gr-qc0202067/text
1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2: %Nature of singularities in anisotropic string cosmology
3: %
4: % AT, ST      
5: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
6: 
7: %\documentstyle[aps,prd,eqsecnum,twocolumn,epsf]{revtex} 
8: \documentstyle[prd,eqsecnum,aps,epsf]{revtex}
9: %\documentclass[draft]{JHEP}
10: %\usepackage{epsfig}
11: %\usepackage{amssymb}
12: %\usepackage{bbm}
13:  
14:  
15: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
16: \newcommand{\beq}{\begin{equation}}
17: \newcommand{\beqn}{\begin{eqnarray}} 
18: \newcommand{\eeq}{\end{equation}}
19: \newcommand{\eeqn}{\end{eqnarray}}
20: \newcommand{\beqa}{\begin{eqnarray}}
21: \newcommand{\eeqa}{\end{eqnarray}}
22: \newcommand{\mpl}{M_p}
23: \newcommand{\k}{{\kappa}}
24: \newcommand{\lmk}{\left(}
25: \newcommand{\rmk}{\right)}
26: \newcommand{\lkk}{\left[}
27: \newcommand{\rkk}{\right]}
28: \newcommand{\lnk}{\left\{}
29: \newcommand{\rnk}{\right\}}
30: \newcommand{\zk}{z_k}
31: \newcommand{\call}{{\cal L}}
32: \newcommand{\calr}{{\cal R}}
33: \newcommand{\half}{\frac{1}{2}}
34: \newcommand{\kc}{\kappa\chi}
35: \newcommand{\bkc}{\beta\kappa\chi}
36: \newcommand{\gkc}{\gamma\kappa\chi}
37: \newcommand{\gbkc}{(\gamma-\beta)\kappa\chi}
38: \newcommand{\dchi}{\delta\chi}
39: \newcommand{\dsigma}{\delta\sigma}
40: \newcommand{\dOmega}{\delta\Omega}
41: \newcommand{\Phibd}{\Phi_{\rm BD}}
42: \newcommand{\echi}{\epsilon_\chi}
43: \newcommand{\esigma}{\epsilon_\sigma}
44: \newcommand{\Phihat}{\hat{\Phi}}
45: \newcommand{\Psihat}{\hat{\Psi}}
46: \newcommand{\ahat}{\hat{a}}
47: \newcommand{\that}{\hat{t}}
48: \newcommand{\Hhat}{\hat{H}}
49: \newcommand{\ab}{q}
50: 
51: \newcommand{\sign}{\mathop{\rm sign}\nolimits}
52: \newcommand{\dett}{\mathop{\rm det}\nolimits}
53: \newcommand{\Ree}{\mathop{\rm Re}\nolimits}
54: \newcommand{\identity}{\mathbbm{1}}
55: \newcommand{\dbar}{\partial \hskip -5.5pt/ }
56: \newcommand{\gsim}{\mbox{\raisebox{-1.ex}{$\stackrel
57:      {\textstyle>}{\textstyle\sim}$}}}
58: \newcommand{\lsim}{\mbox{\raisebox{-1.ex}{$\stackrel
59:      {\textstyle<}{\textstyle \sim}$}}}
60: 
61: %\renewcommand{\theequation}{\thesection.\arabic{equation}}
62: \newcommand{\dddot}[1]{\stackrel{...}{#1}}
63: \newcommand{\pkt}{\; .}
64: \newcommand{\kma}{\; ,}
65: \newcommand{\tr}{{\rm tr \; }}
66: \newcommand{\Tr}{{\rm  Tr  \;}}
67: \newcommand{\re}[1]{{\rm Re }\,#1}
68: \newcommand{\im}[1]{{\rm Im }\,#1}
69: \newcommand{\pslash}{ \slash\hspace{-2.3mm} p}
70: \newcommand{\calF}{{\cal F}}
71: \newcommand{\calC}{{\cal C}}
72: \newcommand{\calS}{{\cal S}}
73: \newcommand{\calE}{{\cal E}}
74: \newcommand{\calL}{{\cal L}}
75: \newcommand{\intk}{\int\frac{d^3k}{(2\pi)^3 2\omega_{k0}}\,}
76: \newcommand{\intkp}{\int\frac{d^3k'}{(2\pi)^3 2\omega_{k'0}}\,}
77: \newcommand{\intp}{\int\frac{d^3p}{(2\pi)^3 2\Ep\,}}
78: \newcommand{\intpp}{\int\frac{d^3p'}{(2\pi)^3 2\Epp\,}}
79: \newcommand{\intppp}{\int\frac{d^3p''}{(2\pi)^3 2\Eppp\,}}
80: \newcommand{\bfk}{{\bf k}}
81: \newcommand{\bfp}{{\bf p}}
82: \newcommand{\bfq}{{\bf q}}
83: \newcommand{\bfx}{{\bf x}}
84: \newcommand{\bgamma}{\mbox{\boldmath{$\gamma$}\unboldmath}}
85: \newcommand{\bsigma}{\mbox{\boldmath{$\sigma$}\unboldmath}}
86: \newcommand{\bSigma}{\mbox{\boldmath{$\Sigma$}\unboldmath}}
87: \newcommand{\be}{\begin{equation}}
88: \newcommand{\ee}{\end{equation}}
89: \newcommand{\bea}{\begin{eqnarray}}
90: \newcommand{\eea}{\end{eqnarray}}
91: \newcommand{\omk}{\omega_{k0}}
92: \newcommand{\omkp}{\omega_{k'0}}
93: \newcommand{\Ep}{E_{p}}
94: \newcommand{\Epp}{E_{p'}}
95: \newcommand{\Eppp}{E_{p''}}
96: 
97: %%%%% singlefig %%%%%
98: \newcommand{\singlefig}[2]{
99: \begin{center}
100: \begin{minipage}{#1}
101: \epsfxsize=#1
102: \epsffile{#2}
103: \end{minipage}
104: \end{center}}
105: %
106: %%%%% figcaption %%%%%
107: \newenvironment{figcaption}[2]{
108:  \vspace{0.3cm}
109:  \refstepcounter{figure}
110:  \label{#1}
111:  \begin{center}
112:  \begin{minipage}{#2}
113:  \begingroup \small FIG. \thefigure: }{
114:  \endgroup
115:  \end{minipage}
116:  \end{center}}
117: %
118: 
119: \begin{document}
120: 
121: \title{Nature of singularities in anisotropic string cosmology} 
122: \author{Alexey Toporensky$^1$ and Shinji Tsujikawa$^2$ } 
123: \address{$^1$ 
124: Sternberg Astronomical Institute, Moscow State University, Universitetsky 
125: Prospekt, 13, Moscow 119899, Russia\\[.3em]} \address{$^2$ Research Center 
126: for the Early Universe, University of Tokyo, Hongo, Bunkyo-ku, Tokyo 
127: 113-0033, Japan\\[.3em]} \date{\today} \maketitle
128: \begin{abstract}
129: We study nature of singularities in anisotropic string-inspired 
130: cosmological models in the presence of a Gauss-Bonnet term.  
131: We analyze two string gravity models-- dilaton-driven and 
132: modulus-driven cases-- in the Bianchi type-I background 
133: without an axion field.  In both 
134: scenarios singularities can be classified in two ways- the determinant 
135: singularity where the main determinant of the system vanishes and the 
136: ordinary singularity where at least one of the anisotropic expansion rates 
137: of the Universe diverges.  In the dilaton case, either of these 
138: singularities inevitably appears during the evolution of the system.  In 
139: the modulus case, nonsingular cosmological solutions exist both in 
140: asymptotic past and future with determinant $D=+\infty$ and $D=2$, 
141: respectively.  In both scenarios nonsingular trajectories in either future 
142: or past typically meet the determinant singularity in past/future when the 
143: solutions are singular, apart from the exceptional case where the sign of 
144: the time-derivative of dilaton is negative.  This implies that the 
145: determinant singularity may play a crucial role to lead to singular 
146: solutions in an anisotropic background.
147: \end{abstract}
148: \vskip 1pc 
149: \pacs{pacs: 98.80.Cq}
150: \vskip 2pc
151: 
152: 
153: \baselineskip = 16pt
154: 
155: %%%%%%%%%%%%%%%%%%%%%%%%%%
156: \section{Introduction}                            
157: %%%%%%%%%%%%%%%%%%%%%%%%%%
158: 
159: Superstring theory continues to be of interest as a possible candidate to 
160: unify all fundamental interactions in nature \cite{GSW}.  
161: It is known that there exist five supersymmetric perturbative string theories 
162: which are classified as the type I, type IIA, type IIB, SO(32) heterotic and 
163: $E_8 \times E_8$ heterotic string.  Recently it was found that these theories
164: are connected by dual symmetries, which leads to the conjecture that each 
165: theory appears as one of five branches of a unified theory, 
166: called M-theory \cite{Witten}. In particular Ho\v{r}ava and 
167: Witten \cite{HW} showed that the 10-dimensional $E_8 \times E_8$ heterotic 
168: string theory is equivalent to an 11-dimensional M-theory compactified 
169: to ${\bf M}^{10} \times {\bf S}^1/Z_2$. 
170:  Then the 10-dimensional spacetime is expected to be 
171: compactified into ${\bf M}^{4}\times {\bf CY}^{6}$, in which case 
172: the standard model particles are confined on the 3-dimensional brane.
173: This gives rise to the well-known brane world scenario \cite{RS} where the 
174: extra dimension is noncompact and gravity is effectively 3-dimensional.
175: 
176: For cosmologists it is very important to test the 
177: viability of string theories by extracting various cosmological 
178: implications from them \cite{review}.  One of such attempts is the 
179: {\it pre-big-bang} (PBB) scenario \cite{PBB} based on the low
180: energy effective action of string theory.
181: In this scenario there exist two branches of solutions by assuming 
182: a $T$-duality.  One of which ($t<0$) corresponds to the 
183: stage of pole-like inflation driven by the kinetic term of the dilaton field, 
184: and another ($t>0$) is the stage where the curvature 
185: continues to decrease.  However it is difficult to smoothly connect these 
186: two branches without singularity in tree-level string action 
187: \cite{tree,Kaloper}.
188: 
189: One is required to take into account quantum loop or derivative corrections
190: in order to overcome such singularity problems. 
191: In fact, Antoniadis, Rizos, 
192: and Tamvakis \cite{oneloop} included a Gauss-Bonnet term to the tree-level 
193: string effective action with dilaton and modulus fields, and showed the 
194: existence of nonsingular cosmological solutions.  
195: In this case nonsingular behavior of solutions is mainly determined by the 
196: evolution of the modulus field.  Therefore
197: allowed ranges of parameters were analyzed in the absence of the dilaton in 
198: the flat Friedmann-Robertson-Walker (FRW) background \cite{RT}
199: (see also ref.~\cite{closed}).  Since it is important to confirm the generality 
200: of singularity avoidance even starting from an anisotropic spacetime, 
201: several authors analyzed nonsingular cosmological solutions in the Bianchi 
202: type-I spacetime without dilaton \cite{KSS,KS} and with dilaton 
203: \cite{yaji}.
204: The presence of the 
205: modulus coupled to the Gauss-Bonnet term allows the existence of 
206: nonsingular solutions unless the dilaton controls the dynamics of the 
207: system.
208: 
209: In order to understand how nonsingular or singular solutions appear,
210: it is necessary to classify nature of singularities in an anisotropic 
211: background.
212: In particular, the main determinant $D$ of the system 
213: is an important quantity to describe the singularities.
214: When only the dilaton field $\phi$ is coupled to Gauss-Bonnet term in 
215: the Bianchi I background, it was conjectured in ref.~\cite{ATU} that 
216: nonsingular cosmological solutions in future crosses the determinant 
217: singularity ($D=0$) in past when $\dot{\phi}$ is positive.  While this 
218: singularity was found more than ten years ago \cite{Nathalie}, only now we 
219: begin to understand its importance.  The similar kind of singularity also 
220: appears in the context of black hole inner solutions in the presence of 
221: dilaton coupled to gravity via the Gauss-Bonnet term \cite{e4,e5}.  In this 
222: paper we shall make detailed analysis about nature of singularities both in 
223: dilaton- and modulus-driven cosmologies in the Bianchi I background. 
224: We do not include an axion field in our analysis, but it is important to 
225: emphasize that its effect is generally vital as studied in Refs.~\cite{axion}.
226: We will classify other kinds of singularities where at least one 
227: expansion rate diverges.  These investigations are important to understand 
228: how nonsingular solutions emerge in the modulus-driven case.  In addition 
229: our analysis will be useful to construct more complicated nonsingular 
230: string-inspired models in the presence of other fields such as axion.
231: 
232: This paper is organized as follows.  In Sec.~II we show background 
233: equations in anisotropic string-inspired models with dilaton or modulus 
234: fields.  In Sec.~III we study nature of singularities in dilaton-driven 
235: cosmology both for positive and negative $\dot{\phi}$ cases.  Sec.~IV 
236: is devoted to the modulus-driven cosmology where both of nonsingular and 
237: singular solutions exist.  We present summary and discussions in the final 
238: section.
239: 
240: 
241: %%%%%%%%%%%%%%%%%%%%%%%%%%
242: \section{The model and background equations}
243: %%%%%%%%%%%%%%%%%%%%%%%%%%
244: 
245: We begin with the action \cite{oneloop,RT,closed,KSS,KS,yaji}, 
246: %
247: \begin{eqnarray}
248:  S = \int d^4 x \sqrt{-g} \left[ \frac12 R -\frac12 (\nabla \phi)^2 +f(\phi) 
249:  R_{\rm GB}^2 \right],
250: \label{lag}
251: \end{eqnarray}
252: %
253: written in the Einstein frame. Here $R$ is the scalar curvature and $\phi$ 
254: denotes a scalar field which is either dilaton or modulus.  
255: $f(\phi)$ depends on string theories, whose explicit 
256: forms are given later.
257: We do not consider the multi-field system of dilaton and modulus 
258: fields \cite{yaji} induced from the one-loop effective action of heterotic 
259: string theory.  In addition we neglect the anti-symmetric tensor 
260: $H_{\mu\nu\lambda}$ and the curvature terms higher than the second order.  
261: The Gauss-Bonnet term, $R_{\rm GB}$, is defined as 
262: %
263: \begin{eqnarray}
264: R_{\rm GB}^2 =R^2-4R^{\mu\nu}R_{\mu\nu}+ 
265: R^{\mu\nu\alpha\beta}R_{\mu\nu\alpha\beta}.
266: \label{gauss}
267: \end{eqnarray}
268: %
269: We normalize time and spatial 
270: coordinates by the string length scale $\sqrt{\lambda_s}$ as 
271: $\bar{x^{\mu}}=x^{\mu}/\sqrt{\lambda_s}$, and the scalar fields as 
272: $\bar{\phi}=\phi \sqrt{\lambda_s}$.  
273: Hereafter we drop bars for simplicity.
274: 
275: Let us consider the Bianchi type-I spacetime whose metric 
276: is given by
277: %
278: \begin{eqnarray}
279: ds^2=-dt^2+a^2(t)dx^2+b^2(t)dy^2+c^2(t)dz^2,
280: \label{metric}
281: \end{eqnarray}
282: %
283: where $a(t), b(t), c(t)$ are the scale factors in an anisotropic background.  
284: We define the anisotropic expansion rates, $p(t), q(t), r(t)$, as 
285: %
286: \begin{eqnarray}
287: p(t)=\frac{\dot{a}}{a}\,,~~~q(t)=\frac{\dot{b}}{b}\,,~~~
288: r(t)=\frac{\dot{c}}{c}\,,
289: \label{Hubble}
290: \end{eqnarray}
291: %
292: where a dot denotes the derivative with respect to $t$.
293: It is also convenient 
294: to introduce new variables $h, \alpha, \beta$\,: 
295: %
296: \begin{eqnarray}
297: p(t)=h+\alpha+\sqrt{3}\beta,~~~ 
298: q(t)=h+\alpha-\sqrt{3}\beta,~~~ 
299: r(t)=h-2\alpha\,.
300: \label{Omega}
301: \end{eqnarray}
302: %
303: Here $h=(p+q+r)/3$ is an average expansion rate, which is the generalization 
304: of the Hubble parameter in an isotropic case, and $\alpha$ and $\beta$ 
305: correspond to parameters of anisotropy.  The triangle in $(\alpha, 
306: \beta)$-plane, 
307: %
308: \begin{eqnarray}
309: \alpha+\sqrt{3}\beta>-h,~~~\alpha-\sqrt{3}\beta>-h,~~~\alpha<h/2\,,
310: \label{triangle}
311: \end{eqnarray}
312: %
313: which extends around the isotropic point $\alpha=\beta=0$, represents 
314: the regions where the Universe expands in all directions.  In the outside 
315: region the Universe is contracting at least in one direction.
316: 
317: The dynamical equations for the background are written as 
318: %
319: \beqa
320: & & (1+8r\dot{f})(\dot{q}+q^{2})
321: +(1+8q\dot{f})(\dot{r}+r^{2})
322: +(1+8\ddot{f})qr
323: +\frac12 \dot{\phi}^2 =0\,,
324: \label{back1}\\
325: & & (1+8r\dot{f})(\dot{p}+p^{2})
326: +(1+8p\dot{f})(\dot{r}+r^{2})
327: +(1+8\ddot{f})rp
328: +\frac12 \dot{\phi}^2 =0\,, 
329: \label{back2} \\
330: & & (1+8q\dot{f})(\dot{p}+p^{2})
331: +(1+8p\dot{f})(\dot{q}+q^{2})
332: +(1+8\ddot{f})pq
333: +\frac12 \dot{\phi}^2 =0\,, 
334: \label{back3} \\
335: & & \ddot{\phi}+(p+q+r)\dot{\phi} -8f' \left\{\dot{p}qr+p\dot{q}r+pq\dot{r} 
336: +pqr(p+q+r)\right\}=0\,,
337: \label{back4} 
338: \eeqa
339: %
340: together with the constraint equation,
341: \beqa
342: pq+qr+rp+24pqr \dot{f}- \frac12 \dot{\phi}^2=0\,.
343: \label{constraint}
344: \eeqa
345: %
346: Defining a 4-dimensional vector, ${\bf x}= (\dot{p}, \dot{q}, \dot{r}, 
347: \ddot{\phi})$, eqs.~(\ref{back1})-(\ref{back4}) can be written in the 
348: matrix form, 
349: %
350: \beqa
351: Z {\bf x}={\bf y}\,,
352: \label{mform}
353: \eeqa
354: %
355: where ${\bf y}={\bf y}(p, q, r, \phi, \dot{\phi})$ and 
356: %
357: \beqa
358:  Z= \left[ {\begin{array}{cccc} 0 & 1 + 8f'\dot\phi r & 1 
359:  + 8f'\dot\phi q 
360:  &8f'qr \\
361: 1+8f'\dot\phi r &0 &1 + 8f'\dot\phi p &8f'rp \\
362: 1 + 8f'\dot\phi q &1 + 8f'\dot\phi p &0 &8f'pq \\
363: -8f'qr & -8f'rp & -8f'pq & 1 \end{array}} \right]\,.
364: \eeqa
365: %
366: The determinant of $Z$ yields
367: %
368: \beqa
369: D=2&+&16f'\dot\phi(p+q+r) -64f'^2 (p^2q^2+q^2r^2+r^2p^2) 
370: +128f'^2pqr(p+q+r) 
371: +128f'^2\dot{\phi}^2 (pq+qr+rp) \nonumber \\
372: &+&1024f'^3\dot{\phi}pqr(pq+qr+rp+\dot{\phi}^2) +12288f'^4\dot{\phi}^2 
373: p^2q^2r^2\,.
374: \label{D} 
375: \eeqa 
376: %
377: In the case of $D \ne 0$, the solutions of eqs.~(\ref{back1})-(\ref{back4}) 
378: are given by ${\bf x}=Z^{-1}{\bf y}$.  
379: When $D$ vanishes, however, we cannot proceed numerical calculations 
380: further.  This ``determinant singularity'' plays an important role in the 
381: anisotropic background \cite{ATU}.
382: 
383: {}From Eq.~(\ref{constraint}) we find the constraint
384: %
385: \beqa
386: \alpha^2+\beta^2 \le h^2+96(pqrf')^2\,, 
387: \label{con2} 
388: \eeqa 
389: %
390: in which case $\dot{\phi}$ is solved as 
391: %
392: \beqa
393: \dot{\phi}=24pqrf' \pm
394: \sqrt{(24pqrf')^2+2(pq+qr+rp)}\,. 
395: \label{con} 
396: \eeqa 
397: %
398: When $f=0$ anisotropy parameters are restricted in the circle, 
399: $\alpha^2+\beta^2 \le h^2$.  If the Gauss-Bonnet term is taken into 
400: account, we have the wider allowed range of anisotropy parameters 
401: given by eq.~(\ref{con2}).
402: 
403: %%%%%%%%%%%%%%%%%%%%%%%%%%
404: \section{Dilaton-driven case}
405: %%%%%%%%%%%%%%%%%%%%%%%%%%
406: 
407: Firstly we consider the dilaton-driven case with 
408: %
409: \beqa
410: f(\phi)=\frac{\lambda}{16}e^{-2\phi},
411: \label{f} 
412: \eeqa 
413: %
414: where the string coupling, $\lambda$, takes a positive value. 
415: We set $\lambda=1$ in our numerical simulations.
416: In the scenario (\ref{f}) nature of singularities was analyzed in 
417: ref.~\cite{ATU} in the Bianchi I background in the case of the plus sign in 
418: the rhs of Eq.~(\ref{con}).  Hereafter we shall make detailed analysis 
419: about the property of singularities in both signs of eq.~(\ref{con}).
420: 
421: The asymptotic behavior of solutions in past and future can be analyzed by 
422: assuming the following power-law forms for the expansion rates: 
423: %
424: \beqa
425: p=c_1 |t|^{s},~~~q=c_2 |t|^{s},~~~r=c_3 |t|^{s}\,.
426: \label{asyexpan}
427: \eeqa 
428: %
429: In order for the $\dot{f}$ term in eqs.~(\ref{back1})-(\ref{back4})
430: to have a power-law dependence, the dilaton is required to take the form 
431: %
432: \beqa
433: \phi=\phi_0+c_4 {\rm ln} |t|\,.
434: \label{phi}
435: \eeqa 
436: %
437: 
438: When the contribution from the Gauss-Bonnet term is negligible,
439: one has $\dot{\phi}^2=2(c_1c_2+c_2c_3+ c_3c_1)|t|^{2s}$ from 
440: eq.~(\ref{constraint}).  Comparing this with 
441: eqs.~(\ref{back1})-(\ref{back4}) and (\ref{phi}), we find 
442: %
443: \beqa
444: s=-1,~~~~c_1+c_2+c_3={\rm sign} (t),~~~~c_1^2+c_2^2+c_3^2+c_4^2=1,~~~~
445: c_4^2=2(c_1c_2+c_2c_3+c_3c_1)\,.
446: \label{coeff}
447: \eeqa 
448: %
449: In the absence of the dilaton ($c_4=0$), 
450: the solution (\ref{coeff}) for $t>0$ represents the vacuum Kasner solution 
451: where the Universe is expanding in two directions and contracting in one 
452: direction.  The interpretation of this solution is that  
453: large anisotropies are required at least in one dimension as $t \to 0$,
454: in order to make the spacetime curved by anisotropies.
455: 
456:  The situation is different when the dilaton is taken into 
457: account.  For example, when $\sqrt{1/2} \le c_4 \le \sqrt{2/3}$ the 
458: Universe is expanding in all directions for $t \to \infty$, while there exist 
459: both Friedmann- and Kasner-type solutions for $0 \le c_4<\sqrt{1/2}$.  
460: Note that in the limit where the Gauss-Bonnet term is 
461: negligible ($f' \to 0$) the determinant approaches a constant value 
462: $D=2$ from eq.~(\ref{D}).
463: 
464: When the Gauss-Bonnet term is dominant in eq.~(\ref{constraint}) 
465: ($|pq+qr+rp| \ll |24pqr \dot{f}|$), one has 
466: $\dot{\phi}=-6\lambda c_1c_2c_3 |t|^{3s}e^{-2\phi}$ by using 
467: (\ref{asyexpan}). Integrating this equation with respect to $t$,
468: we easily find that $\dot{\phi}={\rm sign}(t)\,(3s+1)/(2|t|)$.
469: Combining this with eqs.~(\ref{back1})-(\ref{back4}) gives 
470: %
471: \beqa
472: s=-2,~~~~c_4=-\frac52,~~~~
473: c_1c_2c_3={\rm sign}(t) \frac{5e^{2\phi_0}}{12\lambda}\,.
474: \label{coeff2}
475: \eeqa 
476: %
477: Since $c_1c_2c_3<0$ for $t<0$, the Universe is either contracting in all 
478: directions or expanding in two directions and contracting in one direction.  
479: When $t>0$, the Universe is either expanding in all directions or expanding 
480: in one direction and contracting in two directions.  For the asymptotic 
481: solution (\ref{coeff2}), we have that $\dot{\phi} \propto |t|^{-1}$ and $f' 
482: \propto |t|^5$, in which case the determinant is given by $D \propto |t|^6 
483: \to \infty$ for $|t| \to \infty$.  In spite of this divergent behavior of the 
484: determinant, the solutions are nonsingular with $p, q, r \propto |t|^{-2} 
485: \to 0$ for $|t| \to \infty$.
486: 
487: We shall classify the cases where the solutions of 
488: eqs.~(\ref{back1})-(\ref{back4}) exhibit singular behavior.  
489: When the system passes through the determinant singularity ($D=0$), 
490: eq.~(\ref{mform}) indicates that 
491: ${\bf x}=(\dot{p}, \dot{q}, \dot{r}, \ddot{\phi})$ diverge.  
492: This singularity appears in an anisotropic 
493:  background where three expansion rates are multiple-valued functions
494:  of time \cite{KS}. It is also a physical singurality where the 
495: curvature invariant, $R^{\mu\nu\alpha\beta}
496: R_{\mu\nu\alpha\beta}$, diverges due to the divergence of 
497: the time derivative of the expansion rates. 
498: Near the determinant singularity, the expansion rates and the scalar 
499: field can be expanded as \cite{ATU} 
500: %
501: \beqa
502: h_i &=& h_{is}+h_{i1}\sqrt{|t-t_s|}+h_{i2}(\sqrt{|t-t_s|})^2+ \cdots,  
503: \label{expans2} \\
504: \phi &=& \phi_s+\phi_1 (\sqrt{|t-t_s|})^2+\phi_2 (\sqrt{|t-t_s|})^3+
505: \cdots,
506: \label{expans} 
507: \eeqa 
508: %
509: where $h_i=p, q, r~~(i=1, 2, 3)$, and $t_s$ is the time at singularity. 
510: This means that $\dot{h}_i$ and $\ddot{\phi}$ diverge as $t \to t_s$, 
511: while $p, q, r, \dot{\phi}$ are finite.  This property is different from the 
512: ordinary kind of singularity where $p, q, r$ do not stay finite.
513: The determinant singularity plays a crucial role
514: in an anisotropic background.
515: 
516: The ordinary kind of singularities can be classified as 
517: %
518: \beqa
519: &{\rm (i)}&~p \sim p_0/(t-t_s),~~~q \sim q_0\,,~~~
520: r \sim r_0\,,
521: \label{ordi}
522:  \\ 
523: &{\rm (ii)}&~p \sim p_0/(t-t_s),~~~q \sim p_0/(t-t_s)\,,~~~
524: r \sim r_0\,,
525: \label{ordii}
526: \\
527: &{\rm (iii)}&~p \sim p_0/(t-t_s),~~~q \sim p_0/(t-t_s)\,,~~~
528: r \sim p_0/(t-t_s)\,,
529: \label{ord} 
530: \eeqa 
531: %  
532: where $p_0, q_0$, and  $r_0$ are constants with $p_0>0$.  
533: If the time-direction is futurewards 
534: ($t-t_s \to -0$) one has $p \to -\infty$, while $p \to \infty$ for $t-t_s 
535: \to +0$.  The asymptotic forms of the determinant $(\ref{D})$
536: depend upon the cases presented above.
537: 
538: In the case (i) with a plus sign in rhs of Eq.~(\ref{con}), the signs of 
539: $q_0$ and $r_0$ are the same and the asymptotic form of $\dot{\phi}$ is 
540: given by $\dot{\phi} \sim (q_0+r_0)/(3\lambda e^{-2\phi_s}q_0r_0)$ for 
541: $t-t_s \to +0$, with $\phi_s$ being a constant.  Then the determinant $D$ 
542: yields from eq.~(\ref{D}): 
543: %
544: \beqa
545: D &\sim& -\frac{4\lambda^2 e^{-4\phi_s}p_0^2}{3(t-t_s)^2}  \left[ 
546: \left(q_0-\frac{r_0}{2}\right)^2+\frac34 r_0^2 \right]~\to~-\infty~~~~({\rm 
547: for }~~~q_0r_0>0)\,.
548: \label{Dinf} 
549: \eeqa 
550: % 
551: In the minus sign of eq.~(\ref{con}), one has $D \to -\infty$ 
552: for $q_0r_0<0$ and $t-t_s \to +0$.
553: 
554: In the case (ii) with a plus sign of of eq.~(\ref{con}), one has 
555: $\dot{\phi} \sim 1/(3\lambda r_0 e^{-2\phi_s})$ for $r_0>0$ and $t-t_s \to 
556: +0$.  Then the determinant should asymptotically takes the form 
557: %
558: \beqa
559: D &\sim& -\frac{4\lambda^2p_0^4 e^{-4\phi_s}}{3(t-t_s)^4} 
560: ~\to~-\infty~~~~({\rm for }~~~r_0>0)\,.
561: \label{Din2} 
562: \eeqa 
563: % 
564: In numerical analysis we did not find this case for the parameter ranges and 
565: initial conditions we adopt.  Typically ordinary singularities are dominated 
566: by the cases where one or three expansion rates tend to diverge. 
567: When one chooses the minus sign of eq.~(\ref{con}), the 
568: asymptotic behavior is $D \to -\infty$ for $r_0<0$ and $t-t_s \to +0$.
569: 
570: In the case (iii) with a plus sign of of eq.~(\ref{con}), we find
571: $\dot{\phi} \sim t/(\lambda p_0 e^{-2\phi_s})$ and 
572: %
573: \beqa
574: D~\to~0\,,
575: \label{Din3}
576: \eeqa 
577: % 
578: for $t-t_s \to +0$.  Strictly speaking this holds only for the isotropic case 
579: ($\alpha=\beta=0$) where all expansion rates are the same.  In this case 
580: the solutions do not cross the determinant singularity, although they 
581: approach $D=0$ as $t-t_s \to +0$.  
582: When small anisotropies are included, the 
583: trajectories can pass through $D=0$.  
584: For the minus sign of of eq.~(\ref{con}) 
585: the asymptotic behavior of the determinant is not described by 
586: eq.~(\ref{Din3}), as we will see later.
587: 
588: In another limit, $t-t_s \to -0$, the signs of diverging 
589: expansion rates in eqs.~(\ref{ordi})-(\ref{ord}) are reversed, in which 
590: case the asymptotic forms of $D$ are altered.  Nevertheless the determinant 
591: of the cases (i) and (ii) generally approaches the asymptotic value, 
592: $D=-\infty$.  We shall confirm this by numerical investigations in
593: subsequent sections.
594: 
595: 
596: %%%%%%%%%%%%%%%%%%%%%%%%%%
597: \subsection{Plus sign of eq.~(\ref{con})}
598: %%%%%%%%%%%%%%%%%%%%%%%%%%
599: 
600: We first analyze the case of the plus sign in eq.~(\ref{con}).  When $\phi$ 
601: is largely positive, the term $f'=-(\lambda/8)e^{-2\phi}$ is negligible in 
602: (\ref{con}), implying that $\dot{\phi}$ is positive as long as 
603: $pq+qr+rp>0$.  In this case $\phi$ increases toward the future, which 
604: results in nonsingular asymptotic solutions with determinant, $D \approx 
605: 2$.  When we go back to the past, there are two possibilities for the 
606: evolution of the determinant.  One is the case where the solutions pass 
607: through the determinant singularity ($D=0$) and another is the one where 
608: $D$ goes toward infinity ($D \to +\infty$).  Our numerical investigations 
609: suggest that the latter case does {\em not} occur for the plus sign in 
610: eq.~(\ref{con}), which means that solutions nonsingular in future ($D \to 
611: 2$) meet the determinant singularity in past, irrespective of the initial 
612: values of $\phi$.
613: 
614: In the left panel of Fig.~\ref{diladotppevo} (a) we plot the evolution of the 
615: expansion rates for the anisotropy parameters $\alpha=0.05, \beta=0.05$ 
616: with initial conditions $h=0.16$ and $\phi=0$, corresponding to $D \approx 
617: 1.7$ at $t=0$.  The determinant continues to grow until it approaches the 
618: finite value $D=2$ as $t \to \infty$, which indicates that this solution 
619: belongs to the nonsingular solution in future given by eq.~(\ref{coeff}).  
620: Note that for $t>0$ the Universe is expanding in all directions.  When we 
621: solve the equations of motion pastwards ($t<0$), the solution meets the 
622: determinant singularity around $t=-0.82$, thereby leading to the divergence 
623: of $\dot{p}, \dot{q}, \dot{r}$ and $\ddot{\phi}$.  
624: If we introduce a new time parameter, $\tau$, defined by 
625: %
626: \beqa
627: \tau \equiv \int \frac{dt}{D},
628: \label{tau} 
629: \eeqa 
630: % 
631: it becomes possible to enter the region of the negative 
632: sign of $D$ by overpassing the determinant 
633: singularity \cite{KS}.  This does not mean that 
634: we can remove the singularity  by coordinate 
635: transformations. The determinant singularity is a physical 
636: one where the divergence of the curvature invariant is 
637: unavoidable even in other coordinates. 
638: Then the solution turns back futurewards and the determinant begins to decrease 
639: rapidly toward $D \to -\infty$ [see the right panel of 
640: Fig.~\ref{diladotppevo} (a)].  From Fig.~\ref{diladotppevo} (a) we find 
641: that this belongs to the class of the case (i) with $p \sim p_0, q \sim 
642: q_0, r \sim r_0/(t-t_s)$ and $p_0q_0>0$ [see eq.~(\ref{Dinf})].  The 
643: Universe is rapidly contracting in one direction ($r \to -\infty$) for 
644: $t-t_s \to -0$.  As claimed in ref.~\cite{KS}, these trajectories can be 
645: understood as the pair creation of two branches ($D>0$ and $D<0$) at the 
646: determinant singularity.  For the nonsingular solutions in future, it is 
647: inevitable to cross the determinant singularity in past in an anisotropic 
648: background.  Notice that in the isotropic case $D$ is always positive and 
649: decreases toward zero as $t \to -\infty$.  In the presence of small 
650: anisotropies, however, the solution reaches the determinant singularity at 
651: finite past and fall into the $D=-\infty$ singularity as shown in 
652: Fig.~\ref{diladotppevo} (a).
653: 
654: When $\phi$ is large, the allowed anisotropy parameters in 
655: $(\alpha, \beta)$-plane lie inside a circle, given by 
656: %
657: \beqa
658: \alpha^2+\beta^2~\lsim~h^2,
659: \label{circle} 
660: \eeqa 
661: %
662: which comes from the constraint (\ref{con2}). 
663: In Fig.~\ref{dphi_p} (a) we show nature of singularities in past and 
664: future in the $(\alpha, \beta)$-plane for $h=0.16$ and $\phi=0$ at $t=0$.  In 
665: this case the allowed region is approximately described by 
666: eq.~(\ref{circle}).  We find that all solutions inside this region exhibit 
667: the determinant singularity in past (region II in Fig.~\ref{dphi_p}) while 
668: they are nonsingular ($D \to 2$) in future (region I) .
669: 
670: With the decrease of $\phi$, the allowed region gets larger as found by 
671: eq.~(\ref{con2}).  Nature of singularities becomes more complicated due to 
672: the appearance of ordinary singularities.  If the system is close to the 
673: isotropic case ($|\alpha|, |\beta| \ll 1$), one has $D \to 2$ for $t \to \infty$ 
674: and $D \to 0$ at finite past as in the case of Fig.~\ref{dphi_p} (a).  
675: However, larger anisotropies alter this picture.  For example, we show the 
676: evolution of the expansion rates in the left panel 
677: of Fig.~\ref{diladotppevo} (b) for $\alpha=0.15$ and $\beta=0.05$ 
678: with $h=0.16$ and $\phi=-1$ at $t=0$.  In this case $p$ is singular at finite 
679: past while $q, r$ are positive constants, which means that $D \to -\infty$ 
680: from eq.~(\ref{Dinf}).  The solution exhibits the same kind of singularity 
681: ($D \to -\infty$) in future.  In this case we have 
682: numerically found the determinant is always negative ($D=-1.02$ at $t=0$) 
683: and the solution never crosses $D=0$ in both past and future [see the right 
684: panel of Fig.~\ref{diladotppevo} (b)].  In Fig.~\ref{dphi_p} (b) the 
685: asymptotic property of singularities is presented for $h=0.16$ and 
686: $\phi=-1$ at $t=0$.  There exists the region III where the determinant is 
687: singular ($D \to -\infty$) in both past and future.  We also find that the 
688: region IV with $D \to +\infty$ appears in past when anisotropy parameters 
689: are large.  This corresponds to asymptotic past nonsingular solutions given 
690: by eq.~(\ref{coeff2}) where quadratic curvature corrections are dominant.  
691: It is important to separate this case from the ordinary singularity with 
692: $D=-\infty$, although this classification was not done in ref.~\cite{ATU}.  
693: {}From Fig.~\ref{dphi_p} (b) the solutions with $D=0$ in future correspond 
694: to, in past, the determinant singularity or the regular solutions with $D 
695: \to +\infty$.  Notice that the parameter range of future nonsingular 
696: solutions are smaller compared to the case of Fig.~\ref{dphi_p} (a).
697: 
698: In Fig.~\ref{dphi_p} (c) we show the density plot for $h=0.16$ and $\phi=-2$ at 
699: $t=0$.  This suggests that future nonsingular trajectories are restricted to 
700: be very narrow near the isotropic point, $\alpha=\beta=0$.  We also find 
701: that the determinant singularity in future corresponds to the regular 
702: solutions with $D \to +\infty$ in past.  As one example we plot in 
703: Fig.~\ref{diladotppevo} (c) the evolution of $p, q, r$ for $h=0.16$ and 
704: $\phi=-2$ at $t=0$ with anisotropy parameters, $\alpha=0.1$, and 
705: $\beta=0.1$.  In this case the past asymptotic solution is categorized in 
706: the nonsingular solution given by eq.~(\ref{coeff2}).  Fig.~\ref{dphi_p} 
707: (c) shows that two expansion rates $p, q$ are positive while $r$ is 
708: negative for $t<0$, implying $c_1c_2c_3<0$ in eq.~(\ref{coeff2}).  The 
709: solution comes regularly from the asymptotic past with $D$ being decreased 
710: toward the future [see the right panel of Fig.~\ref{diladotppevo} (c)].  
711: It crosses the determinant singularity around $t=0.3$, after which the 
712: determinant continues to decrease until the solution falls into the 
713: ordinary singularity with $D \to -\infty$.  Note that $p$ diverges as $t-t_s 
714: \to +0$, while $q$ and $r$ approach positive constant values, in which case 
715: one has $D \to -\infty$ by eq.~(\ref{Dinf}).  This trajectory can be 
716: regarded as the pair annihilation of two branches with $D>0$ and $D<0$.  
717: {}From Fig.~\ref{dphi_p} (b) and (c), the solutions nonsingular in past ($D 
718: \to +\infty$) meet the determinant singularity in future.  In this case the 
719: solutions do not go futurewards beyond the determinant singularity.  In 
720: Fig.~\ref{dphi_p} (c) the parameter regions with ordinary singularity in 
721: future ($D=-\infty$) typically correspond to those with the same 
722: singularity in past, in which case the asymptotic behavior of the expansion 
723: rates are the similar as in Fig.~\ref{diladotppevo} (b).  Notice that in 
724: Fig.~\ref{dphi_p} (c) there exist some parameter ranges where the solutions 
725: meet the determinant singularity in both past and future.  Although this 
726: case is very rare, it is still possible to cross the determinant singularity 
727: twice.
728: 
729: {}From Fig.~\ref{dphi_p} we find that nonsingular solutions
730: in future are of the determinant-type in past.
731: This can be understood that the determinant evolves from $D=2$ to $D=0$, 
732: if we solve the equations of motions from asymptotic future to past
733: [see the right panel of Fig.~\ref{diladotppevo} (a)].  
734: We also analyzed other cases varying the values of $h$ and $\phi$ for 
735: $10^{-5}<h<1$ and $-15<\phi<15$.  For a fixed value of $h$
736: there exists a minimal value of $\phi$ which leads to nonsingular solutions 
737: in future [see Fig.~\ref{fh}].  The allowed range of the dilaton for 
738: future nonsingular solutions gets wider with the decrease of $h$ as shown 
739: in Fig.~\ref{fh}.  In all cases analyzed in our numerical simulations, the 
740: past singularity for future nonsingular solutions corresponds to the 
741: determinant-type in an anisotropic background.  We also found that 
742: trajectories with nonsingular past asymptotic meet the determinant 
743: singularity in future.  
744: In addition we showed that the ordinary singularity with (\ref{ordi}) appears, 
745: in which case the determinant is divergent as $D \to -\infty$.  This is 
746: different from the case $D \to +\infty$ where the solutions are nonsingular 
747: in asymptotic past.  In the next subsection we will analyze how the 
748: behavior of the determinant is altered for the different sign in 
749: eq.~(\ref{con}).
750: 
751: 
752: %%%%%%%%%%
753: \begin{figure}
754: \begin{center}
755: \singlefig{14cm}{diladotppevo.eps}
756: \begin{figcaption}{diladotppevo}{14cm}
757: The evolution of $p$, $q$, $r$, $\phi$, and $D$ in the dilaton-driven case 
758: for the plus sign in eq.~(\ref{con}).  Each figure corresponds to initial 
759: conditions with $h=0.16$ and (a)~$\phi=0$, $\alpha=0.05$, $\beta=0.05$, 
760: (b)~$\phi=-1$, $\alpha=0.15$, $\beta=0.05$, (c)~$\phi=-2$, $\alpha=0.1$, 
761: $\beta=0.1$, respectively.  The determinant singularity ($D=0$) can be 
762: passed through by introducing a new time parameter, $\tau=\int dt/D$.
763: \end{figcaption}
764: \end{center}
765: \end{figure}
766: %%%%%%%%%%
767:  
768: %%%%%%%%%%
769: \begin{figure}
770: \begin{center}
771: \singlefig{15cm}{dphi_p.eps}
772: \begin{figcaption}{dphi_p}{15cm}
773: Nature of singularities in $(\alpha, \beta)$-plane with initial 
774: conditions, (a)~$\phi=0$, (b)~$\phi=-1$, and (c)~$\phi=-2$ for the plus 
775: sign in eq.~(\ref{con}).  The left figures correspond to past solutions 
776: while right figures correspond to future solutions.  Each region 
777: corresponds to (I) nonsingular solutions with $D \to 2$, (II) singular 
778: solutions with determinant singularity $(D=0)$, (III) singular solutions 
779: with $D \to -\infty$, (IV) nonsingular solutions with $D \to +\infty$, 
780: respectively.  The black color indicates prohibited regions in the initial 
781: condition space.
782: \end{figcaption}
783: \end{center}
784: \end{figure}
785: %%%%%%%%%%
786: 
787: %%%%%%%%%%
788: \begin{figure}
789: \begin{center}
790: \singlefig{10cm}{fh.eps}
791: \begin{figcaption}{fh}{10cm}
792: The minimal initial values of $\phi$ which allow solutions nonsingular in 
793: future in the dilaton-driven case for the plus sign in eq.~(\ref{con}).
794: \end{figcaption}
795: \end{center}
796: \end{figure}
797: %%%%%%%%%% 
798:  
799: 
800: %%%%%%%%%%%%%%%%%%%%%%%%%%
801: \subsection{Minus sign of eq.~(\ref{con})}
802: %%%%%%%%%%%%%%%%%%%%%%%%%%
803: 
804: In the previous subsection solutions nonsingular  
805: in future meet the determinant singularity in past, due to the fact that the 
806: determinant evolves from the asymptotic value $D=2$ toward 
807: the determinant singularity.  
808: For the minus sign of eq.~(\ref{con}), the determinant can increase 
809: toward the past, implying that the trajectories may not cross the determinant 
810: singularity.  In fact we will show that the past singularity is not 
811: necessarily of the determinant-type.
812: 
813: Let us first analyze the asymptotic behavior of the determinant when all 
814: expansion rates diverge as eq.~$(\ref{ord})$.  In eq.~(\ref{con}) when the 
815: $2(pq+qr+rp)$ term is positive and much larger than $(24pqrf')^2$, one has 
816: $\dot{\phi} \sim -\sqrt{2(pq+qr+rp)} \sim -\sqrt{6}p_0/(t-t_s)$.  Therefore 
817: the dilaton grows as $\phi \sim -\sqrt{6} p_0 {\ln}~(t-t_s)$ for 
818: $t-t_s \to +0$, in which case the asymptotic form of the determinant yields 
819: %
820: \beqa
821: D \sim 2 &+& 6\sqrt{6}\lambda p_0^2 (t-t_s)^{2(\sqrt{6}p_0-1)} 
822: +39\lambda^2 p_0^4 (t-t_s)^{4(\sqrt{6}p_0-1)} \nonumber \\
823: &+& 18 \sqrt{6} \lambda^3 p_0^6 (t-t_s)^{6(\sqrt{6}p_0-1)}
824: +18 \lambda^4 p_0^6 (t-t_s)^{8(\sqrt{6}p_0-1)}.
825: \label{Dnapprox1}
826: \eeqa 
827: % 
828: This indicates that the determinant grows infinitely ($D \to +\infty$)
829: for $p_0<1/\sqrt{6}$, while it decreases toward a finite value $D=2$ for 
830: $p_0>1/\sqrt{6}$.  The latter case corresponds to the one where the average 
831: expansion rates are large initially.  
832: 
833: When  the condition $(24pqrf')^2 \gg |2(pq+qr+rp)|$
834: is satisfied, the asymptotic form of $\dot{\phi}$ is given by 
835: $\dot{\phi} \sim -6\lambda p_0^3e^{-2\phi}/(t-t_s)^3$, 
836: which yields $e^{2\phi} \sim 6\lambda p_0^3/(t-t_s)^2$.  
837: Then one has the following 
838: positive finite value of the determinant in the limit of 
839: $t-t_s \to +0$: 
840: %
841: \beqa
842: D \sim 2+\frac{13}{12p_0^2}+\frac{7}{36p_0^4}+
843:  \frac{5}{432p_0^6}.
844: \label{Dnapprox2}
845: \eeqa 
846: % 
847: Whether the asymptotic form is given by (\ref{Dnapprox1}) 
848: or (\ref{Dnapprox2}) depends on the initial values of the expansion rates 
849: and the dilaton.
850: In the latter case $\phi$ is generally small so that the 
851: condition $(24pqrf')^2 \gg 2(pq+qr+rp)$ is fulfilled.  
852: In both cases we can expect that the trajectories do not cross 
853: the determinant singularity in past.
854: 
855: We show in Fig.~\ref{dphi_n} the density plot of the asymptotic behavior of 
856: the determinant for $h=0.16$ at $t=0$.  
857: When $\phi=0$ at $t=0$ all trajectories satisfying the constraint (\ref{con2}) 
858: are nonsingular in future with $D=2$.  In this case the past solutions are 
859: dominated by the singularity with asymptotic positive determinant 
860: as found from the left panel of Fig.~\ref{dphi_n} (a).  
861: As one example we plot in Fig.~\ref{diladotpnevo} (a) the evolution 
862: of $p$, $q$, $r$, $\phi$, and $D$ for $\phi=0$ at $t=0$ 
863: with anisotropy parameters $\alpha=0.05$ and 
864: $\beta=0.05$.  This belongs to the class (\ref{Dnapprox1}) with 
865: $\alpha<1/\sqrt{6}$ where all expansion rates diverge with determinant 
866: $D \to +\infty$.  The determinant is always positive and continues to grow 
867: pastwards.  The left panel of Fig.~\ref{dphi_n} (a) shows that there exist 
868: some past solutions which meet the determinant singularity when
869: anisotropies parameters are large.  In this case the determinant grows until 
870: some moment of time pastwards, after which it begins to decrease toward the 
871: determinant singularity.  This behavior can be understood that large 
872: anisotropies prevent all expansion rates from evolving almost similarly as 
873: eq.~(\ref{ord}).  However this region is typically small for the 
874: positive values of $\phi$ at $t=0$.
875: 
876: With the decrease of the initial $\phi$, the constraint (\ref{con2}) gives 
877: wider allowed parameter ranges in $(\alpha, \beta)$-plane.  Let us 
878: consider the case $\phi=-1$ and $h=0.16$ at $t=0$.  As is found from 
879: Fig.~\ref{dphi_n} (b) we have additional regions with the $D =-\infty$ 
880: singularity in both past and future.  Nature of singularities around the 
881: isotropic point is similar to the $\phi=0$ case explained above.  For the 
882: solutions nonsingular in future, the past singularity is either the type of 
883: $D>0$ or $D=0$.  Fig.~\ref{diladotpnevo} (b) is the latter case
884: where the trajectory comes regularly from the asymptotic future ($D=2$) and 
885: meets the determinant singularity in past.  This evolution is similar to 
886: the case of large anisotropy parameters in Fig.~\ref{dphi_n} (a).  {}From 
887: Fig.~\ref{dphi_n} (b) the solutions singular in future ($D=-\infty$) 
888: correspond to, in past, the singularity with $D=-\infty$ or the determinant 
889: singularity.  The difference between two cases is whether the determinant 
890: is always negative [like Fig.~\ref{diladotppevo} (b)] or it passes through 
891: $D=0$.
892: 
893: The situation becomes somewhat different with $\phi$ being decreased further. 
894: Fig.~\ref{dphi_n} (c) indicates that nonsingular trajectories with $D 
895: \to +\infty$ appear in future around the isotropic point for 
896: $\phi~\lsim~-2$ at $t=0$.  
897: This is the case all expansion rates are finite as described by 
898: eq.~(\ref{coeff2}). However these solutions are singular in past with all 
899: expansion rates being infinite.  In the region V shown in 
900: Fig.~\ref{dphi_n} (c), the condition $(24pqrf')^2> |2(pq+qr+rp)|$ is typically 
901: satisfied at $t=0$.  Therefore the determinant tends to approach the finite 
902: value (\ref{Dnapprox2}) in past.  
903: We show in Fig.~\ref{diladotpnevo} (c) the evolution of the system for 
904: $\alpha=-0.02$, $\beta=0.02$, $\phi=-2$, and $h=0.16$.
905: While the evolution of the expansion rates 
906: in Fig.~\ref{diladotpnevo} (c) looks similar as in Fig.~\ref{diladotpnevo} 
907: (a), the behavior of the determinant is different in both past and future.  
908: In Fig.~\ref{diladotpnevo} (c) the determinant decreases from infinity to a 
909: finite positive value toward the past.  We have numerically found that two 
910: terms $(24pqrf')^2$ and $2(pq+qr+rp)$ become comparable during the 
911: evolution.  Therefore the asymptotic value of $D$ is somewhat different 
912: from eq.~(\ref{Dnapprox2}), but it is still a finite positive value.  We 
913: find from Fig.~\ref{dphi_n} (c) that future nonsingular solutions with either 
914: $D=2$ or $D=+\infty$ correspond to the singularity (\ref{ord}) with 
915: positive determinant in past.  Figs.~\ref{dphi_n} (b) and (c) also suggest 
916: that as $\phi$ decreases the ordinary singularity with $D \to -\infty$ 
917: appears in both past and future for large anisotropy parameters,
918: 
919: Compared to the case of the previous subsection, the range of nonsingular 
920: solutions in future is not so narrow due to the presence of nonsingular 
921: trajectories with $D \to +\infty$. 
922: However this solution is not appropriate to lead to our present Universe
923: due to the dominance of the quadratic curvature term.
924: In addition these solutions typically approach the ordinary 
925: singularity (\ref{ord}) with all expansion rates infinite in past.  
926: This property is different from the plus 
927: sign of eq.~(\ref{con}) where the past singularity for solutions 
928: nonsingular in future corresponds to the determinant-type.  In both cases 
929: we have found that nonsingular cosmological solutions in both past and 
930: future do not exist for the dilaton-driven case for wide ranges of the 
931: parameter space ($10^{-5}<h<1$ and $-15<\phi<15$).  However the 
932: situation is changed in the modulus-driven case as we will analyze in the 
933: next section.
934: 
935: %%%%%%%%%%
936: \begin{figure}
937: \begin{center}
938: \singlefig{14cm}{diladotpnevo.eps}
939: \begin{figcaption}{diladotpnevo}{14cm}
940: The evolution of $p$, $q$, $r$, $\phi$, and $D$ in the dilaton-driven case 
941: for the minus sign in eq.~(\ref{con}).  Each figure corresponds to initial 
942: conditions with $h=0.16$ and (a)~$\phi=0$, $\alpha=0.05$, $\beta=0.05$, 
943: (b)~$\phi=-1$, $\alpha=0.15$, $\beta=0.05$, (c)~$\phi=-2$, $\alpha=-0.02$, 
944: $\beta=0.02$, respectively.  
945: \end{figcaption}
946: \end{center}
947: \end{figure}
948: %%%%%%%%%%
949: 
950: %%%%%%%%%%
951: \begin{figure}
952: \begin{center}
953: \singlefig{13cm}{dphi_n.eps}
954: \begin{figcaption}{dphi_n}{13cm}
955: Nature of singularities in $(\alpha, \beta)$-plane with initial 
956: conditions, (a)~$\phi=0$, (b)~$\phi=-1$, and (c)~$\phi=-2$ for the minus 
957: sign in eq.~(\ref{con}).  The left figures correspond to past solutions 
958: while right figures correspond to future solutions.  Each region 
959: corresponds to (I) nonsingular solutions with $D \to 2$, (II) singular 
960: solutions with determinant singularity $(D=0)$, (III) singular solutions 
961: with $D \to -\infty$, (IV) nonsingular solutions with $D \to \infty$, (V) 
962: singular solutions with positive determinant, respectively.  The black 
963: color indicates prohibited regions in the initial condition space.
964: \end{figcaption}
965: \end{center}
966: \end{figure}
967: %%%%%%%%%%
968: 
969: 
970: %%%%%%%%%%%%%%%%%%%%%%%%%%
971: \section{Modulus-driven case}
972: %%%%%%%%%%%%%%%%%%%%%%%%%%
973: 
974: In the modulus-driven case the function $f(\phi)$ in eq.~(\ref{lag}) 
975: is expressed as \cite{oneloop,RT,KSS,KS,yaji} 
976: %
977: \begin{eqnarray}
978: f(\phi) = -\frac{1}{16}\delta \xi (\phi)\,,
979: \label{fmodu}
980: \end{eqnarray}
981: %
982: where the coefficient $\delta$ 
983: is determined by the 4-dimensional trace anomaly of the $N=2$ sector.  
984: Here the function $\xi(\phi)$ is defined by 
985: %
986: \begin{eqnarray}
987: \xi(\phi)={\rm ln} 
988: \left[2e^{\phi} \eta^4 (ie^{\phi}) \right]\,,
989: \label{xi}
990: \end{eqnarray}
991: %
992: where $\eta (ie^{\phi})$ is the Dedekind $\eta$ function.
993: Since $\xi(\phi)$ is well approximated as
994: $\xi(\phi) \simeq -(2\pi/3) {\rm cos h} \phi$ \cite{yaji}, 
995: the function $f$ takes the form 
996: %
997: \begin{eqnarray}
998: f(\phi) \simeq \frac{\pi \delta}{48}\left(e^{\phi}+e^{-\phi}\right)\,.
999: \label{fmodu2}
1000: \end{eqnarray}
1001: %
1002: When $\phi$ is largely negative ($|\phi| \gg 1$), 
1003: eq.~(\ref{fmodu2}) reduces to the form (\ref{f})
1004: by setting $\delta=3\lambda/\pi$.
1005: Therefore when $\delta>0$ solutions nonsingular in both past and future do 
1006: not exist in this case. However nonsingular cosmological solutions have been 
1007: found for negative values of $\delta$ \cite{oneloop}.  Hereafter we shall 
1008: focus on the negative $\delta$ case (setting $\delta=-1$ for simplicity) using 
1009: the function $f(\phi)$ given by (\ref{fmodu2}).
1010: 
1011: When the Gauss-Bonnet term dominates in eqs.~(\ref{back1})-(\ref{back4}), 
1012: the asymptotic solution is similarly given as eq.~(\ref{coeff2}), i.e., 
1013: %
1014: \begin{eqnarray}
1015: p=c_1 |t|^{-2},~~~q=c_2 |t|^{-2},~~~r=c_3 |t|^{-2},~~~
1016: \phi=\phi_0 \pm 5 {\rm ln} |t|,~~~
1017: c_1c_2c_3={\rm sign}(t) \frac{5e^{2\phi_0}}{4\pi \delta},
1018: \label{moduco1}
1019: \end{eqnarray}
1020: %
1021: The past asymptotic solutions correspond to $c_1c_2c_3>0$ for negative 
1022: $\delta$.  This sign is different from the dilaton-driven case with 
1023: positive $\lambda$ which we already analyzed in the previous section. 
1024: Note that the determinant is divergent ($D \to +\infty$) for $|t| \to \infty$.
1025: 
1026: For another asymptotic solution where the effect of the Gauss-Bonnet term 
1027: is negligible, the evolution of the background is given by
1028: %
1029: \begin{eqnarray}
1030: & & p=c_1 |t|^{-1},~~~q=c_2 |t|^{-1},~~~r=c_3 |t|^{-1},~~~
1031: \phi=\phi_0 +c_4{\rm ln} |t|, \nonumber \\
1032: & & c_1+c_2+c_3={\rm sign} (t),~~~~c_1^2+c_2^2+c_3^2+c_4^2=1,~~~~
1033: c_4^2=2(c_1c_2+c_2c_3+c_3c_1)\,.
1034: \label{moduco2}
1035: \end{eqnarray}
1036: %
1037: Therefore one has $f' \propto |t|^{|c_4|}$, $\dot{\phi}  \propto |t|^{-1}$,
1038: and $p, q, r \propto |t|^{-1}$ with $|c_4|<1$ in eq.~(\ref{D}), in which case
1039: the solutions are nonsingular with an asymptotic value of the 
1040: determinant, $D =2$. 
1041: 
1042: For the nonsingular cosmological solutions found in 
1043: ref.~\cite{oneloop}, $\dot{\phi}$ does not change
1044: its sign \cite{RT}.  Due to the symmetric structure of the function (\ref{fmodu2})
1045: with respect to $\phi=0$, we will consider the positive $\dot{\phi}$ case where 
1046: the modulus continues to grow from past to future.  For negative 
1047: $\dot{\phi}$ the analysis is essentially the same by changing $\phi$ to $-\phi$.
1048: When the solutions are singular, they meet the determinant 
1049: singularity [see eqs.~(\ref{expans2}) and (\ref{expans})]  or the 
1050: ordinary singularity [see eqs.~(\ref{ordi})-(\ref{ord})].
1051: Let us consider the asymptotic behavior of the determinant 
1052: for the ordinary singularity. 
1053: In the cases (\ref{ordi})-(\ref{ord}),
1054: possible asymptotic behavior of the determinant can be 
1055: summarized as
1056: %
1057: \beqa
1058: &{\rm Case (i)}&~~~ D \sim -\frac{256{f_s'}^2p_0^2}
1059: {3(t-t_s)^2}\left[\left(q_0-\frac{r_0}{2}\right)^2+
1060: \frac34 r_0^2 \right] \to 
1061: -\infty,~~~{\rm with}~~~\dot{\phi} \sim -\frac{1}{24{f_s'}}
1062: \frac{q_0+r_0}{q_0r_0} \sim {\rm const} \,,
1063: \label{mord1}
1064:  \\ 
1065: &{\rm  Case (ii)}&~~~D \sim -\frac{256{f_s'}^2p_0^4}
1066: {3(t-t_s)^4} \to -\infty,~~~{\rm with}~~~\dot{\phi} \sim 
1067: -\frac{1}{24r_0{f_s'}} \sim {\rm const} \,,
1068: \label{mord2}
1069: \\
1070: &{\rm Case (iii)}&~~~ D \to 0,~~~\dot{\phi} \sim 
1071: -\frac{t-t_s}{8p_0{f_s'}},
1072: \label{mor3} 
1073: \eeqa 
1074: %  
1075: where $f_s' \equiv (\pi \delta/48)(e^{\phi_s}-e^{-\phi_s})$ with 
1076: $\phi_s$ being a constant. 
1077: Although the case (ii) is rare, we have numerically checked that 
1078: this asymptotic solution certainly exists for the modulus-driven 
1079: case. The asymptotic (iii) corresponds to the isotropic case, as is similar to 
1080: the dilaton-driven cosmology.  It is easy to verify that this trajectory is 
1081: possible for positive $\phi$ but impossible for negative $\phi$.  In what 
1082: follows we shall analyze how the evolution of the expansion rates and the 
1083: determinant are different from that of the dilaton-driven case paying 
1084: particular attention for nonsingular trajectories.
1085: 
1086: Nonsingular asymptotic solutions for $\dot{\phi}>0$ can be
1087: described as $\phi=\phi_0-5 {\rm ln} |t|$ in past [see 
1088: eq.~(\ref{moduco1})] and $\phi=\phi_0+c_4{\rm ln} |t|$ 
1089: with $0<c_4<1$ in future [see eq.~(\ref{moduco2})]. 
1090: Therefore $\phi$ continues to grow from asymptotic 
1091: past starting from large negative values of $\phi$ toward 
1092: the future. Such examples are plotted in Fig.~\ref{moduevo}
1093: (a) and (b).
1094: In these cases the determinant evolves from $D=+\infty$
1095: (past) to $D=2$ (future) without passing through the 
1096: determinant singularity.
1097: In the dilaton-driven case when the past trajectories are nonsingular 
1098: with $D=+\infty$ they inevitably meet the determinant singularity
1099: in future [see Fig.~\ref{dphi_p} (b) and (c)].
1100: This is mainly due to the fact that $f'$
1101: is negative in the dilaton case while its sign is different in the modulus 
1102: case for $\delta<0$ and $\phi<0$. Therefore the determinant (\ref{D}) is 
1103: dominated by positive terms in the modulus case, 
1104: which provides a way not to pass through the determinant singularity.
1105: Namely negative $\delta$ is crucial for the existence of nonsingular solutions.
1106: 
1107: In Fig.~\ref{modu} we show past and future asymptotic properties 
1108: in three different cases.  Note that we have defined $\tilde{\alpha} \equiv 
1109: \alpha/h$ and $\tilde{\beta} \equiv \beta/h$ in order to compare the cases 
1110: where the average expansion rate $h$ is changed.  When $h=0.05$ and 
1111: $\phi=-5$, nonsingular solutions in past ($D=+\infty$) are not singular 
1112: in future with determinant, $D \to 2$ [see Fig.~\ref{modu} (a)].  The 
1113: Universe exhibits superinflation with growing expansion rates until the 
1114: graceful exit around $t=0$ [see Fig.~\ref{moduevo} (a)].  Notice that we 
1115: have $pqr>0$ in asymptotic past, as predicted by eq.~(\ref{moduco1}) 
1116: for negative $\delta$.  The expansion rates begin to decrease after the 
1117: graceful exit, whose asymptotic solutions in future are given by 
1118: eq.~(\ref{moduco2}).  In Fig.~\ref{moduevo} (a) we find that the future 
1119: solution corresponds to the Kasner-type where the Universe is contracting in 
1120: one direction.
1121: 
1122: Fig.~\ref{modu} (a) indicates that some trajectories which are nonsingular 
1123: in future cross the determinant singularity in past.
1124: The evolution of the background is similar to Fig.~\ref{diladotppevo} (a)
1125: which we already analyzed in the dilaton case.
1126: We also find from Fig.~\ref{modu} (a) that when anisotropy 
1127: parameters are large the solutions meet the ordinary 
1128: singularity with $D=-\infty$ in both past and future.
1129: This is the case of (\ref{ordi}) or (\ref{ordii})
1130: where at least one expansion rate diverges as plotted in  
1131: Fig.~\ref{diladotppevo} (b). 
1132: Although in Fig.~\ref{modu} past nonsingular solutions ($D \to +\infty$)
1133: do not meet the determinant singularity in future, we 
1134: have checked that this singular behavior occurs 
1135: for smaller values of $\phi$ as shown in Fig.~\ref{diladotppevo} (c).
1136: These results imply that the property of singularities is similar to the 
1137: dilaton-driven case described in Sec.~III A.
1138: 
1139: For larger initial values of $\phi$, the allowed region can be 
1140: approximately described as $\tilde{\alpha}^2+\tilde{\beta}^2~\lsim~1$.  When 
1141: $h=0.05$ and $\phi=2$ shown in Fig.~\ref{modu} (b), all future asymptotic 
1142: solutions are nonsingular with determinant, $D \to 2$.  However we find 
1143: from the left panel of Fig.~\ref{modu} (b) that the region of past 
1144: nonsingular solutions gets smaller relative to Fig.~\ref{modu} (a).  For 
1145: large anisotropy parameters, the determinant tends to decrease pastwards 
1146: from asymptotic future with $D=2$, thereby resulting in the singularity 
1147: with $D=0$.  This is not the case for the nonsingular trajectories with 
1148: small anisotropy parameters.  One example is plotted in Fig.~\ref{moduevo} 
1149: (b).  Although the determinant decreases from $t=+\infty$ to $t \sim -6$, 
1150: it begins to grow toward the asymptotic past before crossing the determinant 
1151: singularity.  {}From Fig.~\ref{moduevo} (b) we find that the trajectory 
1152: comes regularly from asymptotic past and connects to the Friedmann-type 
1153: branch where the Universe is expanding in all directions.  When anisotropy 
1154: parameters are small and belong to the region (\ref{triangle}), the future 
1155: solution is of the Friedmann-type.
1156: 
1157: 
1158: %%%%%%%%%%
1159: \begin{figure}
1160: \begin{center}
1161: \singlefig{14cm}{moduevo.eps}
1162: \begin{figcaption}{moduevo}{14cm}
1163: The evolution of $p$, $q$, $r$, $\phi$, and $D$ in the modulus-driven 
1164: case.  Each figure corresponds to initial conditions with (a)~$\phi=-5$, 
1165: $h=0.05$, $\tilde{\alpha} \equiv \alpha/h=-0.5$, $\tilde{\beta}
1166: \equiv \beta/h=0.5$, (b)~$\phi=2$, $h=0.05$, $\tilde{\alpha}=0.3$, 
1167: $\tilde{\beta}=-0.2$, (c)~$\phi=2$, $h=0.1$, $\tilde{\alpha}=0.5$, 
1168: $\tilde{\beta}=-0.5$, respectively.
1169: \end{figcaption}
1170: \end{center}
1171: \end{figure}
1172: %%%%%%%%%%
1173:  
1174: %%%%%%%%%%
1175: \begin{figure}
1176: \begin{center}
1177: \singlefig{12cm}{modu.eps}
1178: \begin{figcaption}{modu}{12cm}
1179: Nature of nonsingular and singular solutions in $(\tilde{\alpha}, 
1180: \tilde{\beta})$-plane with initial conditions, (a)~$h=0.05$, $\phi=-5$, 
1181: (b)~$h=0.05$, $\phi=2$, and (c)~$h=0.1$, $\phi=2$.  The left figures 
1182: correspond to past solutions while right figures correspond to future 
1183: solutions.  Each regions correspond to (I) nonsingular solutions with $D 
1184: \to 2$, (II) singular solutions with determinant singularity $(D=0)$, (III) 
1185: singular solutions with $D \to -\infty$, (IV) nonsingular solutions with $D 
1186: \to \infty$, respectively.  The black color indicates prohibited regions in 
1187: the initial condition space.
1188: \end{figcaption}
1189: \end{center}
1190: \end{figure}
1191: %%%%%%%%%%
1192: 
1193: %%%%%%%%%%
1194: \begin{figure}
1195: \begin{center}
1196: \singlefig{10cm}{modufh.eps}
1197: \begin{figcaption}{modufh}{10cm}
1198: The regions of nonsingular and singular solutions for the modulus-driven 
1199: case in the initial condition space of $h$ and $\phi$.
1200: \end{figcaption}
1201: \end{center}
1202: \end{figure}
1203: %%%%%%%%%% 
1204: 
1205: When the average expansion rate $h$ gets larger, the region of nonsingular
1206: cosmological solutions becomes smaller.  We plot in Fig.~\ref{modu} (c) 
1207: the density plot of nature of the determinant for $h=0.1$ and $\phi=2$.  
1208: The past solutions are dominated by the determinant singularity, whereas 
1209: the future solutions are nonsingular as is similar to Fig.~\ref{modu} (b).  
1210: In this case nonsingular trajectories in both past and future exist in only 
1211: small parameter ranges around the isotropic point.
1212: 
1213: It is also worth mentioning that the ordinary singularity (\ref{ord}) with 
1214: asymptotic determinant (\ref{Dnapprox1}) and (\ref{Dnapprox2})
1215: for the minus sign of eq.~(\ref{con}) does not appear in the present case.
1216: When $\phi$ increases toward the past, $|f'|$ also grows in the modulus
1217: case. Therefore the condition $(24pqrf')^2 \ll |2(pq+qr+rp)|$ is not satisfied,
1218: implying that case of (\ref{Dnapprox1}) does not occur. 
1219: When $(24pqrf')^2 \gg |2(pq+qr+rp)|$, there are two possibilities
1220: for the asymptotic form of $\dot{\phi}$, one of which is  
1221: $\dot{\phi} \simeq \pi \delta (e^{\phi}-e^{-\phi})
1222: p_0^3/(t-t_s)^3$
1223: and another is $\dot{\phi} \simeq C(t-t_s)$ where $C$ is a constant.
1224: In the former case it is easy to show that asymptotic solutions do not 
1225: exist by integrating $\dot{\phi} \simeq \pi \delta
1226: (e^{\phi}-e^{-\phi})p_0^3/(t-t_s)^3$ 
1227: [Note that this is possible in the dilaton-driven case
1228: with determinant (\ref{Dnapprox2})]. 
1229: In the latter case the determinant
1230: approaches $D=0$ as in eq.~(\ref{mor3}).
1231: 
1232: We have done numerical simulations for other cases varying the values of
1233: $h$ and $\phi$.  When $h$ is large, $\phi$ is required to be small for the 
1234: existence of nonsingular cosmological solutions.  This property is found 
1235: in Fig.~\ref{modufh} where we plot the regions of nonsingular and 
1236: singular solutions in the $(h, \phi)$-plane.  
1237: Nonsingular trajectories come regularly from asymptotic past with $D=+\infty$ 
1238: where the quadratic curvature term is dominant [see eq.~(\ref{moduco1})], 
1239: and smoothly connect another nonsingular branch with $D=2$ where the 
1240: Gauss-Bonnet term is negligible [see eq.~(\ref{moduco2})]. 
1241:  For singular 
1242: solutions nature of singularities is found to be similar to that of the 
1243: dilaton-driven case discussed in Sec.~III A.
1244: 
1245:  
1246: %%%%%%%%%%%%%%%%%%%%%%%%%%
1247: \section{Summary and discussions}
1248: %%%%%%%%%%%%%%%%%%%%%%%%%%
1249: 
1250: In this paper we have analyzed past and future asymptotic regimes in 
1251: Bianchi type-I string-inspired cosmological models in the presence of a 
1252: Gauss-Bonnet curvature invariant.  If the loop correction is not taken into 
1253: account, one has no-go results that the initial big-bang singularity can 
1254: not be avoided.  The Gauss-Bonnet term allows the existence of nonsingular 
1255: cosmological solutions, depending on the theories we adopt.  We 
1256: investigated two gravity theories, viz., dilaton- and modulus-driven 
1257: cosmologies.  In the former case the dynamics appears to depend 
1258: significantly on the sign of rhs of eq.~(\ref{con}).  Hence, we treated 
1259: possible signs in eq.~(\ref{con}) separately and constructed three pictures 
1260: of cosmological evolution:
1261: \begin{itemize}
1262: \item[(a)] Dilaton-driven cosmology with plus sign in eq.~(\ref{con})
1263: \item[(b)] Dilaton-driven cosmology with minus sign in eq.~(\ref{con})
1264: \item[(c)] Modulus-driven cosmology with $\delta < 0$\,.
1265: \end{itemize}
1266: 
1267: As the quadratic curvature corrections may provide violations of strong
1268: and week energy conditions \cite{KS,ATU}, the Bianchi type-I 
1269: Universe can recollapse in high-curvature regime.  
1270: This fact gives us a variety of 
1271: possible types of trajectories with different past and future asymptotics.  
1272: Some trajectories can not leave high-curvature regime, some reaches the 
1273: low-curvature future attractor.
1274: 
1275: In the dilaton-driven cosmology the 4-dimensional string coupling, $\lambda$,
1276: is required to be positive.  This forbiddens the existence 
1277: of nonsingular cosmological solutions even when the Gauss-Bonnet 
1278: term is taken into account.  In the case (a) nonsingular trajectories in 
1279: future correspond to the low-curvature solutions with determinant 
1280: $D \to 2$ where the Gauss-Bonnet term is negligible.
1281: These trajectories meet the singularity where the determinant of the system 
1282: vanishes at finite past.  At this determinant singularity, $\dot{p}, 
1283: \dot{q}, \dot{r}, \ddot{\phi}$ diverge in eq.~(\ref{mform}) while $p, q, r, 
1284: \dot{\phi}$ stay finite.  This kind of singularities restricts the presence 
1285: of nonsingular solutions in an anisotropic background.  There exist other 
1286: kind of singularities (we call ordinary singularities) where at least one 
1287: expansion rate diverges [see eqs.~(\ref{ordi})-(\ref{ord})].  When the 
1288: singularity (\ref{ordi}) or (\ref{ordii}) appears at finite past or future, 
1289: the determinant approaches $D=-\infty$.  We also find some trajectories 
1290: which are nonsingular in asymptotic past with determinant, $D \to +\infty$.  
1291: These solutions appear for large anisotropy parameters, whose existence, to 
1292: our knowledge, was not discovered previously.  In the case (a) 
1293: these solutions are found to meet the determinant singularity in future 
1294: (see Fig.~\ref{dphi_p}).
1295: In the case (b) there appears the 
1296: ordinary singularity of eq.~(\ref{ord}) where all expansion rates diverge.  
1297: In this case the determinant is divergent or approaches a positive constant 
1298: value, depending on the initial conditions of $\phi$ and $h$ [see 
1299: (\ref{Dnapprox1}) and (\ref{Dnapprox2})].  The solutions nonsingular in 
1300: future typically correspond to this type of singularity or the determinant 
1301: singularity in past as found by Fig.~\ref{dphi_n}.
1302: 
1303: In the modulus-driven case the coupling, $\delta$, can take either
1304: positive or negative value in eq.~(\ref{fmodu2}).  When $\delta>0$
1305: solutions nonsingular in both past and future ar not found as is similar 
1306: to the dilaton-driven case.  For negative values of $\delta$, however,
1307: there exist nonsingular cosmological solutions where two branches
1308: of superinflation and decreasing curvature can be joined to each other.
1309: In this case the solutions come regularly from asymptotic past
1310: ($D=+\infty$) with determinant being decreased.  The determinant 
1311: approaches the future asymptotic value $D=2$ without crossing the 
1312: determinant singularity.  This is the main difference from the 
1313: dilaton-driven case where the past nonsingular trajectories inevitably 
1314: meet the determinant singularity in future.
1315: When the solutions are singular we find that nature of singularities
1316: is similar to the dilaton-driven case discussed in Sec.~III A.
1317: 
1318: Our numerical investigations show that not all combinations of past and future 
1319: asymptotics are possible.  We pay particular attention to nonsingular past 
1320: and low-curvature nonsingular future regimes (the latter should correspond 
1321: to our present Universe).  For these regimes we get the following results:
1322: \begin{itemize}
1323: \item For the dilaton-driven case trajectories with nonsingular past 
1324: asymptotic meet the determinant singularity in future. 
1325: \item For cases (a) and (c) trajectories with low-curvature future asymptotic 
1326: meet the determinant singularity in past, when the past solutions are singular.
1327: \end{itemize}
1328: 
1329: The first property tells us that the negative sign of the coupling constant
1330: is essentially important for constructing a purely nonsingular string 
1331: cosmological models.  This result is known for FRW Universe 
1332: \cite{oneloop,RT} and, hence, is still valid in anisotropic Bianchi I case, 
1333: despite the existence of a past nonsingular regime in the dilaton-driven 
1334: cosmology.  The second property indicates that the determinant singularity 
1335: may play a crucial role when we try to trace back in time evolution of the 
1336: our present Universe in anisotropic background.
1337: 
1338: Recently string-inspired cosmological models which can avoid the big bang 
1339: singularity have received much attention \cite{BM,FMS,CCM,BEM,ATU2} 
1340: together with the proposal of the Ekpyrotic universe \cite{KOST}.  
1341: In those cases the 
1342: quantum loop corrections or the higher-order derivatives play important 
1343: roles to determine the dynamics before the graceful exit.  It is certainly 
1344: of interest to extend our analysis to more complicated models such as the 
1345: multi-field case in the presence of the Gauss-Bonnet term.  In fact while 
1346: there exist nonsingular solutions in the single-field modulus-driven 
1347: case considered in this work, density perturbations generated by the 
1348: fluctuation of modulus exhibits blue-spectra with a spectral tilt $n=10/3$ 
1349: \cite{KS2,shinji_ekp}.  This contradicts with the observational supported 
1350: flat spectra with $n \simeq 1$.  In the multi-field case, however, it may 
1351: be possible to produce almost scale-invariant spectra if a light scalar 
1352: field such as axion generates a flat isocurvature perturbations during 
1353: superinflation \cite{LW,ES,MT} or if the axion is nonminimally coupled to 
1354: the dilaton with some potential \cite{FB}.  We leave to future work to 
1355: construct such nonsingular cosmological models which are consistent with 
1356: observations.
1357: 
1358: 
1359: %%%%%%%%%%%%%%%%%%%%%%%%%%%%
1360: \section*{ACKNOWLEDGEMENTS}
1361: We thank S. Alexeyev and H. Yajima for useful discussions.  The work of AT is 
1362: supported by Russian Foundation for Basic Researches via grants Ns.  
1363: 00-15-96699 and 02-02-16817.  ST is thankful for financial support from the 
1364: JSPS (No.  04942).  
1365: %%%%%%%%%%%%%%%%%%%%%%%%%%%%
1366: 
1367: 
1368: %%%%%%%%%
1369: % references
1370: %%%%%%%%%
1371: 
1372: \begin{thebibliography}{99}
1373: 
1374: %superstring theory
1375: \bibitem{GSW}
1376: M.  B.  Green, J.  H.  Schwarz, and E.  Witten,
1377: {\it Superstring Theory} (Cambridge University Press,
1378: Cambridge, 1987).
1379: 
1380: \bibitem{Witten}
1381: E.  Witten,
1382: Nucl.  Phys. {\bf B 443}, 85 (1995).
1383: 
1384: \bibitem{HW}
1385: P. Ho\v{r}ava and E. Witten, Nucl. Phys. {\bf B460}, 506 (1996).
1386: 
1387: \bibitem{RS}
1388: L. Randall and R. Sundrum, Phys. Rev. Lett. {\bf 83},  4690 (1999).
1389: 
1390: %superstring cosmology
1391: \bibitem{review} 
1392: J. E. Lidsey, D. Wands and E. J. Copeland, 
1393: %Superstring cosmology
1394: Phys.  Rep.  {\bf 337}, 343 (2000).
1395: 
1396: \bibitem{PBB} 
1397: G. Veneziano, 
1398: %Scale factor duality for classical and quantum strings
1399: Phys.  Lett.  B {\bf 265}, 287 (1991); 
1400: M.  Gasperini and G.  Veneziano, 
1401: %Pre big bang in string cosmology,
1402: Astropart.  Phys.  {\bf 1}, 317 (1993); Mod.  
1403: Phys.  Lett.  {\bf A 8}, 3701 (1993).
1404: 
1405: \bibitem{tree}
1406: R.  Easther, K.  Maeda, and D.  Wands, 
1407: %Tree-level String Cosmology
1408: Phys.  Rev.  D {\bf 53}, 4247 (1996).
1409: 
1410: \bibitem{Kaloper}
1411:  N.  Kaloper, R.  Madden, and K.  A.  Olive, 
1412: %Axions and the Graceful Exit Problem in String Cosmology
1413: Nucl.  Phys.  {\bf B452}, 677 (1995);
1414: Phys.  Lett.  B {\bf 371}, 34 (1996).
1415: 
1416: \bibitem{oneloop}
1417: I.  Antoniadis, J.  Rizos, and K.  Tamvakis, 
1418: %Singularity-free cosmological solutions of the superstring effective action
1419: Nucl.  Phys.  {\bf B415}, 497 (1994).
1420: 
1421: \bibitem{RT}
1422: J.  Rizos and K.  Tamvakis, 
1423: Phys.  Lett.  B {\bf 326}, 57 (1994).
1424: 
1425: \bibitem{closed}
1426: R.  Easther and K.  Maeda, 
1427: %One-Loop Superstring Cosmology and the Non-Singular Universe
1428: Phys.  Rev.  D {\bf 54}, 7252 (1996).
1429: 
1430: \bibitem{KSS}
1431: S.  Kawai, M.  Sakagami, and J.  Soda,
1432: %Instability of 1-loop superstring cosmology
1433: Phys.  Lett.  B {\bf 437}, 284 (1998).
1434: 
1435: \bibitem{KS}
1436: S.  Kawai and J.  Soda, 
1437: %Nonsingular Bianchi type I cosmological solutions
1438: %from 1-loop superstring effective action
1439: Phys. Rev. D {\bf 59}, 063506 (1999).
1440: 
1441: \bibitem{yaji}
1442: H.  Yajima, K.  Maeda, and H.  Ohkubo, 
1443: %Generality of Singularity Avoidance in Superstring Theory: 
1444: %Anisotropic Case
1445: Phys. Rev. D {\bf 62}, 024020 (2000).
1446: 
1447: \bibitem{ATU}
1448: S. Alexeyev, A. Toporensky, and V. Ustiansky,
1449: Phys. Lett. B {\bf 509}, 151 (2001).
1450: %The nature of singularities in Bianchi I cosmological string gravity 
1451: %model with second order curvature corrections
1452: 
1453: \bibitem{Nathalie}
1454: N.  Deruelle and J.  Madore,
1455: %On the Vanishing of the Cosmological Constant
1456: Phys.  Lett.  {\bf 114A}, 185 (1986);
1457: N.  Deruelle, 
1458: %ON THE APPROACH TO THE 
1459: %COSMOLOGICAL SINGULARITY IN QUADRATIC 
1460: %THEORIES OF GRAVITY: THE KASNER REGIMES.  
1461: Nucl.  Phys.  {\bf B327}, 253 (1989); 
1462: N.  Deruelle and P. L. Farina-Busto, 
1463: %The Lovelock Gravitational Field Equations in Cosmology 
1464: Phys.  Rev. D {\bf 41}, 2214 (1990); 
1465: N.  Deruelle and P.  Spindel, 
1466: %KALUZA-KLEIN BIANCHI I COSMOLOGIES
1467: %IN THE R**2 THEORY OF GRAVITY.
1468: Class. Quant. Grav. {\bf 7}, 1599 (1990);
1469: M.  Demianski, Z. Golda, and W. Puszkarz,
1470: %DYNAMICS OF THE D-DIMENSIONAL FRW 
1471: %COSMOLOGICAL MODELS WITHIN THE 
1472: %SUPERSTRING GENERATED GRAVITY MODEL.
1473: Gen.  Rel.  Grav.  {\bf 23}, 917 (1991).
1474: 
1475: \bibitem{e4}
1476: P. Kanti, N. E. Mavromatos, J. Rizos, K. Tamvakis,
1477: and E. Winstanley,
1478: Phys.  Rev.  D {\bf 54}, 5049 (1996); D {\bf 57}, 6255 (1998);
1479: T. Torii, H. Yajima and K. Maeda,
1480: Phys. Rev. D {\bf 55}, 739 (1997).
1481: 
1482: \bibitem{e5}
1483: S. O. Alexeyev and M. V. Pomazanov,
1484: Phys.  Rev.  D {\bf 55}, 2110 (1997);
1485: S. O. Alexeyev and M. V. Sazhin,
1486: Gen. Relativ. and Grav. {\bf 30}, 1187 (1998).
1487: 
1488: \bibitem{axion}
1489: R. R. Metsaev and A. A. Tseytlin,
1490: Nucl. Phys. {\bf B293}, 385 (1987);     
1491: E.  J.  Copeland, A. Lahiri, D. Wands, 
1492: Phys.  Rev.  D {\bf 50}, 4868 (1994); D {\bf 50}, 1569 (1995);
1493: J.  D.  Barrow and K. E.  Kunze, 
1494: Phys.  Rev.  D {\bf 55}, 623 (1997);
1495: J.  D.  Barrow and M.  P.  Dabrowski,  
1496: Phys.  Rev.  D {\bf 55}, 630 (1997);
1497: E.  J.  Copeland, R.  Easther, and
1498: D.  Wands, Phys.  Rev.  D {\bf 56}, 874 (1997).
1499: 
1500: 
1501: \bibitem{BM}
1502: R. Brustein and R. Madden, 
1503: %A MODEL OF GRACEFUL EXIT IN STRING COSMOLOGY 
1504: Phys. Rev. D {\bf 57}, 712 (1998). 
1505: 
1506: \bibitem{FMS}
1507: S.  Foffa, M .Maggiore, and R. Sturani,
1508: %Loop corrections and graceful exit in string cosmology
1509: Nucl.  Phys.  {\bf 552}, 395 (1999).
1510: 
1511: \bibitem{CCM}
1512: C.  Cartier, E.  J.  Copeland, and R.  Madden, 
1513: %THE GRACEFUL EXIT IN STRING COSMOLOGY 
1514: JHEP {\bf 01}, 035 (2000);
1515: C.  Cartier, E.  J.  Copeland, and M.  Gasperini, 
1516: %Gravitational waves in non-singular string cosmologies
1517: Nucl.  Phys. {\bf B607}, 406 (2001).
1518: 
1519: \bibitem{BEM}
1520: R. H.  Brandenberger, R. Easther, and J. Maia, 
1521: %NONSINGULAR DILATON COSMOLOGY 
1522: JHEP {\bf 9808}, 007 (1998); 
1523: D. A. Easson and R. H.  Brandenberger,
1524: %NONSINGULAR DILATON COSMOLOGY 
1525: %in the String Frame
1526: JHEP {\bf 9909}, 003 (1999).
1527: 
1528: \bibitem{ATU2}
1529: S. Alexeyev, A. Toporensky, and V. Ustiansky,
1530: Class.  Quant.  Grav.  {\bf 17}, 2243 (2000).
1531: 
1532: 
1533: %% Ekpyrotic universe
1534: 
1535: \bibitem{KOST}
1536: J.  Khoury, B.  A.  Ovrut, P.  J.  Steinhardt, and N.  Turok,
1537: %The Ekpyrotic Universe: Colliding Branes and the Origin 
1538: %of the Hot Big Bang
1539: Phys. Rev. D {\bf 64}, 123522 (2001); hep-th/0109050.
1540: 
1541: \bibitem{KS2}
1542: S. Kawai and J. Soda, 
1543: %Evolution of Fluctuations during Graceful Exit in String Cosmology
1544: Phys.  Lett.  B {\bf 460}, 41 (1999).
1545: 
1546: \bibitem{shinji_ekp}
1547: S. Tsujikawa, 
1548: %Density perturbations in the Ekpyrotic Universe and string-inspired 
1549: %generalizations
1550: Phys.  Lett.  B {\bf 526}, 179 (2002).
1551: 
1552: \bibitem{LW}
1553: D.  H.  Lyth and D.  Wands,
1554: %Generating the curvature perturbation without an inflaton
1555: Phys.  Lett.  B {\bf 524}, 5 (2002).
1556: 
1557: \bibitem{ES}
1558: E. Enqvist and M. S. Sloth, hep-ph/0109214. 
1559:  
1560: \bibitem{MT}
1561: T. Moroi and T. Takahashi, 
1562: Phys.  Lett.  B {\bf 522}, 215 (2001).
1563: 
1564: \bibitem{FB}
1565: F.  Finelli and R. H. Brandenberger, 
1566: %On the Generation of a Scale-Invariant Spectrum of Adiabatic Fluctuations 
1567: % in Cosmological Models with a Contracting Phase,
1568: hep-th/0112249.
1569: 
1570: 
1571: \end{thebibliography}
1572: 
1573: \end{document}
1574:  
1575: 
1576: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1577: %%% figures (uu-files)%%%%%%%%
1578: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
1579: