton with other particles and take into account its self–interaction only. The Lagrangian of the model is taken in the standard form, 1 m2 λ L = (∂ φ)2 φ2 φ4 µ 2 − 2 − 4! WewillalsoneglecttheexpansionoftheUniverseandworkintheordinaryMinkowskian space–time. At the moment when inflation completes, the inflaton field has a nonzero expectation value φ = φ which is a constant over the whole space. Soon after that 0 the field begins oscillations around φ = 0. For convenience the following picture is used in further discussions: at t < 0 there is an extra force term Jφ in the Lagrangian so that the expectation value of φ is a nonzero constant φ , but at t = 0 this source 0 is suddenly turned off and the field starts oscillate. We will be interested in the time dependence of the mean value of φ, which will be denoted as φ(t). Thelinearized classicalequationforthemodesofmomentumkofthefield, φ(t,k), λ ∂2 +ω2 + φ2(t) φ(t,k) = 0 (1) t k 2 0 ! has the same form as that of the oscillator whose frequency is a periodic function of time, ω2 = ω2+λφ20(t). It is well known that at some values of ω this system exhibits k 2 k parametric resonance: there does not exist any solution to eq.(1) which remains finite on the whole time axis [7]. A typical solution to eq.(1) grows exponentially in the limits t + and/or t . This phenomenon occurs when ω lies in k → ∞ → −∞ 3 certain energetic bands (“resonance bands”). We will consider the situation of broad resonance, which correspond to large values of φ , φ 1/√λ or larger. 0 0 ∼ In the quantum case, the existence of parametric resonance leads to the amplifi- cation of the quantum fluctuations of the modes with ω inside the resonance bands. k When the amplitude of these fluctuations are still small, one can make use of the standard Bogoliubov transformations to find the quantum state of the system [5]. However, the interesting regime is that of late times where the non–linearity becomes essential, and one needs another approach to attack the problem. Thegeneralmethodtofindreal–timeevolutionofquantumfieldsistheSchwinger– Keldysh closed–time–path formalism [8, 9, 10]. At diagrammatic level, the formalism gives rise to the same set of Feynman rules as that for calculating the S–matrix, except that every internal integral in Feynman diagrams is now performed along a contour that goes from t = to t = + and then goes back to t = (Fig.1), −∞ ∞ −∞ and instead of one Feynman propagators G(x,y) there exists four ones depending on whether x and y lie on the upper or the lower parts of the contours, 0 0 G (x,y) = 0 Tφ(x)φ(y) 0 , G (x,y) = 0 φ(y)φ(x) 0 , ++ +− h | | i h | | i G (x,y) = 0 φ(x)φ(y) 0 , G (x,y) = 0 T¯φ(x)φ(y) 0 , −+ −− h | | i h | | i It is easy to see that these propagators arise from the propagators on the contour C, G (x,y) = 0 T φ(x)φ(y) 0 , where T is the notation for time–ordering along C. C C C h | | i Note also that the four propagators are equal at x = y. To derive the Feynman rules, we will decompose the quantum field into a classical ˜ part φ (t) and quantum fluctuation φ, 0 ˜ φ(t) = φ (t)+φ(t) 0 where φ (t) = φ at t < 0 and at t > 0 satisfies the field equation 0 0 ∂2φ +m2φ +λφ3 = 0 µ 0 0 0 0 ˜ The Lagrangian for φ is then 1 m2 λ λ λ L(φ˜) = (∂ φ)2 φ˜2 φ2φ˜2 φ φ˜3 φ˜4 2 µ − 2 − 4 0 − 3! 0 − 4! and corresponds to the Feynman rules shown in Fig.2. Before turning to the calculation of diagrams, let us consider in more details the propagators on the background φ (t). Consider, for instance, G (x,y). It is 0 ++ 4 convenient to use the mixed representation (coordinate in time and momentum in space) where this propagators can be written into the form, dk G (x,y) = eik(x−y)G (x ,y ;k) ++ (2π)3 ++ 0 0 Z where iG (x ,y ;k) = θ(x y )fk(x )fk(y )+θ(y x )fk(x )fk(y ) (2) ++ 0 0 0 − 0 1 0 2 0 0 − 0 2 0 1 0 We have introduced the mode functions fk which are the two linearly independent 1,2 solutions to the equation λ ∂2 +ω2 + φ2(t) fk (t) = 0 (3) t k 2 0 ! 1,2 with the following boundary conditions at t < 0, fk(t) e−iωkt, fk(t) eiωkt 1 ∼ 2 ∼ where ω = k2 +m2 + λφ20. The mode functions are normalized so that f′(t)f (t) k 2 1 2 − f (t)f′(t) =qi. Ifφ (t)is0,fk andfk wouldbeequaltoe−iωkt/√2ω andeiωkt/√2ω , 1 2 − 0 1 2 k k respectively. It is easy to see that fk(t) = (fk(t))∗. 2 1 The formulas for other propagators are similar, iG (x ,y ;k) = fk(x )fk(y ) +− 0 0 2 0 1 0 iG (x ,y ;k) = fk(x )fk(y ) −+ 0 0 1 0 2 0 iG (x ,y ;k) = θ(x y )fk(x )fk(y )+θ(y x )fk(x )fk(y ) (4) −− 0 0 0 − 0 2 0 1 0 0 − 0 1 0 2 0 Consider now a value of k where there is parametric resonance in eq.(3). When parametric resonance is present, eq.(3) possesses two real solutions fk(t) and fk(t) + − which satisfy the following conditions at t > 0 [7], fk(t+T) = eǫkTfk(t) + + fk(t+T) = e−ǫkTfk(t) − − where ǫ is some real, positive constant depending on ω . In particular, f grows as k k + t + while the behavior of f is opposite. The previously defined mode functions − → ∞ f and f are linear combinations of f and f . Not interested in the particular form 1 2 + − of the coefficients, we write, fk(t) = α fk(t)+β fk(t), fk(t) = α∗fk(t)+β∗fk(t), (5) 1 k + k − 2 k + k − 5 where we have made use of the fact that f and f are complex conjugate and f 1 2 + and f are real. Since as f is growing with t while f is falling, at large t the part − + − f (t) dominates over f (t). If in eq.(5) one neglects f and leaves only f , there + − − + is no difference between f and f , except for an overall factor. If one also neglects 1 2 modes outside resonance bands, which is natural since these modes are not enhanced, and substituting eq.(5) to eqs.(2), (4) one finds that the four propagators are equal to each other to the leading order, iG (x,y) = iG (x,y) = iG (x,y) = iG (x,y) = ++ +− −+ −− dk = iG0(x,y) = α 2fk(x )fk(y ) (6) (2π)3 | k| + 0 + 0 Z Now let us consider the one–loop contribution to φ. The only diagram that con- tributes to φ is the tadpole one shown in Fig.3a. This diagram is equal to iλ iλ φ (x) = dy dyG (x,y)φ (y )G(y,y) = dyG (x,y)φ (y )G(y,y) (7) 1 0 C 0 0 R 0 0 2 2 Z Z Z C where we have dropped the indices of G(y,y) since all the four Green functions are equal at coinciding points and introduced the retarded Green function, G = G R ++ − G . One can expect that at late x the integral in eq.(7) is dominated by large +− 0 values of y where G(y,y) can be replaced by the leading term, eq.(6). So one finds, 0 λ dk φ (x ) = dy G (x ,y ;0)φ (y ) α fk(y ) 2 (8) 1 0 2 0 R 0 0 0 0 (2π)3 | k + 0 | Z Z Eq.(8) can be represented graphically as in Fig.3b, where each external line ending with a bullet is associated with the factor α fk(y ) if k is a resonance mode and 0 in | k| + 0 the opposite case. Note that fk grows exponentially with t, so φ is also growing. At + 1 t when φ becomes comparable to φ one can expect that allterms of the perturbative 1 0 series are of the same order, and the perturbation theory breaks down. We will try to extract the main contribution from each order of the perturbation theory in this regime. We will not be interested at latter times when φ φ . 1 0 ≫ Let us move to two–loop diagrams. Consider, for example, the one depicted in Fig.4a, iλ2 dk φ (x ) = dy dz G (x ,y ;0)φ (y )G (y ,z ;k)G (y ,z ;k)φ (z )φ (z ) 2 0 2 0 0 (2π)3 C 0 0 0 0 C 0 0 C 0 0 0 0 1 0 Z C (9) 6 If one is interested large values of x , it can be expected that the important region of 0 integration in r.h.s. of eq.(9) is that of large y and z . So, one may suspect that the 0 0 integral can be calculated by replacing G (x ,y ;k) by its leading order contribution, C 0 0 eq.(6). One obtains after that iλ2 dk φ (x ) = dy dz G (x ,y ;0)φ (y )(fk(y ))2(fk(z ))2φ (z )φ (z ) 2 0 − 2 0 0 (2π)3 C 0 0 0 0 + 0 + 0 0 0 1 0 Z C However, it is easy to see that the integration over z gives zero, since it goes along 0 the contour C and the integrand, (fk(z ))2φ (z )φ (z ), is the same on the upper + 0 0 0 1 0 and lower parts of the contour. So, one should turn to the next–to–leading order. Denoting the next correction to eq.(6) as G1(x,y) = G (x,y) G0(x,y), one writes, C C − dk φ (x ) = λ2 dy dz G (x ,y ;0)φ (y )G1(y ,z ;k)fk(y )fk(y )φ (z )φ (z ) 2 0 0 0 (2π)3 C 0 0 0 0 C 0 0 + 0 + 0 0 0 1 0 Z C dk = λ2 dy dz G (x ,y ;0)φ (y )G (y ,z ;k)fk(y )fk(y )φ (z )φ (z ) (10) 0 0 (2π)3 C 0 0 0 0 C 0 0 + 0 + 0 0 0 1 0 Z C where inthelast equation we have replaced G1 byG , making use ofthefact that the C C leading term vanishes. Eq.(10) can be rewritten in a simpler form using the retarded propagator G = G G , R ++ +− − ∞ dk φ (x ) = λ2 dy dz G (x ,y ;0)φ (y )G (y ,z ;k)fk(y )fk(y )φ (z )φ (z ) 2 0 0 0 (2π)3 R 0 0 0 0 R 0 0 + 0 + 0 0 0 1 0 Z −∞ where the integrations over dy and dz are performed from to + . Eq.(10) can 0 0 −∞ ∞ be represented graphically as in Fig.4b. As in the case of the one–loop graph, the calculation of the loop diagram, thus, is reduced to evaluation of a tree connected graph. This technique canbegeneralized todealwith otherdiagramsandhigher ordersof perturbation theory. Summarily, the technique is the following. To calculate a given Feynman diagram with l loops, one cuts l internal propagators, so that the obtained diagram is tree, and connected. If there are many ways to perform this cutting, one should take a sum over all possibilities. Each propagator that has been cut is associ- ated with two factors of fk attached to the two vertices that the former propagator + had been connecting. The propagators that have not been cut are associated with the retarded propagator G . The obtained diagrams are computed by the standard R way, where the integral are performed along the usual time axis. We note that this 7 reduction of loop graphs to tree ones is very similar to that found in another setting [11]. We emphasize that our method is applicable in the regime when all terms of the perturbation series are of the same order. At latter times when φ φ our approach 1 0 ≫ does not work, since the pieces that have been neglected may become important. 3. The fact that the diagrams obtained after the cutting procedure are tree points to the possibility of classical description of the problem. To find the latter, let us consider the Cauchy initial problem with the classical field equation (∂2 +m2)φ+λφ3 = 0 (11) µ with the following initial condition at the asymptotics t , → −∞ dk φ(t,x) = φ (t)+ c fk(t)eikx (12) 0 (2π)3 k + Z where c are some arbitrary set of complex numbers. For φ to be real, we require that k c∗ = c . Since at t fk(t) is small and satisfies eq.(3), eq.(12) is the solution k −k → −∞ + tothe fieldequation uptothefirst order off . The correctionsto eq.(12), which start + at the second order of f , can be calculated iteratively from the field equation and + the result can be represented in the form of tree Feynman diagrams. For example, the second–order contribution is represented by the first diagram in Fig.5 together some third– and fourth–order graphs. In these graphs each external leg ending on a bullet is associated with the factor c fk, and each internal line corresponds to the k + retarded propagator. Despite the fact that these diagrams are similar to that obtained by the cutting procedure described above, there are two main differences between the two cases. First, the diagrams obtained by cutting loop graphs always contains an even number of external bullet legs. Second, theexternal bullet legs inthe first case can begrouped into pairs so that the momenta running along legs of the same pair are of opposite signs, while in the graphs coming from eq.(12) the only requirement is that the sum of the momenta of the bullet legs is 0. Nevertheless, it is easy to show that there is a simple relation between the two cases. Namely, if one denotes as φ[t,x,c ] the k solution to the field equation with the boundary condition (12), then the quantum average of φ in our problem is given by an integral over c with a Gaussian measure, k dk c 2 k φ(t) = c exp | | φ[t,x,c ] (13) k k Z D −Z (2π)3 2|αk|2! (the integral in the r.h.s. of eq.(13) does not depend on x). To see this, one notices that the integration over c leaves only diagrams with even number of bullet legs k D 8 which can be grouped in paired of opposite momenta. The integration also adds a factor of α 2 to each pair with momentum (k, k). So, the graphical representation k | | − of the r.h.s. of eq.(13) is the same as those obtained by cutting the loop graphs. In eq.(13), c are supposed to be real, Gaussian–distributed around 0 with the k standard deviation α and the c with different k are not correlated. Eq.(13) is not k k | | the only possible, one could write, for example φ(t) = θ φ[t,x, α exp(iθ )] (14) k k k D | | Z where c now are complex numbers with fixed absolute value equal to α but with k k | | random phase. One can write more relations of this type. Eqs.(13,14) give rise to the possibility of numerical simulation of the processes occurring in the Universe after inflation. To do this it is sufficient to take an initial field configuration in the form of a homogeneous oscillations plus appropriate random perturbations, and evolve this configuration in time according to the field equation. The mean value of φ in this classical system is the same as that of the quantum one. It is trivial to derive the same result in theories with more than one scalar fields. Eqs.(13,14) are the manifestation of the fact that when the amplitude of quantum fluctuations is large (roughly speaking, when the occupation numbers are large), the system behaves like a classical one. Our derivation gives this statement a precise meaning. It is also interesting to mention the relation between eq.(13,14) and the funda- mental phenomenon of decoherence of the Universe. In a situation when, as in our case, a closed system starts its evolution from a quantum pure state at t = 0 it will remains in a pure state at any time moment. The system that we are considering can be described, in the pure state language, by the operators obtained by Bogoliubov transformation [5]. However, when the resonance modes become large the modes strongly interact between themselves and the pure state description becomes very complicated. In rescue, another description comes into effect: the system can be considered as a mixed state, i.e. that of an ensemble, of states (semiclassical in our case) characterized by the solution to the field equation with initial condition (12) with c playing the role of the indices numbering the quantum states in the ensemble. k Explicitly, the system is described by the following density matrix, ρˆ φ[t,x,c ] ρ[c ] φ[t,x,c ] k k k ∼ | i h | {Xck} where φ[t,x,c ] is the notation for a (pure) semiclassical quantum state associated k | i 9 withthe classical field configurationφ[t,x,c ](the choice of φ[t,x,c ] is notunique). k k | i Eq.(13,14) ensures that the average of φ in this mixed state is equal to its true quantum average in the pure state with proper choice of ρ[c ]. It can be shown k that the same is valid for other Green functions as well. So, for calculating physical quantities one can consider the system as being in a mixed state, though in fact its true state is a pure one. Note that the time period when c are large, c 1, but k k ≫ still sufficiently small so that the modes are linear, is the matching region where both pure– and mixed–state descriptions can be used. 4. So, wehave shown thattheproblemofevolutionofthehomogeneousoscillating scalar field that has a direct connection with preheating in inflationary cosmology is equivalent to the classical evolution of the scalar field with some random initial conditions. We also show that the parametric resonance phenomenon provides an effective mechanism for the generation of decoherence of the Universe. We have seen a nice feature of this mechanism that the transition from the pure–state to mixed– state description can be traced explicitly. The result of this paper may be useful for numerical modeling of the preheating process andhopefully will leads tothe better understanding of thelatter. Preliminary results seems to agree qualitatively with that of the the Hartree–Fock approximation: the fall of the amplitude of oscillations is saturated at some finite level. Moreover, after this saturation the modes with small k seem to play an important role in the evolution. The work is still continuing, results will be published elsewhere. The author thanks P. Arnold, V. Rubakov, P. Tinyakov and L. Yaffe for valuable discussions. 10