All Vacuum Near-Horizon Geometries in D-dimensions with (D 3) Commuting Rotational Symmetries − Stefan Hollands1,2 and Akihiro Ishibashi2 ∗ † 0 1 0 1School of Mathematics, Cardiff University 2 n Cardiff, United Kingdom a J 2KEK Theory Center, Institute of Particle and Nuclear Studies 7 2 High Energy Accelerator Research Organization (KEK) Tsukuba, Japan ] c q - 19 January 2010 r g [ 3 v Abstract 2 6 4 We explicitly construct all stationary, non-static, extremal near horizon geometries 3 in D dimensions that satisfy the vacuum Einstein equations, and that have D 3 . − 9 commuting rotational symmetries. Ourwork generalizes [arXiv:0806.2051] by Kunduri 0 and Lucietti, where such a classification had been given in D = 4,5. But our method 9 0 is different from theirs and relies on a matrix formulation of the Einstein equations. : Unlike their method, this matrix formulation works for any dimension. The metrics v i that we find come in three families, with horizon topology S2 TD−4, or S3 TD−5, X × × or quotients thereof. Our metrics depend on two discrete parameters specifying the r a topologytype,aswellas(D 2)(D 3)/2 continuousparameters. Notallofourmetrics − − in D 6 seem to arise as the near horizon limits of known black hole solutions. ≥ 1 Introduction Many known families of black hole solutions possess a limit wherein the black hole hori- zon becomes degenerate, i.e. where the surface gravity tends to zero; such black holes are called extremal. While extremal black holes are not believed to be physically realized as ∗[email protected] †[email protected] 1 2 macroscopic objects in nature, they are nevertheless highly interesting from the theoretical viewpoint. Due to the limiting procedure, they are in some sense at the fringe of the space of all black holes, and therefore possess special properties which make them easier to study in various respects. For example, in string theory, the derivation of the Bekenstein-Hawking entropyofblackholesfromcounting microstates(seee.g. [13]forareview) isbest understood for extremal black holes. Furthermore, many black hole solutions that have been constructed in the context of supergravity theories (see e.g. [17, 18]) have supersymmetries, and are thus automatically extremal. Many of the arguments related to the derivation of the black hole entropy—especially in the context of the “Kerr-CFT correspondence” [19, 34, 10, 3, 21, 1]—actually only involve the spacetime geometry in the immediate (actually infinitesimal) neighborhood of the black hole horizon. More precisely, by applying a suitable scaling process to the spacetime metric which in effect blows up this neighborhood, one can obtain in the limit a new spacetime metric, called a “near horizon geometry.” It is the near horizon geometry which enters many of the arguments pertaining to the derivation of the black hole entropy. The near horizon limit can be defined for any spacetime (M,g) with a degenerate Killing horizon, N—not necessarily a black hole horizon. The construction runs as follows1 . First, recall that a spacetime with degenerate Killing horizon by definition has a smooth, codi- mension one, null hypersurface N, and a Killing vector field K whose orbits are tangent N, and which on N are tangent to affinely2 parametrized null-geodesics. Furthermore, by assumption, there is a “cross section”, H, of codimension one in N with the property that each generator of K on N is isomorphic to R and intersects H precisely once. In the vicinity of N, one can then introduce “Gaussian null coordinates” u,v,ya as follows, see e.g. [36]. First, we choose arbitrarily local3 coordinates ya on H, and we Lie-transport them along the flow of K to other places on N, denoting by v the flow parameter. Then, at each point of N we shoot off affinely parametrized null-geodesics and take u to be the affine parameter along these null geodesics. The tangent vector ∂/∂u to these null geodesics is required to have unit inner product with K = ∂/∂v on H, and to be orthogonal to the Lie-transported cross-section H. It can be shown that the metric then takes the Gaussian null form g = 2dv(du+u2αdv +uβ dya)+γ dyadyb, (1.1) a ab where the function α, the one-form β = β dya, and the tensor field γ = γ dyadyb do not a ab dependonv. TheKillinghorizonN islocatedatu = 0, andthecrosssectionH atu = v = 0. The near horizon limit is now taken by applying to g the diffeomorphism v v/ǫ,u ǫu 7→ 7→ 1The general definition of a near-horizon limit was first considered in the context of supergravity black holes in [39], and in the context of extremal but not supersymmetric black holes in [12] for the static case and in [30] for the generalcase. The concept of near-horizongeometry itself has appeared previously in the literature, e.g., [20] for 4-dimensional vacuum case (also see [33] for the isolated horizon case). 2For a non-degenerate horizon, the orbits on N of K would not be affinely parametrized. 3Of course,it will take more thanone patchto coverH, but the fields γ,β,α on H below in eq. (1.1)are globally defined and independent of the choice of coordinate systems. 3 (leaving the other coordinates ya unchanged), and then taking ǫ 0. The so-obtained → metric looks exactly like eq. (1.1), but with new metric functions obtained from the old ones by evaluating them at u = 0. Thus, the fields α,β,γ of the near horizon metric neither depend on v nor u, and can therefore be viewed as fields on H. If the original spacetime with degenerate Killing horizon satisfied the vacuum Einstein equation or the Einstein equation with a cosmological constant, then the near horizon limit does, too. The near horizon limit is simpler than the original metric in the sense that it has more symmetries. For example, if the limit procedure is applied to the extremal Kerr metric in D = 4 spacetime dimensions with symmetry group R U(1), then—as observed4 first by [4] × (see also [5, 7])—the near horizon metric has an enhanced symmetry group of O(2,1) U(1). × ThefirstfactorofthisgroupisrelatedtoanAdS -factorinthemetric. Asimilarphenomenon 2 occurs for stationary extremal black holes in higher dimensions with a comparable amount of symmetry: As proved in [30], if (M,g) is a D-dimensional stationary extremal black hole with isometry group5 R U(1)D−3 and compact horizon cross section H, then the near horizon limit has the enha×nced symmetry group O(2,1) U(1)D−3. In D 5 dimensions, it × ≥ is not known at present what is the most general stationaryextremal black hole solutionwith symmetry group R U(1)D−3, so one can neither perform explicitly their near horizon limits. × Nevertheless, because the near horizon metric has an even higher degree of symmetry—the metric functions essentially only depend non-trivially on one coordinate—one can try to classify them directly. This was done for the vacuum Einstein equations in dimensions D = 4,5 by [29], where a list of all near horizon geometries, i.e. metrics of the form (1.1) with metric functions α,β,γ independent of u,v, was obtained. It is a priori far from obvious that all these metrics are the near horizon limits of actual globally defined black holes. Remarkably though, [29] could prove that the metrics found are indeed the limits of the extremal black ring [14], boosted Kerr string, Myers-Perry [37], and the Kaluza-Klein black holes [38, 32], respectively. In this paper, we give a classification of all possible vacuum near horizon geometries with symmetry group O(2,1) U(1)D−3 in arbitrary dimensions D. The method of analysis × used in [29] seems restricted to D = 4,5, so we here use a different method based on a matrix formulation of the vacuum Einstein equations that works in arbitrary dimensions. The metrics that we find come in three families depending on the topology of H, which can be either S3 TD−5,S2 TD−4 or L(p,q) TD−5, where L(p,q) is a Lens space. The × × × metrics in each of these families depend on (D 2)(D 3)/2 real parameters; they are given − − explicitly in Thm. 1 below. When specialized to D = 5, our first two families of metrics 4By construction, the near horizon geometry has the Killing fields ∂/∂v and u∂/∂u v∂/∂v, which − generate a two-parameter symmetry group. The non-trivial observation by [4] is that this actually gets enhanced to the three-parameter group O(2,1). 5The“rigiditytheorem”[23]guaranteesthatastationaryextremalblackholehasasymmetrygroupthat contains R U(1), i.e. guarantees only one axial Killing field in addition to the assumed timelike Killing field. There×fore, in D 5, assuming a factor of U(1)D−3 is a non-trivial restriction, while it is actually a ≥ consequence of the rigidity theorem in D =4. 4 must coincide with those previously found in [29], whereas the last family is shown to arise from the first one by taking quotients (this last properties generalizes to arbitrary D). In all dimensions, examples for near horizon geometries with topology S2 TD−4 are provided by × the near horizon limit of the “boosted Kerr-branes” see e.g. [30, 15]. This family of metrics depends on (D 2)(D 3)/2 real parameters and it is conceivable that all near horizon − − geometries of this topology can be obtained in this way. The analogous construction is also possible when the horizon topology is S3 TD−5. However, in this case, the resulting metrics × depend on fewer parameters. We should also point out that there are vacuum near-horizon geometries that possess fewer symmetries than R U(1)D−3. For example, the near-horizon geometry of the ex- × tremal Myers-Perry black holes, constructed explicitly in [15], has the smaller symmetry group, R U(1)[(D−1)/2]. In this paper we are not going to classify such less symmetric vac- × uum near-horizon geometries. Also, we are not going to consider the case of a non-vanishing cosmological constant, since, as far as we are aware, there has appeared no successful reduc- tionofthe Einstein gravity with a cosmological constant to a suitable nonlinear sigma model, which is however required in our approach. The same remark would apply to other theories with different matter fields. On the other hand, we expect our approach to be applicable to theories that can be reduced to suitable sigma-models. For D = 5 minimal gauged and ungauged supergravity, the near horizon geometries were classified in [31, 39] using a method different from ours. Also for D = 4 Einstein-Maxwell theory with a cosmological constant, see e.g. [28]. 2 Geometrical coordinates TheaimofthispaperistoclassifythenearhorizongeometriesinD dimensions. Asexplained in the previous section, by this we mean the problem of finding all metrics g of the form (1.1) with vanishing Ricci tensor (i.e. vacuum metrics), where γ = γ dyadyb is a smooth metric ab on the compact manifold H, β = β dya is a 1-form on H and α is a scalar function on a H. These fields do not depend on u,v, and the near horizon geometries therefore have the Killing vectors K = ∂/∂v and X = u∂/∂u v∂/∂v. We do not assume a priori that the − near horizon metrics arise from a black hole spacetime by the limiting procedure described above. Unfortunately, this problem appears to be difficult to solve in this generality, so we will makeasignificant furthersymmetry assumption. Namely, wewillassumethatourmetricsdo not only have the Killing vectors K,X, but in addition admit the symmetry group U(1)D−3, generated by (D 3) commuting Killing fields ψ1,...,ψD−3 that are tangent to H and also commute with K−,X. Thus, the full isometry group of our metric is (at least) G U(1)D−3, 2 × where G denotes the Lie-group that is generated by K,X. This means roughly speaking 2 that the metric functions can nontrivially depend only on a single variable, and our metrics may hence be called “cohomogeneity-one.” As a consequence, Einstein’s equations reduce to 5 a coupled system of non-linear ordinary differential equations in this variable. Our aim is to solve this system in the most general way and thereby to classify all near horizon geometries with the assumed symmetry. It seems that this system becomes tractable only if certain special coordinates are intro- duced that are adapted in an optimal way to the geometric situation under consideration. These coordinates are the well-known Weyl-Papapetrou coordinates up to a simple coor- dinate transformation. However, to introduce these coordinates in a rigorous and careful manner is more subtle in the present case than for non-extremal horizons. These technical difficulties are closely related to the fact that the usual Weyl-Papapetrou coordinates are actually singular onH, the very place we areinterested in most. To circumvent this problem, we follow the elegant alternative procedure introduced in [30, 29]. That procedure applies in the form presented here to non-static geometries, and we will for the rest of this paper make this assumption. The static case has been treated previously in [12, 27]. We first observe that the horizon H is a compact (D 2)-dimensional manifold with an action of U(1)D−3. By general and rather straightforwa−rd arguments (see e.g. [26, 24]) it follows that, topologically, H can only be of the following four types: S3 TD−5, × S2 TD−4, H =  × (2.2) ∼ L(p,q) TD−5,  × TD−2.   Furthermore, in the first three cases, the quotient space H/U(1)D−3 is a closed interval— which we take to be [ 1,1] for definiteness—whereas in the last case, it is S1. We will not − treat the last case in this paper6, but we note that the topological censorship theorem [11] implies that there cannot exist any extremal, asymptotically flat or Kaluza-Klein vacuum black holes with H = TD−2. Thus, while there could still be near horizon geometries with ∼ H = TD−2, they cannot arise as the limit of a globally defined black hole spacetime. ∼ In this paper, we will focus on the first three topology types. In these cases, the Gram matrix f = γ(ψ ,ψ ) (2.3) ij i j is non-singular in the interior of the interval and it has a one-dimensional null-space at each of the two end points [24]. In fact, there are integers ai Z such that ± ∈ f (x)ai 0 at boundary points 1. (2.4) ij ± → ± The integers ai determine the topology of H (i.e. which of the first three cases we are in), ± as we explain more in Thm. 1 below. The first geometric coordinate, x, parametrizes the interval [ 1,+1], and is introduced − as follows. Consider the 1-form on H defined by Σ = (detf) ⋆γ (ψ1 ψD−3), where ∧ ··· ∧ 6See, however, the note added in proof. 6 the Hodge dual is taken with respect to the metric γ on H. Using the fact that the ψ are i commuting Killing fields of γ, one can show that Σ is closed, and that it is Lie-derived by all ψ . Hence Σ may be viewed as a closed 1-form on the orbit space H/U(1)D−3, which, as we i have said, is a closed interval. It can be seen furthermore that Σ does not vanish anywhere within this closed interval, so there exists a function x, such that dx = CΣ. (2.5) The constant C is chosen so that x runs from 1 to +1. We take x to be our first coordinate, and we take the remaining coordinates on H−to be angles ϕ1,...,ϕD−3 running between 0 and 2π, chosen in such a way that ψ = ∂/∂ϕi. In these coordinates, the metric γ on H i takes the form 1 γ = dx2 +f (x)dϕidϕj . (2.6) C2detf ij To define our next coordinate, we consider the 1-form field β on H, see eq. (1.1). Standard results on the Laplace operator ∆ on a compact Riemannian manifold (H,γ) guarantee γ that there exists a smooth function λ on H such that ⋆ d⋆ β = ∆ λ, (2.7) γ γ γ where ⋆ is the Hodge star of γ. The function λ is unique up to a constant. Because β γ and γ are Lie-derived by all the rotational Killing fields ψ , it follows that L λ = c are i ψi i harmonic functions on H, i.e. constants. Furthermore, these constants must vanish, because the ψ have periodic orbits. Thus, λ is only a function of x. We also claim that the 1- i form β dλ has no dx-part. To see this, we let h be the scalar function on H defined by − h = ⋆γ(ψ1 ψD−3 [β dλ]). Using eq. (2.7) and the fact that the ψi are commuting ∧···∧ ∧ − Killing fields of γ, it is easy to show that dh = 0, so h is constant. Furthermore, by eq. (2.4) there exist points in H where the linear combinations ai ψ = 0, and it immediately follows ± i from this that h = 0 on H. This shows that β dλ has no dx-part, hence we can write − β = dλ+Ceλk dϕi, (2.8) i where we have introduced the quantities k := C−1e−λψ β. (2.9) i i · The next coordinate is defined by r := ueλ, (2.10) and we keep v as the last remaining coordinate. The coordinates ϕi,r,x,v are the desired geometrical coordinates. In these, the metric takes the form dx2 g = e−λ[2dvdr+r2(2αe−λ eλk ki)dv2]+ +f (dϕi+Crkidv)(dϕj+Crkjdv). (2.11) − i C2detf ij 7 We have also determined that the quantities ki,f ,α,λ are functions of x only. The indices ij i,j,... are raised with the inverse fij of the Gram matrix, e.g. ki = fijk . j So far, we have only used the symmetries of the metric, but not the fact that it is also required to be Ricci flat. This imposes significant further restrictions [29, 30]. Namely, one finds that ki are simply constants, and that (2αe−λ eλk ki) is a negative7 constant, i − which one may choose to be C2 after a suitable rescaling of the coordinates r,v and the − constants ki, and by adding a constant to λ. Then the Einstein equations further imply that ∂x2(e−λdetf) = −2; hence e−λ = −(x − x−)(x − x+)(detf)−1 for real numbers x±. Furthermore, λ is smooth and detf vanishes only at x = 1 by eq. (2.4), so x± = 1 and ± ± consequently e−λ = (1 x2)(detf)−1. (2.12) − Thus, in summary, we have determined that the near horizon metric is given by 1 x2 dx2 g = − (2dvdr C2r2dv2)+ +f (dϕi +rCkidv)(dϕj +rCkjdv) (2.13) detf − C2detf ij where ki,C are constants, and where f depends only on x. ij Intheremainderofthepaper,wewillworkwithaboveformofthemetric(2.13). However, we will, for completeness, also give the relation to the more familiar Weyl-Papapetrou form: For r > 0 (i.e., strictly outside the horizon), we define new coordinates (t,ρ,z,φi) by the transformation [16] z := rx (2.14) ρ := r√1 x2 (2.15) − t := Cv +(Cr)−1 (2.16) φi := ϕi +C−1kilogr. (2.17) In the new coordinates (t,ρ,z,φi), the metric then takes the Weyl-Papapetrou form ρ2dt2 e−λ g = + (dρ2 +dz2)+f (dφi +rkidt)(dφj +rkjdt), (2.18) −detf C2r2 ij where it is understood that r2 = ρ2+z2. Note that, by contrast with the coordinate system (v,r,x,ϕi), the Weyl-Papapetrou coordinate system does not cover the horizon itself, i.e., it is not defined for r = 0 but only for r > 0. This can be seen in several ways, for example by noting that the coordinate transformation is singular at r = 0, i.e. on the horizon, or alternatively, by noting that the horizon corresponds in the new coordinates to the single point ρ = z = 0. This behavior is characteristic for extremal horizons and does not happen in the non-extremal case. 7Here one must use that the metric is not static, i.e. that not all ki vanish. 8 In obtaining our form (2.13) for the near horizon metric, we have used up all but the ij-components of the Einstein equations. The remaining Einstein equations determine the matrix of functions f (x). As is well-known [35], a beautifully simple form of these equations ij canbeobtainedbyintroducingthetwist potentialsoftherotationalKillingfieldsasauxiliary variables. These potentials χ are defined up to a constant by i dχi = ⋆(ψ1 ψD−3 dψi). (2.19) ∧···∧ ∧ To see that this equation makes sense, one has to prove that the right side is an exact form. Indeed, taking d of the right side and using the vanishing of the Ricci tensor together with the fact that the Killing fields all commute, one gets zero. To see that the right side is even exact, it is best to pass to the orbit space M/(G U(1)D−3) first, which can be identified 2 × with the interval [ 1,1]. Then the χ can be defined on this orbit space and lifted back to i − functions onM. It also follows fromthis construction that χ only depends onthe coordinate i x parametrizing [ 1,1]. Setting − (detf)−1 (detf)−1χ Φ = (detf)−1χ f −+(detf)−1χi χ , (2.20) i ij i j (cid:18)− (cid:19) it is well-known that the vanishing of the Ricci-tensor implies that ∂ [(1 x2)Φ−1∂ Φ]+∂ [r2Φ−1∂ Φ] = 0. (2.21) x x r r − These equations are normally written in the Weyl-Papapetrou coordinates ρ,z (see e.g. [24]), and the above form is obtained simply by the change of variables eq. (2.14). Since Φ is a function of x only in our situation (but would not be e.g. for black holes without the near horizon limit taken) an essential further simplification occurs: The second term in the above set of matrix equations is simply zero! Hence, the content of the remaining Einstein equations is expressed in the matrix of ordinary differential equations ∂ [(1 x2)Φ−1∂ Φ] = 0. (2.22) x x − In fact, this equation could be derived formally and much more directly by simply assuming the Weyl-Papapetrou form of the metric, introducing r,x as above, and then observing that, in the near horizon limit, the dependence on r is scaled away, so that the matrix partial differential equations (2.21) reduce to the ordinary differential equations (2.22). 3 Classification To determine all near horizon metrics (2.13), we must solve the matrix equations (2.22), i.e. find f ,χ . Then the constants ki are given by ij i 1 x2 ki = − fij∂ χ , (3.23) x j detf 9 and this determines the full metric up to the choice of the remaining constant C. We must furthermore ensure that, among all such solutions, we pick only those that give rise to a smooth metric g. The equations (2.22) for Φ are easily integrated to L 1+x Φ(x) = Q exp[2arcth(x) L] = Q . (3.24) · 1 x (cid:18) − (cid:19) Here, Q = Φ(0),L = 1(1 x2)Φ(x)−1∂ Φ(x) are both constant real (D 2) (D 2) 2 − x − × − matrices, and we mean the matrix exponential etc. It follows from the definition that Φ has the following general properties: It is symmetric, detΦ = 1, and it is positive definite. It is an easy consequence of these properties that detQ = 1, TrL = 0 (taking the determinant of the equation), that Q = QT is positive definite, and that LTQ = QL. These relations allow us to write Q = STS for some real invertible matrix S = (s ) of determinant 1, and IJ to conclude that SLS−1 is a real symmetric matrix. By changing S to VS, where±V is a suitable orthogonal transformation, we can achieve that σ 0 ... 0 0 0 σ ... 0 SLS−1 =  . 1 .  (3.25) . . . .   0 0 ... σD−3     is a real diagonal matrix, while leaving Q unchanged. It then follows that Φ(x) = ST exp[2arcth(x) SLS−1]S, that is · D−3 1+x σK Φ (x) = s s . (3.26) IJ KI KJ 1 x K=0(cid:18) − (cid:19) X This is the most general solution to the field equation for Φ in the near horizon limit, and it depends on the real parameters s ,σ , which are subject to the constraints IJ I D−3 det(s ) = 1, σ = 0. (3.27) IJ I ± I=0 X The near horizon metric is completely fixed in terms of Φ. It can be obtained combining eqs. (3.26) with eq. (2.20) to determine f ,χ , which in turnthen fix the remaining constants ij i ki,C in the near horizon metric. In the rest of this section, we explain how this can be done. It turns out that the smoothness of the near horizon metric also implies certain constraints on the parameters σ ,s , and we will derive the form of these. Our analysis applies in I IJ principle to all dimensions D 4. The case D = 4, while being simplest, is somewhat ≥ different from the remaining cases D 5 and would require us to distinguish these cases in ≥ 10 many of the formulae below. Therefore, to keep the discussion simple, we will stick to D 5 ≥ in the following. First, we consider the ij-component of Φ in eq. (3.26). By eq. (2.20) this is also equal to D−3 1+x σI s s = Φ = f +(detf)−1χ χ . (3.28) Ii Ij ij ij i j 1 x I=0 (cid:18) − (cid:19) X Now, the coordinate x [ 1,1] parametrizes the orbit space H/U(1)D−3 of the horizon, ∈ − which is topologically a finite interval. The boundary points x = 1 correspond to points ± on the horizon where an integer linear combination ai ψ of the rotational Killing fields ± i vanishes. This is equivalently expressed by the condition f (x)aj 0 as x 1. By P ij ± → → ± contrast, forallvaluesofx ( 1,+1), nolinearcombinationoftherotationalfieldsvanishes. ∈ − Therefore, detf = 0 for x ( 1,+1), while detf 0 as x 1. In fact, using eq. (2.12) 6 ∈ − → → ± one sees that (detf)−1 = 2c2(1 x)−1 +2c2(1+x)−1 +... as x 1, (3.29) + − − → ± where the dots represent contributions that go to a finite limit, and where c± are non-zero constants related to λ by 4c2 = e−λ(±1) = 0. The twist potentials χ also go to a finite limit ± i 6 as x 1. By adding suitable constants to the twist potentials if necessary, we may achieve → ± that 1 χ µ as x 1, (3.30) i i → c± → ± where µ R are constants. The upshot of this discussion is that, as one approaches the i boundary∈points, the components Φ are dominated by the rank-1 part (detf)−1χ χ , which ij i j diverges as 2(1 x)−1µ µ as x 1. This behavior can be used to fix the possible values i j ∓ → ± of the eigenvalues σ as follows. First, it is clear that at least one of the eigenvalues must I be non-zero, for otherwise the right side of eq. (3.28) would be smooth as x 1, which → ± we have just argued is not the case. Let us assume without loss of generality then that σD−3 σD−3−n > 0 are the n positive eigenvalues. Multiplying eq. (3.28) by 1 x ≥ ··· ≥ − and taking x +1, we see that σD−3 = 1, that µi = s(D−3)i, and that all other remaining positive eigenv→alues must be strictly between 0 and 1. If we now subtract (1 x2)−1µ µ i j − from both sides of the equation, then the right side of eq. (3.28) goes to a finite limit as x 1, and so the left side has to have that behavior, too. This is only possible if there → are no other remaining positive eigenvalues besides σD−3. A similar argument then likewise shows that there is only one negative eigenvalue, which has to be equal to 1 (without loss − of generality we may take σD−4 = 1) and that µi = s(D−4)i. − In summary, we have shown that 0 if I D 5, ≤ − σ = 1 if I = D 4, (3.31) I  − − 1 if I = D 3, −  