Black Holes and Thermodynamics

arXiv:gr-qc/9702022v1 11 Feb 1997

Robert M. Wald Enrico Fermi Institute and Department of Physics University of Chicago 5640 S. Ellis Avenue Chicago, Illinois 60637-1433 July 26, 2011
Abstract We review the remarkable relationship between the laws of black hole mechanics and the ordinary laws of thermodynamics. It is emphasized that in analogy with the laws of thermodynamics the validity the laws of black hole mechanics does not appear to depend upon the details of the underlying dynamical theory (i.e., upon the particular eld equations of general relativity). It also is emphasized that a number of unresolved issues arise in ordinary thermodynamics in the context of general relativity. Thus, a deeper understanding of the relationship between black holes and thermodynamics may provide us with an opportunity not only to gain a better understanding of the nature of black holes in quantum gravity, but also to better understand some aspects of the fundamental nature of thermodynamics itself.

Introduction

Undoubtedly, one of the most remarkable developments in theoretical physics to have occurred during the past twenty ve years was the discovery of a close relationship between certain laws of black hole physics and the ordinary laws of thermodynamics. It appears that these laws of black hole mechanics and the laws of thermodynamics are two major pieces of a puzzle that t together so perfectly that there can be little doubt that this t is of deep signicance. The existence of this close relationship between these laws may provide us with a key to our understanding of the fundamental nature of black holes in a quantum theory of gravity, as well as to our understanding of some aspects of the nature of thermodynamics itself. The aim of this article is to review the nature of the relationship between the black hole and thermodynamics laws. Although some notable progress has been made, many mysteries remain. It was rst pointed out by Bekenstein [1] that a close relationship might exist between certain laws satised by black holes in classical general relativity and the ordinary laws of thermodynamics. The area theorem of classical general relativity [2] states that the area, A, of a black hole can never decrease in any process A 0. (1) Bekenstein noted that this result is closely analogous to the statement of ordinary second law of thermodynamics: The total entropy, S, of a closed system never decreases in any process S 0. (2)

Indeed, Bekenstein proposed that the area of a black hole (times a constant of order unity in Planck units) should be interpreted as its physical entropy. A short time later, the analogy between certain laws of black hole physics in classical general relativity and the laws of thermodynamics was developed systematically by Bardeen, Carter, and Hawking [3]. They proved that in general relativity, the surface gravity, , of a stationary black hole (dened by eq.(19) below) must be constant over the event horizon of the black hole. They noted that this result is analogous to the zeroth law of thermodynamics, which states that the temperature, T , must be uniform over a body in thermal equilibrium. Finally, Bardeen, Carter, and Hawking proved the rst 2

law of black hole mechanics. In the vacuum case, this law states that the dierences in mass, M, area, A, and angular momentum, J of two nearby stationary black holes must be related by M = 1 A + J, 8 (3)

where denotes the angular velocity of the event horizon (dened by eq.(18) below). Additional terms may appear on the right side of eq.(3) when matter elds are present. They noted that this law is closely analogous to the ordinary rst law of thermodynamics, which states that dierences in energy, E, entropy, and other state parameters of two nearby thermal equilibrium states of a system is given by E = T S + work terms. (4)

If we compare the zeroth, rst, and second laws of ordinary thermodynamics with the corresponding laws of black hole mechanics, we see that the analogous quantities are, respectively, E M, T , and S A/8, where is a undetermined constant. Even in the context of classical general relativity, a hint that this relationship might be of physical signicance arises from the fact that E and M represent the same physical quantity, namely the total energy of the system. However, in classical general relativity, the physical temperature of a black hole is absolute zero, so there can be no physical relationship between T and . Consequently, it also would be inconsistent to assume a physical relationship between S and A. For this reason, at the time the paper of Bardeen, Carter, and Hawking appeared, most researchers (with the notable exception of Bekenstein) viewed the analogy between the black hole and thermodynamical laws as a mathematical curiosity, devoid of any physical signicance. That view changed dramatically with Hawkings discovery [4] that, due to quantum particle creation eects, a black hole radiates to innity all species of particles with a perfect black body spectrum, at temperature (in units with G = c = h = k = 1) T = . (5) 2 Thus, /2 truly is the physical temperature of a black hole, not merely a quantity playing a role mathematically analogous to temperature in the 3

laws of black hole mechanics. This left little doubt that the laws of black hole mechanics must correspond physically to the laws of thermodynamics applied to a system consisting of a black hole. As will be discussed further below, it also left little doubt that A/4 must represent the physical entropy of a black hole in general relativity. Thus, Hawkings calculation of particle creation eectively gave a resoundingly positive answer to the question of whether there exists any physical signicance to mathematical the relationship between the laws of black hole mechanics and the laws of thermodynamics. This conclusion is particularly intriguing, since, as I shall review in sections II and III, the the derivations of the laws of black hole mechanics are so dierent in nature from those of the laws of thermodynamics that it is hard to see how it is possible that these laws could really be the same. As I will discuss further in section IV, this conclusion also raises a number of new questions and issues, most prominently that of providing a physical explanation for the origin of black hole entropy.

The nature of the laws of ordinary thermodynamics

The ordinary laws of thermodynamics are not believed to be fundamental laws in their own right, but rather to be laws which arise from the fundamental microscopic dynamics of a suciently complicated system when one passes to a macroscopic description of it. The great power and utility of the laws of thermodynamics stems mainly from the fact that the basic form of the laws does not depend upon the details of the underlying microscopic dynamics of particular systems and, thus, the laws have a universal validity at least for a very wide class of systems. The analysis of how the laws of ordinary thermodynamics arise applies to a classical or quantum system with a large number of degrees of freedom, whose time evolution is governed by Hamiltonian dynamics. It is important to emphasize that it is crucial here, at the outset, that there be a well dened notion of time translations (to which the Hamiltonian is conjugate), and that the Hamiltonian, H, (and, thus, the dynamics) be invariant under these time translations. It then follows that the total energy, E, of the system (i.e.,

the value of H) is conserved. We now shall focus on the case of a classical dynamical system. By the previous remark, it follows that each dynamical orbit in phase space is conned to its energy shell E , i.e., the hypersurface in phase space dened by the equation H(x) = E = constant. The crucial assumption needed for the applicability of thermodynamical laws to such a system is that generic orbits in phase space behave ergodically in the sense that they come arbitrarily close to all points of E , spending equal times in equal volumes; equivalently, the total energy of the system is the only nontrivial constant of motion for generic orbits. (By a slight modication of these arguments, the laws also can accomodate the presence of a small number (compared with the number of degrees of freedom) of additional constants of motion such as the angular momentum of a rotationally invariant system.) Of course, even when such ergodic behavior occurs, it would take a dynamical orbit an innite amount of time to completely sample E . The degree of sampling which is actually needed for the applicability of thermodynamics depends upon what observable (or collection of observables) of the dynamical system is being measured or, in more common terminology, the amount of coarse graining of phase space that one does. For a ne-grained observable (corresponding to measuring detailed information about the microscopic degrees of freedom of the system), the sampling of the energy shell must be extremely good, and the amount of time needed for this sampling will be correspondingly long, thereby making the laws of thermodynamics inapplicable or irrelevant for the system. However, for the types of macroscopic, coarse-grained observables, O, usually considered for systems with a huge number of degrees of freedom, the sampling need only be quite modest and the required sampling time (= the timescale for the system to reach thermal equilibrium) is correspondingly short. One may then get considerable predictive power from the laws of thermodynamics about the values of O which one would expect to observe for the system under various physical conditions. The statistical entropy, SO , of a classical dynamical system relative to a macroscopic, coarse-grained observable (or collection of observables), O, is dened to be the observable whose value at point x on E is the logarithm of the volume of the region of the energy shell at which O takes the same

value as it does at x, i.e., SO (x) = ln[vol(Rx )] where Rx = {y E |O(y) = O(x)}. (7) For the types of coarse-grained observables O which are normally considered, the largest region, Rmax , of the form (7) will have a volume nearly equal to that of the entire energy shell E . If dynamical orbits sample E and spend equal times in equal volumes, then we would expect SO to increase in value until it reaches its maximum possible value, namely, ln[vol(Rmax )] ln[vol(E )]. Subsequently, SO should remain at that value for an extremely long time. During this extremely long period, the value of O remains unchanged, so no change would be perceived in the system, and the system would be said to have achieved thermal equilibrium. The thermodynamic entropy, S, of the system is dened by S = ln[vol(E )] (8) (6)

Unlike SO , S is not an observable on phase space but rather a function on a low dimensional thermodynamic state space comprised by the total energy, E, of the system, and any parameters (such as, for example, an external magnetic eld) appearing in the Hamiltonian which one might contemplate varying, together with any additional constants of motion for the system (such as, the total angular momentum for a rotationally invariant system). These variables characterizing thermodynamic state space are usually referred to as state parameters. The temperature, T , is dened by S 1 = , T E (9)

where the remaining state parameters are held xed in taking this partial derivative. Like S, the temperature, T , is a function on thermodynamic state space, not an observable on phase space. It follows from the above discussion that when a system is in thermal equilibrium, its statistical entropy, SO , equals its thermodynamic entropy, S. Similarly suppose a system consists of weakly interacting subsystems, so that each subsystem can be viewed as an isolated system in its own right. Suppose 6

that O is comprised by a collection of observables, Oi , for each subsytem, and suppose, in addition, that each subsystem (viewed as an isolated system) is in thermal equilibrium although the entire system need not be in thermal equilibrium. Then SO will equal the sum of the thermodynamic entropies, Si , of the subsystems Si . (10) SO =
i

We now are in a position to explain the origin of the laws of thermodynamics. As argued above, if SO is less than its maximum possible value (for the given value of E and the other state parameters), we should observe it to increase until thermal equilibrium is reached. In particular, in the case where the total system consists of subsytems and these subsystems are individually in thermal equilibrium at both the beginning and the end of some process (but not necessarily at the intermediate stages), we should have
i

(Si )1

(Si )0 ,
i

(11)

where (Si )0 and (Si )1 denote, respectively, the initial and nal thermodynamic entropies of the ith subsystem. This accounts for the second law of thermodynamics, eq.(2). It should be noted that the time asymmetry present in this formulation of the second law arises from the implicit assumption that, commonly, SO is initially below its maximum possible value. (Otherwise, a more relevant formulation of the second law would merely state that only very rarely and/or briey would we expect to observe SO to uctuate below its maximum possible value.) The fact that we do commonly observe systems with SO below its maximum possible value shows that the present state of our universe is very special. The zeroth law of thermodynamics is an immediate consequence of the fact that if the subsystems appearing in eq.(10) are at dierent temperatures, then SO can be increased by transferring energy from a subsystem of high temperature to a subsystem of low temperature. (This fact follows immediately from the denition, (9), of T .) Thus, for a thermal equilibrium state where, by denition, SO achieves its maximum value it is necessary that T be uniform. Finally, since S is a function on thermodynamic state space, its gradient 7

can be written as dS =

1 dE + T

Xj dj
j

(12)

where j denotes the state parameters other than E, and Xj S/j (where E and the state parameters other than j are held xed in taking this partial derivative). Using Liouvilles theorem, one can argue that S should be constant when suciently slow changes are made to parameters appearing in the Hamiltonian. This fact gives T Xj the interpretation of being the generalized force conjugate to j (at least for the case where j is a parameter appearing in the Hamiltonian), and it gives T Xj dj the interpretation of being a work term. This accounts for the rst law of thermodynamics, eq.(4). Thus far, our discussion has been restricted to the case of a classical dynamical system. However, as discussed in more detail in [5], a completely parallel analysis can be for a quantum system. In this analysis, the classical coarse-grained observable O on phase space is replaced by a self-adjoint operator O acting on the Hilbert space of states with energy between E and E + E. The spectral decomposition of O takes the form O= m Pm (13)

where the {Pm } are a family of orthogonal projection operators, and it is assumed in correspondence with the assumptions made about coarse-graining in the classical case that the degeneracy subspaces of O are large. The sta tistical entropy, SO is then dened to be the quantum observable SO = ln(dm )Pm (14)

where dm is the dimension of the mth degeneracy subspace of m , i.e., dm = tr(Pm ). (15)

Again, it is assumed that the maximum value of dm is essentially the dimension of the entire Hilbert space of states of energy between E and E + E. The corresponding denition of the thermodynamic entropy, S, of a quantum system is S = ln n (16) 8

where n denotes the dimension of the Hilbert space of states between E and E + E, i.e., n is proportional to the density of quantum states per unit energy. Again, S is not a quantum observable, but rather a function on thermodynamic state space. Arguments for the zeroth, rst, and second laws of thermodynamics then can be made in parallel with the classical case. It should be noted that, thus far, I have made no mention of the third law of thermodynamics. In fact, there are two completely independent statements which are referred to as the third law. The rst statement consists of the rather vague claim that it is physically impossible to achieve T = 0 for a (sub)system. To the extent that it is true, I would view this claim as essentially a consequence of the second law, since it always is highly entropically favorable to take energy away from a subsystem at nite temperature and add that energy to a subsystem whose temperature is very nearly 0. The second statement, usually referred to as Nernsts theorem, consists of the claim that S 0 as T 0. This claim is blatantly false in classical physics it fails even for a classical ideal gas but it holds for many quantum systems (in particular, for boson and fermion ideal gases). Clearly, the Nernst theorem is actually a claim about the behavior of the density of states, n(E), as the total energy of the system goes to its minimum possible value. Indeed, more precisely, as explained in section 9.4 of [6], it should be viewed as a statement about the extrapolation to minimum energy of the continuum approximation to n(E). Elsewhere, I shall give some examples of quantum systems which violate the Nernst theorem [7] (see also Section IV below). Thus, while the Nernst theorem holds empirically for systems studied in the laboratory, I do not view it as a fundamental aspect of thermodynamics. In particular, I do not feel that the well known failure of the analog of the Nernst theorem to hold for black hole mechanics where there exist black holes of nite area with = 0 should be viewed as indicative of any breakdown of the relationship between thermodynamics and black hole physics. The above discussion explains the nature and origin of the laws of thermodynamics for ordinary classical and quantum systems. However, as I shall now briey describe, when general relativity is taken into account, a number of new issues and puzzles arise. In the rst place, it should be noted that general relativity is a eld theory and, as such, it ascribes innitely many degrees of freedom to the spacetime metric/gravitational eld. If these degrees of freedom are treated classi9

cally, no sensible thermodynamics should be possible. Indeed, this situation also arises for the electromagnetic eld, where a treatment of the statistical physics of a classical electromagnetic eld in a box yields the Rayleigh-Jeans distribution and its associated ultraviolet catastrophe. As is well known, this diculty is cured by treating the electromagnetic eld as a quantum eld. I see no reason not to believe that similar diculties and cures will occur for the gravitational eld. However, it should be emphasized at the outset that one should not expect any thermodynamic laws to arise from a statistical physics treatment of classical general relativity; a quantum treatment of the degrees of freedom of the gravitational eld should be essential. A much more perplexing issue arises from the fact that, as emphasized above, the arguments for the validity of thermodynamics for ordinary systems are based upon the presence of a well dened notion of time translations, which are symmetries of the dynamics. Such a structure is present when one considers dynamics on a background spacetime whose metric possesses a suitable one-parameter group of isometries, and when the Hamiltonian is invariant under these isometries. However, such a structure is absent in general relativity, where no background metric is present. Furthermore, when the degrees of freedom of the gravitational eld are excited, one would not expect the dynamical spacetime metric to possess a time translation symmetry. The absence of any rigid time translation structure in general relativity can be viewed as being responsible for making notions like the energy density of the gravitational eld ill dened in general relativity. Notions like the entropy density of the gravitational eld are not likely to fare any better. It may still be possible to use structures like asymptotic time translations to dene the notion of the total entropy of an (asymptotically at) isolated system. (As is well known, total energy can be dened for such systems.) However, for a closed universe, it seems unlikely that any meaningful notion will exist for the total entropy of the universe (including gravitational entropy). If so, it is far from clear how the second law of thermodynamics is to be formulated for a closed universe. This issue appears worthy of further exploration. Another important issue that arises in the context of general relativity involves ergodic behavior. As discussed above, ordinary thermodynamics is predicated on the assumption that generic dynamical orbits sample the entire energy shell, spending equal times in equal volumes. However, gross violations of such ergodic behavior occur in classical general relativity on 10

account of the irreversible tendency for gravitational collapse to produce singularities from which one cannot then evolve back to uncollapsed states. Interestingly, however, there are strong hints that ergodic behavior could be restored in quantum gravity. In particular, the quantum phenomenon of black hole evaporation (see Section IV below) provides a means of evolving from a collapsed state back to an uncollapsed conguration. Finally, as noted above, the fact that we commonly observe the increase of entropy shows that the present state of the universe is very special. As has been emphasized by Penrose [8], the specialness of the present state of the universe traces back, ultimately, to extremely special initial conditions for the universe at the big bang. This specialness of the initial state of the universe should have some explanation in a complete, fundamental theory. However, at present, it remains a matter of speculation as to what this explanation might be. The above comments already give a clear indication that there are deep and fundamental issues lying at the interface of gravitation and thermodynamics. As we shall see in the next two sections, the theory of black holes gives rise to very signicant further relationships between gravitation and thermodynamics.

The nature of the laws of classical black hole mechanics

As previously indicated in the Introduction, it appears overwhelmingly likely that the laws of classical black hole mechanics must arise, in a fundamental quantum theory of gravity, as the classical limit of the laws of thermodynamics applied to a system comprised by a black hole. However, as we shall see in this section, the present derivations of the laws of classical black hole mechanics could hardly look more dierent from the arguments for the corresponding laws of thermodynamics, as given in the previous section. Nevertheless, as I shall emphasize here, the derivations of the laws of black hole mechanics appear to share at least one important feature of the thermodynamic arguments: There appears to be a universality to the laws of black hole mechanics in that the basic form of the laws appears to be independent of the details of the precise Lagrangian or Hamiltonian of the underlying

11

where I + denotes future null innity and I denotes the chronological past. The event horizon, H, of a black hole is dened to be the boundary of B. If an asymptotically at spacetime (M, gab ) contains a black hole B, then B is said to be stationary if there exists a one-parameter group of isometries on (M, gab ) generated by a Killing eld ta which is unit timelike at innity. The black hole is said to be static if it is stationary and if, in addition, ta is hypersurface orthogonal in which case there exists a discrete time reection isometry about any of the orthogonal hypersurfaces. The black hole is said to be axisymmetric if there exists a one parameter group of isometries which correspond to rotations at innity. A stationary, axisymmetric black hole is said to possess the t orthogonality property if the 2-planes spanned by ta and the rotational Killing eld a are orthogonal to a family of 2-dimensional surfaces. In this case, there exists a discrete t reection isometry about any of these orthogonal 2-dimensional surfaces. For a black hole which is static or is stationary-axisymmetric with the t orthogonality property, it can be shown [9] that there exists a Killing

theory of gravity in a manner analogous to the universality of the form of the laws of ordinary thermodynamics. In this section, we will consider theories of gravity which are much more general than general relativity, but we shall restrict attention to geometric theories, wherein spacetime is represented by a pair (M, gab ), where M is a manifold and gab is a metric of Lorentzian signature. Other matter elds also may be present on spacetime. For deniteness, I will assume that M is 4-dimensional, but all results below generalize straightforwardly to any dimension n 2. For our discussion of the rst and second laws of black hole mechanics, it will be assumed, in addition, that the eld equations of the theory have been obtained from a dieomorphism covariant Lagrangian. In physical terms, a black hole in a spacetime, (M, gab ), is a region where gravity is so strong that nothing can escape. In order to make this notion precise, one must have in mind a region of spacetime to which one can contemplate escaping. For an asymptotically at spacetime (representing an isolated system), the asymptotic portion of the spacetime near innity is such a region. The black hole region, B, of an asymptotically at spacetime, (M, gab ), is dened as B M I (I + ), (17)

12

eld a of the form a = ta + a (18) which is normal to the event horizon, H. The constant dened by eq.(18) is called the angular velocity of the horizon. (For a static black hole, we have = 0.) A null surface whose null generators coincide with the orbits of a one-parameter group of isometries is called a Killing horizon, so the above result states that the event horizon of any black hole which is static or is stationary-axisymmetric with the t orthogonality property must always be a Killing horizon. A stronger result holds in general relativity, where, under some additional assumptions, it can be shown that the event horizon of any stationary black hole must be a Killing horizon [10]. From this result, it also follows that in general relativity, a stationary black hole must be nonrotating (from which staticity follows [11], [12]) or axisymmetric (though not necessarily with the t orthogonality property). Now, let K be any Killing horizon (not necessarily required to be the event horizon, H, of a black hole), with normal Killing eld a . Since a ( b b ) also is normal to K, these vectors must be proportional at every point on K. Hence, there exists a function, , on K, known as the surface gravity of K, which is dened by the equation a ( bb ) = 2 a (19)

It follows immediately that must be constant along each null geodesic generator of K, but, in general, can vary from generator to generator. It is not dicult to show (see, e.g., [13]) that = lim(V a) (20)

where a is the magnitude of the acceleration of the orbits of a in the region o of K where they are timelike, V ( a a )1/2 is the redshift factor of a , and the limit as one approaches K is taken. Equation (20) motivates the terminology surface gravity. Note that the surface gravity of a black hole is dened only when it is in equilibrium (i.e., stationary), analogous to the fact that the temperature of a (sub)system in ordinary thermodynamics is dened only for thermal equilibrium states. In the context of an arbitrary metric theory of gravity, the zeroth law of black hole mechanics may now be stated as the following theorem [9], [14]: 13

For any black hole which is static or is stationary-axisymmetric with the t orthogonality property, the surface gravity , must be constant over its event horizon H. The key ingredient in the proof of this theorem is the identity [14] 1 (21) [a b] = abcd c d 4 which holds on an arbitrary Killing horizon, where a abcd b c d denotes the twist of the Killing eld a . For a static black hole, we have a = 0, and the constancy of on H follows immediately. Further arguments [14] similarly establish the constancy of for a stationary-axisymmetric black hole with the t orthogonality property. It should be emphasized that this result is purely geometrical, and involves no use of any eld equations. A stronger version of the zeroth law holds in general relativity. There it can be shown [3] that if Einsteins equation holds with the matter stressenergy tensor satisfying the dominant energy condition, then must be constant on any Killing horizon. In particular, one need not make the additional hypothesis that the t orthogonality property holds. An important consequence of the zeroth law is that if = 0, then in the maximally extended spacetime representing the black hole, the event horizon, H, comprises a branch of a bifurcate Killing horizon. (A precise statement and proof of this result can be found in [14]. Here, a bifurcate Killing horizon is comprised by two Killing horizons, HA and HB , which intersect on a spacelike 2-surface, C, known as the bifurcation surface.) As stated above, the event horizon of any black hole which is static or is stationary-axisymmetric with the t orthogonality property, necessarily is a Killing horizon and necessarily satises the zeroth law. Thus, the study of such black holes divides into two cases: degenerate black holes (for which, by denition, = 0), and black holes with bifurcate horizons. Again, this result is purely geometrical involving no use of any eld equations and, thus, it holds in any metric theory of gravity. We turn, now, to the consideration of rst law of black hole mechanics. For this analysis, it will be assumed that the eld equations of the theory arise from a dieomorphism covariant Lagrangian 4-form, L, of the general structure L = L (gab ; Rabcd , a Rbcde , ...; , a , ...) (22) where a denotes the derivative operator associated with gab , Rabcd denotes 14

the Riemann curvature tensor of gab , and denotes the collection of all matter elds of the theory (with indices surpressed). An arbitrary (but nite) number of derivatives of Rabcd and are permitted to appear in L. Here and below we use boldface letters to denote dierential forms and we will surpress their indices. We also shall denote the complete collection of dynamical elds, (gab , ) by (thereby surpressing the indices of gab as well). Our treatment will follow closely that given in [15]; much of the mathematical machinery we shall use also has been extensively employed in analyses of symmetries and conservation laws of Lagrangian systems (see, e.g., [16] and references cited therein). The Euler Lagrange equations of motion, E = 0, are obtained by writing the variation of the Lagrangian in the form L = E() + d(, ). (23)

where no derivatives of appear in the rst term on the right side. Usually, the manipulations yielding eq.(23) are performed under an integral sign, in which case the second term on the right side becomes a boundary term, which normally is discarded. In our case, however, our interest will be in the mathematical structure to the theory provided by . Indeed, the precise form of and the auxilliary structures derived from it will play a crucial role in our analysis, whereas the precise form of the eld equations, E = 0, will not be of interest here. The sympletic current 3-form on spacetime, which is a local function of a eld conguration, , and two linearized perturbations, 1 and 2 o of is obtained by taking an antisymmetrized variation of (, 1 , 2 ) = 2 (, 1 ) 1 (, 2 ) (24)

The (pre-)symplectic form, which is a map taking eld congurations, , together with a pairs of linearized perturbations o of , into the real numbers is obtained by integrating over a Cauchy surface, (, 1 , 2 ) =

(25)

(This integral is independent of choice of Cauchy surface when 1 and 2 satisfy the linearized eld equations.) The (pre-)symplectic form, , provides the structure needed to dene the phase space of the theory [17]. It also 15

where the denotes the contraction of the vector eld a into the rst index of the dierential form L. One can show [18] that j always can be written in the form j = dQ + a Ca , (28)

where L denotes the Lie derivative with respect to the vector eld a . (Equation (26) can be put in the more familiar form of Hamiltons equations of motion by solving it for L , thus expressing the time derivative of in terms of functional derivatives of H.) On account of the dieomorphism covariance of L, the innitesimal diffeomorphism generated by an arbitrary vector eld, a , is a local symmetry of the theory. Hence, there is an associated, conserved Noether current 3-form, j, dened by j = (, L ) L (27)

provides the structure needed to dene the notion of a Hamiltonian, H, conjugate to an arbitrary vector eld, a , on spacetime1 : H is a function on phase space satisfying the property that about any solution, , the variation of H satises H = (, , L ), (26)

where Ca = 0 when the equations of motion hold, i.e., Ca corresponds to constraints of the theory. Equation (28) denes the Noether charge 2-form Q, which is unique up to Q Q + X() + Y(, L ) + dZ(, ). (29)

where X, Y, and Z are arbitrary forms which are locally constructed from the elds appearing in their arguments (and with Y being linear in L and Z being linear in ). Here the term X arises from the ambiguity L L + dX in the choice of Lagrangian, the term Y(, L ) arises from the ambiguity + dY in eq.(23), and the term dZ arises directly from eq.(28). The rst law of black hole mechanics is a direct consequence of the variational identity j = (, , L ) + d( ), (30)
As discussed in [17], it will, in general, be necessary to choose a to be eld dependent, in order that it project to phase space.
1

16

which follows directly from eqs.(23), (24) and (27) above. One immediate consequence of this identity, together with eq.(26), is that if a Hamiltonian, H, conjugate to a exists, it must satisfy H =

[j d( )]

(31)

From eq.(28) it follows, in addition, that on shell i.e., when the equations of motion hold and, hence, Ca = 0 we have H =

d[Q ],

(32)

so, on shell, H is given purely by surface terms. In the case of an asymptotically at spacetime, the surface term, H , arising from innity has the interpretation of being the total canonical energy (conjugate to a ) of the spacetime. Now, let be any solution to the eld equations E = 0 with a Killing eld a , and let be any solution to the linearized eld equations o (not necessarily satisfying L = 0). It follows immediately from eq.(30) (with a = a ) together with the variation of eq.(28) that d[Q ] = 0 (33)

We apply this equation to a spacetime containing a black hole with bifurcate Killing horizon, with a taken to be the Killing eld (18) normal to the horizon, H. (As mentioned above, the assumption of a bifurcate Killing horizon involves no loss of generality [14] if the zeroth law holds and = 0.) We integrate this equation over a hypersurface, , which extends from the bifurcation surface, C, of the black hole to innity. The result is H = Q
C

(34)

where the fact that a = 0 on C has been used. We now evaluate the surface terms appearing on each side of eq.(34). As noted above, H has the interpretation of being the canonical energy conjugate to a . For a of the form (18), we have (see [11]) H = M J + ... 17 (35)

where the ... denotes possible additional contributions from long range matter elds. On the other hand, it is possible to explicitly compute Q, and thereby show that [15] Sbh , (36) Q= 2 C where L Sbh 2 nab ncd . (37) C Rabcd Here nab is the binormal to C (normalized so that nab nab = 2), L is the Lagrangian (now viewed as a scalar density rather than a 4-form), and the functional derivative is taken by formally viewing the Riemann tensor as a eld which is independent of the metric in eq.(22). Combining eqs.(34), (35), and (36), we obtain Sbh + J + ..., (38) M = 2 which is the desired rst law of black hole mechanics. Indeed, this result is actually stronger than the form of the rst law stated in the Introduction, since eq.(38) holds for non-stationary perturbations of the black hole, not merely for perturbations to other stationary black hole states. For the case of vacuum general relativity, where L = R g, a simple calculation yields Sbh = A/4. (39) However, if one considers theories with non-minimally-coupled matter or higher derivative theories of gravity, additional curvature contributions will appear in the formula for Sbh . Nevertheless, as eq. (37) explicitly shows, in all cases, Sbh is given by an integral of a local, geometrical expression over the black hole horizon. The above analysis also contains a strong hint that the second law of black hole mechanics may hold in a wide class of theories. Consider a stationary black hole with bifurcate Killing horizon, but now let us normalize the Killing eld, a , normal to the horizon by the local condition that a b = nab on C (or, equivalently, a b a b = 2 on H), rather than by the asymptotic behavior (18) of a at innity. With this new normalization, eq.(36) (with the s removed) becomes Sbh = 2 18 Q[ a ].
C

(40)

It is easy to show that for a stationary black hole, this equation continues to hold when C is replaced by an arbitrary cross-section, , of H. Now, consider a process in which an initially stationary black hole evolves through a non-stationary era, and then settles down to another stationary nal state. Let a be any vector eld which coincides with the Killing eld normal to H (with the above, new normalization) in the two stationary regimes. Then, by eqs.(40) and (28) (together with Ca = 0), we obtain Sbh = 2 = 2
H 1

Q[ a ] 2 j[ ],
a

Q[ a ] (41)

where 0 and 1 are, respectively, cross-sections of H in the initial and nal stationary regimes. Equation (41) states that the change in black hole entropy is proportional the net ux of Noether current (conjugate to a ) through H. Now, in many circumstances, the Noether current conjugate to a suitable time translation can be interpreted as the 4-density of energymomentum. Thus, eq.(41) suggests that the second law of black hole mechanics, Sbh 0, may hold in all theories which have suitable positive energy properties (such as, perhaps, a positive Bondi energy ux at null innity). However, I have not, as yet obtained any general results along these lines. Nevertheless, it has long been known that the second law holds in general relativity [2], provided that the matter present in spacetime satises the following positive energy property (known as the null energy condition): for any null vector k a , the matter stress-energy tensor, Tab , satises Tab k a k b 0.

Quantum black hole thermodynamics

In the previous section, we used a purely classical treatment of gravity and matter elds to derive analogs of the laws of thermodynamics for black holes. However, as already noted in the Introduction, in classical physics, these laws of black hole mechanics cannot correspond physically to the laws of thermodynamics. It is only when quantum eects are taken into account that these subjects appear to merge. The key result establishing a physical connection between the laws of black hole mechanics and the laws of thermodynamics is, of course, the thermal particle creation eect discovered by Hawking [4]. This result is derived 19

in the context of semiclassical gravity, where the eects of gravitation are still represented by a classical spacetime (M, gab ), but matter elds are now treated as quantum elds propagating in this classical spacetime. In its most general form, this result may be stated as follows (see [19] for further discussion): Consider a black hole formed by gravitational collapse, which settles down to a stationary nal state. By the zeroth law of black hole mechanics, the surface gravity, , of this stationary black hole nal state will be constant over its event horizon. Consider a quantum eld propagating in this background spacetime, which is initially in any (non-singular) state. Then, at asymptotically late times, particles of this eld will be radiated to innity as though the black hole were a perfect black body at the Hawking temperature, eq. (5). Thus, a stationary black hole truly is a state of thermal equilibrium, and /2 truly is the physical temperature of a black hole. It should be noted that this result relies only on the analysis of quantum elds in the region exterior to the black hole. In particular, the details of the gravitational eld equations play no role, and the result holds in any metric theory of gravity obeying the zeroth law. The physical connection between the laws of black hole mechanics and the laws of thermodynamics is further cemented by the following considerations. If we take into account the back reaction of the quantum eld on the black hole (i.e., if the gravitational eld equations are used self-consistently, taking account of the gravitational eects of the quantum eld), then it is clear that if energy is conserved in the full theory, an isolated black hole must lose mass in order to compensate for the energy radiated to innity in the particle creation process. As a black hole thereby evaporates, Sbh will decrease, in violation of the second law of black hole mechanics. (Note that in general relativity, this can occur because the stress-energy tensor of quantum matter does not satisfy the null energy condition even for matter for which this condition holds classically in violation of one of the hypotheses of the area theorem.) On the other hand, there is a serious diculty with the ordinary second law of thermodynamics when black holes are present: One can simply take some ordinary matter and drop it into a black hole, where, classically at least, it will disappear into a spacetime singularity. In this latter process, one loses the entropy initially present in the matter, but no compensating gain of ordinary entropy occurs, so the total entropy, S, decreases. Note, however, that in the black hole evaporation process, although Sbh decreases, there is signicant amount of ordinary entropy generated outside the black 20

hole due to particle creation. Similarly, when ordinary matter (with positive energy) is dropped into a black hole, although S decreases, by the rst law of black hole mechanics, there will necessarily be an increase in Sbh . The above considerations motivated the following proposal [1], [20]. Although the second law of black hole mechanics breaks down when quantum processes are considered, and the ordinary second law breaks down when black holes are present, perhaps the following law, known as the generalized second law always holds: In any process, the total generalized entropy never decreases S 0, (42) where the generalized entropy, S , is dened by S S + Sbh . (43)

A number of analyses [21], [22], [23], [24] have given strong support to the generalized second law. Although these analyses have been carried out in the context of general relativity, the arguments for the validity of the generalized second law should be applicable to a general theory of gravity, provided, of course, that the second law of black hole mechanics holds in the classical theory. The generalized entropy (43) and the generalized second law (42) have obvious interpretations: Presumably, for a system containing a black hole, S is nothing more than the true total entropy of the complete system, and (42) is then nothing more than the ordinary second law for this system. If so, then Sbh truly is the physical entropy of a black hole. Although I believe that the above considerations make a compelling case for the merger of the laws of black hole mechanics with the laws of thermodynamics, there remain many puzzling aspects to this merger. One such puzzle has to do with the existence of a thermal atmosphere around a black hole. It is crucial to the arguments for the validity of the generalized second law (see, in particular, [21]) that near the black hole, all elds are in thermal equilbrium with respect to the notion of time translations dened by the horizon Killing eld a (see eq.(18) above). For an observer following an orbit of a just outside the black hole, the locally measured temperature (of all species of matter) is T = , (44) 2V 21

where V = ( a a )1/2 . Note that, in view of eq.(20) above, we see that T a/2 as the black hole horizon, H, is approached. Thus, in this limit eq.(44) corresponds to the at spacetime Unruh eect [25]. Since T as the horizon is approached, this thermal atmosphere has enormous entropy. Indeed, if no cut-o is introduced, the entropy of the thermal atmosphere is divergent.2 . However, inertial observers do not see this thermal atmosphere, and would attribute the physical eects produced by the thermal atmosphere to other causes, like radiation reaction eects [21]. In particular, with respect to a notion of time translations which would be naturally dened by inertial observers who freely fall into the black hole, the entropy of quantum elds outside of a black hole should be negligible. Thus, it is not entirely clear what the quantity S appearing in eq.(43) is supposed to represent for matter near, but outside of, a black hole. Does S include contributions from the thermal atmosphere? If so, S is divergent unless a cuto is introduced although changes in S (which is all that is needed for the formulation of the generalized second law) could still be well dened and nite. If not, what happens to the entropy in a box of ordinary thermal matter as it is slowly lowered toward the black hole? By the time it reaches its oating point [21], its contents are indistinguishable from the thermal atmosphere, so has its entropy disappeared? These questions provide good illustrations of some of the puzzles which arise when one attempts to consider thermodynamics in the framework of general relativity, as previously discussed at the end of Section II. However, undoubtedly the most signicant puzzle in the relationship between black holes and thermodynamics concerns the physical origin of the entropy, Sbh , of a black hole. Can the origin of Sbh be understood in essentially the same manner as in the thermodynamics of conventional systems (as suggested by the apparently perfect merger of black hole mechanics with thermodynamics), or is there some entirely new phenomena at work here (as suggested by the radical dierences in the present derivations of the laws
If a cut-o at the Planck scale is introduced, the entropy of the thermal atmosphere agrees, in order of magnitude, with the black hole entropy (39) There have been a number of attempts (see [24] for further discussion) to attribute the entropy of a black hole to this thermal atmosphere, or to quantum hair outside the black hole. However, if any meaning can be given to the notion of where the entropy of a black hole resides, it seems much more plausible to me that it resides in the deep interior of the black hole (corresponding to the classical spacetime singularity).
2

22

of black hole mechanics and the laws of thermodynamics)? As already remarked near the end of Section II, a classical treatment (as in Section III), or even a semiclassical treatment (as above in this Section), cannot be adequate to analyze this issue; undoubtedly, a fully quantum treatment of all of the degrees of freedom of the gravitational eld will be required. Our present understanding of quantum gravity is quite rudimentary. Most approaches used to formulate the quantum theory of elds propagating in a Minkowski background spacetime either are inapplicable or run into severe diculties when one attempts to apply them to the formulation of a quantum theory of the spacetime metric itself. However, several approaches have been developed which have some appealing features and which hold out some hope of overcoming these diculties. The most extensively developed of these approaches is string theory, and, very recently, some remarkable results have been obtained in the context of string theory relevant to understanding the origin of black hole entropy. I will very briey describe some of the key results here, referring the reader to the contribution of Horowitz [26] for further details and discussion. In the context of string theory, one can consider a low energy limit in which the massive modes of the string are neglected. In this limit, string theory should reduce to a 10-dimensional supergravity theory. If one treats this supergravity theory as a classical theory involving a spacetime metric, gab , and other classical elds, one can nd solutions describing black holes. These classical black holes can be interpreted as providing a description of states in string theory which is applicable at low energies. On the other hand, one also can consider a weak coupling limit of string theory, wherein the states are treated perturbatively about a background, at spacetime. In this limit, the dynamical degrees of freedom of the theory are described by perturbative string states together with certain soliton-like congurations, known as D-branes. In the weak coupling limit, there is no literal notion of a black hole, just as there is no notion of a black hole in linearized general relativity. Nevertheless, certain weak coupling states comprised of D-branes can be identied with certain black hole solutions of the low energy limit of the theory by a correspondence of their energy and charges. Now, the weak coupling states are, in essence, ordinary quantum dynamical degrees of freedom in a at background spacetime. The ordinary entropy, S, of these states can be computed using eq.(16) of Section II. The 23

corresponding classical black hole states of the low energy limit of the theory have an entropy, Sbh , given by eq.(39) of Section III. The remarkable results referred to above are the following: For certain classes of extremal ( = 0) [27] and nearly extremal [28] black holes, the ordinary entropy, (16), of the weak coupling D-brane states agrees exactly with the entropy, (39), of the corresponding classical black hole states occurring in the low energy limit of the theory.3 Since these entropies have a nontrivial functional dependence on energy and charges, it is hard to imagine that this agreement could be the result of a random coincidence. Furthermore, for low energy scattering, the absorption/emission coecients (gray body factors) of the corresponding D-brane and black hole congurations also agree [29]. In particular, since the temperatures of the corresponding D-brane and black hole congurations agree (by virtue of the equality of entropies), it follows that the Hawking radiation from the corresponding D-brane and black hole congurations also agree, at least at low energies. Since the entropy of the D-brane congurations arises entirely from conventional state counting, the above results strongly suggest that the physical origin of black hole entropy should be essentially the same as for conventional thermodynamic systems. However, undoubtedly, a much more complete understanding of the nature of black holes in quantum gravity when one is neither in a weak coupling nor low energy limit will be required before a denitive answer to this question can be given. As I have tried to summarize in this article, a great deal of progress has been made in our understanding of black holes and thermodynamics, but many unresolved issues remain. I believe that the attainment of a deeper understanding of the nature of the relationship between black holes and thermodynamics is our most promising route toward an understanding of the fundamental nature of both quantum gravity and thermodynamics.

References
[1] J.D. Bekenstein, Phys. Rev. D7, 2333 (1973).
Note that since the D-brane states are, in essence, ordinary quantum systems, they thereby provide an example of a quantum system which violates the Nernst theorem formulation of the third law of thermodynamics.
3

24

[2] S.W. Hawking, Phys. Rev. Lett. 26, 1344 (1971). [3] J.M. Bardeen, B. Carter, and S.W. Hawking, Commun. Math. Phys. 31, 161 (1973). [4] S.W. Hawking, Commun. Math. Phys. 43, 199 (1975). [5] R.M.Wald, Phys. Rev. D20, 1271 (1979). [6] K. Huang, Statistical Mechanics, John Wiley & Sons (New York, 1963). [7] R.M.Wald, to be published. [8] R. Penrose, in General Relativity, an Einstein Centennary Survey, ed. by S.W. Hawking and W. Israel, Cambridge University Press (Cambridge, 1979). [9] B. Carter, in Black Holes, ed. by C. DeWitt and B.S. DeWitt, Gordon and Breach (New York, 1973). [10] S.W. Hawking and G.F.R. Ellis, The Large Scale Structure of Spacetime, Cambridge University Press (Cambridge, 1973). [11] D. Sudarsky and R.M.Wald, Phys. Rev. D46, 1453 (1992). [12] P.T. Chrusciel and R.M. Wald, Commun. Math Phys. 163, 561 (1994). [13] R.M.Wald, General Relativity, University of Chicago Press (Chicago, 1984). [14] I. Racz and R.M.Wald, Class. Quant. Grav. 13, 539 (1996). [15] V. Iyer and R.M.Wald, Phys. Rev. D50, 846 (1994). [16] M. Ferraris and M. Francaviglia, in Mechanics, Analysis, and Geometry: 200 Years after Lagrange, North -Holland (Amsterdam, 1991). [17] J. Lee and R.M.Wald, J. Math. Phys. 31, 725 (1990). [18] V. Iyer and R.M.Wald, Phys. Rev. D52, 4430 (1995). [19] R.M.Wald, Quantum Field Theory in Curved Spacetime and Black Hole Thermodynamics, University of Chicago Press (Chicago, 1994). 25

[20] J.D. Bekenstein, Phys. Rev. D9, 3292 (1974). [21] W.G. Unruh and R.M.Wald, Phys. Rev. D25, 942 (1982). [22] W.H. Zurek and K.S. Thorne, Phys. Rev. Lett. 54, 2171 (1986). [23] V.P. Frolov and D.N. Page, Phys. Rev. Lett. 71, 3902 (1993). [24] R. Sorkin, contribution to this volume. [25] W.G. Unruh, Phys. Rev. D14, 870 (1976). [26] G. Horowitz, contribution to this volume. [27] A. Strominger and C. Vafa, Phys. Lett. B379, 99 (1996). [28] C. G. Callan, Jr. and J. M. Maldacena, Nucl. Phys. B472, 591 (1996); G. T. Horowitz and A. Strominger, Phys. Rev. Lett. 77, 2368 (1996). [29] J.M. Maldacena and A. Strominger, hep-th/9609026.

26