4.3 Model with a scalar field and Perfect Fluid
4.3.2 The T = 1 Boundary
solution, i.e. the center manifold originating from dS0 whose asymptotics are given by n= 1 : H ∼ −t, ϕ∼ −t, σ ∼0, ρpf ∼(−t)−2, as t→ −∞ (4.55a) n= 2 : H ∼e−23t, ϕ∼e−3t, σ∼0, ρpf ∼e−43 t, as t→ −∞ (4.55b) n≥3 : H ∼(−t)n−2n , ϕ∼(−t)n−22 , σ ∼0, ρpf ∼(−t)−n−22n , as t→ −∞ (4.55c)
(a)The Kasner circle in the Ωϕ= 0 plane in theT = 1 boundary.
FL1
Ωφ=1( pf=0(
Ω P1
21
P
14
P
ΣΦ
X
(b)Σ = 0 plane in theT = 1 bound-ary..
Figure 4.4: he invariantT = 0 boundary and the invariant subset for a monomial potential V =14(λϕ)4.
Proof. To prove this theorem we need to use Lemma4.1alongside with generalized averaging techniques based on the methods used in [50,66,78]. We make the same approach regarding ϵ as previous seen in Chapters 2-3. So given real function f, its average over a time period associated to Ωϕis given by
⟨f⟩= 1 P(Ωϕ)
Z τ0+P(Ωϕ) τ0
f(τ)dτ. (4.59)
Taking the time averaging for dXdτ in (4.56) and using the equation for Σϕ it gives d
dτ
XdX dτ
− dX
dτ 2
+nX2n= 0. (4.60)
Taking the time averaging for the orbit we get
*dX dτ
2+
=⟨Σ2ϕ⟩=n⟨X2n⟩. (4.61)
Thus, using this result for a periodic orbit onT = 1 in (4.10) we get
⟨γϕ⟩= 2n
n+ 1. (4.62)
We now set ϵ(τ) = 1−T(τ) and consider the system dΩϕ
dτ = 3ϵh(γpf−γϕ)(1−Ωϕ) + (2−γpf)Σ2σiΩϕ (4.63a) dΣσ
dτ =−3
2ϵh(2−γpf)(1−Σ2σ) + (γpf−γϕ)Ωϕ
iΣσ (4.63b)
dϵ dτ =−1
nϵ2(1−ϵ)(1 +q) (4.63c)
where (X,Σϕ,Σ+,Σ−) solves (4.13) and q+ 1 = 3
2(2Σ2ϕ+ 2Σ2++ 2Σ2−+γpf(1−Ωϕ−Σ2σ)). (4.64) The general idea of this averaging method is based on making the near identity transformation
Ωϕ(τ) =y(τ) +ϵ(τ)w(y, z, τ, ϵ) (4.65a) Σσ(τ) =z(τ) +ϵ(τ)g(y, z, τ, ϵ), (4.65b) and then prove that the evolution of the variables y and z are approximated, at first order, by the solution of ¯y and ¯z respectively of the averaged equation.
So starting withythe evolution equation for this new variable can be obtained using equations (4.63a) and (4.63c) together with the evolution equation for Ωϕ. This then gives
dy
dτ = 1 +ϵ∂w
∂y −1"
dΩϕ
dτ −
w+ϵ∂w
∂ϵ dϵ
dτ −ϵ∂w
∂τ −ϵ∂w
∂z dz dτ
#
= 1 +ϵ∂w
∂y −1"
3ϵ(γpf − ⟨γϕ⟩)y(1−y) + (⟨γϕ⟩ −γϕ)y(1−y) + (2−γpf)yz2
− 3ϵ2(γpf−γϕ) (1−2y)w+w(2−γpf)z2+g(2−γpf)y (4.66) + 3ϵ3(2−γpf)gw−(γpf −γϕ)w2+ (2−γpf)g2y−
w+ϵ∂w
∂ϵ dϵ
dτ −ϵ∂w
∂τ −ϵ∂w
∂z dz dτ
# .
Forz we made a similar thing using the equations (4.63b) and (4.63c) and we get dz
dτ = 1 +ϵ∂g
∂z −1"
dΣσ
dτ −
g+ϵ∂g
∂ϵ dϵ
dτ −ϵ∂w
∂τ −ϵ∂g
∂y dy dτ
#
= 1 +ϵ∂g
∂z −1"
− 3
2ϵ(2−γpf)z(1−z2) + (γpf − ⟨γϕ⟩)yz+ (⟨γϕ⟩ −γϕ)yz
− 3
2ϵ2(2−γpf)g(1−z)2−2+ (γpf −γϕ)wz+ (γpf−γϕ)gy (4.67)
+ 3
2ϵ3(2−γpf)zg2+ (2−γpf)g3−(γpf −γϕ)gw−
g+ϵ∂g
∂ϵ dϵ
dτ −ϵ∂g
∂τ −ϵ∂g
∂y dy dτ
# . Setting
∂w
∂τ =f1(y, z, τ, ϵ)− ⟨f1(y, z, .,0)⟩= 3⟨γϕ⟩y−2Σ2ϕ
y(1−y) (4.68a)
∂g
∂τ =f2(y, z, τ, ϵ)− ⟨f2(y, z, .,0)⟩= 3 (γϕ− ⟨γϕ⟩)yz (4.68b)
and expanding (4.66) and (4.67) in powers of ϵsmall enough, we get dy
dτ =ϵ⟨f1⟩(y, z) +ϵ2h1(y, z, w, g, τ, ϵ) +O(ϵ3) (4.69a) dz
dτ =ϵ⟨f2⟩(y, z) +ϵ2h2(y, z, w, g, τ, ϵ) +O(ϵ3) (4.69b) where
⟨f1⟩(y, z) =⟨f1(y, z, .,0)⟩= 3(γpf− ⟨γϕ⟩)y(1−y) + 3(2−γpf)yz2 (4.70a)
⟨f2⟩(y, z) =⟨f2(y, z, .,0)⟩=−3
2(2−γpf)z(1−z2)−3
2(γpf− ⟨γϕ⟩)yz (4.70b) h1(.) =3(γpf−γϕ)(1−2y)w+w(2−γpf)(wz2+gy)+ 1
n(1 +q)(1−ϵ)w (4.70c) 3
2
∂w
∂z
(2−γpf)z(1−z2) + (γpf− ⟨γϕ⟩)yz−3∂w
∂y
(γpf− ⟨γϕ⟩)y(1−y) + (2−γpf)yz2 h2(.) =− 3
2
(2−γpf)g(1−z)2−2+ (γpf−γϕ)(wz−gy)+ 1
n(1 +q)(1−ϵ)w
(4.70d)
−3∂g
∂y
(γpf − ⟨γϕ⟩)y(1−y) + (2−γpf)yz2+3 2
∂g
∂z
(2−γpf)z(1−z2) + (γpf − ⟨γϕ⟩)yz. Notice that for large times,T ≈1,ϵ≈0, the right-hand side of (4.69a) and (4.69b) is almost
periodic meaning that ⟨γϕ⟩ −γϕ ≈0 which implies that w and g are bounded. So it follows from (4.65) that y and z are also bounded. We can now drop the high order terms in ϵ in (4.69) and study the truncated averaged equation, which leads to then system
d¯y
dτ = 3ϵ(γpf− ⟨γϕ⟩) ¯y(1−y) + 3ϵ¯ (2−γpf) ¯yz¯2 (4.71a) dz¯
dτ =−3
2ϵ(2−γpf) ¯z1−z¯2−3
2ϵ(γpf − ⟨γϕ⟩) ¯yz¯ (4.71b) dϵ
dτ =−1
nϵ2(1−ϵ)(1 +q). (4.71c)
Without loss of generality we can introduce a new time variable 1
ϵ d dτ = d
dτ¯, (4.72)
that will reduce the system. The new system reads dy¯
dτ = 3 (γpf− ⟨γϕ⟩) ¯y(1−y¯) + 3 (2−γpf) ¯yz¯2 (4.73a) dz¯
dτ =−3
2(2−γpf) ¯z1−z¯2−3
2(γpf− ⟨γϕ⟩) ¯yz¯ (4.73b) dϵ
dτ =−1
nϵ(1−ϵ)(1 +q). (4.73c)
Fixed points and stability in the case γpf ̸=⟨γϕ⟩
For γpf ̸= ⟨γϕ⟩ the dynamical system admits four fixed points. One in scalar field subset (Ωϕ= 1), two equivalent fixed points in the pure anisotropic subset (Σσ = 1) and one isolated fixed point in the pure matter subset (Ωpf = 1). The first fixed point is located on the intersection ofϵ= 0 (T = 1) with the pure scalar field subset and is
F1 : y¯= 1, z¯= 0, ϵ= 0. (4.74) The linearisation around this fixed point yields to the eigenvalues−3(γpf−⟨γϕ⟩),−32(2−⟨γϕ⟩), and −3⟨γ2nϕ⟩ were the associated eigenvectors are the canonical basis ofR3.
The two equivalent fixed points located in the intersection ofϵ= 0 with the pure anisotropic subset are
F2± : y¯= 0, z¯=±1, ϵ= 0. (4.75) The linearisation around this fixed points yields to the eigenvalues 3(2− ⟨γϕ⟩), 3(2−γpf) and
−n3 whose eigenvectors are (∓2,1,0), (0,1,0) and (0,0,1).
The isolated fixed point is
F3 : y¯= 0, z¯= 0, ϵ= 0, (4.76) the linearisation around this fixed points yields to the eigenvalues 3(γpf− ⟨γϕ⟩), −32(2−γpf), and −3γ2npf whose eigenvectors are the canonical basis of R3.
Regarding the eigenvalues of each fixed point we need to split the analysis depending on the sign ofγpf− ⟨γpf⟩.
• Ifγpf >⟨γϕ⟩, F1 has three negative eigenvalues since ⟨γϕ⟩ ∈(0,2) so it is a hyperbolic sink. For F2± since γpf ∈ (0,2) we see that two eigenvalues are positive and the one in the ϵ-direction is negative, so F2± is an hyperbolic saddle, being a source in the {ϵ= 0}-subset. LastlyF3 as two negative eigenvalues and one positive eigenvalue being a hyperbolic saddle. So in this case F1 is the future attractor, so the universe will asymptotically approach the pure scalar field solution when ¯τ →+∞.
• If γpf < ⟨γϕ⟩, F1 has two negative eigenvalues an one positive eigenvalue so it is a hyperbolic saddle. Regarding F2± we have two positive eigenvalues and one negative eigenvalue, moreover in the {ϵ = 0} subset F2± is a hyperbolic source. Lastly, F3 has three negative eigenvalues being a hyperbolic sink. In this caseF3is the future attractor, so the universe will asymptotically approach the the flat FL solution when ¯τ →+∞.
Convergence of the solutions in the case γpf− ⟨γϕ⟩
Now we need to prove that the variables y and Ωϕ follows this evolution and the same need to be done forz and Σσ.
We start by introducing two new variables
|η(τ)|:=|y(τ)−y¯(τ)|, |ζ(τ)|:=|z(τ)−z¯(τ)| (4.77) that we need to estimate. In order to do so we define the sequences{τn}and {ϵn} such that ϵn=ϵ(τn), with n∈Nand
τn+1−τn= 1 ϵn
, τ0 = 0, ϵ0>0 (4.78)
where limτn= +∞ and limϵn= 0. So starting with the estimation of|η(τ)|we get
|η(τ)| = Z τ
τn
3ϵ(γpf − ⟨γϕ⟩)y(1−y) + 3ϵ(2−γpf)yz2+ϵ2h1(y, z, w, g, s, ϵ)ds
− Z τ
τn
(3ϵ(γpf− ⟨γϕ⟩) ¯y(1−y¯) + 3ϵ(2−γpf)¯yz¯2
≤ 3ϵn
Z τ τn
|γpf− ⟨γϕ⟩|
| {z }
|.|≤C1
|(y−y)¯ | |1−(y+ ¯y)|
| {z }
|.|≤1
ds+ 3ϵn
Z τ τn
|2−γpf|
| {z }
C2
|yz2−y¯z¯2|
| {z }
|.|≤|y−¯y|+|z2−¯z2|
ds + ϵ2n
Z τ τn
|h1(y, z, w, g, s, ϵ)|
| {z }
|.|M1
ds+O(ϵ3n)
≤ 3ϵn(C1+C2)Z τ
τn
|η(s)|ds+ 3ϵn Z τ
τn
|z2−z¯2|
| {z }
|.|≤|z−¯z||z+¯z|≤2|z−¯z|
+ϵ2nM1(τ−τn) +O(ϵ3n)
≤ 3ϵn(C1+C2)Z τ
τn
|η(s)|ds+ 6C2ϵn
Z τ τn
|ζ(s)|ds+ϵ2nM1(τ −τn) +O(ϵ3n). (4.79) For|ζ(τ)|=|z(τ)−z¯(τ)|, in this case we get
|ζ(τ)| = Z τ
τn
−3
2ϵ(2−γpf)z(1−z2)−3
2ϵ(γpf− ⟨γϕ⟩)yz+ϵ2h2(y, z, w, g, s, ϵ)ds
− Z τ
τn
−3
2ϵ(2−γpf) ¯z(1−z¯2)−3
2ϵ(γpf − ⟨γϕ⟩) ¯y¯z
≤ 3 2ϵn
Z τ τn
|2−γpf|
| {z }
|.|≤C2
|z−z| |¯ 1−zz¯−z2−z¯2|
| {z }
|.|≤2
ds+3 2ϵn
Z τ τn
|γpf− ⟨γϕ⟩|
| {z }
|.|≤C1
|yz−y¯z|¯
| {z }
|.|≤|y−¯y|+2|z−¯z|
ds + ϵ2n
Z τ τn
|h2(y, z, w, g, s, ϵ)|
| {z }
|.|≤M2
ds+O(ϵ3n)
≤ 3ϵn(C1+C2)Z τ
τn
|ζ(s)|ds+3 2ϵnC1
Z τ τn
|η(s)|ds+ϵ2nM2(τ−τn) +O(ϵ3n). (4.80) whereC1,C2,M1, andM2 are positive constants.
Let’s introduce a new function|ψ(τ)|=|η(τ)|+|ζ(τ)|so adding the two above equations to each other we get
|ψ(τ)| = 3ϵn(C1+C2)Z τ
τn
|η(s)|ds+ 6C2ϵn
Z τ τn
|ζ(s)|ds+ϵ2nM1(τ −τn) + 3ϵn(C1+C2)Z τ
τn
|ζ(s)|ds+3 2ϵnC1
Z τ τn
|η(s)|ds+ϵ2nM2(τ −τn) +O(ϵ3n)
= 3ϵn(C1+C2)Z τ
τn
(|η(s)|+|ζ(s)|)ds+ 6C2ϵn
Z τ τn
|ζ(s)|ds+3 2ϵnC1
Z τ τn
|η(s)|ds + ϵ2n(M1+M2)(τ −τn) +O(ϵ3n)
Using the fact that 6C2
Z τ τn
|ζ(s)|ds+3 2C1
Z τ τn
|η(s)|ds≤2 max{6C2,3 2C1}
Z τ
τn
|ζ(s)|ds+Z τ
τn
|η(s)|ds
we get
|η(τ)| ≤ ϵnC∗
Z τ τn
|ψ(s)|ds+ϵ2nM∗(τ−τn) (4.81)
≤ ϵn
M∗
C∗
eC∗ϵn(τ−τn)−1+O(ϵ2n) (4.82) whereC∗= 3 (C1+C2) + 2 max{6C2,32C1}and M∗ =M1+M2 are positive constants . So for τ−τn∈[0,1/ϵn] the above inequality becomes
|ψ(τ)| ≤ϵnK (4.83)
where K > 0. Since |ψ(τ)|= |η(τ)|+|ζ(τ)| → 0 as ϵn → 0 it follows that |η(τ)| → 0 and
|ζ(τ)| →0 asϵn→0. It follows than from (4.65a) , the triangle inequality, and the fact that ϵ→0 asτ →+∞, it follows from that Ωϕhas the same limit as ¯yand, therefore converges for 0 or 1 depending on the sign ofγpf− ⟨γϕ⟩. Regarding Σσ using (4.65b), the triangle inequality and ϵ→ 0 as τ →+∞, it follow that Σσ goes to zero since the system will converge to the point (Ωϕ,Σσ, ϵ) = (1,0,0) if γpf >⟨γϕ⟩ and will converge to the point (Ωϕ,Σσ, ϵ) = (0,0,0) ifγpf <⟨γϕ⟩. This completes the proof for the cases (i) and (ii) of the theorem.
Fixed points and stability in the case γpf =⟨γϕ⟩
Now we analyse the caseγpf =⟨γϕ⟩. In this case our averaged system (4.73) reads d¯y
dτ = 3(2−γpf)¯y¯z2 (4.84a)
dz¯ dτ =−3
2(2−γpf)¯z(1−z¯2) (4.84b) dϵ
dτ =−1
nϵ(1−ϵ)(1 +q). (4.84c)
The system admits a line of fixed points in the intersection of the scalar field subset with ϵ= 1 that is given by
Ly : y¯=y0, z¯= 0, ϵ= 0 (4.85) the linearisation around this line of fixed points yields to the eigenvalues 0,−n+13 , and −n+13 whose eigenvectors are the canonical basis ofR3. In this case we have two negative eigenvalues (sinceγpf ∈(0,2)) and a zero eigenvalue corresponding to a center manifold. To solve this we will introduce adapted variables
˜
y= ¯y−y¯0, z˜= ¯z, ˜ϵ=ϵ (4.86) which leads to the adapted system
dy˜
dτ =F(˜y,z, ϵ¯ ), dz¯ dτ =−3
2(2−γpf)¯z+G(˜y,z, ϵ¯ ), dϵ
dτ =−3γpf
2n ϵ+N(˜y,z, ϵ¯ ) (4.87) where F,G and N are high order functions. With this new adapted variables we relocated the fixed points on the lineLy to the origin. The 1-dimensional center manifold can be locally represented by the graph h :Ec→ Es, i.e., (¯z, ϵ) = (h1(˜y), h2(˜y)) satisfying the fixed point, h(0) = 0 and the tangency, dh(0)d˜y = 0 conditions which solves the following equations
F(˜y, h1(˜y), h2(˜y))h′1(˜y) =−3
2(2−γpf)h1(˜y) +G(˜y, h1(˜y), h2(˜y)) (4.88a) F(˜y, h2(˜y), h2(˜y))h′2(˜y) =−3γpf
2n h2(˜y) +N(˜y, h1(˜y), h2(˜y)) (4.88b) In general is not possible to find the exact solution for hi and Taylor expansion are used, however in this particular case the can find an exact solution for h1 andh2 so we get
h1(˜y) =±q1 +e2Cz(˜y+y0), h2(˜y) = 1 1 +1+C y+y˜ 0
ϵe2Cz( ˜y+y0)
γpf 2n(2−γpf )
(4.89)
Therefore, it follows that the center manifold reads dy˜
dτ = 3y(1 +e2Cz(y+y0))(2−γpf) (4.90) which shows explicitly that each point on the lineLy are center saddles.
The system also admits two more fixed points in the intersection of the subset{z= 1} and {ϵ= 0}given by
F2± : y= 0, z=±1, ϵ= 0. (4.91)
The linearisation around this fixed points yields to the eigenvalues n+13 , n+13 , and−n3 were the associated eigenvectors are the canonical basis of R3. In this case we have two positive eigenvalues and a negative eigenvalue, so we are in a presence of a saddle. Moreover in the {ϵ= 0}subset we see thatF2± are sources.
Therefore the line of fixed is normally hyperbolic and each point on the line is the ω-limit point of a unique interior orbit. So there exists an orbit of the dynamical system (4.84) with ϵ >0 initially that converges to (y0,0,0) for each y0 asτ →+∞.
Convergence of solutions in the case γpf =⟨γϕ⟩
Just as in the proof of the cases (i) and (ii), we can again estimate|η(τ)|and|ζ(τ)|however the estimation is very similar to the ones introduced in (4.79) and (4.80) since the only difference is thatγpf− ⟨γϕ⟩= 0. Asϵn→0 then|η(τ)| →0 and so yand ¯yhave the same limit and the same can be said regarding z and ¯z. Finally, from equation (4.65a), the triangle inequality, and the fact that ϵ → 0 as τ → +∞, it follows from that Ωϕ has the same limit as ¯y and therefore will converge to a point between (0,1). In the case of Σσ using (4.65b), the triangle inequality andϵ→0 asτ →+∞, it follow that Σσ goes to zero. This concludes the proof of the case (iii).
The global profile for our solutions in the (X,Σϕ, T)-plane can be seen in Fig. 4.5. While the qualitative solutions for the (Σ+,Σ−, T)-plane can be found in Fig.4.6
(a)Global picture forγpf= 1. (b)Global picture forγpf=43. (c)Global picture forγpf=32.
Figure 4.5: Qualitative solutions for the scalar field potential V(ϕ) = 14(λϕ)4 in the (X,Σϕ, T)-plane for various matter of state.
Figure 4.6: Qualitative solutions for the scalar field potential V(ϕ) = 14(λϕ)4 in the (Σ+,Σ−, T)-plane forγpf = 43.