gr-qc0304056/le.tex
1: \documentstyle[aps,prd,eqsecnum]{revtex}
2: \input psfig.sty
3: 
4: \begin{document}
5: %\voffset 1.0truein
6: %\hoffset -0.5truein
7: 
8: \title{Lyapunov timescales and black hole binaries}
9: \author{Neil J. Cornish${}^{*}$ and Janna Levin${}^{**}$}
10: \address{${}^{*}$ Department of Physics, Montana State University, Bozeman, MT 59717}
11: \address{${}^{**}$ DAMTP, Cambridge University,Wilberforce Rd., Cambridge CB3 0WA }
12: 
13: 
14: 
15: \twocolumn[\hsize\textwidth\columnwidth\hsize\csname
16:            @twocolumnfalse\endcsname
17: 
18: \maketitle
19: \widetext
20: 
21: \begin{abstract}
22: 
23: Black holes binaries support unstable orbits
24: at very close separations. In the simplest case of geodesics around
25: a Schwarzschild black hole the orbits, though
26: unstable, are regular. Under perturbation the unstable orbits
27: can become the locus of chaos.
28: All unstable orbits, whether regular or chaotic, can be quantified by
29: their Lyapunov exponents.
30: The exponents are
31: observationally relevant since the phase of gravitational
32: waves can decohere in a Lyapunov time.
33: If the timescale for dissipation due to gravitational waves is shorter
34: than the Lyapunov time, chaos will be damped and essentially unobservable.
35: We find the two timescales can be comparable.
36: We emphasize that the Lyapunov exponents must only be used
37: cautiously for several reasons: they are relative and depend on the
38: coordinate system used, 
39: they vary from orbit to orbit, and finally they can be deceptively
40: diluted by transient behaviour for orbits which pass in and out of
41: unstable regions.
42: 
43: 
44: \end{abstract}
45: 
46: \medskip
47: \noindent{04.30.Db,97.60.Lf,97.60.Jd,95.30.Sf,04.70.Bw,05.45}
48: \medskip
49: ]
50: 
51: 
52: \narrowtext
53:  
54: \setcounter{section}{1}
55: 
56: Newtonian gravity predicts the elliptical planetary orbits 
57: around the sun which Kepler
58: described. Einstein gravity predicts precessing
59: elliptical orbits around a central star, thereby reconciling the 
60: precession of the perihelion of Mercury with general relativity.
61: In the extreme case of a central black hole, there are also 
62: a simple set of {\it unstable} circular orbits 
63: in addition to the usual stable circular orbits. 
64: Related to these are the homoclinic orbits which lie on the boundary
65: between dynamical stability and instability \cite{{bc},{loc}}.
66: To the list of possible orbits, a set of chaotic
67: orbits has recently been added for rapidly
68: spinning black holes \cite{{sm},{me1},{me2}} (and 
69: for the interesting but less physically realistic
70: Majumdar-Papapetrou black hole pairs of equal mass and charge
71: \cite{{bhs},{frank},{cf},{meglow}}).  
72: With the future gravitational wave experiments 
73: LIGO and LISA we hope to see these innermost orbits and 
74: reconstruct a map of spacetime around black holes.
75: 
76: The instability of the innermost orbits around black holes will etch
77: certain landmarks in a gravitational wave map.
78: Gravitational waveforms of neighboring orbits
79: will dechohere in a time scale set by the instability \cite{njc1}.
80: If the timescale for dissipation through gravitational radiation is faster
81: than the instability timescale, then chaos will be damped and the 
82: gravitational wave signal will not observably decohere.
83: 
84: The simple set of unstable circular orbits around a Schwarzschild
85: black hole are a consequence of the
86: nonlinearity of general relativity. 
87: Their instability can be quantified
88: by a positive Lyapunov exponent \cite{njc1}.
89: Although Lyapunov exponents are often associated with chaotic dynamics,
90: the geodesics around a Schwarzschild black hole are not chaotic:
91: the orbits are fully soluble and therefore integrable.
92: However,
93: under perturbation, chaos is likely to develop along the unstable
94: circular and homoclinic orbits.
95: An example of this has been found when
96: the black holes spin.
97: The nonlinearity degenerates to a nonintegrability and chaos
98: \cite{{sm},{me1},{me2}}. 
99: The number of unstable periodic orbits 
100: proliferates so that they have to pack themselves into a fractal
101: in order to crowd into that region of phase \cite{{frank},{me2}}.
102: The unstable orbits will have positive Lyapunov exponents 
103: \cite{{rs},{usl}} and will emerge as fractals in phase space \cite{{me1},{me2}}.
104: 
105: The Lyapunov exponents, while a seemingly useful tool, have uncomfortable
106: shortcomings in the context of general relativity.
107: Firstly, the Lyapunov exponents vary from orbit to orbit and so do not
108: have the surveying power to scan the collective
109: behaviour of all orbits that
110: fractals methods do.
111: Secondly, the Lyapunov exponents are a measure of the deviation of two
112: neighboring orbits in time and therefore overtly depend on the time coordinate
113: used. Since time is relative so too are the Lyapunov exponents.
114: The relativity of the Lyapunov exponents has been known to erroneously lead 
115: to zero Lyapunov exponents for truly
116: chaotic systems \cite{{barrow},{book},{mixm},{sk}}.
117: Importantly, topological measures of chaos such as 
118: fractals are coordinate invariant and are not plagued by the relativism
119: of space and time
120: \cite{{frank},{mixm}}.
121: 
122: 
123: In rare cases when there is a prefered time direction the ambiguity
124: of time can be avoided.
125: For the simplest case of a Schwarzschild black hole
126: there is a timelike Killing vector which selects a prefered time direction.
127: In other words, from our position asymptotically far away from the black
128: hole,
129: we use a well defined time coordinate in our observations. As long as we
130: conscientiously compare all timescales in the same coordinate system,
131: we should get meaningful comparisons.
132: 
133: We investigate the stability of three types of black hole binary:
134: (i) Schwarzschild black hole (non-spinning, test
135: particle motion), (ii) the Post-Newtonian (PN) expansion of the
136: two-body problem (non-spinning black holes), and 
137: (iii) the chaotic orbits of spinning black holes in the PN-expansion.
138: 
139: The stability analysis begins with
140: the equations of motion summarized as
141: \begin{equation}
142: \frac{d X_{i}}{dt} = H_{i}(X_{j}) \, .
143: \end{equation}
144: To analyze the stability of a given orbit we linearize the equations
145: of motion about that orbit
146: \begin{equation}
147: \frac{d\,  \delta \! X_{i}(t)}{dt} = K_{ij}(t)\,  \delta \! X_{j}(t) \, ,
148: \end{equation}
149: with
150: \begin{equation}
151: K_{ij}(t) = \left. \frac{\partial H_{i}}{\partial X_{j}}\right|_{X_{i}(t)}
152: \end{equation}
153: the linear stability matrix.
154: The solution to the linearized equations can be written as
155: \begin{equation}
156: \delta \! X_{i}(t) = L_{ij}(t)\, \delta\! X_{j}(0) \, 
157: \end{equation}
158: in terms of the evolution matrix which must obey
159: 	\begin{equation}\label{evo}
160: 	\dot L_{ij}(t)=K_{im} L_{mj}(t)\, 
161: 	\end{equation}
162: and $L_{ij}(0) = \delta_{ij}$.
163: A determination of the eigenvalues of $L_{ij}$ 
164: leads to the principal Lyapunov exponent.
165: Specifically
166: 	\begin{equation}\label{L}
167: 	\lambda = \lim_{t \rightarrow \infty} \frac{1}{t} \log \left(
168: 	\frac{ L_{jj} (t)}{L_{jj} (0)} \right) \, .
169: 	\end{equation}
170: 
171: \section{Schwarzschild orbits}
172: 
173: \subsection{Circular orbits}
174: 
175: Here we evaluate the Lyapunov exponent of unstable
176: orbits around the Schwarzschild black hole. We consider the usual
177: geodesics of a non-spinning, light companion.
178: We work
179: in Schwarzschild time, the time measured by an observer asymptotically
180: far from the black hole. The Lyapunov exponent was already evaluated in
181: Ref.\ \cite{njc1} in a different time coordinate system. 
182: This exemplifies the ambiguity of time. 
183: Still, the timescales which were compared 
184: in that paper were all measured in the same coordinate system
185: and therefore the general conclusions of Ref.\ \cite{njc1} still hold.
186: 
187: To isolate $\lambda$,
188: we begin with the  Lagrangian for a (non-spinning) 
189: test particle in the Schwarzschild
190: spacetime 
191: \begin{eqnarray}
192: {\cal L} &=& \frac{1}{2}\left(- \frac{(r-2)}{r}\left(\frac{ d t}{ds}\right)^2
193: +\frac{r}{r-2}\left(\frac{ d r}{ds}\right)^2 \right. \nonumber \\
194: && \quad \left. + r^2 \left(\frac{ d \theta}{ds}\right)^2
195: +r^2\sin^2\theta \left(\frac{ d \phi}{ds}\right)^2 \right).
196: \end{eqnarray}
197: The black hole mass has been set to unity.
198: We consider motion in an equatorial plane to eliminate the 
199: cylic $\theta$ variable by setting it equal to $\pi/2$ and
200: define the canonical momenta by $\delta {\cal L}/\delta(dq/ds)=p_q$:
201: 	\begin{eqnarray}\label{p}
202: 	-p_t &=& \frac{r-2}{r}\frac{dt}{ds} = E\nonumber \\
203: 	p_\phi &=& {r^2}\frac{d\phi}{ds} = L \nonumber \\
204: 	p_r &=& \frac{r}{r-2}\frac{dr}{ds} \, .
205: 	\end{eqnarray}
206: To change into Schwarzschild time $t$, we use
207: eqn.\ (\ref{p}) to define the transformation
208: 	\begin{equation}
209: 	\frac{d}{ds}=\frac{Er}{r-2}\frac{d}{dt} \, .
210: 	\end{equation}
211: Notice that in Ref.\ \cite{njc1} an unusual time coordinate $t^\prime$
212: was used instead which was defined by the transformation
213: 	\begin{equation}
214: 	\frac{d}{ds}=\frac{r-2}{E r}\frac{d}{dt^\prime} \, .
215: 	\end{equation}
216: We will redo the stability analysis in Schwarzschild time $t$.
217: The equations of motion can be derived through
218: \begin{equation}
219: 	\frac{\delta{\cal L}}{\delta dr/ds}-\frac{\delta {\cal L}}
220: {\delta r}=0 \, 
221: 	\end{equation}
222: and reduce to a two-dimensional system:
223: 	\begin{eqnarray}
224: 	\dot p_r &=& -\frac{E}{r(r-2)}-\frac{r-2}{r^3}\frac{p_r^2}{E}
225: 		+\frac{r-2}{r^4}\frac{L^2}{E}\nonumber \\
226: 	\dot r &=& \left (\frac{r-2}{r}\right )^2\frac{p_r}{E} 
227: 	\end{eqnarray}
228: where an overdot denotes differentiation with respect 
229: to Schwarzschild time $t$.
230: To compare with Ref.\ \cite{njc1}, we first consider circular orbits.
231: Linearizing the equations of motion 
232: with $X_i(t)=(p_r, r)$ about orbits of constant $r$ gives
233: \begin{equation}\label{kcirc}
234: K_{ij} = \pmatrix{0 & \frac{2(r-1)E}{r^2(r-2)^2}-\frac{L^2}{E}\frac{(3r-8)}{r^5}\cr
235: \left (\frac{r-2}{r}\right )^2\frac{1}{E}  &
236: 0
237: }
238: \end{equation}
239: The eigenvalues along circular orbits are
240: 	\begin{eqnarray}\label{lamcirc}
241: 	\lambda_{\pm}= \pm 
242: 	 \left [\frac{2(r-1)}{r^4}
243: 	-\frac{(3r-8)(r-2)^2}{r^7}\frac{L^2}{E^2}\right ]^{1/2}
244: 	\, .
245: 	\end{eqnarray}
246: 
247: 
248: For the unstable circular orbit at $r=4$ the angular momentum 
249: and energy are $L=4,E=1$ respectively and the eigenvalues of
250: $K_{ij}$ are
251: 	\begin{equation}\label{lpm}
252: 	\lambda_{\pm}=\pm\frac{1}{8\sqrt{2}} \, .
253: 	\end{equation}
254: The conservation of energy ensures that in these canonical
255: coordinates, the Lyapunov exponents must come in plus-minus pairs
256: to conserve the volume of phase space.
257: 
258: The unit normalized eigenvectors corresponding to $\lambda_\pm$ are
259: 	\begin{eqnarray}
260: 	{\bf e}_+ &=&\frac{1}{3} \left(-1, 2\sqrt{2}\right) \nonumber \\
261: 	{\bf e}_- &=&\frac{1}{3} \left(1, 2\sqrt{2}\right) \, .
262: 	\end{eqnarray}
263: In this eigenbasis $K_{ij}$ is diagonal with
264: 	\begin{equation}
265: 	K_{ij}=\pmatrix{\frac{1}{8\sqrt{2}} & 0 \cr
266: 	0 & -\frac{1}{8\sqrt{2}} 
267: 	}
268: 	\end{equation}
269: so that
270: 	\begin{equation}
271: 	L_{ij}=\pmatrix{ \exp(\frac{1}{8\sqrt{2}}t) & 0 \cr
272: 	0 & \exp(-\frac{1}{8\sqrt{2}}t) 
273: 	}\, .
274: 	\end{equation}
275: Notice that in Ref.\ \cite{njc1}, the analysis was carried out in a 
276: four-dimensional coordinate system: $X^\prime_i=(p_r,p_\phi,r,\phi)$.
277: The additional coordinates are unimportant in the dynamical study and 
278: only hampered the diagonalization of $K_{ij}$ \cite{njc1}. If we 
279: were to redo the stability analysis in three coordinates
280: $X^{\prime \prime}=(p_r,p_\phi,r)$, we would get the same eigenvalues
281: (\ref{lpm}) and an additional $\lambda=0$. If we move up to the four
282: coordinates of $X_i^\prime$ we add yet another $\lambda=0$ giving
283: a degenerate set of eigenvalues. The matrix $K_{ij}$ cannot be
284: diagonalized in the event of degenerate eigenvalues which leads to
285: an unnecessary complication. For this reason, we stick to the pertinent
286: two-dimensional system $X_i=(p_r,r)$.
287: 
288: The relativity of time and the influence on the Lyapunov exponent is 
289: apparent at this stage. In Schwarschild time $t$ we find
290: $\lambda_{\pm}=\pm\frac{1}{8\sqrt{2}}$ while in the time $t^\prime$
291: used in Ref.\ \cite{njc1}, at $r=4$ the exponents were
292: found to be $\lambda^\prime_{\pm}=\pm \frac{1}{2\sqrt{2}}$. The Lyapunov
293: exponents are not coordinate invariant. However at $r=4$, 
294: $t^\prime = t/4$ and it follows that the combination
295: 	\begin{equation}
296: 	\lambda t = \lambda^\prime t^\prime \, 
297: 	\end{equation}
298: is invariant.
299: 
300: We compare the Lyapunov timescale $T_\lambda=1/\lambda$
301: to the gravitational wave timescale $T_w=2\pi/\dot \phi$.
302: For the orbit at $r=4$, 
303: $T_\lambda/T_w=\sqrt{2}/\pi \approx 0.45$. The Lyapunov timescale
304: is less than about one orbit around the central black hole.
305: Notice that even though Ref.\ \cite{njc1} operated in an unusual
306: time, the calculations were performed self-consistently so that
307: the ratio of $T_\lambda$ to $T_w$ is correct.
308: The Lyapunov timescale is shorter than the gravitational wave timescale,
309: making the instability observationally relevant.
310: 
311: In principle we could also compare $T_\lambda$ to the decay time due to 
312: energy lost in the form of gravitational radiation. For a test-particle
313: in a circular orbit around a Schwarzschild black hole the decay time is
314: $T_d=(5/256)r^4/\mu$ where $\mu$ is the reduced mass. At $r=4$ this is
315: $T_d=5/(3\mu)\gg 1$ since in the test-particle limit $\mu \ll 1$
316: and $T_\lambda $ will be shorter than the decay time, again making the
317: instability observationally relevant.
318: 
319: 
320: 
321: \subsection{Homoclinic orbits}
322: 
323: Although the unstable circular obits are often emphasized
324: they are actually a subset
325: of the pertinent orbits. The division between stability and instability
326: for a Schwarzschild black hole is often taken to be the innermost
327: stable circular orbit (ISCO). The ISCO is actually the saddle point
328: at which the unstable circular orbit coincides with the stable circular
329: orbit. From a dynamical systems point of view, the true division 
330: between stability and instability occurs more generally
331: along the homoclinic orbits \cite{{bc},{loc}}.
332: 
333: The stable manifold of a periodic orbit is defined as 
334: the set of points in phase
335: space that, when evolved forwards in time, approach the periodic orbit. 
336: The unstable manifold is the set of
337: points in phase that, when evolved backwards in time, approach the periodic 
338: orbit.
339: For an integrable, nonchaotic system,
340: the stable and unstable manifolds can intersect each other
341: along a single orbit. This orbit is called homoclinic if it 
342: approaches the same fixed point in the past and in the future.
343: For black holes the 
344: homoclinic orbits begin at an unstable 
345: circular orbit, roll out to a maximum radius
346: and fall back in to the same unstable circular orbit.
347: An example is shown in fig. \ref{orbit}.
348: The homoclinic orbits are sometimes called zoom-and-whirl orbits in 
349: the gravitational wave literature because the orbits whirl around the
350: center of mass and then zoom out into an ellipse before whirling in again.
351: 
352: 
353: \begin{figure}
354: \centerline{\psfig{file=orbit_.25.ps,width=2.5in}}
355: \caption{A segment of the homoclinic orbit with $\beta=1/4$.
356: \label{orbit}}  \end{figure} 
357: 
358: Under perturbation the homoclinic orbits can become the site of
359: a homoclinic tangle. The tangle occurs when the stable and unstable
360: manifolds intersect transversely at an infinite number of points.
361: The intersection will no longer occur along a simple line but will
362: instead define a fractal set of chaotic orbits. In this section, we 
363: study the stability of the nonchaotic, simple set of homoclinic orbits
364: around a Schwarzschild black hole. In \S \ref{chaos} we study chaotic
365: orbits of spinning black holes.
366: 
367: As is well known, 
368: orbital motion around a Schwarzschild black hole conveniently reduces
369: to one-dimensional motion in an effective
370: potential
371: 	\begin{equation}\label{cons}
372: 	\frac{1}{2}\dot r^2 + V_{\rm eff}(r)=E
373: 	\end{equation}
374: with
375: 	\begin{equation}\label{eff}
376: 	V_{\rm eff}(r)=E+\frac{(r-2)^3}{2E^2r^3}\left 
377: 	({1+\frac{L^2}{r^2}
378: 	}\right )-\frac{(r-2)^2}{2r^2}\, .
379: 	\end{equation}
380: Circular orbits are solutions of $V_{\rm eff}=E$. For large enough
381: angular momentum there are two circular orbits, one unstable and one 
382: stable. As found in Ref.\ \cite{bc}, the homoclinic orbits have $E<1$ and
383: are described by the solution
384: 	\begin{eqnarray}\label{homo}
385: 	p_r &=& \pm \frac{r}{r-2}\left [ E^2-\frac{r-2}{r}\left (1+\frac{L^2}{r^2}\right )\right ]^{1/2} \nonumber \\
386: 	\frac{1}{r} & = & \frac{1-2\beta}{6} + \frac{\beta}{2} \tanh^2(\sqrt{\beta} \phi/2)  \nonumber \\
387: 	\end{eqnarray}
388: where
389: $0 \le \beta \le 1/2$ and $t(\phi)$ is a complicated function \cite{bc}.
390: \footnote{There appears to be typo in eqn. (1.5) of \cite{loc}.
391: Eqn. (\ref{homo}) has a factor 1/2 which is missing from the second term 
392: in eqn. (1.5).}
393: The circular orbits can also be parameterzied by $\beta$ as
394: 	\begin{eqnarray}
395: 	\label{circ}
396: 	 r_{\rm unstable}&=&6/(1+\beta)	\nonumber \cr
397: 	r_{\rm stable}&=&6/(1-\beta) 	\nonumber \cr
398: 	L&=&2\sqrt{3/(1-\beta^2)} \nonumber \cr
399: 	E&=&\frac{2-\beta}{3}\sqrt{2/(1-\beta)}\, .
400: 	\end{eqnarray}
401: 
402: \begin{figure}
403: \centerline{\psfig{file=veff.ps,width=2.5in}}
404: \caption{The effective potential for orbits with 
405: $\beta=1/4$. The segment in bold marks the corresponding homoclinic trajectory.
406: The homoclinic orbit begins at the unstable circular orbit at the top 
407: of the hill ($r=4.8$), rolls out past the stable circular orbit in the valley
408: and on out to the maximum radius ($r=12$) before rolling
409: in again and climbing back up 
410: to the unstable orbit.
411: \label{veff}}  \end{figure} 
412: 
413: A homoclinic oribt starts at $r_{\rm unstable}$ and rolls out to 
414: 	\begin{equation}\label{max}
415: 	r_{\rm max}=6/(1-2\beta)\, 
416: 	\end{equation}
417: winding around the black hole as it does so and then 
418: drops back in to $r_{\rm unstable}$. The ISCO is a homoclinic orbit
419: with $\beta=0$.  The first homoclinic
420: orbit at $\beta=1/2$ starts at the unstable orbit
421: at $r=4$ and rolls out to infinity before returning.
422: For $\beta=1/4$, $r_{\rm unstable}=4.8$, $r_{\rm stable}=8$ and
423: $r_{\rm max}=12$. 
424: The effective potential for the homoclinic orbit 
425: is drawn in fig.\ \ref{veff}. A segment of the orbit in the equatorial
426: plane is represented in fig.\ \ref{orbit}.
427: 
428: To analyze the stability we linearize to find
429: \begin{equation}\nonumber
430: K_{ij} = \pmatrix{
431: -\frac{2(r-2)}{r^3}\frac{p_r}{E} & \frac{2(r-1)E}{r^2(r-2)^2}+\frac{2(r-3)}{r^4}\frac{p_r^2}{E} -\frac{L^2}{E}\frac{(3r-8)}{r^5}\cr
432: \left (\frac{r-2}{r}\right )^2\frac{1}{E}  &
433: \frac{4(r-2)}{r^3}\frac{p_r}{E}  
434: }\, .
435: \end{equation}
436: The most general eigenvalues are
437: 	\begin{eqnarray}\label{leg}
438: 	\ell_{\pm}=\left (\frac{r-2}{r^2}\right )\frac{p_r}{E}
439: 	& \pm &
440: 	 \left [(2r-5)\frac{(r-2)^2}{r^6}\frac{p_r^2}{E^2} \right. \cr
441: &+& \left. 2\frac{(r-1)}{r^4}
442: 	-\frac{(3r-8)(r-2)^2}{r^7}\frac{L^2}{E^2}\right ]^{1/2}
443: 	\end{eqnarray}
444: Strictly speaking, $\ell$ is a stability exponent and is not identical
445: to the time averaged Lyapunov exponent defined in eqn. (\ref{L}).
446: Figure \ref{beta.25} shows the real and imaginary
447: parts of the positive stability exponent. As expected, the exponent
448: is postive near the unstable inner radius and becomes imaginary in the
449: vicinity of the stable circular radius dropping down to nearly zero as
450: it reaches the apihelion and then runs back through these values as it
451: moves back in to perihelion.
452: 
453: 
454: \begin{figure}
455: \centerline{\psfig{file=lam_.25.ps,width=2.5in}}
456: \caption{$\beta=1/4$ The solid line is the Real part of the 
457: positive stability exponent and the dotted line is the Imaginary part.
458: \label{beta.25}}  \end{figure} 
459: 
460: Because of this time variability in the stability along the orbit,
461: we have to be cautious in interpreting the timescales. 
462: The gravitational wave frequency will jag up and down as the orbit
463: zooms and whirls \cite{loc}. And it isn't obvious which timescales
464: to compare. Instead of using the variable, analytic result we could try 
465: a time average.
466: 
467: To this end we compare the analytic value of $\ell$ from eqn.\ (\ref{leg})
468: to the
469: time average Lyapunov exponent defined from eqn.\ (\ref{L})
470: 	\begin{equation}
471: 	\lambda = \lim_{t \rightarrow \infty} \frac{1}{t} \log \left(
472: 	\frac{ L_{jj} (t)}{L_{jj} (0)} \right) \, .
473: 	\end{equation}
474: For comparison we first look along the unstable circular orbit at $r=4.8$.
475: The upper panel of 
476: figure \ref{le_b.25} shows the Lyapunov exponent as defined by eqn.\ (\ref{L})
477: for the unstable
478: circular orbit. The exponent was obtained
479: by numerically integrating the $L_{ij}$ using eqn.\ (\ref{evo}) and 
480: figure \ref{le_b.25}
481: shows $\lambda t$ versus $t$ from eqn.\ (\ref{L}).
482: The numerically calculated value shown
483: in figure \ref{le_b.25} is identical to the analytic value given by 
484: eqn.\ (\ref{lamcirc}) of $\lambda \approx 0.475$.
485: 
486: 
487: However for a homoclinic orbit which begins at $r=4.8$,
488: the time averaged (\ref{L}) behaves as though there is no 
489: instability (lower panel of figure \ref{le_b.25}), 
490: when we know from the analytic result shown in fig.\ \ref{beta.25}
491: that there is.
492: The time averaged Lyapunov exponent
493: will vanish along this orbit even though it clearly has an unstable 
494: segment. 
495: We have to be very cautious therefore when we interpret the Lyapunov
496: exponent. 
497: 
498: \begin{figure}
499: \centerline{\psfig{file=le_b.25_c.ps,width=2.in}}
500: \centerline{\psfig{file=le_b.25.ps,width=2.in}}
501: \caption{$\beta=1/4$. The slope of the line in the top panel is 
502: the numerically calculated Lyapunov exponent $\lambda \approx 0.475$
503: for the unstable circular orbit at $r=4.8$ using eqn.\ (\ref{L}). 
504: This matches exactly the analytic value predicted from 
505: eqn.\ (\ref{lamcirc}).
506: The slope of the line in the lower panel shows
507: zero Lyapunov exponent as calculated by eqn.\ (\ref{L}) for the homoclinic
508: orbit. This is to be contrasted with the analytic value of 
509: the stability exponent shown in fig.\ \ref{beta.25} which has positive segments
510: \label{le_b.25}}  \end{figure} 
511: 
512: For emphasis, if one was just scanning numerically, the mistaken
513: conclusion could be drawn that these orbits were dynamically simple.
514: This may turn out to be particularly important for the chaotic orbits
515: of \S \ref{chaos}.
516: 
517: \section{Post-Newtonian Orbits}
518: 
519: To move beyond the test particle limit, the two-body problem
520: has been expanded in a Post-Newtonian (PN) expansion approximation
521: to the fully relativistic two-body problem \cite{{kww},{linw},{extra1}}.
522: In this section we consider two black holes
523: which are not spinning.
524: In Ref.\ \cite{loc} the stability of the fixed points
525: in the PN equations to second-order (2PN) was tested
526: following \cite{kww}. 
527: We quote the results
528: of Ref.\ \cite{loc} here.
529: To second-order in the PN expansion,
530: the center of mass equations of motion for the binary orbit can be
531: written in harmonic coordinates as \cite{{kww},{linw},{extra1}}
532: 	\begin{eqnarray}
533: 	\ddot r_h &=&r_h \dot \phi^2-{1\over r_h^2}\left (A+B\dot
534: 	r_h\right ) \label{eom1}\\
535: 	\ddot \phi &=& -\dot \phi \left ({1\over r_h^2}B+2{\dot r_h
536: 	\over r_h}\right )\label{eom2}
537: 	\end{eqnarray}
538: where $M=1$ is the total mass of the pair.
539: The transformation between harmonic coordinates and Schwarzschild coordinates
540: is
541: $r_h =r -m$.  The form of $A(r_h,\dot r_h, \dot \phi)$ and $B(r_h,\dot r_h, \dot \phi)$
542: depends on the relative masses of the two black holes and on
543: the order of the PN expansion and can be found in 
544: Ref.\ \cite{kww}. 
545: 
546: As in Ref.\ \cite{kww}, 
547: the stability of the fixed points is tested by perturbing eqns.\
548: (\ref{eom1})-(\ref{eom2})
549: about a circular orbit to obtain,
550: 	\begin{equation}
551: 	K_{ij}
552: 	=\pmatrix{0 & 1 & 0 \cr
553: 	a & 0 & b \cr
554: 	0 & c & 0}
555: 	\label{mat}
556: 	\end{equation}
557: with
558: 	\begin{eqnarray}
559: 	a &=&3\dot \phi_o^2-{m\over r_{ho}^2}
560: 	\left ({\partial A\over \partial r_h}\right )_o  \nonumber \\
561: 	b&=&2r_{ho}\dot \phi_o -{m\over r_{ho}^2}\left (	
562: 	\partial A\over \partial \dot \phi\right )_o
563: 	\nonumber \\
564: 	c&=&-\dot \phi_o
565: 	\left ({2\over r_{ho}} +{m\over r_{ho}^2}
566: 	\left (\partial B\over \partial \dot r_h\right )_o\right )
567: 	\, ,
568: 	\label{abc}
569: 	\end{eqnarray}
570: where $r_{ho}$ is the radius of the circular orbit in harmonic coordinates
571: and $\dot \phi^2_o=mA_o/r^3_{ho}$ is a function of the radius of the orbit
572: and is found explicitly in Ref.\ \cite{loc}.
573: The eigenvalues of (\ref{mat}) are,
574: 	\begin{equation}
575: 	\ell=0,\quad \ell_\pm=\pm(a+bc)^{1/2} \, .
576: 	\end{equation}
577: Stable oscillations about a circular orbit correspond to imaginary
578: $\ell$ so that $a+bc<0$. 
579: Unstable orbits correspond to real positive $\ell$ and so have $a+bc>0$. 
580: The value of $\ell^2$ for equal mass binaries as a function 
581: of the circular radius is plotted in fig.\ \ref{le_2pn}.
582: 
583: \begin{figure}
584: \centerline{\psfig{file=le_2pn.ps,angle=0,width=2.5in}}
585: \caption{$\ell^2$ as a function of the constant circular
586: radius in harmonic coordinates, $r_{ho}$. $\ell^2>0$ corresponds to
587: unstable circular orbits and $\ell^2<0$ corresponds to
588: stable circular orbits.
589: \label{le_2pn}}  \end{figure} 
590: 
591: 
592: Using the results of Ref.\ \cite{loc} we deduce that
593: in the test-mass limit,
594: the innermost {\it unstable}
595: circular orbit occurs at Schwarzschild radius 
596: $r_o=r_{ho}+1=4.96$. A comparison 
597: to the gravitational wave timescale in the PN expansion then gives 
598: $T_\lambda/T_w\approx 0.158$.
599: In the opposite extreme of equal mass binaries,
600: the innermost {\it unstable}
601: circular orbit occurs at $r_o=r_{ho}+1=5.78$ and
602: $T_\lambda/T_w\approx 0.21$.
603: A direct comparison to the Schwarzschild case isn't that
604: valuable. What is noteworthy is that the Lyapunov timescales
605: are again less than about one orbit around the center of mass.
606: Consequently, one expects the decoherence of the gravitational
607: waveform to be observationally significant. The instability
608: timescales are comparable to the decay times although of course
609: even a small loss in energy can induce merger for such an unstable
610: orbit.
611: 
612: The analysis extended to homoclinic orbits can be gleaned from 
613: Ref.\ \cite{loc}. 
614: The homoclinic orbits to 2PN order 
615: show similar features to the homoclinic
616: orbits of the Schwarzschild spacetime. The analytic Lyapunov exponent
617: will pass from positive to imaginary values as the orbit winds around
618: the center of mass. However, the time-averaged exponent will dilute 
619: these critical features.
620: 
621: None of these orbits are chaotic although they are unstable.
622: We turn to the chaotic orbits of rapidly spinning binaries next.
623: 
624: \section{Chaotic orbits}\label{chaos}
625: 
626: The dynamics can become chaotic when the homoclinic orbit is perturbed
627: leading to a homoclinic tangle. The intersection of the stable and unstable
628: manifold will no longer occur along a line but will intersect transversally
629: an infinite number of times. The fractal set of unstable chaotic
630: orbits lies along this tangled intersection. Ref.\ \cite{bc} studied generic
631: gravitational perturbations along the homoclinic orbits and found,
632: as they expected, that the dynamics could become chaotic. The physical
633: significance of the perturbations however wasn't clear and therefore
634: the observational consequences were difficult to assess. 
635: 
636: \begin{figure}
637: \centerline{\psfig{file=leL_pns.ps,width=2.5in}}
638: \caption{The slope of the line is 
639: the Lyapunov exponent for the chaotic orbit of fig.\ \ref{orb_pns}.
640: \label{leL_pns}}  \end{figure} 
641: 
642: The chaotic dynamics discovered in 
643: Ref.\ \cite{sm} for a supra-maximally spinning test-particle
644: and for rapidly spinning pairs in the Post-Newtonian expansion
645: \cite{{me1},{me2}}, may occur along these homoclinic orbits. At the least
646: the chaotic behaviour kicks up most conspicuously in the vicinity of
647: these orbits.
648: 
649: We cannot determine the Lyapunov exponents analytically since the orbits
650: are not analytically soluble - the very meaning of nonintegrability.
651: We can however use eqn.\ (\ref{L}) to numerically determine the exponents
652: as we have done in Ref.\ \cite{usl}.
653: The Lyapunov exponent for a maximally spinning pair of black holes
654: is shown in fig.\ \ref{leL_pns}.
655: %$(x,\dot x)=(5.5,0,0,0,0.4,0), \theta_1=\pi/2,\theta_2=\pi/6$ and 
656: %mass ratio 1:3.
657: The value read from this 
658: is $T_\lambda \approx 11$
659: in units of windings around the center of mass. 
660: A segment of the orbit projected onto a plane
661: is shown in fig.\ \ref{orb_pns}.
662: 
663: 
664: \begin{figure}
665: \centerline{\psfig{file=orbit_pns.ps,width=2.5in}}
666: \caption{A projection onto the plane 
667: of two maximally spinning black holes in a chaotic orbit.
668: \label{orb_pns}}  \end{figure} 
669: 
670: 
671: For such an erratic orbit, it is not simple to define the 
672: gravitational wave timescale or the
673: radiation
674: reaction timescale. By starting the numerical simulation at a radius 
675: greater than 20 we found a rough estimate of $T_d\sim 4-5$ orbits
676: so that $T_\lambda/T_d > 2$ and the Lyapunov timescale is longer than
677: the dissipation timescale. For this 
678: orbit the gravitational waveform will not have time to decohere before
679: plunge and observations will not be severly disrupted by irregularity.
680: However a factor of 2 is a tight margin especially since the PN-expansion
681: is being pushed to extremes at such small separations.
682:  
683: It is important to stress that this is but one specific orbit and the ratio
684: of $T_\lambda/T_w$ will vary from orbit to orbit. Some chaotic orbits 
685: will undoubtedly have a longer Lyapunov timescale to dissipation timescale.
686: A broad survey of the irregular region of phase space would be valuable.
687: At 2PN order this is not so useful since the PN-expansion converges very
688: slowly to the full relativistic problem. Notice for instance that 
689: $T_\lambda$ is longer for this two-body approximation than it is for
690: the fully relativistic Schwarzschild orbits. 
691: This may be a consequence
692: of the slow convergence of the expansion and may show an underestimate
693: of the instability.
694: It is also possible that, like the homoclinic orbits, the measure of
695: instability is diluted by the time average across the orbit.
696: A survey at higher orders
697: may be useful but requires a greater than 3PN-expansion
698: that includes all spin arrangements and is not restricted to circular or 
699: quasi-circular orbits. We hope these higher orders will be available 
700: imminently and a survey of orbits will be viable in the near future.
701: 
702: \section*{Summary}
703: 
704: The Lyapunov exponents can be very useful for comparisons of physical
705: scales. However, Lyapunov exponents also have some shortcomings which
706: require we tread cautiously:
707: 
708: \noindent $\bullet $ Lyapunov exponents are relative. They depend on the
709: worldline of the observer and the time they measure.\par
710: \noindent $\bullet$ They vary from orbit to orbit and may not contain
711: generic information.\par
712: \noindent $\bullet$ They can give zero when averaged over orbits which move
713: in and out of unstable regions.\par
714: \noindent Therefore, while important and useful, the Lyapunov exponents
715: can be misleading and can only be used cautiously.
716: 
717: As far as we can trust them, the Lyapunov exponents
718: give an estimate of the importance of instability to observations
719: of gravitational waves. If the Lyapunov timescale is short compared
720: to the inverse 
721: frequency of the gravitational waves emitted and is short compared
722: to the
723: dissipation timescale then instability will cause an
724: observable decoherence of gravitational waves.
725: We found that \par
726: \noindent $\bullet$ the Lyapunov timescale
727: is shorter than both the gravitational wave timescale and the
728: dissipation timescale for unstable circular orbits in the 
729: approximation of a test-particle around a Schwarzschild black hole,\par
730: \noindent $\bullet$ the Lyapunov timescale
731: is shorter than the gravitational wave timesecale and comparable to the
732: dissipation timescale for unstable circular orbits in the 
733: 2PN approximation in the absence of spins, \par
734: \noindent $\bullet $
735: and that the Lyapunov time was about a factor of 2 larger than the decay 
736: time for one
737: randomly sampled chaotic orbit of a pair of maximally spinning black 
738: holes in the 2PN expansion. 
739: 
740: The longer Lyapunov time for orbits in the 2PN approximation versus the 
741: test-particle approximation may be a real effect or 
742: it may be due to the slow convergence of the 2PN expansion to the full
743: nonlinear problem or finally it may be due to the time average over such 
744: a varied orbit.
745: In short, dissipation due to gravitational waves does abate chaos 
746: although the competition between chaos and dissipation is close.
747: Better approximations to the two-body problem
748: are needed to determine conclusively 
749: if chaos will affect observations of gravitational waves.
750: 
751:  
752: \section*{Acknowledgements}
753: NJC is supported in part by National Science Foundation Grant No. PHY-0099532.
754: JL is supported by a PPARC Advanced Fellowship and an award from NESTA.
755: 
756: 
757: \begin{references}
758: 
759: \bibitem{bc}
760: L.Bombelli and E.Calzetta, Class. Quantum. Grav.
761: {\bf 9} 2573 (1992).
762: 
763: \bibitem{loc}
764: J. Levin, R. O'Reilly, \& E.J. Copeland, Phys. Rev. D{\bf 62},
765: 024023 (2000).
766: 
767: \bibitem{sm} S. Suzuki \& K. Maeda, Phys. Rev. D{\bf 55}, 4848 (1997).
768: 
769: \bibitem{me1} J. Levin, Phys. Rev. Lett. {\bf 84}, 3515 (2000).
770: 
771: 
772: \bibitem{me2} J. Levin, gr-qc/0010100 (2001).
773: 
774: 
775: \bibitem{bhs}
776: G. Contopolous, Proc. R. Soc. {\bf A431} 183 (1990);
777: Proc. R. Soc. {\bf A435} 551 (1990).
778: 
779: \bibitem{frank} C.P. Dettmann, N. E. Frankel \& N.J. Cornish, Phys. Rev. D.{\bf 50} R618 (1994).
780: 
781: \bibitem{cf} N.J. Cornish and N.E.Frankel, Phys. Rev. D
782: {\bf 56} 1903 (1997).
783: 
784: \bibitem{meglow} J.Levin, Phys. Rev. D. {\bf 60} 64015 (1999).
785: 
786: \bibitem{njc1} Neil J. Cornish, gr-qc/0206062.
787: 
788: 
789: \bibitem{rs} J.D. Schnittman \& F.A. Rasio, Phys. Rev. Lett. {\bf 87}, 121101 (2001).
790: 
791: \bibitem{usl} N.J. Cornish and J. Levin, gr-qc/0207016;
792: N.J. Cornish and J. Levin, gr-qc/0207020.
793: 
794: \bibitem{barrow} J.D. Barrow, Phys. Rev. Lett. {\bf 46} 963 (1981); Phys. Rep.
795: {\bf 85} 1 (1982).
796: 
797: \bibitem{book} {\it Deterministic Chaos in General Relativity} eds. D. Hobill, A. Burd and A. Coley, (Plenum Press, New York, 1994) and references therein.
798: 
799: \bibitem{mixm} N.J. Cornish \& J.J. Levin,
800: Phys. Rev. Lett. {\bf 78} 998 (1997); Phys. Rev. D.{\bf 55} 7489 (1997).
801: 
802: \bibitem{sk} See also, 
803: O. Semerak and V. Karas, Astron. Astrophys. 343 (1999) 325.
804: 
805: \bibitem{kww} L.E.Kidder, C.M.Will and A.G.Wiseman, Phys. Rev. D
806: {\bf 47} 3281 (1993).
807: 
808: \bibitem{linw}  C.W.Lincoln and C.M.Will, Phys. Rev. D. {\bf 42}
809: 1123 (1990).
810: 
811: \bibitem{extra1} T. Damour and N. Deruelle, C. R. Acad. Sci. Paris {\bf
812: 293} 537 (1981); {\bf 293} 877 (1981).
813: 
814: \end{references}
815: 
816: \end{document}
817: 
818: 
819: 
820: