Properties of fluctuating states in loop quantum cosmology

In loop quantum cosmology, the values of volume fluctuations and correlations determine whether the dynamics of an evolving state exhibits a bounce. Of particular interest are states that are supported only on either the positive or the negative part of the spectrum of the Hamiltonian that generates this evolution. It is shown here that the restricted support on the spectrum does not significantly limit the possible values of volume fluctuations.


Introduction
A solvable model [1] that captures basic features of classical and quantum cosmology is given by two canonical variables, Q and P with Poisson bracket {Q, P } = 1, and a 1parameter family of Hamiltonians, H δ = |Q sin(δP )|/δ with δ ≥ 0. In the limit δ → 0, H 0 = |QP | is quadratic up to the absolute value, and a system close to an upside-down harmonic oscillator is obtained. Since QP and therefore sgn(QP ) is preserved by equations of motion generated by the auxiliary Hamiltonian H ′ 0 = QP , the set of regular solutions (such that QP = 0) of the classical H 0 -system is given by the union of two disjoint sets: solutions of the H ′ 0 -system with initial values Q(0), P (0) such that sgn(Q(0)P (0)) = 1, and solutions of the −H ′ 0 -system with initial values Q(0), P (0) such that sgn(Q(0)P (0)) = −1. All classical solutions can therefore be obtained from a quadratic Hamiltonian.
In a simple cosmological interpretation, |Q| is proportional to the volume of an expanding or collapsing universe, while P is proportional to the Hubble parameter. According to the Friedmann equation of classical cosmology for flat spatial slices, H 0 = |QP | can be interpreted as the momentum canonically conjugate to a free, massless scalar source φ, whose energy density ρ ∝ 1 2 p 2 φ /Q 2 with the momentum p φ canonically conjugate to φ is then required to be proportional to P 2 . Solutions Q(φ) and P (φ) of Hamilton's equations of motion generated by ±H 0 ∝ p φ therefore describe how Q and P change in relation to the "internal time" φ. If δ = 0, H δ can still be interpreted in this way, but only if the Friedmann equation is modified such that P 2 is replaced by sin 2 (δP )/δ 2 . This modification may be motivated by the appearence of holonomies in loop quantum gravity [2,3] and loop quantum cosmology [4], and is supposed to describe an implication of quantum geometry.
Replacing the unbounded function P 2 with a bounded function sin 2 (δP )/δ 2 , still proportional to the energy density of a matter source, suggests that the classical big-bang singularity, at which the energy density diverges, could be avoided by quantum-geometry effects [5]. Indeed, solutions for Q(φ) of equations of motion generated by ±H are superpositions of real exponential functions. If the condition Q 2 − |J δ | 2 = 0 is imposed, which ensures that P in the definition of J δ is real, the equation which is by definition positive for regular solutions, implies that Q(φ) must be cosh-like and ReJ δ (φ) sinh-like. The eternally collapsing behavior of the volume Q(φ) approaching zero if δ = 0, Q(φ) = Q(0) exp(±φ), is then replaced by a "bounce" at the non-zero minimum of cosh. The preceding argument ignores quantum fluctuations, which may be expected to be significant in a discussion of big-bang solutions. If (∆Q) 2 is large, it could conceivable change the balance of signs in (3), in which Q 2 = Q 2 + (∆Q) 2 would take the place of Q 2 . For states with (∆Q) 2 ≥ (δH ′ δ ) 2 + (∆ReJ δ ) 2 , the right-hand side of (3), written for expectation values, is no longer positive, and Q (φ) would not be cosh-like. The possibility of such non-bouncing solutions in loop quantum cosmology has been demonstrated using canonical effective methods [6], in particular for small δH ′ δ relevant for an understanding of generic spacelike singularities [7,8].
However, for quantum states the absolute value in H δ has to be treated with greater care than in the case of classical solutions. Solutions of quantum evolution generated by an operatorĤ δ via a Schrödinger equation for wave functions can be expressed as superpositions of solutions of quantum evolution generated by an operatorĤ ′ δ , provided the latter are supported solely on the positive or negative part of the spectrum ofĤ ′ δ . (See Sec. 2.3 below for a demonstration.) This condition is a straightforward replacement of the classical restriction on initial values. But it may have more significant ramifications, in particular when quantum fluctuations are taken into account that may be larger than the expectation value Ĥ δ , as required to change the signs in (3). A state that is supported only on the positive part of the spectrum ofĤ ′ δ and has an expectation value of |Ĥ ′ δ | close to zero may not have arbitrarily large fluctuations ofĤ ′ δ . The question to be addressed in this paper is whether this restriction also limits the size of fluctuations ofQ.

Eigenstates
We will first determine the spectra ofĤ ′ 0 andĤ ′ δ and then discuss relevant properties of states obtained from superpositions of their positive parts.

Eigenstates ofĤ
We use the symmetric orderingĤ to quantize H ′ 0 = QP on the standard L 2 -Hilbert space. Eigenstates of this operator in the Q and P -representations are determined by the same type of first-order differential equation, in the Q-representation, and in the P -representation. For every λ, there are in each representation two orthogonal solutions ψ λ± (Q) and φ λ± (P ), respectively, given by It is obvious that ψ λ 1 + and ψ λ 2 − are orthogonal to each other for any λ 1 and λ 2 , and so are φ λ 1 + and φ λ 2 − . Moreover, where the substitutions q = |Q|, x = log q, p = |P | and y = log p have been used. For real λ, all eigenstates are delta-function normalizable, fixing the coefficients c λ± = 1/ √ 2π = d λ± . The spectrum ofĤ ′ 0 is therefore real, continuous, and twofold degenerate.

Eigenstates ofĤ ′ δ
For δ = 0, the Hamiltonian is periodic in P with period 2π/δ. To be specific, we will assume that the basic operators are represented on a separable Hilbert space of square-integrable functions periodic in P , such thatQ has a discrete spectrum given by δZ. Inequivalent representations, such as states which are periodic only up to a phase factor exp(iǫ), for which the spectrum ofQ is shifted by ǫ, or non-separable Hilbert spaces as used often in loop quantum cosmology [9], would not change our results. In the Q-representation, our states therefore obey an ℓ 2 inner product such that We writeĤ ′ δ aŝ Since exp(iδP ) is a translation operator in the Q-representation, eigenstates ofĤ ′ δ in this representation are determined by a difference equation where Q takes the values n δ with integer n. This equation with non-constant coefficients does not have straightforward solutions. It is, however, possible to show that eigenstates obey a similar twofold degeneracy as in the case ofĤ ′ 0 : Lemma 1 For given λ, there are two orthogonal solutions ψ λ± , one of which is supported on positive values of Q (and Q = 0), and one on negative values of Q. They are related by Proof: Let us first look for solutions such that ψ λ+ (− δ) = 0. Using the equation (15) for Q = − δ, we obtain ψ λ+ (−2 δ) = 2i(λ/ )ψ λ+ (− δ) = 0. Moreover, if ψ λ+ (−(n − 1) δ) = 0 and ψ λ+ (−n δ) = 0, using the equation for Q = −n δ shows that ψ λ+ (−(n + 1) δ) = 0. By induction, ψ λ+ (Q) = 0 for all integer Q/ δ < 0. However, if ψ λ+ (0) = 0 for such a solution, ψ λ+ ( δ) = −2i(λ/ )ψ λ+ (0) = 0, using (15) for Q = 0. The solution therefore is not identically zero, and it is unique up to multiplication with a constant ψ λ+ (0). A similar line of arguments, starting with the assumption that ψ λ− (0) = 0, implies that ψ λ− (Q) = 0 for all integer Q/ δ ≥ 0, while assuming ψ λ− (− δ) = 0 guarantees that the solution is not identically zero. Since the supports of any ψ λ 1 + and ψ λ 2 − are disjoint, the two states are orthogonal with respect to the inner product (13).
In the P -representation, eigenstates of (14) obey the first-order differential equation This equation is solved by The substitution x = log | cot(δP/2)| shows that these states are delta-function normalized.
The spectrum therefore has the same properties as in the case ofĤ ′ 0 , being real, continuous, and twofold degenerate.

Existence of positive-energy solutions with large fluctuations
For any δ, completeness of the eigenstates of a self-adjoint operator shows that any state ψ(Q) has an expansion of the form in terms of eigenstates ofĤ ′ δ , for some c λ± normalized such that ∞ −∞ |c λ± | 2 dλ = 1. It evolves according to (22) The actual dynamics in our models of interest is generated by the HamiltonianĤ δ = |Ĥ ′ δ |. This operator has the same eigenstates ψ λ± (Q), but with eigenvalues |λ|. Its spectrum is therefore four-fold degenerate and positive. Dynamical solutions in these models are given by and The decomposition into positive-energy solutions ψ + and negative-energy solutions ψ − simply rewrites generic wave functions and does not restrict their fluctuations of Q or P . However, it is sometimes preferred [10] (although not required [11]) to discard negativeenergy solutions and consider only positive-energy solutions ψ + (or vice verse, but no superpositions of solutions with opposite signs of the energy). A question of interest in quantum cosmology is whether this restriction in any way limits the possible magnitude of fluctuations of Q or P , which would then have consequences for bouncing or non-bouncing behavior according to [6]. Using the spectral properties derived in the preceding section, we now show that this is not the case.
In particular, for potential non-bouncing behavior, we are interested in solutions with small Ĥ δ , such that If Ĥ δ is small, given the positivity of the spectrum ofĤ δ , the range of possible values of ∆H δ seems to be limited because the state in the λ-representation can spread out only to one side of Ĥ δ . However, the twofold degeneracy of the spectrum ofĤ ′ δ , of the specific form derived in the preceding section, in particular in Lemma 1, shows that there is no such limitation for fluctuations ∆Q even if Q is required to be small: In order to construct a state, supported only on the positive part of the spectrum ofĤ ′ δ , such that it has a small expectation value and large fluctuations ofQ, we choose some c λ such that ∞ 0 |c λ | 2 dλ = 1, and define ψ c+ (Q) = ∞ 0 c λ ψ λ+ (Q)dλ. This state is supported on the positive part of the spectrum ofĤ ′ δ , by construction, and has a certain expectation value Q c + > 0 and fluctuations ∆ c+ Q > 0. Similarly, the state ψ c− (Q) = ∞ 0 c λ ψ λ− (Q)dλ, using the transformation (16), has expectation value Q c − = − Q c + − δ < 0 and fluctuations ∆ c− Q = ∆ c+ Q > 0. The state with some |α| ≤ √ 2, then has expectation value and fluctuations given by For α = 1, the result can also be written as Since Q c+ is not restricted by the positivity condition, ∆Q is unlimited even on states with small expectation value Q .

Moments
Since H ′ 0 is a function of Q and P , H ′ 0 -moments in a given state are related to Q and P -moments in the same state. There may therefore be restrictions on the magnitude of Q or P -fluctuations if a state is required to have small Ĥ ′ 0 and small H ′ 0 -fluctuations. We will now demonstrate that Q and P -fluctuations are indeed restricted in such a state, but only if additional assumptions on the QP -covariance are made.

Relationships between moments
BecauseĤ ′ 0 is quadratic inQ andP , ∆H ′ 0 is related to moments of up to fourth order in Q and P . In the following calculations, we will be using the notation of [12], as in Definition 1 Given a set of operatorsÂ i , i = 1, . . . , n, and integers k 1 , . . . , k n ≥ 0 such that i k 1 ≥ 2, the moments of a state are where ∆Â i =Â i − Â i , all expectation values are taken in the given state, and the subscript "symm" indicates that all products of operators are taken in totally symmetric (or Weyl) ordering: The following reordering relations will be useful: Lemma 2 For two operatorsQ andP such that [Q,P ] = i ,

QP +
PQ − Q P and zero skewness (third-order moments), then the relative fluctuations of Q andP are bounded from above by the relative fluctuation ofĤ = 1 2 (QP +PQ): If ∆(QP ) = 0 and ∆(Q 2 P ) = 0 = ∆(QP 2 ), we obtain This result shows that a state with small relative H-fluctuations but large relative Q-fluctuations must have non-zero covariance or skewness.

Example
As shown in [13], the right-hand side of (26) is strictly negative for a Gaussian state in Q. This inequality then cannot be fulfilled. The same paper showed that the right-hand side of (26) is approximately zero for a state given by if Q > 0 and ψ(Q) = 0 otherwise, with constantsQ > 0, σ > 0 andλ. We now demonstrate that such a state can be approximated by a state supported only on the positive part of the spectrum ofĤ ′ 0 , which then provides an example of how the restriction given by Proposition 1 can be overcome by states with non-zero covariance.
Let us choose a Gaussian for λ > 0 and c λ = 0 otherwise, where normalizes c λ restricted to positive λ and is close to N 2 ≈ 1 forλ ≫ σ, or ∆H ′ 0 / Ĥ ′ 0 ≪ 1. Using the definition (27) with α = √ 2, we consider the state ψ c+ (Q) = ∞ 0 c λ ψ λ+ (Q)dλ. The integral can be approximated by extending the integration over positive λ to all real λ, which is valid provided c λ is negligible for λ < 0. Given (51), the approximation can be used if the λ-variance σ is much less than the λ-expectation value, σ ≪λ. The same condition allows us to approximate N ≈ 1, and we obtain for Q > 0. DefiningQ = exp(−p), the result equals (50). The resulting state (53) shows that the log |Q|-variance is given by ∆ log |Q| = /(2σ), while the log |Q|-expectation value is log |Q| = −p. We can therefore maintain the conditionλ ≫ σ, or ∆H ′ 0 / Ĥ ′ 0 ≪ 1, for the approximation in (53) to be valid, and choose a small Q with large ∆Q.