1: \documentclass[12pt,letterpaper]{JHEP3} \usepackage{cite}
2: \usepackage{epsfig}
3: %\usepackage{showkeys}
4: %Repot #'s: YITP-SB-07-04
5:
6: \newcommand{\solar}{\odot}
7: \newcommand{\ra}{\rightarrow}
8:
9:
10: \title{Predicting the Cosmological Constant from the Causal Entropic
11: Principle}
12:
13: \author{%
14: Raphael Bousso$^a$, Roni Harnik$^b$,
15: Graham D. Kribs$^c$, and Gilad Perez$^d$\\
16: \llap{$^a$}
17: Center for Theoretical Physics, Department of Physics,
18: University of California,\\
19: Berkeley, CA 94720-7300, U.S.A. {\em and} \\
20: Lawrence Berkeley National Laboratory,
21: Berkeley, CA 94720-8162, U.S.A.
22: \vspace*{2mm} \\
23: \llap{$^b$}
24: Stanford Linear Accelerator Center, Stanford University, \\
25: Stanford, CA 94309, U.S.A. {\em and} \\
26: Physics Department, Stanford University, Stanford, CA 94305, U.S.A.
27: \vspace*{2mm} \\
28: \llap{$^c$}
29: Department of Physics and Institute of Theoretical Science,
30: University of Oregon, \\ Eugene, OR 97403, U.S.A.
31: \vspace*{2mm} \\
32: \llap{$^d$}
33: C.N. Yang Institute for Theoretical Physics, State University of New York, \\
34: Stony Brook, NY 11794-3840, U.S.A.}
35:
36:
37: \abstract{We compute the expected value of the cosmological constant
38: in our universe from the Causal Entropic Principle. Since observers
39: must obey the laws of thermodynamics and causality, the principle
40: asserts that physical parameters are most likely to be found in the
41: range of values for which the total entropy production within a
42: causally connected region is maximized. Despite the absence of more
43: explicit anthropic criteria, the resulting probability distribution
44: turns out to be in excellent agreement with observation. In
45: particular, we find that dust heated by stars dominates the entropy
46: production, demonstrating the remarkable power of this thermodynamic
47: selection criterion. The alternative approach---weighting by the
48: number of ``observers per baryon''---is less well-defined, requires
49: problematic assumptions about the nature of observers, and yet
50: prefers values larger than present experimental bounds.}
51:
52: %\reportnumber{YITP-SB-07-04}
53: \preprint{\hepth{0702115}}
54:
55:
56:
57: \begin{document}
58:
59: \section{Introduction}
60: \label{sec-intro}
61:
62: The discovery that the universe is in a period of accelerated
63: expansion~\cite{Per98,Rie98}, combined with an accurate accounting of
64: the total, matter, and radiation components of the energy density
65: \cite{Spergel:2006hy}, provide overwhelming evidence for dark energy.
66: These measurements are completely consistent with the interpretation
67: of dark energy as a non-zero cosmological constant, $\Lambda$. This
68: has undermined the hope that the energy of our vacuum is uniquely
69: determined by fundamental theory. Instead, it lends credence to the
70: hypothesis that the cosmological constant is an environmental
71: variable, which takes on different values in widely separated regions
72: of the universe.
73:
74: The observed vacuum energy density\footnote{Unless indicated
75: otherwise, all observed values in this paper are taken from
76: Ref.~\cite{Tegmark:2005dy}. Where no explicit units are given, we set
77: $\hbar=G_{\rm N}=c=k_{\rm B}=1$.}
78: \begin{equation}
79: \rho_\Lambda=\frac{\Lambda}{8\pi} = (1.25\pm 0.25)\times 10^{-123}~,
80: \label{eq-cc}
81: \end{equation}
82: is at least $55$ orders of magnitude smaller than what would be
83: expected from the standard model of particle physics (see, e.g.,
84: Ref.~\cite{Polchinski:2006gy} for a recent review). The environmental
85: approach does not assert that this tiny value is inevitable, or even
86: typical among all possible values. Rather, it aims to show that it is
87: not atypical among values measured by observers.
88:
89: A number of conditions must be satisfied for the environmental
90: approach to work. The first, obviously, is that fundamental theory
91: must {\em admit}\/ the observed value of the vacuum energy. This can
92: happen without explicit tuning if the theory gives rise to an enormous
93: number $N$ of different vacua. Of course, typical values will be of
94: order unity, but if they are randomly distributed, they can form a
95: dense spectrum, or ``discretuum'', with average spacing of order
96: $1/N$. If $N\gg 10^{123}$, then it is likely that the observed value
97: is included in the spectrum. Thus, the approach really depends on
98: whether fundamental theory (which, presumably, is more or less unique)
99: admits a sufficiently dense discretuum.
100:
101: The second condition is that the observed value must be dynamically
102: {\em attainable\/}, starting from generic initial conditions. With
103: $N\gg 10^{123}$ possibilities, there is no reason for the universe to
104: start out in a vacuum like ours. The environmental approach therefore
105: depends on a means to start from some generic initial value and later
106: realize the observed value, either as a branch in the wavefunction or
107: as a particular spacetime region embedded in a vast universe.
108:
109: Finally, the environmental approach requires an explanation of why we
110: happen to {\em observe\/} such an unusually small vacuum energy. Most
111: values of $\rho_\Lambda$ in the discretuum will be of order unity, and
112: only a fraction (in the simplest case, a fraction of order
113: $10^{-123}$) will have a magnitude as small as the observed value. It
114: is not enough to show that the small value given by Eq.~(\ref{eq-cc})
115: is possible; one must also show that it is {\em not unlikely\/} to be
116: observed.
117:
118: The first condition appears to be satisfied by string theory, which
119: admits as many as $10^{500}$ long-lived metastable
120: vacua~\cite{BP,KKLT,DenDou04b,Sil01,DouKac06} (see Ref.~\cite{Sch06}
121: for a discussion of earlier work). The second condition can then be
122: met because the vacua are metastable and can decay into one
123: another.\footnote{In general, long-lived metastability implies that
124: all matter is diluted before the next decay occurs, so the mechanism
125: depends on efficient reheating in the new region. This rules out
126: models that reduce the cosmological constant
127: gradually~\cite{BT1,BT2}. In the string landscape, the vacuum
128: preceding ours was likely to have had an enormous cosmological
129: constant. Its decay acted like a big bang for the observed universe
130: and allowed for efficient reheating~\cite{BP}.}
131:
132: In this paper we address the third condition. We will use a novel
133: approach, the Causal Entropic Principle, to argue that the observed
134: value of $\rho_\Lambda$ is not unlikely. Our main result is shown in
135: Fig.~\ref{fig-showmethemoney}.
136: %
137: \EPSFIGURE[!t]{stars_PS_hopkins.eps,width=.8\textwidth}{\label{fig-showmethemoney}
138: The probability distribution of the vacuum energy measured by
139: typical observers, computed from the Causal Entropic Principle, is
140: shown as a solid curve. The values consistent with present
141: cosmological data, in the shaded vertical bar, are well inside the
142: $1\sigma$ region (shown in white), and hence, not atypical. For
143: comparison, the dashed line shows the distribution derived by
144: estimating the number of observers per baryon. Unlike our curve, it
145: assumes that galaxies are necessary for observers; yet, the observed
146: value is very unlikely under this distribution. For more details
147: about both curves, see Figures 2 and 8.}
148: %
149:
150: The Causal Entropic Principle is based on two ideas: any act of
151: observation increases the entropy, and spacetime regions that are
152: causally inaccessible should be disregarded. It assumes that on
153: average, the number of observations will be proportional to the amount
154: of matter entropy produced in a causally connected region, $\Delta S$.
155: Vacua should be weighted by this factor to account for the rate at
156: which they will be observed.
157:
158: Crucially, the size of the causal diamond is inversely proportional to
159: the vacuum energy, so smaller values of $\rho_\Lambda $ allow for
160: greater complexity. This compensates for the scarcity of vacua with
161: small $\rho_\Lambda $. As a result, $\rho_\Lambda$ prefers to take a
162: value such that vacuum energy begins to dominate near the time of peak
163: entropy production.
164:
165: We will find that entropy production in the causal diamond is
166: dominated by dust particles heated by stars. This is an important
167: result in its own right: our weight, $\Delta S$, is a simple physical
168: quantity that turns out to be sensitive to the existence of galaxies,
169: stars, and heavy elements.
170:
171: We will show that the entropy production rate peaked approximately 2
172: to 3.5 billion years after the big bang. It is this time-scale,
173: rather than the time of galaxy formation, which governs our prediction
174: of the cosmological constant, and it prefers a range of $\rho_\Lambda
175: $ that is in very good agreement with observation.
176:
177: The same result also explains the so-called coincidence problem or
178: ``why now'' problem. According to the Causal Entropic Principle,
179: typical observers will exist when most of the entropy production in
180: the causal diamond occurs. Our result ensures that this happens
181: during the era when the matter and vacuum energy densities are
182: comparable.
183:
184:
185: \paragraph{Outline}
186:
187: Historically, discussions of the third condition---why do we observe
188: an unusually small $\rho_\Lambda$---have focussed on anthropic
189: selection effects, which we discuss in Sec.~\ref{sec-weighting}. Long
190: before the discovery of the string landscape, it was noted that not
191: all values of $\rho_\Lambda$ are compatible with the existence of
192: observers~\cite{DavUn,Linde84,Sak84,Ban85,BarTip,Linde}. This culminated in Weinberg's
193: successful prediction~\cite{Wei87} that a small non-zero value of
194: $\rho_\Lambda$ would be observed, which we review in
195: Sec.~\ref{sec-weinberg}. Weinberg's assumption was relatively modest:
196: Observers require galaxies. But astronomers have since discovered
197: galaxies that would have formed even if $\rho_\Lambda$ had been more
198: than a thousand times larger than the observed value. This leaves a
199: large discrepancy between theory and observation.
200:
201: A possible resolution is to ask not only about the existence of
202: observers, but to weight vacua by the {\em number\/} of observers they
203: contain. This number is generically infinite or zero, so a
204: regularization scheme is needed. A popular approach is to weight by
205: the number of observers per baryon. In Sec.~\ref{sec-opb}, we argue
206: that this approach is both poorly motivated and, in a realistic
207: landscape, poorly defined. Moreover, it does not resolve the conflict
208: with observation. To mitigate the discrepancy, one is forced to posit
209: increasingly specific conditions for life, such as the chemical
210: elements required. Indeed, to do reasonably well, one must suppose
211: that observers can only arise in galaxies as large as ours---a very
212: strong assumption, for which there appears to be no evidence.
213:
214: In Sec.~\ref{sec-cep}, we motivate and discuss a different approach to
215: this problem. The Causal Entropic Principle weights each vacuum by
216: the amount of entropy, $\Delta S$, produced in a causally connected
217: region~\cite{Bou06}. This is the largest spacetime region that can be
218: probed and across which matter can interact. Since observation
219: requires free energy, it is natural to expect that the number of
220: observers will scale, on average, with $\Delta S$. In other words, we
221: demand nothing more than that observers obey the laws of
222: thermodynamics. This is far weaker even than Weinberg's criterion,
223: and one might be concerned that it will not be sufficiently
224: restrictive. Yet, in combination with the causal cutoff, this minimal
225: requirement becomes a powerful predictive tool.
226:
227: This is demonstrated in Sec.~\ref{sec-calculation}, where we use the
228: Causal Entropic Principle to derive the probability distribution for
229: $\rho_\Lambda$ over universes otherwise identical to ours. We begin,
230: in Sec.~\ref{sec-metric}, by computing the geometry of the causal
231: diamond for general $\rho_\Lambda $. We are mainly interested in its
232: comoving volume as a function of time, $V_{\rm c} (t)$. In
233: Sec.~\ref{sec-entropy}, we consider important mechanisms by which
234: entropy is produced within our causal diamond. We estimate
235: contributions from stars, quasars, supernovae, and other processes.
236: We find that the leading contribution to $\Delta S$ comes from
237: infrared photons emitted by interstellar dust heated by stars.
238:
239: In Sec.~\ref{sec-rate}, we analyze this leading contribution in more
240: detail. We compute $d S/(d V_{\rm c} \, dt)$, the rate at which
241: entropy is produced per unit comoving volume per unit time. This rate
242: will depend on $\rho_\Lambda$ because large $\rho_\Lambda$ disrupts
243: galaxy formation and thus star formation. Interestingly, however, it
244: turns out that this dependence is not important for our final result.
245: In Sec.~\ref{sec-total}, we integrate the rate found in
246: Sec.~\ref{sec-rate} over the causal diamond determined in
247: Sec.~\ref{sec-metric}. This yields the weight factor, $\Delta
248: S(\rho_\Lambda )$. We display the resulting probability distribution
249: for $\rho_\Lambda$, and we note that the observed vacuum energy lies
250: in the most favored range.
251:
252: In Sec.~\ref{sec-discussion}, we pinpoint the origins and discuss some
253: implications of our main findings. We also identify important
254: intermediate results.
255:
256: \paragraph{Extensions}
257:
258: In the interest of time and clarity, we have limited our task. We use
259: the Causal Entropic Principle solely to compute a probability
260: distribution over positive values of the vacuum energy, holding all
261: other physical parameters fixed. This is the case most frequently
262: studied in the literature, making it straightforward to compare our
263: result with those obtained from the traditional method of weighting by
264: observers-per-baryon~\cite{MarSha97,GarLiv99,Efs95,Vil04}.
265:
266: In other words, our distribution is conditioned on the assumptions
267: that $\rho_\Lambda>0$, and that all low energy physics is the same as
268: in our vacuum. We ask only about the probability distribution of
269: $\rho_\Lambda$ in this subspace of the landscape. This is a valid
270: consistency check: Suppose that the observed value were disfavored on
271: a subspace picked out by other observed parameters. Then it would
272: only become less likely when the distribution is extended over the
273: entire landscape, and so the model would conflict with observation.
274:
275: Because negative values of $\rho_\Lambda$ are tightly constrained by
276: standard (though questionable) anthropic arguments, the main challenge
277: for environmental approaches has been to suppress large positive
278: values of $\rho_\Lambda$. For this purpose, it suffices to
279: concentrate on the subset of vacua with $\rho_\Lambda>0$. This
280: simplifies our analysis, since negative $\rho_\Lambda$ lead to a
281: different class of metrics. (Interestingly, a preliminary analysis
282: indicates that negative values will be somewhat favored by the Causal
283: Entropic Principle, though not by enough to render the observed value
284: unlikely.)
285:
286: It will be interesting to use the Causal Entropic Principle to compute
287: the probability distribution of other parameters, such as the
288: amplitude of primordial density perturbations, $\delta\rho/\rho$, the
289: spatial curvature, $k$, or the baryon to photon ratio, $\eta$. A
290: crucial task will be to estimate the probability distribution over
291: multiple parameters at once, since this is a much more stringent test
292: for the environmental approach. For example, consider a distribution
293: over two parameters, $\rho_\Lambda$ and $\delta\rho/\rho$. When
294: weighting by observers-per-baryon, the upper bound on $\rho_\Lambda$
295: arises from the requirement that galaxies can form. Hence, it would
296: seem highly favorable to increase $\delta\rho/\rho$, since
297: $\rho_\Lambda$ could then be increased by the third power of the same
298: factor. This would render the observed values of both $\rho_\Lambda$
299: and $\delta\rho/\rho$ exceedingly unlikely. As we will discuss in
300: Sec.~\ref{sec-understand}, however, galaxy formation is not a
301: significant constraint on $\rho_\Lambda$ in our approach. Hence, we
302: expect this problem to be virtually absent.
303:
304: Given a multivariate distribution, one can ask about the probability
305: distribution over one parameter (say, $\log\rho_\Lambda$) with other
306: parameters integrated out. As more parameters are allowed to vary,
307: the distribution for $\log\rho_\Lambda$ is likely to become broader
308: after they are integrated out. Yet, the observed value must remain
309: typical if the environmental approach is to succeed. The most radical
310: choice is to study the distribution of $\log\rho_\Lambda$ after
311: integrating out all other parameters characterizing the landscape.
312: This would be tantamount to deriving the typical range of $\log
313: \rho_\Lambda $ from fundamental theory alone, without reference to
314: parameters specific to our own vacuum. This would have been
315: impossible in conventional approaches. But as we discuss in
316: Sec.~\ref{sec-deltas}, $\Delta S$ may depend simply on $\rho_\Lambda $
317: when averaged over the entire landscape. Hence, the Causal Entropic
318: Principle puts this task within our reach.
319:
320:
321: \section{Approaches to weighting vacua}
322: \label{sec-weighting}
323:
324: \subsection{The Weinberg bound}
325: \label{sec-weinberg}
326:
327: Weinberg~\cite{Wei87} estimated the range of $\rho_\Lambda$ compatible
328: with galaxy formation. No galaxies form in regions where
329: $\rho_\Lambda$ exceeds the matter density $\rho_{\rm m}$ at the time
330: when the corresponding density perturbations become nonlinear
331: (assuming otherwise identical physics). If we grant that galaxies are
332: a prerequisite for the existence of observers, then these regions will
333: not contain observers, and such values of $\rho_\Lambda$ will not be
334: observed. Combined with a similar argument\footnote{With
335: $\rho_\Lambda<0$, the universe will recollapse after a time of order
336: $|\rho_\Lambda|^{-1/2}$. If one assumes that most observers emerge
337: only after several billion years, then an upper bound on
338: $(-\rho_\Lambda)$ results by requiring that the universe should not
339: recollapse too soon~\cite{BarTip}.---We will consider only positive
340: values of $\Lambda$ in this paper.} for negative values of
341: $\rho_\Lambda$, Weinberg~\cite{Wei87} found that only values in the
342: interval
343: \begin{equation}
344: -1<\frac{\rho_\Lambda}{\rho_{\rm m}}<550
345: \label{eq-littlesteve}
346: \end{equation}
347: will be observed.
348:
349: The upper bound is larger than 1 because the matter density today,
350: $\rho_{\rm m}$, has been diluted since galaxy formation, by the
351: redshift factor $(1+z_{\rm gal})^3$. Weinberg used $z_{\rm
352: gal}\approx 4.5$, but in the meantime, dwarf galaxies have been
353: discovered at redshifts as high as $z_{\rm gal}\approx
354: 10$~\cite{Loeb:2006en}, raising the upper bound:
355: \begin{equation}
356: -1<\frac{\rho_\Lambda}{\rho_{\rm m}}<5000~.
357: \label{eq-bigsteve}
358: \end{equation}
359:
360: The observed ratio of vacuum to matter energy density is much smaller
361: than the upper bound:
362: \begin{equation}
363: \frac{\rho_\Lambda}{\rho_{\rm m}}\approx 2.3~.
364: \end{equation}
365: In other words, it would appear that the observed value of
366: $\rho_\Lambda$ is in fact quite unlikely, even allowing for the
367: anthropic constraint, Eq.~(\ref{eq-bigsteve}).
368:
369: Let us rephrase Weinberg's argument in a more general language. The
370: probability distribution for observed values of $\rho_\Lambda$ can be
371: written as
372: \begin{equation}
373: \frac{dP}{d\rho_\Lambda}\propto
374: w(\rho_\Lambda) \frac{dp}{dN} \frac{dN}{d \rho_\Lambda}~,
375: \end{equation}
376: where $N$ is the number of vacua with vacuum energy smaller than
377: $\rho_\Lambda$, and $p$ is the total prior probability for these
378: vacua. Since $\rho_\Lambda=0$ is not a special point, vacua in the
379: landscape are uniformly distributed when averaged over intervals of
380: $\rho_\Lambda$ of order $10^{-123}$ or smaller near $\rho_\Lambda=0$:
381: \begin{equation}
382: \frac{dN}{d \rho_\Lambda}=\mbox{const}~.
383: \end{equation}
384: Before anthropic selection, it is reasonable to assume that all vacua
385: are equally likely:\footnote{Dynamical effects can modify this flat
386: prior~\cite{SchVil06}. We shall assume that the resulting
387: distribution remains effectively flat, at least for small
388: $|\rho_\Lambda|$. Models violating this assumption are unlikely to
389: solve the cosmological constant problem.}
390: \begin{equation}
391: \frac{dp}{dN}=\mbox{const}~.
392: \end{equation}
393: The anthropic condition of galaxy formation assigns a weight $w=1$ to
394: vacua in the range of Eq.~(\ref{eq-bigsteve}), and $w=0$ to all other
395: vacua. Thus, all values of $\rho_\Lambda$ in the interval of
396: Eq.~(\ref{eq-bigsteve}) are equally likely,
397: $dP/d\rho_\Lambda=\mbox{const}$.
398:
399: Restricting to vacua with $\rho_\Lambda>0$, it is instructive to
400: consider the probability distribution as a function of $\log
401: \rho_\Lambda$,
402: \begin{equation}
403: \frac{dP}{d\log\rho_\Lambda} = \rho_\Lambda\frac{dP}{d\rho_\Lambda}
404: \propto \rho_\Lambda w(\rho_\Lambda) ~.
405: \label{eq-plog}
406: \end{equation}
407: With the above, ``binary'' weight, this distribution will be a growing
408: exponential of $\log \rho_\Lambda$, with a sharp cutoff at $\log
409: \rho_\Lambda\approx -120$ from the upper bound in
410: Eq.~(\ref{eq-bigsteve}). The observed value, $\log
411: \rho_\Lambda\approx -123$, is suppressed by about 3 orders of
412: magnitude compared to values near the cutoff.
413:
414:
415: \subsection{Weighting by observers per baryon}
416: \label{sec-opb}
417:
418: In order to reduce this discrepancy, one would need to go beyond the
419: binary question of whether or not there are observers. Surely the
420: {\em number\/} of observers will depend on the cosmological constant
421: and will begin to decrease for values of $\rho_\Lambda$ smaller than the
422: upper bound in Eq.~(\ref{eq-bigsteve}). If we weight vacua
423: by this number, perhaps this will be more effective at suppressing
424: large values of $\rho_\Lambda$ than a simple binary filter.
425:
426: Unfortunately, this strategy is not well-defined without a
427: regularization scheme. The dynamics of eternal inflation results in a
428: multiverse containing an infinite number of regions for every value of
429: $\rho_\Lambda$. Each region is an open universe with infinite spatial
430: volume at all times. (The hyperbolic spatial geometry reflects the
431: symmetries preserved by a vacuum bubble formed in a first-order phase
432: transition from a higher metastable vacuum.) If a vacuum admits any
433: observers at all, their number will be infinite.
434:
435: Various authors~\cite{MarSha97,GarLiv99,Efs95} have proposed to weight
436: vacua by the number of observers per baryon, or per photon, or per
437: unit matter mass. But none of these choices are particularly well
438: motivated. If there are infinitely many baryons, why should it matter
439: how efficiently they are converted to observers? Why is a vacuum less
440: likely to be observed if a smaller fraction of its mass becomes
441: observers, as long as there are infinitely many of them?
442:
443: More importantly, these regularization methods are not universally
444: defined. This makes them inapplicable in a rich landscape, where we
445: will eventually be forced to consider vacua with very different low
446: energy physics. Two vacua may have different baryon-to-photon ratios,
447: so that the above weighting methods are inequivalent; which should we
448: choose? Indeed, it seems unlikely that a standard definition of
449: ``baryon'' can be given that would be meaningful in all
450: vacua.\footnote{Ref.~\cite{Vilenkin} proposes an interesting method
451: for defining a unit comoving volume in different vacua, in the limit where
452: bubbles preserve an exact $SO(3,1)$ symmetry.}
453:
454: These difficulties arise because the regularization refers to a
455: reference particle species such as ``baryons''. But it also refers to
456: ``observers'', and this leads to additional problems. It seems
457: virtually impossible to define what an observer is in vacua with
458: different low energy physics. Even in our own universe, it is unclear
459: how to estimate the number of observers per baryon. One approximation
460: is to assume that it will be proportional to the fraction of baryons
461: that end up in galaxies. But this fraction depends strongly on the
462: minimum mass of a galaxy capable of harboring observers, $M_*$, which
463: is not known.
464:
465: %
466: \EPSFIGURE[t]{anthropic.eps,width=.7\textwidth}{\label{anthrocurves}
467: Weighted by ``observers per baryon'', the probability distribution
468: for $\rho_\Lambda$ depends strongly on specific assumptions about
469: conditions necessary for life. Three curves are shown,
470: corresponding to different choices for the minimum required mass of
471: a galaxy: $M_* = (10^7, 10^9, 10^{12}) M_\odot$. In neither case is
472: the observed value (vertical bar) in the preferred range. The
473: choice $M_*=10^7M_\odot$ (also shown in Fig.~1) corresponds to the
474: smallest observed galaxies. The choice $M_* = 10^{12} M_\odot$
475: minimizes the discrepancy with observation but amounts to assuming
476: that only the largest galaxies can host observers. By contrast, the
477: Causal Entropic Principle does not assume that observers require
478: structure formation, let alone galaxies of a certain mass; yet its
479: prediction is in excellent agreement with the observed value (see
480: Fig.~8).}
481: %
482: Figure~\ref{anthrocurves} shows probability distributions for
483: $\rho_\Lambda$, under various assumptions for $M_*$. Dwarf galaxies
484: as small as $M_*\sim 10^7 M_\odot$ have been detected. With this
485: choice, the observed value of $\Lambda$ is nearly three orders of
486: magnitude, or 3.5$\sigma$, below the median value.
487:
488: There is some evidence that galaxies with mass below $10^9 M_\odot$
489: will not retain the heavy elements produced in supernova
490: explosions~\cite{Vil04}. Under the {\em additional assumption\/} that
491: such elements are required for life, one may then set $M_*$ to this
492: larger value. But the resulting prediction remains unsatisfactory.
493: The median exceeds the observed value by a factor of more than two
494: orders of magnitude, or 2.9$\sigma$.
495:
496: One can speculate that for some reason, life requires a galaxy as
497: large as our own, or perhaps even a larger group~\cite{MarSha97,Vil04}
498: ($M_*=10^{12} M_\odot$). Then the observed value is about
499: $1.8\sigma$, or a factor of $22$, below the median of the
500: predicted distribution. However, at present we can see no compelling
501: arguments for this extreme choice. Thus, the weighting by
502: observers-per-baryon leads to a dilemma: Either it requires overly
503: specific and questionable assumptions, or else it faces a real
504: conflict with the data.
505:
506: As these difficulties demonstrate, weighting by observers-per-baryon
507: may not be the correct way to compute probabilities in the landscape.
508: We will now argue for a different approach, which is always
509: well-defined. It will allow us to assume nothing more about observers
510: than that they respect the laws of thermodynamics.
511:
512:
513: \subsection{Weighting by entropy production in the causal diamond}
514: \label{sec-cep}
515:
516: \paragraph{Causal Entropic Principle}
517: In this paper we will compute the probability distribution for
518: $\rho_\Lambda$ based on the Causal Entropic Principle, which is
519: defined by the following two conjectures~\cite{Bou06}:
520: \begin{itemize}
521: \item[(1)]{The universe consists of one causally connected region, or
522: ``causal diamond''. Larger regions cannot be probed and should
523: not be considered part of the semi-classical geometry.}
524: \item[(2)]{The number of observations is proportional to $\Delta S$,
525: the total entropy production in the causal diamond.}
526: \end{itemize}
527: %
528: \EPSFIGURE[t]{DS0fig.eps,width=.6\textwidth}{\label{fig-deltas} A
529: causal diamond is the largest spacetime region over which matter can
530: interact. It is delimited by a future lightcone from a point on the
531: reheating surface (orange/light), and by a past lightcone from a late-time
532: event (blue/dark); in the case of de~Sitter vacua this is the
533: cosmological horizon. A vacuum should be weighted by the number of
534: observations made in this spacetime region. Since observation
535: requires free energy, we expect that on average, this number will be
536: proportional to the amount of entropy, $\Delta S$, produced in the
537: causal diamond. Entropy entering through the bottom cone (bottom
538: arrow), such as the CMB, does not contribute to this entropy
539: difference.}
540: %
541:
542: Before motivating the two conjectures, let us first clarify the key
543: terms---causal diamond and entropy---they refer to.
544:
545: The causal diamond~\cite{Bou00a} is the largest region of spacetime
546: causally accessible to a single observer. It is defined by the
547: intersection of the past light-cone of a late-time point on the
548: worldline with a future light-cone of an early-time point, shown in
549: Fig.~\ref{fig-deltas}. We choose this time to be the time of
550: reheating, since no matter existed before then. (In vacua with no
551: reheating, $\Delta S$ vanishes independently of the choice of the
552: causal diamond. This case does not arise here, since we are holding
553: all parameters other than $\rho_\Lambda$ fixed.) Only after reheating
554: can matter begin to interact and commence the formation of complex
555: structures, at most at the speed of light.
556:
557: In a vacuum with negative cosmological constant, the tip of the top
558: cone would be on the big crunch. In any metastable vacuum with
559: positive cosmological constant, like ours, the top cone is given by
560: the de~Sitter event horizon.
561:
562: One may question the universality of ``reheating surface'' or our use
563: of an event horizon (a global concept) in a vacuum with finite
564: lifetime, so let us give a more careful definition. A causal diamond
565: can be triggered (that is, a bottom cone drawn) as soon as there is
566: entropy in the matter sector. Reheating is a special case: during
567: inflation, all entropy is in the gravitational sector (the growing
568: horizon of the inflationary universe), but reheating generates matter
569: entropy (mostly radiation, which we include in this class). After
570: vacuum domination in a metastable de~Sitter region, the diamond
571: empties out at an exponential rate. If we enlarge the diamond by
572: choosing a later point for the tip, and the additional spacetime
573: volume contains no matter (which will be generic at late times), there
574: is no point in going further. The total amount of matter enclosed by
575: the top cone will be the same as if the vacuum had been completely
576: stable. Once the metastable vacuum decays, reheating may again occur,
577: in which case a new diamond is triggered. This definition implies
578: that if a decay or phase transition happens while there is still
579: matter around, the two vacua should not be considered separate, and a
580: single causal diamond will enclose both. Thus, we are fundamentally
581: defining the range of a causal diamond in terms of the presence of
582: excitations in the matter sector. This is well-defined in the entire
583: regime of semi-classical gravity, independently of the details of the
584: particle physics.
585:
586: In a de~Sitter vacuum, the cosmological horizon has entropy given by
587: one quarter of its surface area. The relevant area is the
588: cross-section of the top-cone of the causal diamond, which grows as
589: matter crosses the horizon. This production of Bekenstein-Hawking
590: entropy would contribute enormously to $\Delta S$. However, it is
591: difficult to see the relevance of horizon entropy to the existence of
592: observers. For the same reason, we will ignore the Bekenstein-Hawking
593: entropy produced in black hole formation. This is also natural since
594: we have defined the causal diamond through this distinction. As we
595: have emphasized above, horizons are a gravitational phenomenon and can
596: always be distinguished from the matter sector in the semiclassical
597: regime. Hence, this restriction does not affect the universality of
598: our method.\footnote{A different question is whether the exclusion of
599: black hole horizon entropy actually makes a qualitative difference
600: to the results of this paper. It seems likely that it would not.
601: According to the Causal Entropic Principle, the preferred value of
602: $\rho_\Lambda$ is set by the time of maximum entropy production.
603: The growth of supermassive black holes, and of their entropy, is
604: largest during the era a few Gyr after the big bang and eventually
605: slows down. Thus, it appears to set a similar timescale to the one
606: we obtain from stellar entropy production.}
607:
608: To summarize, we will consider exclusively the production of entropy
609: in ordinary matter. We will weight a vacuum by the total entropy
610: increase in the causal diamond:
611: \begin{equation}
612: w=\Delta S~.
613: \label{eq-jkl}
614: \end{equation}
615:
616:
617: \paragraph{Motivation}
618: From a pragmatic point of view, one can simply regard this proposal as
619: an attractive alternative to weighting by observers-per-baryon. The
620: causal diamond is defined independently of low-energy physics, and the
621: entropy increase is a well-defined quantity that replaces more
622: specific assumptions about observers. However, there are additional,
623: more fundamental reasons to embrace the Causal Entropic Principle.
624: Let us discuss each of the two conjectures put forward at the
625: beginning of this subsection.
626:
627: The first conjecture is motivated by the study of black holes (see
628: \cite{Bou00a,Bou06,Bou06b} for details). There is now considerable
629: evidence that black hole formation and evaporation is a unitary
630: process~\cite{StrVaf96,Mal97}. This means that there will be two
631: copies of the initial state at the same instant of time, one inside
632: the black hole, and the other in the Hawking radiation outside. This
633: would appear to violate the linearity of quantum mechanics. However,
634: no actual observer can verify the problem~\cite{SusTho93,Pre92}, since
635: the two copies reside in causally disconnected regions of the
636: spacetime. Hence, we can escape from the apparent paradox by
637: abandoning the global viewpoint, and be content with merely predicting
638: the observations of (any) one observer.
639:
640: However, it would be unnatural for such a radical revision of our view
641: of spacetime to be confined to the context of black holes. In many
642: cosmological solutions, a single observer can access only a small
643: portion of the global spacetime. Hence, it is equally important to
644: restrict our attention to only one (cosmological) observer at a time,
645: which is what we do in this paper. Descriptions of cosmology from the
646: local viewpoint can be found, for example, in
647: Refs.~\cite{Ban00,Bou00a,Bou00b,DysKle02,Bou05}. Its implications for
648: eternal inflation were studied in Ref.~\cite{BouFre06}.
649:
650: The second conjecture replaces far more specific conditions assumed to
651: be necessary for observers, such as the existence of galaxies, stable
652: planetary orbits, suitable chemistry, etc., which were needed in the
653: observers-per-baryon approach. The basic idea is that observers,
654: whatever their form, have to obey the laws of thermodynamics.
655: Observation requires free energy and is clearly incompatible with
656: thermal equilibrium or an empty universe. The free energy, divided by
657: the temperature at which it is radiated, can be regarded as a measure
658: of the potential complexity arising in a spacetime region. This is
659: equal to the number of quanta produced, or more fundamentally, the
660: entropy increase $\Delta S$.
661:
662: It seems plausible that there might be an absolute minimum complexity
663: necessary for observers, so that subcritical weights $\Delta S<\Delta
664: S_{\rm crit}$ should be replaced by zero. For example, it seems
665: likely that vacua with $\rho_\Lambda$ of order unity, which can
666: contain only a few bits, have strictly zero probability of hosting
667: observers (see also Ref.~\cite{HarKri06}). However, any such cutoff
668: does not appear to play an important role for the questions studied
669: here. We choose the weight factor to be simply $\Delta S$, with no
670: cutoff.
671:
672: To avoid confusion, we stress that our weighting has nothing to do
673: with the Hartle-Hawking amplitude, $\exp(S_{\rm dS})$, which describes
674: the number of quantum states associated with a de~Sitter
675: horizon~\cite{HarHaw83}. The number of observers, and of
676: observations, is naturally proportional to an entropy difference,
677: $\Delta S$, not to an absolute entropy or the exponential of an
678: entropy. (Despite our appropriation of the term ``entropic
679: principle'', for which we apologize to the authors, there is also
680: little relation to Ref.~\cite{hepth0509109}.)
681:
682: We could have adopted only the first conjecture, and continued to
683: estimate the number of observers by more explicit anthropic criteria.
684: This would not have changed our final result significantly. But why
685: make a strong assumption if a more conservative one suffices? Our
686: results suggest that the poor predictions from weighting by
687: observers-per-baryon stemmed not from a lack of specificity in
688: characterizing observers, but from the regularization scheme (``per
689: baryon''). The causal diamond cutoff solves this problem, and allows
690: us to weaken anthropic assumptions to the level of an inevitable
691: thermodynamic condition.
692:
693: Moreover, $\Delta S$ is a well-defined weight in any vacuum, however
694: different from ours. It will be a powerful tool in future work, when
695: parameters other than $\rho_\Lambda$ are allowed to vary
696: simultaneously. We are thus motivated to use our second conjecture
697: throughout. We will find that in our own universe, it captures
698: conditions for observers remarkably well.
699:
700:
701: \section{Computing $\rho_\Lambda$ from the Causal Entropic Principle}
702: \label{sec-calculation}
703:
704: In this section, we will compute the probability distribution over
705: $\rho_\Lambda$, holding all other physical parameters fixed. We do
706: this in four steps. First, we will compute the geometry of the causal
707: diamond as a function of $\rho_\Lambda$. Next, we will identify
708: important effects that produce entropy within the causal diamond.
709: Then we will determine the time-dependence of the entropy production
710: rate per unit comoving volume, as a function of $\rho_\Lambda$.
711: Finally, we will fold this together with the time-dependence of the
712: comoving volume contained in the causal diamond to obtain the weight
713: factor, $\Delta S(\rho_\Lambda)$.
714:
715: \subsection{Metric and causal diamond}
716: \label{sec-metric}
717:
718: Current data are consistent with a spatially flat universe. Hence, we
719: will assume that since the time of reheating, the large scale
720: structure of our universe is described by a spatially flat FRW
721: model:
722: \begin{equation}
723: ds^2 = -dt^2 + a(t)^2 d {\mathbf x} ^2~.
724: \end{equation}
725: (Actually, we are making the stronger assumption that the cosmological
726: constant dominates before curvature does, for {\em all\/} values of
727: $\rho_\Lambda$ considered here. Thus, we are assuming that the
728: universe is flatter than necessarily required by current constraints.)
729:
730: After reheating, the universe will be dominated first by radiation,
731: then by matter, and finally by vacuum energy. The reheating
732: temperature will not be important; in fact, we will neglect the
733: radiation era altogether. Instead, we extrapolate the
734: matter-dominated era all the way back to the big bang ($t=0$), where
735: we will place the bottom tip of the causal diamond. This
736: approximation is justified because the radiation era contributes only
737: a small fraction of conformal time in the range of values of $\rho_\Lambda$
738: that have any significant probability. This is shown in detail in
739: Appendix~\ref{sec-radiation}.
740:
741: Thus, we treat the universe as containing only pressureless matter and a
742: cosmological constant. At early times (matter domination), the scale
743: factor will be proportional to $t^{2/3}$, independently of $\rho_\Lambda$.
744: At late times (vacuum domination) it will grow like
745: $\exp(t/t_\Lambda)$, where
746: \begin{equation}
747: t_\Lambda= \sqrt{\frac{3}{\Lambda}} =
748: \sqrt{\frac{3}{8\pi\rho_\Lambda}}~,
749: \label{eq-tlambda}
750: \end{equation}
751: (In our universe, with $\rho_\Lambda$ given by Eq.~(\ref{eq-cc}), we
752: have $t_\Lambda\approx 0.98 \times 10^{61} = 16.7$ Gyr).
753:
754: An exact solution (aside from the neglected radiation era) that
755: includes both regimes is
756: \begin{eqnarray}
757: a(t) & = & \left[t_\Lambda\sinh\left(\frac{3}{2}
758: \frac{t}{t_\Lambda}\right)\right]^{2/3}\,,
759: \label{eq-aexact}\\
760: \rho(t) & = & \rho_\Lambda+\rho_{\rm m} =
761: \rho_\Lambda \left[1 + \frac{1}{\sinh^2\left(\frac{3}{2}
762: \frac{t}{t_\Lambda}\right)}\right]~.
763: \label{eq-rhoexact}
764: \end{eqnarray}
765: The prefactor, $t_\Lambda^{2/3}$, is arbitrary and can be changed by
766: rescaling the spatial coordinates. Our choice ensures that for
767: solutions with different values of $\Lambda$, the scale factors will
768: agree at early times not only by diffeomorphism, but explicitly. This
769: is convenient because $\Lambda$ is dynamically irrelevant at early
770: times.
771:
772: Vacuum energy begins to dominate the density when $\rho_\Lambda
773: =\rho_{\rm m}$, at $t=0.59\, t_\Lambda$ (in our universe, 9.8 Gyr
774: after the big bang). Acceleration begins earlier, when
775: $\rho+3p=\rho_{\rm m}-2\rho_\Lambda =0$, at $0.44\, t_\Lambda$ (in our
776: universe, at 7.3 Gyr).
777:
778: In order to compute the boundaries of the causal diamond, it is
779: convenient to work in conformal time, $\tau=\int dt/a(t)$. The metric
780: becomes
781: \begin{equation}
782: ds^2 = a(\tau)^2 [-d\tau^2+d {\mathbf x}^2]~.
783: \end{equation}
784: Light-rays obey $ds=0$, and hence for radial light-rays, $d\tau=\pm
785: dr$, where $r=|{\mathbf x}|$.
786:
787: Using our solution, Eq.~(\ref{eq-aexact}), conformal time will be
788: given by
789: \begin{eqnarray}
790: \tau(t) & = & - t_\Lambda^{1/3}
791: \frac{1}{\cosh^{2/3}\left(\frac{3t}{2t_\Lambda}\right)}
792: F\left(\frac{5}{6},\frac{1}{3},\frac{4}{3},
793: \frac{1}{\cosh^2\left(\frac{3t}{2t_\Lambda}\right)}\right)\\ & = & -
794: \frac{t_\Lambda}{a(t)}
795: F\left(\frac{1}{3},\frac{1}{2},\frac{4}{3},
796: \frac{-t_\Lambda^2}{a(t)^3}\right)\,.
797: \end{eqnarray}
798: It has finite range, running from
799: \begin{equation}
800: \tau(0) =
801: -\frac{\Gamma(\frac{4}{3})\Gamma(\frac{1}{6})}{\Gamma(\frac{1}{2})}
802: t_\Lambda^{1/3} \approx -2.804\, t_\Lambda^{1/3}
803: \label{eq-tau0}
804: \end{equation}
805: at the big bang, to $\tau(\infty)=0$ at asymptotically late times.
806: The total conformal lifetime of the universe is $\Delta\tau=
807: \tau(\infty)-\tau(0) = 2.804\, t_\Lambda^{1/3}$.
808:
809: The causal diamond is given by the region delimited by
810: \begin{equation}
811: r= \frac{\Delta\tau}{2}-\left|\frac{\Delta\tau}{2}+\tau\right|~.
812: \label{eq-cd}
813: \end{equation}
814: The comoving volume inside the causal diamond at conformal time $\tau$
815: is simply
816: \begin{equation}
817: \label{covol}
818: V_{\rm c}=\frac{4\pi}{3} r^3(\tau)\,.
819: \end{equation}
820: This is shown in Fig.~\ref{fig-covol}, as a function of proper time,
821: %
822: \EPSFIGURE[!t]{fig4.eps,width=.8\textwidth}{\label{fig-covol} The
823: lower plot shows comoving volume in the causal diamond, $V_{\rm c}$, as a function of time
824: for $\rho_\Lambda = (0.1, 1, 10)$ times the observed value given in
825: Eq.~(\ref{eq-cc}). The kink in the comoving volume corresponds to
826: the ``edge'' of the causal diamond, where the top and bottom cone meet (see Fig. 3). The
827: upper plot shows the rate of entropy production computed in
828: Sec.~\ref{sec-rate}, which peaks around 2 to 3.5 Gyr. As explained
829: in Sec.~\ref{sec-understand}, the Causal Entropic Principle prefers
830: values of $\rho_\Lambda$ such that the ($\rho_\Lambda$-dependent)
831: peak of the comoving volume coincides with the (nearly
832: $\rho_\Lambda$-independent) peak of the entropy production rate (see
833: Fig.~9.}
834: %
835: for several values of $\rho_\Lambda$. The maximum comoving volume
836: occurs at the edge of the causal diamond at conformal time
837: $\tau(0)/2$, or equivalently,
838: \begin{equation}
839: t_{\rm edge} \approx 0.23 t_{\Lambda}
840: \label{eq-edge}
841: \end{equation}
842: (in our universe, $t_{\rm edge} \approx 3.9$ Gyr). The maximum
843: comoving volume itself is $V_{\rm c}(\tau(0)/2) \approx 11.6 t_\Lambda$.
844:
845: At late times, the comoving volume goes to zero. This reflects the
846: exponential dilution of all matter, which is expelled from the diamond
847: by the accelerated expansion. The physical radius approaches a
848: constant, $t_\Lambda$, the horizon radius of the asymptotic de~Sitter
849: space. Note that the ``comoving four-volume'',
850: \begin{equation}
851: V_4\equiv\int_0^\infty V_{\rm c}(t) dt~,
852: \end{equation}
853: is finite. It is proportional to $t_\Lambda^2$, and hence, inversely
854: proportional to the cosmological constant. This will be important,
855: since it means that smaller values of $\rho_\Lambda$ are rewarded by a
856: larger causal diamond, and thus, potentially greater complexity. This
857: can compensate for their rarity.
858:
859:
860: \subsection{Major sources of entropy production}
861: \label{sec-entropy}
862:
863: To calculate $dP/d(\log \rho_\Lambda)$, we need to calculate the total
864: entropy production in the causal diamond as a function of the
865: cosmological constant, $\Delta S(\rho_\Lambda)$. We have determined
866: above how the causal diamond depends on $\rho_\Lambda$, but we must
867: also understand how the rate of entropy production depends on
868: $\rho_\Lambda$. We begin by identifying the major sources
869: contributing to $\Delta S$ in the causal diamond in our own vacuum.
870:
871: First, let us discuss sources which can be neglected. Because the
872: causal diamond is small at early times ($\frac{\tau}{\tau_\infty}\ll
873: 1$), and empty at late times ($1-\frac{\tau}{\tau_\infty}\ll 1$), the
874: most important contributions will be produced in the era between 0.1
875: Gyr to 100 Gyr, when the comoving volume shown in Fig.~\ref{fig-covol}
876: is large. Hence, we can disregard the entropy produced at reheating,
877: at phase transitions, or by any other processes in the early universe.
878: Virtually all of this entropy (in particular, the cosmic microwave
879: background) entered the causal diamond through the bottom cone and
880: does not contribute to $\Delta S$.
881:
882: For the same reason, we neglect the entropy in Hawking radiation
883: produced by supermassive black holes. (One might contemplate the
884: possibility of a ``planet'' orbiting such an object and exploiting its
885: very low temperature radiation as a source of free energy.) However,
886: the timescale for the evaporation of a black hole is enormous ($M^3$).
887: By the time this entropy would be produced, a typical causal diamond,
888: on which the measure for prior probabilities is based in the local
889: approach~\cite{Bou06}, will be completely empty.
890:
891: Having dismissed effects at small comoving volume, we turn to
892: processes which operate from 0.1 Gyr to 100 Gyr, when the comoving
893: volume is large. In this era, entropy is produced by baryonic
894: systems, and we can gauge the importance of a given process by its
895: total entropy production per baryon, or equivalently, per unit mass,
896: or unit comoving volume,\footnote{The choice of reference unit does
897: not affect our weighting, because it drops out after integrating
898: over the causal diamond.} $dS/dV_{\rm c}$. This can be estimated as
899: the ratio between the amount of energy released per baryon, and the
900: temperature at which this energy is dominantly radiated.
901:
902: \paragraph{Stars} Ten percent of baryons end up in galaxies, and thus
903: in stars. Approximately ten percent of those baryons actually burn in
904: the course of the lifetime of the star. The energy released is about
905: 7 MeV for each baryon in the reaction 4p $\rightarrow$ $^4$He, or about
906: $0.7 \times 10^{-2}$ of the rest mass of the proton. Stars radiate at
907: varying temperatures corresponding largely to optical wavelengths
908: (about 0.2 to 3 eV). However, more than half of this radiation is
909: reprocessed by dust~\cite{astroph9812182}.\footnote{We thank Eliot
910: Quataert for pointing this out to us.} It is re-emitted in the
911: infrared, at approximately $60\,\mu$m, or $20$ meV. Hence, there will
912: be more than 100 infrared photons per optical photon~\cite{DolLag06},
913: and the infrared emission will dominate the entropy production. In
914: summary, stellar burning produces an entropy of order $10^6$ per
915: baryon, after reprocessing by dust.
916: %from .1 .1 6.75MeV .5 / 20meV = 1.6 E6
917:
918: \paragraph {Active galactic nuclei} Active galactic nuclei (AGNs)
919: appear to be the main competitor to stars in terms of luminosity and
920: entropy production.\footnote{We thank Petr Horava for drawing our
921: attention to AGNs.} Approximately $10^{-3}$ of the total stellar
922: mass (i.e., $10^{-4}$ of baryonic mass) ends up in supermassive black
923: holes, and perhaps 10\% of this energy is released during the violent
924: accretion process. This suggests that AGNs release at most about one
925: sixth of the energy as compared with stellar burning.
926: %from .1 .001 .1 1GeV / .1 .1 6.75MeV = 1 MeV / 6.75 MeV
927: The effective temperature will again be 20 meV, since a large fraction
928: of the radiation is reprocessed by dust. Hence, AGNs appear to
929: produce somewhat less total entropy per baryon than stars, and we will
930: neglect their contribution in the present paper.
931:
932: In Ref.~\cite{Hopkins:2006fq}, the intrinsic luminosities of all AGNs
933: above observational limits were compiled to create a quasar luminosity
934: function applicable back to about 1 Gyr after the big bang. This
935: luminosity function suggests that the entropy production rate from
936: quasars is more narrowly peaked than that estimated for stars in
937: Fig.~\ref{fig-dSdVdt}. But even at the peak, near a redshift of
938: $z=2$, the quasar luminosity is a factor of 3 lower than the {\em
939: present\/} stellar luminosity~\cite{Kashlinsky:2004jt} (which, in
940: turn, is about one order of magnitude lower than the peak stellar
941: luminosity). This seems to rule out the possibility that AGNs ever
942: dominated the entropy production rate.
943:
944: Other potentially important contributions come from galaxy cooling and
945: from supernovae. Even assuming reprocessing by dust, neither of these
946: phenomena can compete with stellar burning, because they run on less
947: energy but not at lower temperature.
948:
949: \paragraph{Supernovae}
950: We focus on core collapse supernovae since they are more abundant than
951: type Ia supernovae (by factor of 4-5) while producing comparable
952: luminosity per event. They occur in all stars with more than eight
953: solar masses, which constitute 1\% of the total stellar mass [see
954: Eq.~(\ref{imf-eq})]. The collapse of an iron core into a neutron
955: star releases gravitational binding energy not much smaller than the
956: core mass (1.4 solar masses). Thus roughly 10\% of the total mass of
957: the progenitor is released. Most of this energy is carried away by
958: high energy neutrinos, producing little entropy. Only 1\% produces
959: optical photons, and is reprocessed by dust into infrared photons.
960: Altogether, supernovae thus convert a fraction of $10^{-5}$ of stellar
961: baryonic mass into soft photons.
962:
963: Further quantitative analysis confirms the above estimate. We find
964: that entropy production from supernovae is more than order of
965: magnitude below the contribution from dust heated by stars.
966:
967:
968: \paragraph{Galaxy cooling} A typical galaxy like ours has a
969: mass-to-radius fraction of approximately $10^{-6}$. This fraction of
970: the galaxy mass is converted into kinetic energy at virialization.
971: This energy, about $1$~keV per stellar baryon, is converted into
972: radiation as the galaxy cools. The virial temperature is about $10^5$
973: K, or 10 to 100 times greater than the temperature of a star. Even
974: assuming reprocessing by dust, galaxy cooling will produce less than
975: $10^4$ photons per baryon. This is more than two orders of magnitude
976: below the entropy production from dust heated by stars.
977:
978:
979: \subsection{Entropy production rate}
980: \label{sec-rate}
981:
982: We have argued that the dominant source of entropy production in our
983: universe is dust heated by starlight. In this subsection we will
984: consider the rate at which this entropy is produced. We will ask how
985: it depends on time and on $\rho_\Lambda$.
986:
987: \paragraph{Time dependence}
988: Deriving $dS/dV_{\rm c}\, dt$ from first principles would require a
989: detailed description of all of the dynamics that led up to stellar
990: burning, such as non-linear evolution, gas cooling and disk
991: fragmentation. Instead, we will take advantage of observations that
992: quantify the time-dependence of the star formation rate. This will
993: allow us to obtain the entropy production from dust heated by stars.
994: We will then estimate how this rate changes in universes with
995: different cosmological constants. Surprisingly, this latter
996: dependence will not be important for our final result.
997:
998: In recent years, observations of the extragalactic background
999: radiation in a large range of wavelengths have improved our
1000: understanding of the galaxy luminosity function. This has allowed
1001: astronomers to infer the star formation rate (SFR) in galaxies, and
1002: its evolution in time~\cite{Madau:1996yh,Lilly:1996ui}
1003: (see~\cite{Kennicutt:1998zb} for a review). The SFR is defined as the
1004: rate of stellar mass production per comoving volume
1005: \begin{equation}
1006: \dot \rho_\star (t)\equiv
1007: \frac{\partial^2 M_\star}{\partial V_{\rm c} \partial t}\,.
1008: \end{equation}
1009: This function is constrained by observation through a variety of
1010: techniques. For example, UV emission from star-forming galaxies is
1011: dominated by massive stars that are short-lived. Due to the short
1012: lifetime, luminosities in these wavelengths track the birth rate of
1013: stars~\cite{Madau2}. In addition, detailed surveys of the local
1014: universe constrain the SFR at low
1015: redshift~\cite{Cole:2000ea,Hopkins:2003am}. Bounds on the rate of
1016: type II supernovae from Super-Kamiokande and KamLAND also place an
1017: indirect bound on stellar birth~\cite{Fukugita:2002qw}.
1018:
1019: The combination of these measurements constrain the SFR back to
1020: redshifts of about $z\sim 6$, when the universe was 1 Gyr old. Since
1021: then, the SFR may be grossly described as a smooth function that peaks
1022: at around 2.5 Gyr and subsequently decreases exponentially with
1023: time. The SFR today is roughly one order of magnitude less than its
1024: peak value. As we shall see, the era of peak stellar formation, which
1025: we will call $t_\star$, will play an essential role in determining the
1026: cosmological constant using the Causal Entropic Principle.
1027:
1028: Several authors have postulated models or functional forms for the
1029: SFR, fitting model parameters to the data, for
1030: example~\cite{Cole,Gal-Yam:2003ed,Nagamine:2006kz,Hopkins:2006bw}.
1031: Variations between the fits lead to a range for $t_\star \sim 1.5
1032: - 3$ Gyr.
1033:
1034: We will illustrate our computation using two different SFRs, in order
1035: to illustrate the systematic dependence of our calculation on the time
1036: dependence of star formation. The first SFR is from
1037: Ref.~\cite{Nagamine:2006kz} (labeled ``N'' in plots) and has
1038: $t_\star\sim 1.7$ Gyr. The second SFR, from Ref.~\cite{Hopkins:2006bw}
1039: (labeled ``H''), peaks at $t_\star\sim 2.8$ Gyr. We will find in both
1040: cases that the observed value of $\rho_\Lambda$ lies well inside the
1041: $1\sigma$ region of the resulting probability distribution.
1042:
1043: %
1044: \EPSFIGURE[!t]{sfr.eps,width=.8\textwidth}{\label{fig-SFR}
1045: The star
1046: formation rate as a function of time from Nagamine {\em et
1047: al.}~\cite{Nagamine:2006kz}, labeled ``N'', and from Hopkins and
1048: Beacom~\cite{Hopkins:2006bw}, labeled ``H''.
1049: They are peaked at $t_\star\sim$1.7 Gyr and 2.8
1050: Gyr respectively. We have normalized both SFRs such that the stellar
1051: luminosity density calculated below agrees with the bolometric
1052: luminosity observed today. It should be noted, however, that our
1053: result is not sensitive to this re-normalization.}
1054: %
1055: Both SFRs are shown in Fig.~\ref{fig-SFR} re-normalized. One of the
1056: biggest uncertainties regarding the determination of the star
1057: formation rate is the overall normalization of the curve. Fortunately
1058: our result is \emph{completely} insensitive to this overall
1059: normalization; it depends only on the shape of the SFR.
1060:
1061: For concreteness, we have normalized the SFR such that the implied
1062: stellar luminosity today matches the observed bolometric luminosity
1063: from low-redshift stars, $10^{8.6} L_\odot/\mathrm{Mpc}^3$. Our
1064: normalizations differs from those in the original works since we have
1065: used rather rudimentary formulae to compute the present luminosity
1066: from the SFR. However, the shape of the total entropy production rate
1067: derived with these simple formulae agrees well with that in
1068: Ref.~\cite{Nagamine:2006kz} when this SFR is used.
1069:
1070: With the rate of star formation in hand, one can estimate the rate of
1071: entropy production by stars. Let us first consider a single star of
1072: mass $M$ born at a time $t'$. Its entropy production rate is the
1073: stellar luminosity divided by the temperature at which photons are
1074: radiated,
1075: \begin{equation}
1076: \frac{d^2s}{dN_\star dt}(M)= \frac{L_\star}{T_{\rm eff}}\sim
1077: \frac{1}{2}\left(\frac{M}{M_\solar}\right)^{3.5}\frac{L_\solar}{20\,
1078: \mathrm{meV}} = \frac{1}{2}\left(\frac{M}{M_\solar}\right)^{3.5}
1079: 3.7 \times 10^{54} \ \mathrm{yr}^{-1}\,.
1080: \end{equation}
1081: Here we have assumed a mass-luminosity relation
1082: \begin{equation}
1083: \label{eq-masslum}
1084: L_\star\propto M^{3.5}\,.
1085: \end{equation}
1086: (We use $M_\star$ to refer to total stellar mass, and $M$ to refer to
1087: the mass of a specific star. $N_\star$ denotes star number.) The
1088: effective temperature of 20 meV is that of the dust which reprocesses
1089: about one half (hence the prefactor) of the starlight and dominates
1090: photon number.
1091:
1092: The star will only produce entropy over a finite time $t_\star$ that
1093: also depends on $M$:
1094: \begin{equation}
1095: \label{stellarlifetime}
1096: t_\star (M)\sim \left(\frac{M_\solar}{M}\right)^{2.5} 10^{10} \; \mathrm{yr}.
1097: \end{equation}
1098:
1099: Now let us consider a whole population of stars that are formed at a
1100: time $t'$. At birth, stellar masses are observed to be distributed
1101: according to a universal, time-independent function known as the
1102: initial mass function (IMF). The Salpeter
1103: distribution~\cite{Salpeter:1955it}, which goes as $M^{-2.35}$, agrees
1104: reasonably with observation, but has since been refined by many
1105: authors. In the present calculation we will use a modified Salpeter
1106: IMF of the form~\cite{Hopkins:2006bw}
1107: \begin{equation}
1108: \xi_{\mathrm{IMF}}(M)\equiv \frac{dN_\star}{dM} = \left\{
1109: \begin{array}{ll}
1110: C_1M^{-2.35} & \mbox{for}\ M\ge 0.5 M_\solar \\
1111: C_2M^{-1.5} & \mbox{for}\ M<0.5 M_\solar
1112: \end{array} \right.~, \label{imf-eq}
1113: \end{equation}
1114: where the constants $C_{1,2}$ are set by requiring that the IMF is
1115: continuous and that it integrate to one over the range $0.08
1116: M_\solar<M<100 M_\solar$.
1117:
1118: We now have all the ingredients in place to calculate $d^2S/(dM_\star
1119: dt)$, the contribution of a stellar population that is born at time
1120: $t'$ to the entropy production rate at some later time $t>t'$, per
1121: unit initial stellar mass. This rate is a function only of the time
1122: difference $t-t'$
1123: \begin{equation}
1124: \frac{d^2S}
1125: {dM_\star dt} (t-t') =\frac{1}{\langle M \rangle}
1126: \int_{M_{\mathrm{min}}}^{M_{\mathrm{max}}(t-t')} dM\,
1127: \xi_{\mathrm{IMF}} \frac{d^2s}{dN_\star
1128: dt}(M)~,
1129: \label{eq-lkj}
1130: \end{equation}
1131: with the average initial mass (defined at $t'$), $\langle M \rangle$,
1132: defined as
1133: \begin{equation}
1134: \langle M \rangle = \int_{0.08 M_\odot}^{100 M_\odot} dM\,
1135: \xi_{\mathrm{IMF}} M\approx 0.48 M_\odot ~.
1136: \end{equation}
1137: The time dependence enters Eq.~(\ref{eq-lkj}) through the death of stars
1138: of various masses at different times and thus appears in the upper
1139: limit of the upper integral. It is derived by inverting
1140: equation~(\ref{stellarlifetime}):
1141: \begin{equation}
1142: M_{\mathrm{max}}(t-t')=\left\{
1143: \begin{array}{ll}
1144: 100 M_\solar & \ \mbox{for}\ t-t'<10^5\ \mbox{yr} \\
1145: \left(\frac{10^{10} \; \mathrm{yr}}{t-t'}\right)^{2/5} M_\solar&
1146: \ \mbox{for}\ t-t'>10^5\ \mbox{yr} \\
1147: \end{array}\right.\,.
1148: \end{equation}
1149: The lower limit of the integral is set by the minimal mass of stars
1150: that can sustain nuclear burning,
1151: \begin{equation}
1152: M_{\mathrm{min}}=0.08 M_\solar\,.
1153: \end{equation}
1154: These stars burn for $\sim 5000$ Gyr, living well into vacuum domination in
1155: our universe.
1156:
1157: %
1158: \EPSFIGURE[!t]{dsvt_linlin.eps,width=.8\textwidth}{\label{fig-dSdVdt} The
1159: entropy production rate in our universe as a function of time, from
1160: dust heated by starlight. The two curves shown correspond to
1161: different models of the star formation history of the
1162: universe~\cite{Nagamine:2006kz,Hopkins:2006bw}; see Fig.~5.}
1163: %
1164: In order to calculate the entropy production rate in the universe at
1165: time $t$ we must convolve $d^2S/(dM_\star dt)$ with the SFR. That is,
1166: we integrate over the birth times $t'$ of all populations of stars
1167: born before the time $t$:
1168: \begin{eqnarray}
1169: \frac{d^2 S}{d V_{\rm c} d t}(t)&=&
1170: \int_0^t dt'\, \frac{d^2 S}{d M_\star d t}(t-t')
1171: \frac{d^2 M_\star}{d V_{\rm c} d t'}(t') \nonumber \\
1172: &=& \int_0^t dt'\,\frac{d^2 S}{d M_\star d t}(t-t')\, \dot \rho_\star(t')\,.
1173: \label{dSdVdt}
1174: \end{eqnarray}
1175:
1176: This function is plotted in Fig.~\ref{fig-dSdVdt} for the two forms of
1177: the SFR. It is a smooth function that peaks when the universe is
1178: about 2.3 Gyr old and 3.3 Gyr old for the two curves plotted. This
1179: timescale is set by the peak of the star formation rate and the mean
1180: lifetime of a star in our universe. The entropy production rate
1181: decreases as stars born during the peak of the SFR begin to die. But
1182: due to the high abundance of long-lived low mass stars,
1183: $\partial_{V_{\rm c}} \dot S$ maintains a finite value long after star
1184: formation has ceased.
1185:
1186: In our approximation, the entropy production rate is half the
1187: luminosity of stars at the time $t$, divided by the effective
1188: temperature (dust at 20 meV). Modulo this rescaling,
1189: Fig.~\ref{fig-dSdVdt} thus also shows our estimate for the luminosity,
1190: which was used for normalization as described above.
1191:
1192: We caution that we assumed overly simplistic formulae for the
1193: luminosity and the lifetime of a star. For example, we ignored the
1194: dependence on metallicity, and a dependence of the exponent in
1195: Eq.~(\ref{eq-masslum}) on the mass. This will likely affect the shape
1196: of the entropy production rate somewhat, as will corrections from
1197: other sources of entropy. However, our prediction for $\rho_\Lambda$
1198: depends only on the roughest features: the width and position of the
1199: peak of the entropy production rate. Hence, it is unlikely that our
1200: result would be qualitatively affected by our simplifications.
1201:
1202:
1203: \paragraph{Dependence on $\rho_\Lambda$}
1204: The entropy production rate calculated above is that in our universe.
1205: In order to determine $\Delta S(\rho_\Lambda)$, we will need to
1206: calculate this function for universes with different values of the
1207: cosmological constant. Interestingly, this dependence will be of
1208: little importance for our final result.
1209:
1210: Stars have decoupled from the global expansion of the universe, so
1211: their internal dynamics is unaffected by variations of $\rho_\Lambda$.
1212: However, the value of $\rho_\Lambda$ affects the fraction of baryons
1213: that are incorporated in halos large enough to form stars at any given
1214: time, thus affecting the star formation rate, and ultimately the
1215: entropy production. For example, if $\rho_\Lambda$ is large enough to
1216: violate Weinberg's anthropic bound, no baryons will be in star forming
1217: halos and $\Delta S$ will be very suppressed.
1218:
1219: In a universe with a vacuum energy density $\rho_\Lambda$, the
1220: fraction of matter that is incorporated in halos of a mass $M_G$ by
1221: time $t$ or above, $F(\rho_\Lambda, M_G, t)$, is easily calculated in
1222: the Press-Schechter (PS) formalism~\cite{Press:1973iz}. The formulae
1223: for the PS fraction are summarized in~\cite{Tegmark:2005dy}. (This
1224: fraction is also a function of the amplitude of density perturbations
1225: and the temperature at matter-radiation equality, but since these are
1226: held fixed in this calculation we will suppress them.)
1227:
1228: For the purpose of our calculation, we will consider a galaxy to be
1229: star-producing if the mass of its host halo is $10^7 M_\solar$ or
1230: above. Note that this choice involves no speculation about what
1231: observers need. The Causal Entropic Principle requires us to compute
1232: the entropy production rate as a function of $\rho_\Lambda$. This
1233: rate depends on whether or not stars, the dominant contributors of
1234: entropy, actually form. For $M_G>10^7 M_\odot$, the virial
1235: temperature of the halo is above $10^3$ K, enough to support rapid
1236: line cooling and efficient stellar production. (The first generation
1237: of stars---Population III---were formed in galaxies with even lower
1238: masses, but since these stars have been neither observed nor accounted
1239: for in the observation-based SFRs, we will not consider them further.)
1240: In any case, as we shall see, our final result will be quite
1241: insensitive to this choice.
1242:
1243: Based on the SFR $\dot \rho_\star (t)$ in our universe, we will now
1244: estimate the SFR in a universe with a different vacuum energy, $\dot
1245: \rho_\star (\rho_\Lambda,t)$. The SFR in our universe is peaked at
1246: about $t_\star\sim 1.5-3$ Gyr, which is still in the matter dominated
1247: era. The cosmological constant played no dynamical role and cannot
1248: have anything to do with this peak. Rather, this time scale is set by
1249: astrophysical dynamics, such as galaxy formation, cooling, and
1250: feedback.
1251:
1252: %
1253: \EPSFIGURE[!t]{entprod.eps,width=.8\textwidth}{\label{fig-eprrl} The
1254: entropy production rate of Fig.~\ref{fig-dSdVdt} depends only mildly
1255: on the vacuum energy. Hence, dependence on the vacuum energy enters
1256: our calculation mainly through the size of the causal diamond (see
1257: Figures 4 and 9). The rate of Ref.~\cite{Nagamine:2006kz} is shown
1258: here for $\rho_\Lambda = (0.1, 1, 10)$ times the observed value
1259: (from top to bottom) using the approximation of
1260: equation~\ref{fig-approx}.}
1261: %
1262: Therefore, the star formation rate depends mainly on the mass fraction
1263: in star forming galaxies at the critical time $t_\star$. Strictly
1264: speaking, the SFR will be sensitive to the PS fraction at times before
1265: $t_\star$ because of the cooling period that is required between the
1266: time a baryonic structure goes non-linear and the time it collapses
1267: into stars. Baryons that burn during the peak of the SFR actually
1268: fell into non-linear halos a cooling time earlier. We leave a more
1269: careful analysis of these and other effects to further investigation.
1270:
1271: Let us define $F_\star (\rho_\Lambda)$ as the PS fraction that is most
1272: relevant for star formation in a universe with cosmological constant
1273: $\rho_\Lambda$, namely that evaluated at $t_\star$, with a minimum
1274: mass of $10^7 M_\solar$. In order to capture the mild sensitivity of
1275: the SFR to changes of the cosmological constant, we rescale the SFR by
1276: the appropriate $F_\star$:
1277: \begin{equation}
1278: \label{fig-approx}
1279: \dot \rho_\star (\rho_\Lambda, t)=
1280: \dot\rho_\star (t)\times
1281: \frac{F_\star(\rho_\Lambda)}
1282: {F_\star(1.25\times 10^{-123})}\,.
1283: \label{eq-sfrps}
1284: \end{equation}
1285:
1286: Because the observed value of $\rho_\Lambda$ is far from disturbing
1287: the formation of $10^7 M_\odot$ galaxies, small variations of
1288: $\rho_\Lambda$ will barely affect the mass fraction $F_\star$; see
1289: Fig.~\ref{fig-eprrl}. But they do affect the geometry of the causal
1290: diamond. This is the reason why the latter effect will be important
1291: for our final result, while Eq.~(\ref{eq-sfrps}) gives only a tiny
1292: correction.
1293:
1294:
1295: \subsection{Total entropy production in the causal diamond}
1296: \label{sec-total}
1297:
1298: From the above results, we can compute the total entropy production
1299: \begin{equation}
1300: \Delta S(\rho_\Lambda) =
1301: \int_0^\infty dt\, V_{\rm c}(\rho_\Lambda,t) \partial_{V_{\rm c}} \dot S
1302: (\rho_\Lambda,t)~.
1303: \end{equation}
1304: Here, $V_{\rm c}$ is the comoving volume in the causal diamond at the time
1305: $t$, given in Eq.~(\ref{covol}). $\partial_{V_{\rm c}} \dot S$ is the rate
1306: of entropy production per unit comoving volume, given in
1307: Eq.~(\ref{dSdVdt}). The dependence on $\rho_\Lambda$ enters mainly
1308: through $V_{\rm c}$. It is straightforward to perform the integrals
1309: numerically.
1310:
1311: By Eqs.~(\ref{eq-plog}) and (\ref{eq-jkl}), the probability
1312: distribution $dP/d(\log \rho_\Lambda)$ is proportional to
1313: $\rho_\Lambda \Delta S(\rho_\Lambda)$. We show this distribution in
1314: Fig.~\ref{fig-ae} for both the Nagamine {\em et
1315: al.}~\cite{Nagamine:2006kz} as well as the Hopkins {\em et
1316: al.}~\cite{Hopkins:2006bw} SFRs.
1317: %
1318: \EPSFIGURE[!t]{hopkins_nagamine.eps,width=.8\textwidth}{\label{fig-ae}
1319: The probability distribution over $\log(\rho_\Lambda)$ computed from
1320: the Causal Entropic Principle. The two curves shown arise
1321: from two different models of the star formation rate (see
1322: Figs.~\ref{fig-SFR} and~\ref{fig-dSdVdt}). Their differences hint
1323: at the systematic uncertainties in our calculation that arise since
1324: the history of entropy production is not known to arbitrary
1325: precision. These uncertainties are apparently irrelevant to our main conclusion:
1326: either way, the observed value of $\rho_\Lambda$ (vertical bar) is
1327: not unlikely.}
1328: %
1329: For the Nagamine {\em et al.} SFR, the median value is $\rho_\Lambda
1330: = 5.6 \times 10^{-123}$ with the $1\sigma$ error band between $4.2
1331: \times 10^{-124}$ to $5.8 \times 10^{-122}$. For the Hopkins et
1332: al.\ SFR, we find the median value $\rho_\Lambda = 2.1 \times
1333: 10^{-123}$ with the $1\sigma$ error band between $2.4 \times
1334: 10^{-124}$ to $1.4 \times 10^{-122}$.
1335:
1336: There are ``systematic uncertainties'' in our calculation of the
1337: probability distribution, which come from our lack of knowledge of the
1338: precise history of entropy production in our universe. The comparison
1339: between the two models for the SFR gives a good sense of their size.
1340: It shows that these uncertainties do not affect our conclusion: with
1341: either choice, the observed value is well within the $1\sigma$ region,
1342: and hence, not unlikely.
1343:
1344:
1345: \section{Discussion}
1346: \label{sec-discussion}
1347:
1348: \subsection{$\Delta S$ captures complexity}
1349: \label{sec-deltas}
1350:
1351: Our main quantitative result is the probability distribution for
1352: $\rho_\Lambda$. However, we have also discovered an important
1353: intermediate result: in our own causal diamond, the dominant
1354: contribution to entropy production comes from the infrared glow of
1355: dust heated by starlight.\footnote{It is amusing to note that this
1356: class of entropy producers includes the authors and the reader, in
1357: the sense that the Earth, like dust, is made of heavy elements and also
1358: absorbs starlight and re-emits in the infrared.}
1359:
1360: This is remarkable. It shows that a seemingly primitive quantity,
1361: $\Delta S$, captures many of the conditions that are usually demanded
1362: explicitly by anthropic arguments. $\Delta S$ would drop sharply if
1363: galaxies, stars, or heavy elements were absent. According to the
1364: Causal Entropic Principle, the weight carried by such a vacuum would
1365: be suppressed.
1366:
1367: For example, consider a universe like ours, except without heavy
1368: elements. (This could be arranged by adjustments in the standard
1369: model.) Galaxies would still form, and stars would burn, but there
1370: would be no dust available to convert optical photons into a much
1371: larger number of infrared photons~\cite{astroph9812182}. The Causal
1372: Entropic Principle assigns a weight 100 times larger to our vacuum
1373: than to this one---simply based on the entropy production, without
1374: knowing anything about the potential advantages of heavy elements
1375: often claimed in anthropic arguments.
1376:
1377: This demonstrates that $\Delta S$ can be an effective and
1378: very simple substitute for a number of dubious anthropic conditions.
1379: More importantly, our result lends credibility to $\Delta S$ as a
1380: weighting factor for vacua with very different low-energy physics.
1381: Estimating the number of observers in such vacua, even averaged over a
1382: large parameter space, appears wholly intractable, but estimating
1383: $\Delta S$ may not be.
1384:
1385: Thus, the Causal Entropic Principle may allow us, for the first time,
1386: to predict probability distributions over the entire landscape, rather
1387: than just over subspaces constrained to coincide with much of our
1388: low-energy physics. As discussed in Sec.~\ref{sec-intro}, this could
1389: lead to a breakthrough on extracting predictions directly from the
1390: underlying theory (the string landscape), without conditioning on
1391: parameters specific to our own vacuum.
1392:
1393:
1394: \subsection{Understanding our distribution}
1395: \label{sec-understand}
1396:
1397: Now, let us turn to our main result. In our approach, the most likely
1398: range of $\log\rho_\Lambda$ is set not by the time of galaxy
1399: formation, but by the time at which the {\em rate\/} of entropy
1400: production, per unit time and unit comoving volume, is largest. This
1401: can be understood as follows.
1402:
1403: Consider, for the sake of argument, an entropy production rate that is
1404: independent of time and of $\rho_\Lambda$. Then the total entropy
1405: $\Delta S$ produced within the causal diamond is proportional to $\int
1406: V_{\rm c}(t)\, dt$, where $V_{\rm c}$ is the comoving volume (or
1407: equivalently, the mass) present inside the causal diamond at the time
1408: $t$. This integral is the area under the curves shown in the lower
1409: panel of Fig.~\ref{fig-covol}.
1410:
1411: At small times (near the bottom tip) the causal diamond is small and
1412: $V_{\rm c}$ is negligible. After vacuum domination, at a time of order
1413: $t_\Lambda=(3/8\pi\rho_\Lambda)^{1/2}$, the top cone, which contains
1414: one de~Sitter horizon region, quickly empties out and $M(t)$ vanishes
1415: exponentially. Thus, only the era around the time of matter/vacuum
1416: equality contributes significantly to the integral.
1417:
1418: Up to a $\rho_\Lambda$-independent factor of order unity, the above
1419: integral is therefore the product of $t_\Lambda M(t_\Lambda)$. But
1420: $M(t_\Lambda)$ is just the total mass inside the horizon around the
1421: time when $\Lambda$ begins to dominate. This is again of order
1422: $t_\Lambda$ and thus proportional to $\rho_\Lambda^{-1/2}$. Hence the
1423: total entropy produced, $\Delta S$, scales like $1/\rho_\Lambda$ in
1424: our hypothetical case of a constant entropy production rate. This is
1425: also clear by inspecting the area under the different curves in
1426: Fig.~\ref{fig-covol}.
1427:
1428: Assuming a flat prior ($dp/d \rho_\Lambda = \mbox{const}$, or
1429: equivalently, $dp/d\log \rho_\Lambda \propto \rho_\Lambda$), the
1430: observer-weighted probability distribution is
1431: \begin{equation}
1432: \frac{dP}{d\log \rho_\Lambda} \propto w(\rho_\Lambda ) \rho_\Lambda ~,
1433: \end{equation}
1434: and the Causal Entropic Principle states that the weight is
1435: \begin{equation}
1436: w=\Delta S~.
1437: \end{equation}
1438:
1439: For the hypothetical, constant entropy production rate, we have
1440: $w\propto \rho_\Lambda^{-1}$, and hence
1441: \begin{equation}
1442: \frac{dP}{d\log \rho_\Lambda} = \mbox{const} ~.
1443: \end{equation}
1444: The weight $\Delta S$ in this case takes a prior distribution that was
1445: flat in $\rho_\Lambda $ into an observer-weighted distribution that is
1446: flat in $\log \rho_\Lambda $, showing no preference between, say,
1447: $\rho_\Lambda =10^{-121}$ and $\rho_\Lambda = 10^{-123}$.
1448:
1449: In the prior distribution, there are more vacua at large
1450: $\rho_\Lambda$, so exponentially small values of $\rho_\Lambda $ are
1451: very unlikely. The above example shows that the Causal Entropic
1452: Principle captures an important compensating factor: vacua with
1453: smaller $\rho_\Lambda$ give rise to a larger causal diamond, i.e., to
1454: a bigger de~Sitter horizon and a longer time until vacuum domination.
1455: This allows for greater complexity and compensates for the rarity of
1456: such vacua.
1457:
1458: Next, let us consider the time-dependent entropy production rate we
1459: found in Sec.~\ref{sec-rate}. We found that the entropy production
1460: due to stars has a fairly broad peak around $t_{\rm peak}\sim 2$ to 3.5 Gyr
1461: after the big bang. At earlier times, it is lower because fewer stars
1462: have formed; at late times, it is lower because few new stars form
1463: while older ones have burned out.
1464:
1465: Because of the time dependence, $\rho_\Lambda \Delta S$ will no longer
1466: be constant. For large values of $\rho_\Lambda$, the causal diamond
1467: is small, and it will contain only a small comoving volume by the time
1468: the entropy production peaks [$t_{\rm peak}\gg t_{\rm edge}=0.23
1469: t_\Lambda$; see Eq.~(\ref{eq-edge})]. In this regime, $dP/d\log
1470: \rho_\Lambda$ will decrease with $\rho_\Lambda $. For small
1471: $\rho_\Lambda$, the causal diamond is very large, but the entropy
1472: production rate peaks early, when the comoving volume is still
1473: relatively small ($t_{\rm peak}\ll t_{\rm edge}$). In this regime,
1474: $dP/d\log\rho_\Lambda$ will increase with $\rho_\Lambda $. This is
1475: illustrated in Fig.~\ref{fig-goldilocks}.
1476:
1477: %
1478: \EPSFIGURE[!t]{goldilocks.eps,width=1.0\textwidth}{\label{fig-goldilocks}
1479: This cartoon demonstrates how the Causal Entropic Principle leads to
1480: a preferred value of the cosmological constant. The horizontal band
1481: represents the rate of entropy production; darker areas correspond
1482: to a higher rate. Vacua are weighted by $\Delta S$, the total
1483: amount of ``darkness'' inside a causal diamond. Vacua with large
1484: cosmological constant are plentiful in the landscape, but they lead
1485: to small causal diamonds, which capture virtually no entropy
1486: production (right). For some smaller value, the diamond will be
1487: just large enough to capture the bulk of the entropy production
1488: (center). This is the preferred cosmological constant. Larger
1489: diamonds may capture slightly more $\Delta S$ (left), but not in
1490: proportion to their size. They correspond to vacua with very small
1491: cosmological constant which are much rarer in the landscape.
1492: Therefore they will be suppressed.}
1493: %
1494: Therefore, $dP/d\log \rho_\Lambda $ will be maximal for values of
1495: $\rho_\Lambda$ such that
1496: \begin{equation}
1497: t_{\rm edge}(\rho_\Lambda)=t_{\rm peak}~.
1498: \label{eq-fast}
1499: \end{equation}
1500: By Eqs.~(\ref{eq-tlambda}) and (\ref{eq-edge}), the observed value of
1501: $\log\rho_\Lambda$ should be near
1502: \begin{equation}
1503: \log\rho_{\Lambda, {\rm peak}}\approx
1504: \log (0.006/t_{\rm peak}^2)\approx -123~.
1505: \end{equation}
1506: This rough estimate is borne out by our more careful calculation in
1507: Sec.~\ref{sec-calculation}. The excellent agreement of the observed
1508: $\log\rho_\Lambda$, Eq.~(\ref{eq-cc}), with this prediction is
1509: reflected in Fig.~\ref{fig-covol}, where it can be seen that the edge
1510: time and the peak time really coincide for our universe.
1511:
1512: The width of our distribution can also be understood in this manner.
1513: Let $t_{\rm on}$ and $t_{\rm off}$ be the times at which the entropy
1514: production rate is at half of its peak rate. Using those values in
1515: Eq.~(\ref{eq-fast}) gives roughly the $1\sigma$ boundaries we found
1516: for our distribution in Sec.~\ref{sec-total}. To summarize, the peak
1517: and the width of the probability distribution for $\rho_\Lambda$ are
1518: related to the peak and width of the entropy production rate by
1519: Eq.~(\ref{eq-fast}).
1520:
1521: Our distribution has a greater width than the distribution obtained
1522: from the number of observers-per-baryon; this can be seen clearly in
1523: Fig.~\ref{fig-showmethemoney}. This is also not hard to understand.
1524: In the traditional approach, nothing compensates for the exponential
1525: growth of $dP/d\log \rho_\Lambda $ with $\log \rho_\Lambda$, until a
1526: fairly sharp cutoff occurs when $\log \rho_\Lambda $ becomes large
1527: enough to disrupt galaxy formation. Hence, the preferred values of
1528: $\log\rho_\Lambda $ are squeezed into a narrow interval, and the
1529: observed value is strongly excluded. In our approach, the spacetime
1530: volume of the causal diamond depends inversely on $\rho_\Lambda$,
1531: cancelling the pressure towards large values of $\rho_\Lambda$. The
1532: width of the probability curve is set only by the shape of the peak of
1533: the entropy production rate (Fig.~6), which is fairly wide.
1534:
1535: In this discussion we have pretended that $\rho_\Lambda$ does not
1536: affect the entropy production rate. In fact, this is an excellent
1537: approximation. In the vicinity the observed value of $\rho_\Lambda$,
1538: the total entropy production depends on $\rho_\Lambda$ mainly through
1539: the geometry of the causal diamond. The probability density decreases
1540: away from this maximum. As a result, values of $\rho_\Lambda$ large
1541: enough to disrupt galaxy formation are highly suppressed {\em even
1542: before we take into account the suppression of the entropy
1543: production rate resulting from this disruption}.
1544:
1545: This points at another crucial difference between weighting by entropy
1546: production in the causal diamond, and weighting by
1547: observers-per-baryon: the preferred $\rho_\Lambda$ is set by
1548: completely different physical processes, and hence, by essentially
1549: unrelated timescales. In the latter approach, one assumes that
1550: observers require galaxies. Then the disruption of galaxy formation
1551: cuts off the exponential growth of $dP/d\log \rho_\Lambda$. As a
1552: result, the preferred $\log \rho_\Lambda$ is set by the time when
1553: galaxies first form, and this gives a value that is too large compared
1554: to Eq.~(\ref{eq-cc}).
1555:
1556: In our approach, we do not assume that observers require galaxies.
1557: The size of the causal diamond depends inversely on $\rho_\Lambda$,
1558: allowing the preferred range of values for $\log\rho_\Lambda$ to be
1559: set by the time-dependence of the entropy production rate. The time
1560: of peak entropy production by dust heated by starlight picks out the
1561: value $\log \rho_\Lambda \approx -123$. The time-scale when galaxies
1562: form does not enter directly. In our universe, the difference between
1563: the two timescales amounts to ``only'' 3 orders of magnitude in the
1564: preferred $\rho_\Lambda$, but it is easy to imagine other vacua in the
1565: landscape where the era of peak entropy production is parametrically
1566: separated from the era when galaxy halos become nonlinear (for
1567: example, by a large galaxy cooling time).
1568:
1569:
1570: \subsection{Statistical interpretation}
1571: \label{sec-interpret}
1572:
1573: It is worth emphasizing that it is entirely irrelevant whether the
1574: observed $\rho_\Lambda$ is, say, $0.5\sigma$ above or $0.6\sigma$
1575: below the median of our distribution. We get to make only one
1576: measurement. There is no reason to expect this one data point to be
1577: on the median (or on the peak) of the probability distribution. But
1578: we can expect that it will not be a very unlikely value. Any value in
1579: the $1\sigma$ region certainly qualifies as not unlikely. The success
1580: of the Causal Entropic Principle, its formal advantages aside, is not
1581: that it predicts the precise value of $\rho_\Lambda$, but that our
1582: distribution shows that the observed value was {\em not unlikely\/} to
1583: have been observed.
1584:
1585: Physicists have a great degree of confidence in certain theories that
1586: make only statistical predictions, even though we are unable to make
1587: more than a finite number of measurements, let alone test all the
1588: consequences of a theory. In this spirit, our result improves our
1589: confidence in the Causal Entropic Principle and the underlying
1590: landscape. To improve our confidence further, we cannot repeat the
1591: measurement of the cosmological constant, but we can extract other
1592: predictions or postdictions and compare those to observation.
1593:
1594: \acknowledgments We thank T.~Abel, E.~Baltz, L.~Bildsten, G.~Bothun, D.~Croton,
1595: M.~Davis, L.~Hall, P.~Horava, C.~McKee, G.~Smoot, A.~Vilenkin,
1596: M.~White, and especially E.~Quataert for helpful conversations. We
1597: are grateful to P.~Hopkins for providing more detail and numerical
1598: values on quasar luminosities from Ref.~\cite{Hopkins:2006fq}, to
1599: D.~Scott for an up-to-date plot of the grand unified photon spectrum, and A. Vilenkin for pointing out an error in the plots in earlier versions of this paper. We also thank Massimo Porrati for extensive discussions at the outset of this project. RH, GDK, and GP thank the Aspen Center for Physics
1600: where part of this work was completed. The work was supported by the
1601: Berkeley Center for Theoretical Physics, by a CAREER grant of the
1602: National Science Foundation (RB), and by DOE grants DE-AC03-76SF00098
1603: (RB), DE-AC02-76SF00515 (RH), and DE-FG02-96ER40969 (GDK).
1604:
1605: \appendix
1606:
1607: \section{The radiation era}
1608: \label{sec-radiation}
1609:
1610: In this Appendix, we justify our neglect of the radiation era. The
1611: metric, conformal time, and density during this era are
1612: \begin{eqnarray}
1613: a_{\rm rad}(t) &=& c (t-t_0)^{1/2}~,
1614: \label{eq-arad}\\
1615: \tau_{\rm rad}(t) &=& 2c^{-1} (t-t_0)^{1/2}~, \\
1616: \rho_{\rm rad}(t) &= & \frac{3}{32\pi(t-t_0)^2}~.
1617: \label{eq-rhorad}
1618: \end{eqnarray}
1619: The constants $c$ and $t_0$ are determined by matching the Hubble
1620: constant and the scale factor to the metric Eq.~(\ref{eq-aexact}),
1621: which becomes
1622: \begin{equation}
1623: a(t) = \left(\frac{3t}{2}\right)^{2/3}~
1624: \end{equation}
1625: for $t\ll t_\Lambda$. They must agree at the time $ t_{\rm eq}$, when
1626: the matter and radiation densities are equal, i.e.,
1627: when~\cite{Tegmark:2005dy}
1628: \begin{equation}
1629: \rho_{\rm rad}=\rho_{\rm eq}\equiv 0.0026 \xi^4= 3.1\times 10^{-113}~,
1630: %.00260042 (this is pure math)
1631: \end{equation}
1632: where
1633: \begin{equation}
1634: \xi \approx 3.3 \times 10^{-28} % +/- .3
1635: \end{equation}
1636: is the observed mass of pressureless matter per photon. This yields
1637: \begin{eqnarray}
1638: t_0 &=& \frac{t_{\rm eq}}{4} =
1639: \frac{1}{6}\left(\frac{3}{8\pi\rho_{\rm eq}}\right)^{1/2}~,\\
1640: c &=& \left(\frac{24}{\pi\rho_{\rm eq}}\right)^{1/12}~.
1641: \end{eqnarray}
1642:
1643: By Eq.~(\ref{eq-cd}), the size of the causal diamond is set by the
1644: total conformal time duration of the universe since reheating, which
1645: is finite. In Sec.~\ref{sec-metric}, we neglected the radiation era
1646: and extended the matter/vacuum solution all the way back to the big
1647: bang ($t=0$). This yielded a total conformal time
1648: \begin{equation}
1649: \Delta\tau = 2.804 \left(\frac{3}{8\pi \rho_\Lambda}\right)^{1/6}~,
1650: \end{equation}
1651: from Eqs.~(\ref{eq-tlambda}) and (\ref{eq-tau0}).
1652:
1653: In order to correct for the presence of the radiation era, we should
1654: subtract the conformal time interval $\Delta\tau'$ corresponding to
1655: the era $0<t<t_{\rm eq}$ that should be excised from the matter/vacuum
1656: solution. It should be replaced by the conformal time interval
1657: $\Delta\tau''$ corresponding to the radiation dominated era (the
1658: portion of the metric (\ref{eq-arad}) between reheating and matter
1659: domination).
1660:
1661: Using the above results, however, it is easy to show
1662: that
1663: \begin{equation}
1664: 2\Delta\tau''<\Delta\tau' =
1665: \left(\frac{\rho_\Lambda}{\rho_{\rm eq}}\right)^{1/6}
1666: \frac{\Delta\tau}{2.804}\,.
1667: \end{equation}
1668: Thus, the corrections to the conformal time, and thus to the size of
1669: the causal diamond, are negligible for $\rho_\Lambda<\rho_{\rm eq}$.
1670: For example, with the observed value of $\rho_\Lambda$, the correction
1671: is less than 1\%. The probability of values of
1672: $\rho_\Lambda>10^{-120}$ almost vanishes according to our calculation;
1673: yet this is still 7 orders of magnitude below $\rho_{\rm eq}$. Hence,
1674: our approximation is good in the entire range of $\rho_\Lambda$ in
1675: which our probability distribution has support.
1676:
1677:
1678:
1679:
1680: \bibliographystyle{JHEP}
1681: \bibliography{all2}
1682: \end{document}
1683:
1684: