Local Fluid Dynamical Entropy from Gravity
Sayantani Bhattacharyya, Veronika E Hubeny, R. Loganayagam, Gautam, Mandal, Shiraz Minwalla, Takeshi Morita, Mukund Rangamani, Harvey S. Reall

TL;DR
This paper constructs and analyzes the event horizons of geometries dual to fluid flows in strongly coupled N=4 super Yang Mills theory, defining a local entropy current via the horizon's area form.
Contribution
It demonstrates the regularity of horizons in these geometries and establishes a local entropy current in the dual field theory using a boundary-to-horizon map.
Findings
Event horizons are regular in these geometries.
A local entropy current is defined from the horizon area.
The divergence of the entropy current is positive, consistent with the area theorem.
Abstract
Spacetime geometries dual to arbitrary fluid flows in strongly coupled N=4 super Yang Mills theory have recently been constructed perturbatively in the long wavelength limit. We demonstrate that these geometries all have regular event horizons, and determine the location of the horizon order by order in a boundary derivative expansion. Intriguingly, the derivative expansion allows us to determine the location of the event horizon in the bulk as a local function of the fluid dynamical variables. We define a natural map from the boundary to the horizon using ingoing null geodesics. The area-form on spatial sections of the horizon can then be pulled back to the boundary to define a local entropy current for the dual field theory in the hydrodynamic limit. The area theorem of general relativity guarantees the positivity of the divergence of the entropy current thus constructed.
Click any figure to enlarge with its caption.
Figure 1
Figure 2
Figure 3| Symbol | Definition | Symbol | Definition |
|---|---|---|---|
| dimensions of boundary | The event horizon (d-dimensional) | ||
| Bulk co-ordinates | Boundary co-ordinates | ||
| A spatial slice of | Co-ordinates on | ||
| Bulk metric,(2.2) | Boundary metric (Minkowski) | ||
| Induced metric on | Metric on | ||
| Horizon function, (2.7) | Induced metric on | ||
| Eqn. of Horizon | Entropy (d-1)-form on | ||
| Normal vector to the Horizon (2.9) | See (2.17) | ||
| See (2.2), (A.5), (5.1) | See (2.2), (A.5), (5.1) | ||
| See (2.2), (A.10), (5.1) | See (2.2), (5.1) | ||
| Fluid temperature | Shear viscosity | ||
| Energy-momentum tensor | Entropy current | ||
| Fluid velocity () | Projection tensor, | ||
| Fluid acceleration, (A.2) | Fluid expansion, (A.2) | ||
| Shear strain rate, (A.2) | Fluid vorticity, (A.2) | ||
| Visco-elastic stress | |||
| Weyl-covariant derivative | See (B.4) | ||
| Riemann tensor | |||
| Ricci tensor/scalar | See (B.4) | ||
| Einstein tensor | See (B.4) | ||
| Weyl curvature |
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
\captionnamefont\captiontitlefont\captiondelim
: \hangcaption
Local Fluid Dynamical Entropy from Gravity
Sayantani Bhattacharyyaa, Veronika E Hubenyb, R. Loganayagama
Gautam Mandala, Shiraz Minwallaa, Takeshi Moritaa,
Mukund Rangamanib, and Harvey S. Reallc[email protected], [email protected], [email protected],
[email protected], [email protected], [email protected],
[email protected], [email protected]
a Department of Theoretical Physics,Tata Institute of Fundamental Research,
Homi Bhabha Rd, Mumbai 400005, India
b Centre for Particle Theory & Department of Mathematical Sciences,
Science Laboratories, South Road, Durham DH1 3LE, United Kingdom
c Department of Applied Mathematics and Theoretical Physics, Centre for Mathematical Sciences,
Wilberforce Road, Cambridge CB3 0WA, United Kingdom
TIFR/TH/08-06DCPT-08/15
Spacetime geometries dual to arbitrary fluid flows in strongly coupled super Yang Mills theory have recently been constructed perturbatively in the long wavelength limit. We demonstrate that these geometries all have regular event horizons, and determine the location of the horizon order by order in a boundary derivative expansion. Intriguingly, the derivative expansion allows us to determine the location of the event horizon in the bulk as a local function of the fluid dynamical variables. We define a natural map from the boundary to the horizon using ingoing null geodesics. The area-form on spatial sections of the horizon can then be pulled back to the boundary to define a local entropy current for the dual field theory in the hydrodynamic limit. The area theorem of general relativity guarantees the positivity of the divergence of the entropy current thus constructed.
Contents
-
4.1 Classification of ingoing null geodesics near the boundary
-
6.1 The most general Weyl covariant entropy current and its divergence
-
6.2 Positivity of divergence of the gravitational entropy current
-
B.1 Constraints on the entropy current: Weyl covariance and the second law
1 Introduction
Over the last few years, a special class of strongly coupled -dimensional conformal field theories have been “solved” via the AdS/CFT duality. Quite remarkably, the solution to these theories is given by the equations of dimensional gravity (interacting with other fields) in AdSd+1 spacetime. Since the long distance dynamics of any genuinely interacting field theory is well described by the equations of relativistic hydrodynamics, it follows as a prediction of the AdS/CFT correspondence that at long distances, the equations of gravity in an AdSd+1 background should reduce to the (relativistic) Navier-Stokes equations in dimensions. There is now substantial direct evidence for the connection between the long distance equations of gravity on AdSd+1 spacetime and dimensional relativistic fluid dynamics; cf., [1, 2, 3, 4, 6, 5, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38] for a sampling of the literature on the subject.
In particular, it was noted in [33] that the equations of pure gravity with a negative cosmological constant form a universal subsector in any theory of gravity on AdS spacetime. Following up on earlier work [20, 21, 23], it was demonstrated in [33] (for AdS5) and more recently in [37] (for AdS4) that Einstein’s equations in this universal sector may be recast, order by order in a boundary derivative expansion, into equations of motion for two collective fields, namely – the ‘temperature’ and the ‘velocity’. These new equations of motion turn out to be simply the relativistic Navier-Stokes equations of fluid dynamics.
The gravitational solutions of [33] and [37] constitute an explicit map from the space of solutions of the hydrodynamic equations to the space of long wavelength gravitational solutions (which are asymptotically AdS).222By ‘long wavelength’ solutions, we mean solutions whose spacetime variations are slow on a scale set by their respective boundary extrinsic curvature. Via the AdS/CFT dictionary this is the same as the requirement that the solutions vary slowly on the scale of the inverse temperature associated with the local energy density of the solution. Subject to a regularity condition that we will discuss further below, the solutions of [33, 37] are locally exhaustive in solution space i.e., all long wavelength solutions to Einstein’s equations that lie nearby in solution space to a metric dual to a particular fluid flow are themselves metrics dual to slightly perturbed fluid flows. This at first sight surprising result is a consequence of the requirement of regularity. This requirement cuts down the 9-parameter space of Fefferman-Graham type solutions of AdS5 spacetime – parameterized by a traceless boundary stress tensor – to the 4-parameter set of solutions of fluid dynamics.
We believe the local exhaustiveness of the gravity solutions dual to fluid dynamics, described in the previous paragraph, in fact generalizes to a global statement. We think it likely, in other words, that the solutions of [33, 37] in fact constitute all long wavelength asymptotically AdS solutions of gravity with a cosmological constant; we pause here to explain why. Every state in a conformal field theory has an associated local energy density and a consequent associated mean free path length scale , the inverse of the temperature that is thermodynamically associated with this energy density. As a consequence of interactions every state is expected to evolve rapidly – on the time scale – towards local thermodynamical equilibrium, in an appropriate coarse grained sense,333The limit of the field theory (dual to the classical limit in gravity) justifies this coarse graining and supresses consequent fluctuations (which are dual to quantum fluctuations in gravity). at the local value of the temperature. This approach to local equilibrium is not long wavelength in time and is not well described by fluid dynamics. The dual bulk description of this (short wavelength) phenomenon is presumably gravitational collapse into a black hole. On the other hand, once local equilibrium has been achieved (i.e., a black hole has been formed) the system (if un-forced) slowly relaxes towards global equilibrium. This relaxation process happens on length and time scales that are both large compared to the inverse local temperature, and is well described by fluid dynamics and therefore by the solutions of [33, 37]. In other words it seems plausible that all field theory evolutions that are long wavelength in time as well as space are locally equilibriated, and so are well described by fluid dynamics. The discussions of this paragraph, coupled with the AdS/CFT correspondence, motivate the conjecture that the solutions of [33, 37] are the most general regular long wavelength solutions to Einstein’s equations in a spacetime with negative cosmological constant in five and four spacetime dimensions respectively.
We pause here to note two aspects of the solutions of [33, 37] that we will have occasion to use below. First, it is possible to foliate these solutions into a collection of tubes, each of which is centered about a radial ingoing null geodesic emanating from the AdS boundary. This is sketched in Fig. 1 for a uniform black brane, where we indicate the tubes on a local portion of the spacetime Penrose diagram.444This is a causal diagram which captures the entire globally extended spacetime (note that in order for the boundaries to be drawn straight, the singularities are curved, as discussed in [39]). For a realistic collapse scenario, described by the nonuniform solutions of [33], only the right asymptotic region and the future horizon and singularity are present. As we will explain below, the congruence of null geodesics (around which each of our tubes is centered) yields a natural map from the boundary of AdS space to the horizon of our solutions. When the width of these tubes in the boundary directions is small relative to the scale of variation of the dual hydrodynamic configurations, the restriction of the solution to any one tube is well-approximated tube-wise by the metric of a uniform brane with the local value of temperature and velocity. This feature of the solutions – the fact that they are tube-wise indistinguishable from uniform black brane solutions – is dual to the fact that the Navier-Stokes equations describe the dynamics of locally equilibrated lumps of fluid.
Second, the gravitational solutions constructed in [33] are regular everywhere away from a spacelike surface, and moreover the authors conjectured that this singularity is shielded from the boundary of AdS space by a regular event horizon. We will prove this conjecture by explicitly constructing the event horizon of the solutions of [33] order by order in the derivative expansion. It should be possible to carry out a parallel study for the solutions presented in [37] for four dimensions. We will not carry out such a study here; however, aspects of our discussion are not specific to AdS5 and can be used to infer the desired features of dimensional hydrodynamics. We expect that the results of such a study would be similar to those presented in this paper.
As we have explained above, we study the causal properties – in particular, the structure of the event horizon for the solutions presented in [33]. We then proceed to investigate various aspects of the dynamics – specifically, the entropy production – at this event horizon. In the rest of the introduction, we will describe the contents of this paper in some detail, summarizing the salient points.
As we have discussed above, [33] provides a map from the space of solutions of fluid dynamics to a spacetime that solves Einstein’s equations. The geometry we obtain out of this map depends on the specific solution of fluid dynamics we input. In this paper we restrict attention to fluid dynamical configurations that approach uniform homogeneous flow at some fixed velocity and temperature at spatial infinity. It seems intuitively clear from the dissipative nature of the Navier-Stokes equations that the late time behaviour of all fluid flows with these boundary conditions will eventually become and ; we assume this in what follows.555This is true for conformal fluid dynamics in spacetime dimensions for . Conformal fluid dynamics in dimensions is not dissipative (for instance it is non-viscous since here the shear tensor does not exist). More generally, there are as many degrees of freedom in a traceless dimensional stress tensor as in temperature and velocity fields in dimensions. Consequently, the most general solution to dimensional ‘fluid dynamics’ is simply given by and for arbitrary functions and . This solution describes left and right moving waves that maintain their shape forever, propagating non-dissipatively at the speed of light. We thank A. Strominger discussions on this point. The gravitational dual to such an equilibrated fluid flow is simply the metric of a uniformly boosted black brane.
The causal structure of the uniform black brane is given by the Penrose diagram plotted in Fig. 1 (see [39]). In particular, the equation for the event horizon of a uniform black brane is well known. The event horizon of the metric dual to the full fluid flow is simply the unique null hypersurface that joins with this late time event horizon in the asymptotic future.666Note that for a generic hydrodynamic solution, the bulk spacetime has no manifest isometries; the event horizon is therefore not a Killing horizon. It turns out to be surprisingly simple to construct this hypersurface order by order in the boundary derivative expansion used in [33]. In this paper, we perform this construction up to second order in derivatives. Within the derivative expansion it turns out that the radial location of the event horizon is determined locally by values and derivatives of fluid dynamical velocity and temperature at the corresponding boundary point. This is achieved using the boundary to horizon map generated by the congruence of ingoing null geodesics described above (see Fig. 1).777This map may be motivated as follows. Consider perturbing the fluid at a boundary point , e.g., by turning on some local operator of Yang Mills theory. This perturbation instanteneously alters all fluid quantities, including the entropy, at . However, it only alters the geometry near the horizon at and within the lightcone emanating from at the boundary. It is therefore plausible that local properties of the spacetime in the neighbourhood of ingoing geodesics that emanate from capture properties of the fluid at .
However, while locality is manifest in a derivative expansion, upon summing all orders we expect this local behaviour to transmute into a limited nonlocality: the radial position of the event horizon at a given point should be determined by the values of fluid dynamical variables in a patch of size centered around the associated boundary point. The ateleological behaviour of the event horizon is a surprising feature of these solutions and implies that the event horizon behaves as a ‘membrane’ whose vibrations are a local mirror of fluid dynamics. Our explicit construction of the event horizon of the metrics dual to fluid dynamics is one of the key results of our paper; cf., (5.4).
We now turn to a description of our second main result; the construction of an entropy current with non-negative divergence for a class of asymptotically AdS solutions of gravity, and its explicit evaluation for the solutions of [33] at second order in the derivative expansion.888We have been informed that A. Strominger and S. Hartnoll have independently discussed the construction of a positive divergence entropy current on the horizon, utilizing Brown York stress tensor. As we will see in § 3, it is possible to define a natural area -form on any event horizon in a dimensional spacetime in general relativity. This form is defined to ensure that its integral over any co-dimension one spatial slice of the horizon is simply the area of that submanifold. It follows almost immediately from the definition of this form and the classic area increase theorems of general relativity that the exterior derivative (on the event horizon) of this -form, viewed of as a top dimensional form on the horizon, is ‘positive’ (we explain what we mean by the positivity of a top form on the horizon in § 3.1).
The positivity of the exterior derivative of the area -form is a formally elegant restatement of the area increase theorem of general relativity that is local on the horizon. Hence we would like to link this statement to the positivity of the entropy production in the boundary theory. However, at least naively, the CFT fluid dual to our solutions lives at the boundary of AdS space rather than on its horizon. If we wish to study the interplay between the local notion of entropy of the fluid and the fluid equations of motion, it is important for these quantities to be defined on the same space. In order to achieve this, in § 4 we use a congruence of null geodesics described above to provide a ‘natural’ map from the boundary to the horizon for a class of asymptotically AdS solutions of gravity (which include but are slightly more general than those of [33]). The pullback of the area -form under this map now lives at the boundary, and also has a ‘positive’ exterior derivative. Consequently, the ‘entropy current’, defined as the boundary Hodge dual to the pull-back of the area -form on the boundary (with appropriate factors of Newton’s constant), has non-negative divergence, and so satisfies a crucial physical requirement for an entropy current of fluid dynamics.
In § 5, we then proceed to implement the construction described in the previous paragraph for the solutions of [33]. This enables us to derive an expression for the entropy current, , with non-negative divergence, valid up to second order in the derivative expansion. As a check of our final result, we use the equations of fluid dynamics to independently verify the non-negativity of divergence of our entropy current at third order in the derivative expansion. An example of an entropy current for a conformal fluid with non-negative divergence was first described in [40].
We also take the opportunity to extend and complete the analysis presented in [40] to find the most general Weyl covariant two derivative entropy current consistent with the second law. Note that the requirement of pointwise non-negativity of the entropy production – which we impose as a physical constraint of acceptable entropy currents – carries useful information even within the derivative expansion, though this is a little subtle to unravel. In particular, in § 6 we present a parameterization of the most general (7 parameter) class of Weyl invariant candidate entropy currents that has the correct equilibrium limit, to second order in the derivative expansion. We also demonstrate that only a five dimensional sub-class of these currents is consistent with the requirement of pointwise non-negativity of to third order in derivatives. We then turn our attention to the arbitrariness of our gravitational construction of the entropy current and demonstrate that there appears to be a two parameter family of physically acceptable generalizations of this bulk construction (associated with physically acceptable generalizations of the boundary to horizon map and also the generalisations of the area -form itself). As a result, we conclude that the gravitational construction presented in this paper yields a two dimensional sub-class in the five dimensional space of entropy currents with non-negative divergence. It would interesting to understand directly from field theory, what principle (if any) underlies the selection of this distinguished class of entropy currents. It would also be interesting to investigate whether the remaining positive entropy currents may be obtained from a generalized gravitational procedure, perhaps involving apparent and other local horizons.999We thank A. Strominger and M. Van Raamsdonk for discussions on this point.
This paper is organized as follows. In § 2 below, we present an order by order construction of the event horizon in a class of metrics that include those of [33]. Following that, in § 3, we present our construction of a local entropy -form for the event horizons in gravity, and then implement our constructions in detail for the class of metrics studied in § 2. In § 5, we specialize the results of § 2 and § 3 to the metrics dual to fluid dynamics [33], using a map for translating horizon information to the boundary developed in § 4. We obtain explicit formulae, to second order in the derivative expansion, for the event horizon and the entropy current in the geometries of [33]. In § 6, we explain in detail the nature of the constraint imposed on second order terms in the expansion of the entropy current by the requirement of non-negativity of entropy production at third order in the derivative expansion. We also investigate the relationship between the geometrically constructed entropy current and the general entropy current of non-negative divergence generalizing the analysis of [40]. Finally, in § 7, we end with a discussion of our results and open questions. Some technical results regarding the computations are collected in various Appendices.
2 The Local Event Horizon
As we have explained in the introduction, in this paper we will study the event horizon of the metrics dual to fluid dynamics presented in [33]. In that reference the authors construct an explicit classical spacetime dual to an arbitrary solution of fluid dynamics, accurate to second order in the boundary derivative expansion. While the explicit solutions of [33] are rather involved, we will see below that the structure of the event horizons of these solutions are insensitive to many of these details. Consequently, in this section we will describe the metric of [33] only in general structural form, and carry out all our computations for an arbitrary spacetime of this form. In § 5 we will specialize these calculations to the detailed metrics of [33]. We start by presenting a geometric interpretation for the coordinate system used in [33].
2.1 Coordinates adapted to a null geodesic congruence
Consider a null geodesic congruence (i.e., a family of null geodesics with exactly one geodesic passing through each point) in some region of an arbitrary spacetime. Let be a hypersurface that intersects each geodesic once. Let be coordinates on . Now ascribe coordinates to the point at an affine parameter distance from , along the geodesic through the point on with coordinates . Hence the geodesics in the congruence are lines of constant . In this chart, this metric takes the form
[TABLE]
where the geodesic equation implies that is independent of . It is convenient to generalize slightly to allow for non-affine parametrization of the geodesics: let be a parameter related to by . Then, in coordinates , the metric takes the form101010We use upper case Latin indices to denote bulk directions, while lower case Greek indices will refer to field theory or boundary directions. Furthermore, we use lower case Latin indices to denote the spatial directions in the boundary. Finally, we use to indicate the dependence on the four coordinates . Details regarding the conventions used in this paper can be found in Appendix A.
[TABLE]
Note that could be spacelike, timelike, or null. We shall take to be timelike.
This metric has determinant , where is the inverse of . Hence the metric and its inverse will be smooth if , and are smooth, with , invertible, and timelike. These conditions are satisfied on, and outside, the horizons of the solutions that we shall discuss below.
2.2 Spacetime dual to hydrodynamics
The bulk metric of [33] was obtained in a coordinate system of the form (2.2) just described, where the role of is played by the conformal boundary and the null geodesics are future-directed and ingoing at the boundary. The key assumption used to derive the solution is that the metric is a slowly varying function of or, more precisely, that the metric functions have a perturbative expansion (with a small parameter ):
[TABLE]
[TABLE]
The function above has the form ; however, the only property of that we will use is that . The remaining functions (, ) are all local functions of the inverse temperature and the velocity and the coordinate , whose form was determined in [33] and is indicated below in (5.1) and given explicitly in Appendix A; we however will not need the specific form of these functions for the present discussion. As far as the calculations in this section are concerned, the expressions and may be thought of as arbitrary functions of and . The tensor is a co-moving spatial projector.
In the above formulae, is a formal derivative counting parameter. Any expression that multiplies in (2.3) and (2.4) is of order in boundary field theory derivatives. Note that any boundary derivative of any of the functions above is always accompanied by an additional explicit power of . As in [33], all calculations in this paper will be performed order by order in which is then set to unity in the final results. This is a good approximation when field theory derivatives are small in units of the local temperature.
As we have explained in the Introduction, the metrics presented in [33] simplify to the uniform black brane metric at late times. This metric describes a fluid configuration with constant and . As the derivative counting parameter vanishes on constant configurations, all terms in the summation in (2.3) and (2.4) vanish on the uniform black brane configuration. The event horizon of this simplified metric is very easy to determine; it is simply the surface . Consequently, the event horizon of the metric (2.2) has a simple mathematical characterization; it is the unique null hypersurface that reduces exactly, at infinite time to .
In § 2.3 we will describe a local construction of a null hypersurface in the metric (2.2). Our hypersurface will have the property that it reduces exactly to when and are constants, and therefore may be identified with the event horizon for spacetimes of the form (2.2) that settle down to constant and at late times, as we expect for metrics dual to fluid dynamics. We will evaluate our result for the metrics of [33] in § 5 where we will use the explicit expressions for the functions appearing in (2.2).
2.3 The event horizon in the derivative expansion
When is set to zero and and are constants, the surface is a null hypersurface in metrics (2.2). We will now determine the corrected equation for this null hypersurface at small , order by order in the expansion. As we have explained above, this hypersurface will be physically interpreted as the event horizon of the metrics presented in [33].
The procedure can be illustrated with a simpler example. Consider the Vaidya spacetime, describing a spherically symmetric black hole with ingoing null matter:
[TABLE]
Spherical symmetry implies that the horizon is at , with normal . Demanding that this be null gives , a first order ODE for . Solving this determines the position of the horizon non-locally in terms of . However, if we assume that is slowly varying and approaches a constant for large , i.e.,
[TABLE]
then we can solve by expanding in derivatives. Consider the ansatz, , for some constants ; it is easy to show that the solution for the horizon is given by , , , etc.. Hence we can obtain a local expression for the location of the horizon in a derivative expansion.
Returning to the spacetime of [33], let us suppose that the null hypersurface that we are after is given by the equation
[TABLE]
As we are working in a derivative expansion we take
[TABLE]
Let us denote the normal vector to the event horizon by : by definition,
[TABLE]
which also has an expansion. We will now determine and order by order in . In order to compute the unknown functions we require the normal vector to be null, which amounts to simply solving the equation
[TABLE]
order by order in perturbation theory. Note that
[TABLE]
In particular, to order , only the functions for appear in (2.11). However, the LHS of (2.10) includes a contribution of two factors of contracted with the metric. This contribution is equal to evaluated at the horizon. Expanding this term to order we find a contribution
[TABLE]
where is defined in (2.15) below, together with several terms that depend on for . It follows that the expansion of (2.10) to order in yields a simple algebraic expression for , in terms of the functions which are determined from lower order computations.
More explicitly, equation (2.10) gives , with the inverse metric given by:
[TABLE]
where the ‘inverse -metric’ is defined via . Hence the expression for the location of the event horizon (2.10) to arbitrary order in is obtained by expanding
[TABLE]
to the requisite order in , using the expansion of the individual quantities and specified above, as well as of .
2.4 The event horizon at second order in derivatives
The equation (2.10) is automatically obeyed at order . At first order in we find that the location of the event horizon is given by with111111We have used here the fact that and which follow from the solution of [33]. We also restrict to solutions which are asymptotically AdS5 in this section.
[TABLE]
where we define
[TABLE]
At next order, , we find
[TABLE]
where we have121212It is important to note that in our expressions involving the boundary derivatives we raise and lower indices using the boundary metric ; in particular, and with this defintion .
[TABLE]
As all functions and derivatives in (2.14) and (2.16) are evaluated at and the point and we retain terms to and respectively.
It is now simple in principle to plug (2.16) into (2.2) to obtain an explicit expression for the metric of the event horizon.131313There are thus three metrics in play; the bulk metric defined in (2.2), the boundary metric which is fixed and chosen to be and finally the metric on the horizon , , which we do not explicitly write down. As a result there are differing and often conflicting notions of covariance; we have chosen to write various quantities consistently with boundary covariance since at the end of the day we are interested in the boundary entropy current. We will choose to use the coordinates to parameterize the event horizon. The normal vector is a vector in the tangent space of the event horizon (this follows since the hypersurface is null), i.e.,
[TABLE]
which is easily obtained by using the definition (2.9) and the induced metric on the event horizon; namely
[TABLE]
Before proceeding to analyze the entropy current associated with the local area-form on this event horizon, let us pause to consider the expression (2.16). First of all, we see that for generic fluids with varying temperature and velocity, the radial coordinate of the horizon varies with , which, to the first order in the derivative expansion, is given simply by the local temperature. The constraints on this variation are inherited from the equations of relativistic fluid dynamics which govern the behaviour of these temperature and velocity fields, as discussed above. Note that the variation of at a given and as a function of time, can of course be non-monotonic. As we will see in the next section, only the local area needs to increase. This is dual to the fact that while a local fluid element may warm up or cool down in response to interacting with the neighbouring fluid, the local entropy production is always positive.
An example of the behaviour of is sketched in the spacetime diagram of Fig. 2, with time plotted vertically and the radial coordinate as well as one of the spatial coordinates plotted horizontally.
3 The Local Entropy Current
Having determined the location of the event horizon, it is a simple matter to compute the area of the event horizon to obtain the area of the black brane. However, as we wish to talk about the spatio-temporal variation of the entropy, we will first describe entropy production in a local setting. This will allow us to derive an expression for the boundary entropy current in § 5.
3.1 Abstract construction of the area -form
In this brief subsection we present the construction of the area form on the spatial section of any event horizon of a dimensional solution of general relativity.
First, recall that the event horizon is a co-dimension one null submanifold of the dimensional spacetime. As a result its normal vector lies in its tangent space. The horizon generators coincide with the integral curves of this normal vector field, which are in fact null geodesics141414This follows from the fact that the event horizon is the boundary of the past of future infinity together with the fact that boundaries of causal sets are generated by null geodesics [41]. We pause here to note a technical point regarding the behaviour of the horizon generators: While by definition these null geodesics generating the event horizon have no future endpoints [42], they do not necessarily remain on the event horizon when extended into the past. This is because in general dynamical context, these geodesics will have non-zero expansion, and by Raychaudhuri’s equation they must therefore caustic in finite affine parameter when extended into the past. Hence, although the spacetime, and therefore the event horizon, are smooth, the horizon generators enter the horizon at points of caustic. However, since the caustic locus forms a set of measure zero on the horizon, in the following discussion we will neglect this subtlety. that are entirely contained within the event horizon. Let us choose coordinates , with