ApJ, accepted

Preprint typeset using LATEX style emulateapj v. 12/14/05

A FAST NEW PUBLIC CODE FOR COMPUTING PHOTON ORBITS IN A KERR SPACETIME
Jason Dexter
Department of Physics, University of Washington, Seattle, WA 98195-1560, USA

Eric Agol
Department of Astronomy, University of Washington, Box 351580, Seattle, WA 98195, USA
ApJ, accepted
arXiv:0903.0620v1 [astro-ph.HE] 3 Mar 2009

ABSTRACT
Relativistic radiative transfer problems require the calculation of photon trajectories in curved
spacetime. We present a novel technique for rapid and accurate calculation of null geodesics in the
Kerr metric. The equations of motion from the Hamilton-Jacobi equation are reduced directly to
Carlson’s elliptic integrals, simplifying algebraic manipulations and allowing all coordinates to be
computed semi-analytically for the first time. We discuss the method, its implementation in a freely
available FORTRAN code, and its application to toy problems from the literature.
Subject headings: accretion — black hole physics — radiative transfer — relativity

1. INTRODUCTION Noble et al. (2007) created images of galactic center black

Efficient and accurate computation of null geodesics hole candidate Sagittarius A* (Sgr A*) using axisymmet-
in the vicinity of spinning black holes is important for ric general relativistic MHD (GRMHD) simulations, and
studies of active galaxies, X-ray binaries, and other ac- Bromley et al. (2001) studied its polarization from a sim-
creting black hole systems. The radiated flux from ac- plified accretion model. Broderick & Loeb (2006) mod-
cretion disks mostly originates in the innermost radii, eled the frequency dependence of its centroid position,
where relativistic effects are important for understand- and Reid et al. (2008) used ray tracing to compare hot
ing observations. Proper calculation of the bending of spot accretion models with the observed astrometric mo-
light requires integration along rays (Broderick 2006). In tion of its mean position as a function of wavelength. Fi-
general, propagation through the plasma will influence nally, although the spacetime surrounding neutron stars
the photon trajectories, leading to non-geodesic paths only asymptotically approaches the Kerr metric, using
(Broderick & Blandford 2003, 2004). However, these ef- its null geodesics for ray tracing has still found applica-
fects are mostly important at low frequencies, compa- tion in modeling spectra of neutron stars (Braje et al.
rable to the expected plasma and cyclotron frequency. 2000).
When plasma effects can be neglected, the rays corre- Despite all of this work, numerical integration of Kerr
spond to null geodesics, and these circumstances are as- null geodesics is computationally expensive in certain ap-
sumed throughout this paper. plications. Rauch & Blandford (1994) (hereafter RB94)
The first applications of general relativistic radiative described a method for calculating null geodesics in the
transfer to accreting systems were of two main types. Kerr metric semi-analytically using the Hamilton-Jacobi
Cunningham (1975) packaged all radiative effects for op- formulation of the equations of motion and used it to
tically thick, geometrically thin disks as a transfer func- study the primary caustic. Bozza (2008) used a simi-
tion to go from local emissivity to that observed at in- lar method to investigate caustics of all orders, build-
finity. Luminet (1979) used the simple relationships be- ing on earlier analytic work (Bozza 2002). Fanton et al.
tween impact parameters at infinity and constants of the (1997) used a fast analytic version for creating line pro-
motion to shoot rays backwards in time from an ob- files and accretion disk images, and Agol (1997) applied
server’s photographic plate to the object under study. this method to the case of polarization from thin disk
More recently, Viergutz (1993) and Beckwith & Done accretion. Falcke et al. (2000) went on to use this code
(2005) considered the so-called emitter-observer prob- along with a simple model for the Galactic center black
lem. That is, given locations of the emitter and the hole to create images of its accretion flow.
observer, determine the constants of the motion for null All of this work used Legendre’s formulation of elliptic
geodesics connecting the two. This approach is much integrals (e.g., Abramowitz & Stegun 1965), and treated
more efficient when the source is highly localized, such as the φ and t coordinates numerically, if at all. The ta-
an orbiting star or hotspot. Here, backwards ray shoot- bles given in Carlson (1988, 1989, 1991, 1992) greatly
ing is impractical since most of the rays miss the target. simplify the reductions of the equations of motion to el-
Such techniques have been applied to the study of emis- liptic integrals. The primary aim of this paper is to use
sion lines and spectra from active galactic nuclei (AGN) Carlson’s integrals to calculate all geodesic coordinates
accretion disks and tori (Cadez et al. 1998; Wu & Wang semi-analytically for the first time.
2007) as well as their quasi-periodic oscillations (QPOs) Section 2 gives the geodesic equations in Kerr space-
(Schnittman et al. 2006). Li et al. (2005) used a ray time. Sections 3 and 4 present the reductions to elliptic
tracing approach to study the spectra of X-ray binaries. integrals and the specifics of our implementation. Sec-
tion 5 outlines a variety of checks performed to ensure its
Electronic address: [email protected] validity and accuracy, and discusses the speed improve-
2 Dexter and Agol

ment that should be expected from using an analytic The signs of the integrals in r and θ are independent
code. Section 6 provides an overview of our code for and arbitrary, but are fixed for a given geodesic. It may
readers who are not interested in all of its detail, and the seem odd that these equations lend themselves to the
code is applied to toy problems and test cases in Sec- choice of r or θ as independent variable to determine the
tion 7. Finally, Section 8 discusses future work both in cyclic coordinates t and φ. However, this is the natural
extending the code and in applying it to more realistic outcome of the separation of the Hamilton-Jacobi equa-
astrophysical situations. tion.

2. GEODESIC EQUATIONS OF MOTION 3. REDUCTION TO CARLSON INTEGRALS

In Boyer-Lindquist coordinates (t,r,θ,φ), the Kerr line In reducing the equations of motion from the previ-
element can be written, ous section, we follow closely the treatment given in Ap-
pendix A of RB94. First change variables to (t, u, µ,
2 φ) with µ = cos θ, u = 1/r. This set is more useful
∆ 2 Σ2

2ar
ds2 = −ρ2 dt + 2 dφ − 2 dt sin2 θ + computationally, since the location of an observer at in-
Σ2 ρ Σ finity is mapped to u = 0. The domain of u is then
ρ2 2 0 ≤ u ≤ u+ ≤ 1, where u+ is the location of the event
dr + ρ2 dθ2 , (1) horizon. Similarly, −1 ≤ µ ≤ 1. Then the definitions
∆
q 2 ≡ Q/E 2 , l ≡ Lz /E, and γ ≡ E/m put the equations
with the definitions of motion in dimensionless form. The integral equation
relating u and µ is
∆ = r2 − 2r + a2 , ρ2 = r2 + a2 cos2 θ, (2)
dµ du
Z Z
Σ2 = (r2 + a2 )2 − a2 ∆ sin2 θ, (3) sµ p = su p ; (11)
M (µ) U (u)
where a is the angular momentum of the black hole and
we use units with G = c = M = 1. where
Carter (1968) demonstrated the separability of the
Hamilton-Jacobi equation for geodesics, M = q 2 + (ã2 − q 2 − l2 )µ2 − ã2 µ4 (12)
−2 −2 2 2 2 2
∂S ∂S ∂S U = (1 − γ ) + 2γ u + (ã − q − l )u +
−2 = g µν µ ν , (4)
∂λ ∂x ∂x 2[(a − l)2 + q 2 ]u3 − a2 q 2 u4 , (13)
where S is Hamilton’s principal function (the classical and ã2 = (1 − γ −2 )a2 . This paper only considers null
action) and λ is an affine parameter. The separation geodesics, so that γ −2 = 0 throughout. The arbitrary
reduces the equations of motion to quadratures (Chan- signs have been written explicitly, and are chosen to be
drasekhar 1983) relating the coordinates r and θ: sx = sign(ẋ), where a dot refers to a derivative with
Z r Z θ respect to affine parameter. This is done so that both
dr dθ sides of (11) are always positive. The equations for the
√ = √ , (5)
R Θ other coordinates become
where
dµ
Z
t − t0 = s µ a 2 µ2 √ +
R = [(r2 + a2 )E − aLz ]2 − (6) M
2a(a − l)u3 + a2 u2 + 1 du
Z
2 2
∆ [Q + (Lz − aE) + δ1 r ] su √ (14)
u2 (u/u+ − 1)(u/u− − 1) U
Θ = Q− [ a2 (δ1 − E 2 ) + L2z csc2 θ] cos2 θ; (7)
lµ2 dµ
Z
and the constants of the motion are the angular momen- φ − φ0 = sµ √ +
1 − µ2 M
tum about the black hole spin axis, Lz , the energy, E,
2(a − l)u + l du
Z
and Carter’s constant Q. δ1 = 0 (1) for null (timelike) su √ , (15)
geodesics. (u/u+ − 1)(u/u− − 1) U
The equations of motion for the cyclic coordinates are √
where u± = [1 ± 1 − a2 ]−1 . The limits of integration
r have been omitted due to complications in accounting for
dr
Z
t = λE + 2 r[r2 E − a(Lz − aE)] √ (8) turning points. This is discussed in more detail below.
∆ R Given initial and final values of u and µ, we can com-
Z r
dr pute t and φ. Since the µ integral is easier to invert
φ = a [(r2 + a2 )E − aLz ] √ + and this method is of more general utility, u is taken
∆ R as the independent variable and the goal is to solve for
Z θ
dθ µf given µ0 , u0 and uf . In certain applications it is
(Lz csc2 θ − aE) √ , (9) more convenient to choose µ as the independent vari-
Θ
able. For example, in the case of thin disk accretion we
with know the inclination angle as well as the value of µ where
r θ
r2 cos2 θ
Z Z
the geodesic intersects the disk. Section 4 gives solutions
λ= √ dr + a2 √ dθ. (10)
R Θ for uf given µ0 , µf and u0 to handle these cases.
 Analytic Kerr Photon Orbits 3

TABLE 1
Reduction of Iu

Case Parameter Range Arguments (ai , bi );(fj , gj , hj ) u ǫ [, ] RB94

b
1 Cubic (3 real) a = 0, q 2 + l2 ≥ 27 or u ≤ u2 (−u1 , 1), (u2 , −1), (u3 , −1) [0, u2 ] 1,3,8,10
a 6= 0, q 2 = 0, |l| 6= |a,
c
2 Cubic (3 real) a = 0, q 2 + l2 ≥ 27 or u ≥ u3 (−u1 , 1), (−u2 , 1), (−u3 , 1) [u3 , u+ ) 2,4,9,11
a 6= 0, q 2 = 0, |l| 6= |a,
a = 0, q 2 + l2 < 27 or a 6= 0, q 2 = 0, l 6= a (−u1 , 1);(2u1 [(a − l)2 + q 2 ]
−1
3 Cubic (1 real) , f /u1 , 1) [0, u+ ) 5,7,12
4 No roots q 2 = 0, l = a ... [0, u+ ) 6
d
5 Quartic (2 real) a 6= 0, q 2 > 0 (−u1 , 1),q 2 2 −1 −1
“ s (u4 , −1) ;([−a aq”u1 u4 ] , [u1 + u4 ]f, 1)
−1 [0, u+ ) 13,19
d
6 Quartic (0 real) a 6= 0, q 2 < 0 e −1/2 , e(h −h ) , h1 , (e −1/2 −1
, −g1 , h1 ) [0, u+ ) 14
2 1
b
7 Quartic (4 real) a 6= 0, q 2 6= 0, u ≤ u2 (−u1 , 1), (u2 , −1), (u3 , −1),(u4 , −1) [0, u2 ] 15,17
c
8 Quartic (4 real) a 6= 0, q 2 =6 0, u ≥ u3 (−u1 , 1), (−u2 , 1), (−u3 , 1),(u4 , −1) [u3 , u+ ) 16,18

ah is found from solving Eq. (21) and selecting one of the two real roots. d, e are defined in Eq. (22).
1
b When u = u , the domain of u is [0, u ).
2 3 2
c When u = u , the domain of u is (u , u ).
2 3 3 +
d q = sign(q 2 ).
s

3.1. Reduction of Iu
Z x
Call the left-hand side (LHS) and right-hand side [p] = (f + gt + ht2 )p2 /2
Y
(ai + bi t)pi /2 dt (18)
(RHS) of (11) Iµ and Iu respectively, and start with the y i=1,4,5
reduction of Iu :
for one pair of complex roots or
du
Z
Iu = su p . (16) Z 2
xY
U (u)
[p] = (fi + gi t + hi t2 )pi /2 (a5 + b5 t)p5 /2 dt (19)
y i=1
2
Except in the special case with a = l, q = 0, U (u) is
either a quartic or cubic and its roots are denoted ui with for two. In using this form, it is assumed that each power
i = 1 − 3, 4, and ordered increasingly. If real, u1 < 0 and pi of an irreducible quadratic is written twice in the vec-
in the quartic case, u4 > 1 or u4 < 0. They are of no tor [p]. In other words, when one pair of roots is com-
physical significance. When all roots are real, the allowed plex, p2 = p3 . When all roots are complex, p2 = p3 and
regions for the integrand are u > u3 and u < u2 so that U p1 = p4 .
is positive. Thus the roots are the turning points for null To ensure that x > y in cases where a turning point
geodesics starting outside u2 and inside u3 respectively may be present, integrals are written in pieces involving
in both the cubic and quartic cases. There can be no the relevant turning point, u∗ , and the number of turning
more than one turning point, since the allowed region points along the portion of the geodesic being followed,
is bounded on one side either by infinity or the event Nu (either 0 or 1):
horizon. When one or both pairs of roots are complex, !
there is no turning point in u. u∗ u∗
du du
Z Z
Upon encountering a turning point, the sign of u is Iu = su √ − (−1)Nu √ . (20)
reversed, so that the total integral is the sum of the inte- u0 U uf U
gral from u0 to the turning point and that from uf to the The Carlson papers reduce all elliptic forms to a set
turning point. The idea is to ensure that the integrals of four fundamental integrals, known as the R-functions
in u and µ monotonically increase along a geodesic. In (Press et al. 1992), which replace Legendre’s integrals of
a sense this allows the independent variable to take the the first, second and third kind. They are all integrals
place of the affine parameter, which cannot be used since from 0 to ∞ and hence don’t require a limit of integration
it is a function of u and µ. to be a turning point, greatly simplifying complex root
Carlson (1988, 1989) contain formulas for evaluation cases where no physical turning point is present. This is
of integrals of the form one of many advantages of Carlson’s approach. As is the
case for Legendre’s formulation, any elliptic integral can
Z 5
xY be reduced to a sum of Carlson’s R-functions. Where
[p] = (ai + bi t)pi /2 dt; (17) Legendre integrals are used in this paper, they are cal-
y i=1 culated in terms of the R-functions using the formulas
in Press et al. (1992). The integrals encountered in this
with all quantities real, x > y, and ai + bi t > 0 for paper are always of the form p = [−1, −1, −1, −1, p5] for
y < t < x. The form of a given integral is described quartic cases and p = [−1, −1, −1, p5] for cubic cases.
by the vector [p], which contains the powers, pi , of the Thus the form of coordinate integrals in the following
factored roots. Cases with one or two pairs of complex will be specified by p5 alone.
roots are handled in Carlson (1991, 1992), where they To maintain as much generality as possible, all inte-
are written in terms of real quantities as grals are written as above in terms of their roots. In cubic
4 Dexter and Agol
p p
cases the roots are found from solving the cubic equation, ξ= f + gx + hx2 , η = f + gy + hy 2 , (24)
while for quartic cases they are found numerically using q
the routine zroots.f from Press et al. (1992). Finally, cij = 2f bi bj − g(ai bj + aj bi ) + 2hai aj , (25)
instead of writing out the explicit formulas from Carl- (X1 Y4 + Y1 X4 ) p
son’s papers and going through the algebra separately in M= (ξ + η)2 − h(x − y)2 , (26)
each case, we have written routines for each case. This is x−y
much simpler and of more general utility, since numerous L2± = M 2 + c214 ± c11 c44 . (27)
integrals must be done to calculate the coordinates of a
Then,
point along a geodesic.
The integral Iu has p5 = 0 and is given by Carlson 4
(1989) Eq. (2.12) for real roots for cubic cases. Quar- Iu = p RF (M 2 , L2− , L2+ ). (28)
tic cases are found in Carlson (1988) Eq. (2.13) for real |e|
roots and Carlson (1992) Eq. (2.36) for all complex roots. RF is computed using the routine from Press et al.
The quartic and cubic cases with a single pair of complex (1992). Equations for Carlson elliptic integrals with
roots are given by Carlson (1989) Eq. (3.8). The nec- p5 6= 0 can similarly be found in the Carlson papers
essary arguments to the Carlson routines are listed by listed above.
case in Table 1, along with case definitions, appropriate
domains of u, and the corresponding cases in Appendix 3.2. Inversion of Iµ
A of RB94. Next, the Iµ integral needs to be inverted to solve for
As can be seen from Table 1, writing formulas in terms µf . As with U (u), the roots of the biquadratic M (µ),
of the roots of U has the advantage of unifying many M± , determine the physical turning points in µ. When
disparate cases from previous work. Equal roots cases, M− > 0, there are four real roots and the orbit cannot
which describe orbits approaching the unstable circular cross the equatorial plane. The physical turning points
photon orbits, cannot strictly speaking be treated iden- correspond to the two roots with
tically to other real roots cases as shown in the table. p the same sign as µ0
and are denoted µ± = sign(µ0 ) M± . When M− < 0,
Here, integration to the turning point diverges. The code p
flags for these cases and integrates them directly from the physical turning points are µ± = ± M+ and are
u0 to uf , and the arguments listed in the table are still symmetric about the equatorial plane. We can calculate
valid. In practice, however, except for the well known the number of times the geodesic has crossed a µ turning
Schwarzschild unstable circular orbits with q 2 + l2 = 27, point from the magnitude of the Iu integral.
Rµ This
R µ is done
equal roots cases are almost impossible to trigger. This by noting that the maximum value of µf+ is µ−+ and its
is because the Carlson routines as written maintain ac- minimum
√ value is zero. In this derivation the integrand
curacy until |u2 − u3 | . 10−12 , which is usually more dµ/ M , common to all integrals, is omitted. Then, for
precise than the determination of the imaginary parts of sµ = 1,
the roots.
For one pair of complex roots, the arguments f , g Z µ+ Z µ+ Z µ+ Z µ+
and h are found by setting U (u) = qs (u4 − u)(u − +(N − 1) ≤ Iu ≤ +N , (29)
u1 )(f + gu + hu2 ), where qs = sign(q 2 ), and match- µ0 µ− µ0 µ−
ing powers of u. When all roots are complex, setting where N is the number of turning points reached in µ,
U (u) = (f1 +g1 u+h1 u2 )(f2 +g2 u+h2 u2 ) yields five non- and µ± are the upper and lower turning points in µ.
linear equations for our six unknown coefficients. The The integrals are written in these pieces so that they
degree of freedom is used to simplify the equations, and are always positive, as required for use with Carlson’s
a sixth degree polynomial is solved numerically for h1 : integrals. This condition can be written more concisely
" as
 2 #
c √ c d c Rµ '
h61 − √ h51 −h41 + h31 −h21 − √ h1 +1 = 0,
&
e 2 − Iu − µ0+
e e e e N= R µ+ , (30)
(21) µ−
where where ⌈ ⌉ is the ceiling function. If sµ = −1, then the
first turning point reached is µ− . The condition can then
c = a2 − l2 − q 2 , d = 2[(a − l)2 + q 2 ], e = −a2 q 2 . (22) be written
The only pair of real solutions to this equation corre- "Z # "Z #
spond to the values of h1 , h2 . µ− Z µ− µ− Z µ−
As a full example of one of these reductions, consider − +(N − 1) ≤ Iu ≤ − +N .
case 5 from Table 1 with u0 < uf (su = 1). This is µ0 µ+ µ0 µ+
the Kerr case with no physical turning points. From 18, Rµ Rµ Rµ (31)
we see that b1 = 1, b4 = −qs , a1 = −u1 , a4 = qs u4 , Using µ0− = µ0+ − µ−+ , we can rewrite this in terms of
x = uf , y = u0 . The sign qs is used to keep each factor the same integrals used above:
positive. Matching the powers of U (u) as described above
gives f = −qs /(u1 u4 e), g = (u4 + u1 )/(u1 u4 )f , h = 1. Z µ+ Z µ+ Z µ+ Z µ−
Following Carlson (1991), we define − +N ≤ Iu ≤ − +(N + 1) . (32)
µ0 µ− µ0 µ+
p p
Xi = ai + bi x, Yi = ai + bi y, (23) Finally,
 Analytic Kerr Photon Orbits 5

$ Rµ % 3.2.2. M− < 0
Iu + µ0+
N= R µ+ , (33) When M− < 0, y < 0 in (38) so that (39) is no longer
µ−
valid. Since the integrand is an even function of µ, we
can write
and ⌊ ⌋ is the floor function. To write out the general so- Z µ+
lution for Iu = Iµ for arbitrary number of turning points 1 dµ
and sµ , we include coefficients for the various pieces of I= q , (41)
|a| −µf (µ2 − µ2 )(µ2 − M )
the Iµ integral: + −

Z µ+ Z µf Z µ+ which is in the correct form, except that −µf can be

Iu = α1 +α2 +α3 . (34) negative. This causes no problems. In this case,
µ0 µ− µ−

The coefficients are functions of sµ and N determined

q µ2+
µf = µ− cn(J, k), J = µ2+ − M− |a|I, k 2 = ,
by writing down specific cases. For example, α1 reflects − M− µ2+
whether the integration is positive or negative from µ0 (42)
to µf and is easily seen to be α1 = sµ . Similarly, α2 and we’ve used µ− = −µ+ for M− < 0. The µ terms in I
reflects whether the last turning point reached is µ− or are computed the same as in (40), with k defined in (39),
µ+ . Thus the coefficient is α2 = sµ (−1)N . The third
r
µ20 p
coefficient is slightly more complicated and turns out to x = 1 − µ2 and A = |a| M+ − M− . The integral
+
be between the turning points here is twice the complete

2N + 3 − sµ
 elliptic integral K(k).
α3 = 2 − 1. (35)
4 3.2.3. q 2 = 0
Armed with the number of turning points and the co- A special case is encountered when q 2 = 0. M (µ) has
efficients, we solve for µf by inverting the second integral a double root at µ = 0, causing Iµ to diverge there,
on the RHS of (34): and preventing these orbits from reaching the equatorial
plane. Hence, they have at most one physical turning
µf µ+ µ+
! point. In this case Iµ is elementary, and the solution for
dµ 1
Z Z Z
√ = Iu − α1 −α3 . (36) µf is
µ− M α2 µ0 µ−
µf = µ+ sech |aµ+ |Iu − sµ s1 sech−1 (µ0 /µ+ ) ,
 
(43)
Calling the RHS I and writing out the square root on
the LHS for the general case (a 6= 0, q 2 6= 0) gives where s1 = sign(µ0 ).

1
Z µf
dµ 3.2.4. a = 0
I= . (37) Finally, when a = 0 (the Schwarzschild case) the µf
|a| µ−
p
(M+ − µ2 )(µ2 − M− )
integral is again elementary. The solution for µf is then,
Carlson (2005) contains a table for inverting integrals
of the form
" r   !#
1 d −1 µ0
Z x µf = µ− cos Iu − α1 cos − α3 π .
dt α2 2 µ+
I= , (38)
(44)
p
y (a1 + b1 t2 )(a2 + b2 t2 )
where all quantities are real, x > y, 0 ≤ y < x and either 3.3. t and φ coordinate integrals
y = 0, x = ∞ or one limit is a root of the integrand. The Given the solution for µf , equations for the coordinates
latter case applies here. t and φ can be reduced to elliptic integrals as well. Each
coordinate is expressed as a sum of integrals over u and
3.2.1. M− > 0 µ. As is done above, the u terms are reduced to Carlson’s
When M− > 0, all requirements are met as written, formulation, and the µ terms to Legendre’s.
and The µ integral term in (14), which we’ll denote Tµ , can
be written as a single Legendre integral of the 2nd kind.
µ2− For example, the µ0 term in the M− < 0 case is reduced
µf = µ− nd (J, k), J = µ+ |a|I, k 2 = 1 − , (39) as follows: Z
µ2+ µ+
µ2 dµ
Tµ = |a|
where nd (J, k) = 1/dn(J, k) and dn is a Jacobi-Elliptic
q
µ0 (µ2+ − µ2 )(µ2 − M− )
function. The µ integral terms in I are calculated as
Z x
Z µf 1 − t2
dµ 1 = |a|µ+ dt q
p = F (x, k) (40) 0 (1 − t2 )(1 − t2 − M−
)
µ0 M (µ) A µ2 +

2 M−
where F (x, k) is Legendre’s integral
q of the first kind Z x 1−t − µ2+
M+ −µ20 =A dt q + a2 M− Iu
(Abramowitz & Stegun 1965), x = M+ −M− , A = |a|µ+ 0 (1 − t2 )(1 − t2 − M−
)
µ2+
and k is the same as above. The integral between turning
points is the complete elliptic integral K(k). = AE(x, k) + a2 M− Iu , (45)
6 Dexter and Agol

where E(x, k) is the Legendre integral of the second kind The first term is from Tu and the second term is Tµ .
with arguments x andpk defined in the previous section. Component integrals are calculated the same way as
The substitution t = 1 − µ0 /µ+ is made between lines Iu or Iµ respectively. That is, µ component integrals are
one and two, and M− /µ2+ is added and subtracted from calculated in pieces using the appropriate coefficients as
the numerator between lines two and three. In the M− > described above while u component integrals are calcu-
0 case, Tµ is given by the first term of the above formula, lated with reference to the physical turning point, if one
with the arguments A, k, x for that case given with the exists. These are all the integrals required to compute
solution for µf in Subsection 3.2.1. null geodesics in Kerr spacetime. These equations for the
The µ term in the φ component formula (15) can be re- φ, t coordinates are written in Boyer-Lindquist coordi-
duced to a Legendre integral of the 3rd kind in analogous nates. For certain applications, Kerr-Schild coordinates
fashion. For the M− < 0 case we proceed as follows: are used instead. We note here for completeness the ana-
lytic transformations between our Boyer-Lindquist coor-
µ+ dinates and these Kerr-Schild coordinates (t̃,ũ,µ̃,φ̃) (Font
l 1 dµ
Z
Φµ = −lIu + et al. 1999),
2
p
|a| µ0 1−µ (µ+ − µ2 )(µ2 − M− )
2

= −lIu + √ !
Z x 1 − u[1 + 1 − a2 ]
l 1 dt t̃ = t + log ∆ − ur log √ , (50)
1 − u[1 − 1 − a2 ]
|a|µ+ 0 1 − µ2+ + µ+ 2 t2 (1 − t2 )((1 −
q
M−
µ2+
) − t2 ) √ !
1 − u[1 + 1 − a2 ]
l φ̃ = φ − a ur log √ , (51)
= −lIu + Π(n; x, k) (46) 1 − u[1 − 1 − a2 ]
A(1 − M+ )
ũ = u, (52)
where Π(n; x, k) is the Legendre integral of the 3rd kind µ̃ = µ. (53)
µ2+
and n = 1−µ2+
. The formula for the M− > 0 case is the
The transformations are valid outside the event horizon,
M+ −M−
same, with n = 1−M+ and the other arguments defined where ∆ and the numerator of the other log terms are
in Subsection 3.2.1 above. Note that we are using the positive.
sign convention for n from Press et al. (1992), which is
opposite that in Abramowitz & Stegun (1965). 4. SOLUTION FOR UF
Tu , the u integral term in (14), is expanded with partial For some applications, it is preferable to use µ as the
fractions, and after a little algebra is written independent variable and solve for uf given u0 . In par-
ticular, consider geodesics connecting an observer at in-
finity with a thin, equatorial accretion disk. The initial
a2
 Z
1 1 du
Tu = su ur 2a(a − l) + + 3 √ polar angle is the inclination of the observer. The final
u+ u+ (u/u+ − 1) U polar angle is π/2 (µf = 0), and we solve for the ra-
2 dial coordinate where the ray intersects the disk. This
 Z
a 1 1 du
− 2a(a − l) + + 3 √ method, however, is of less general utility than that de-
u− u− (u/u− − 1) U
 Z  scribed above. Even in simple geometries, the number
1 1 du 1 du
Z
√ √ of turning points in µ along a geodesic is not known in
+ − + , (47)
u2− u2+ u U ur u2 U advance as it must be to use µ as the independent vari-
√ able. One way around this is to calculate all geodesics
where ur ≡ uu++−uu−
−
= −(2 1 − a2 )−1 is negative. Three connecting the observer with the disk for a fixed num-
of the terms have p5 = −2 and one has p5 = −4. When ber of µ turning points (Cunningham & Bardeen 1973;
a limit of integration is at infinity (u = 0), this integral Viergutz 1993).
blows up, as it should. In practice, the code picks a non- The approach in solving for uf is the same as in solv-
infinite starting radius large enough that the geodesic ing for µf . The integral Iµ is computed as a Legendre
trajectories from infinity to the starting radius differ neg- integral of the first kind. Given the number of turning
ligibly from their flat space counterparts. points, Iµ is computed in pieces as shown above using
Then, the coefficients α1,2,3 .
After finding Iµ , we invert Iu . This inversion
 Z ranges from relatively straightforward to algebraically
l 1 du formidable. As examples, we discuss cubic and quar-
Φu = su u r + 2(a − l) √ −
u+ (u/u+ − 1) U tic real roots cases in detail. Table 2 gives the solution

l
Z
1 du
 for uf in all cases. This problem was first addressed by
+ 2(a − l) √ ,(48) Agol (1997) and the solutions here are from its Table 5.2
u− (u/u− − 1) U
with some modification.
where both integrals are already calculated as part of Tu . For our first example, consider the first two cases of
Finally, the dimensionless affine parameter can also be Table 1 where there are three real roots. The integral to
calculated along the path from (10) without any addi- invert is
tional integrals:
!
u+ u+
du du
Z 2 Z Z
du µ dµ
Z
λ′ = su √ + a2 sµ √ . (49) Iµ = su p ± p , (54)
u2 U M u0 U (u) uf U (u)
 Analytic Kerr Photon Orbits 7

TABLE 2
Solution for uf

a c
uf J m1 c1 c2 c3
√
u32 u31 d
1 u1 + u21 cd2 J c1 [Iµ − Iu (u0 , u2 )] u31
... ...
√ 2
u32 u31 d
2 u1 + u31 dc2 J c1 [Iµ + Iu (u3 , u0 )] u31 2
... ...
c2 +u1 −(c2 −u1 )cnJ √
1
+ 6u8c
1 +c3 a+l
p
3 1+cnJ
c1 [Iµ + Iu (u1 , u0 )] 2
2dc2 u1 (3u1 + c3 ) a−l
2
4 ... ... ... ... ... ...
b
u4 c5 +qs u1 c4 −(qs u4 c5 −u1 c4 )cnJ )2 −(u4 −u1 )2 √
5 (c4 −qs c5 )cnJ +qs c4 +c5
c1 [Iµ + Iu (ub , u0 )] qs (c4 +qs c54c ec4 c5 ... ...
4 c5 r
”2 √
n(1+c2 4n2 −(c4 −c5 )2 d
“
2 )scJ c4 −c5 e
6 c3 + 1−c2 scJ
su c1 [Iµ + Iu (c3 , u0 )] c4 +c5 2
(c4 + c5 ) (c4 +c5)2 −4n2
m + c2 n
√
u2 −c2 u3 sn2 J u41 u32 ec3 u21
7 1−c2 sn2 J
c1 [Iµ − Iu (u0 , u2 )] u42 u31
u42 u31
√2 u31
u3 −c2 u2 sn2 J u41 u32 ec3 u43
8 1−c2 sn2 J
c1 [Iµ + Iu (u3 , u0 )] u42 u31 2 u42
u42 u31

a snJ = sn(J, m ), cnJ = cn(J, m ), scJ = sn(J, m )/cn(J, m ) and m = 1 − k2 is used instead of k. u
1 1 1 1 1 xy ≡ ux − uy .
b q = sign(q 2 ). If q = 1 then u = u , u = u . Otherwise, u = u , u = u .
s s a 4 b 1 a 1 b 4
c I (y, x) = s R x √du .
u u y
U (u)
d Complex roots are written as m ± in, p ± ir and are ordered so that m > p and n > 0.

TABLE 3
Auxillary Constants used in table 2

c4 c5
p p
5 2 2 2 2
p (m − u4 ) + n p (m − u1 ) + n
6 (m − p)2 + (n + r)2 (m − p)2 + (n − r)2

where u+ is the relevant turning point: u2 or u3 . Denote

the first term Iu+ , and write the second term in terms of √ u2 − u1
the roots of the integrand: J ≡ u3 − u1 I, , k2 =
(60)
u3 − u1
dn(J, k) cn(J, k)
su
Z u+
du dc(J, k) ≡ , cd(J, k) ≡ ,
Iµ − Iu+ = ± √ p , cn(J, k) dn(J, k)
d uf (u − u 1 )(u − u2 )(u − u3 ) where cn, dn are Jacobi-Elliptic functions. Note that the
(55) result does not depend on whether or not a turning point
where d = 2[(a − l)2 + q 2 ]. This
√ can be put in the form has been reached, since both cn and dn are even in J.
(38) with the substitution z = u − u1 : When U (u) has four real roots and u ≤ u2 ,
√ u2
su du
Z
u+ −u1
dz
Z
±I = √ , Iµ − Iu+ = ±√ p ,
p
[z 2 + (u1 − u2 )][z 2 + (u1 − u3 )] e uf (u − u1 )(u − u2 )(u − u3 )(u4 − u)
uf −u1
(56) q (61)
u2 −u
where where e = |aq|. With the substitution z = u3 −u , this
√ becomes
d

I≡ Iµ − Iu+ , (57)
2 r
u2 −uf
dz
Z u3 −uf
and Iu+ is determined from the same Carlson formulas ±I = p ,
as for Iu above. 0 [z + (u1 − u2 )][z 2 + (u1 − u3 )]
2

Comparing (56) with (38), we see that a1 = u1 − u2 , (62)

a2 = u1 − u3 and b1 = b2 = 1. If u+ = u3 , then the where
limits of integration must be switched, since by definition √
x > y. These integrals correspond to the third row, third e

I≡ Iµ − Iu+ . (63)
and fourth columns of Table 1 from Carlson (2005), and 2
the solutions for uf are Again comparing with (38) and using Carlson (2005),
we find
uf = u1 + (u2 − u1 ) cd2 (J, k), u0 ≤ u2 (58) u3 (u2 − u1 )sn2 − u2 (u3 − u1 )
2
= u1 + (u3 − u1 ) dc (J, k), u0 ≥ u3 , (59) uf = , (64)
(u2 − u1 )sn2 − (u3 − u1 )
with where
8 Dexter and Agol

than that one by a factor of about 5, due to the fact that

our code computes the minimum number of R-functions
p
sn = sn(J, k), J = (u4 − u2 )(u3 − u1 )I (65)
possible, and shares them between routines when neces-
(u4 − u3 )(u2 − u1 ) sary. The code from Agol (1997) was found to be ∼ 100
k2 = . (66)
(u4 − u2 )(u3 − u1 ) times faster than numerical integration in the case of
Again the result is independent of whether or not a tracing geodesics from infinity to a thin disk. This is an
turning point is present. In (54) above, the sign of the optimal problem for an analytic code, since we can solve
second term on the RHS depends on whether a turning for the point where µ = 0, whereas a numerical code
point is present. This allows us to determine the number must integrate the geodesic from infinity until it reaches
of turning points in u. that point, and then zoom in on the intersection to find
When complex roots are present, the reduction to an approximate solution to the desired accuracy. In addi-
standard form (38) is much more difficult. It is dis- tion, this example didn’t include the t and φ coordinates,
cussed in Erdélyi et al. (1981), and relevant formulas for which are sped up by a much smaller factor than u and
the inversion can be found there and in Byrd & Fried- µ.
man (1971). In particular, our cases 3,5 are from Byrd As a lower bound for the speed improvement of our
& Friedman (1971) equations 239.00 (p86) and 259.00, code over numerical integration, a routine was written
260.00 (p133,135). Our formula for case 6 is based on to integrate the photon four momentum for all coor-
Erdélyi et al. (1981) Table 2, p310-311. The intricacy of dinates with respect to affine parameter using the im-
these reductions demonstrates the advantage of Carlson’s plementation of the Bulirsch-Stoer method from Press
method. The computation of integrals is equally efficient et al. (1992). We then compared the integration of many
with complex or real roots. Unfortunately, when inver- points along a single geodesic starting from infinity with
sion is required, Carlson (2005) is only a somewhat more this numerical code and our analytic one. This is the
compact version of Legendre’s original notation and of- ideal case for numerical integration, since the intermedi-
fers no real advantage over previous work. ate points calculated along the ray are no longer wasted
as in the first example. For the case considered with no
5. CODE CHECKS AND SPEED TESTS turning points in u or µ, the analytic code was found to
Using the solution for µf , the equality of Iµ and Iu be faster by a factor of ∼ 3.
has been checked to machine accuracy (at least 14 sig- However, our numerical code for integrating geodesics
nificant digits in all cases). Once found, µf can be used is much simpler than a complete code would have to be.
as an input to recover uf . In this way, the two routines It cannot handle turning points, and requires knowledge
have been shown to agree in all cases. Precision in the of the affine parameter on the ray in order to know where
calculation is limited by error in the determination of the region of interest in the integration is. In practice,
the roots of U (u). An advantage of using Carlson’s for- turning points would have to be detected and a scheme
mulation is that all component integrals are computed for determining the region of interest in affine parameter
without reference to the complex roots, which often have implemented. Alternatively, a somewhat more compli-
less numerical precision than real ones. Formulas for uf , cated scheme such as the Hamiltonian method described
discussed in Section 4, are also written in terms of real in Schnittman & Bertschinger (2004a) could be adopted.
quantities leading to higher accuracy. In any case, these additions would slow down geodesic
Certain special cases can be integrated analytically for computation. We conservatively estimate, then, that the
all components, providing independent checks on com- lower bound for the speed advantage of our analytic code
ponent integral formulas. These include µ cases with over numerical integration is a factor of ∼ 5. A similar
q 2 = 0, u cases with q 2 = 0, l = a and u cases with test for the case of integrating down to the thin disk
equal physical real roots, corresponding to unstable cir- yielded a speed difference of a factor of ∼ 300, leading to
cular photon orbits. In all of these cases, the component an upper bound on the speed advantage of ∼ 500, which
integral formulas above agree with the analytic results is in good agreement with the naive estimate of multi-
to machine accuracy. The component integral formu- plying the speedup found by Agol (1997) by the speedup
las above also reduce to those derived separately for the factor between our code and the one presented there.
Schwarzschild case, a = 0. All of the formulas given Then the range of speedup that can be expected by us-
here for µf agree with those in tables 1-2 of RB94. The ing the analytic code described here is a factor between
Schwarzschild formulas were also tested against the ap- 5 − 500 depending mostly on the application, but also on
proximate formulas given by Beloborodov (2002). the specific implementation of the numerical integration
Further, the implementations of Carlson’s integral ta- code.
bles have been checked extensively using the Mathemat- In addition to being faster, the analytic formulation is
ica NIntegrate function. The same is true of the t and much more flexible. It can calculate an arbitrary num-
φ formulas, as well as the individual integral components ber of points beginning and ending anywhere on any
Iu , Tu , and Iµ , Tµ , Φµ . geodesic, provided that the constants of the motion can
The R-function routines maintain accuracy until a . be calculated. This is exploited in the thin disk toy mod-
10−5 or q 2 . 10−10 . If such parameters are encountered, els below, where we solve for the point µf = 0. It could
the code will give a warning and set the offending value also allow, for example, a calculation of Compton scat-
to zero. tering by tracing rays out from every point on a geodesic,
The geodesic computations have been checked against and computing the scattered intensity into that point as
calculations done by the code used in Falcke et al. (2000) a separate ray tracing computation. In any event, the
and are found to be in excellent agreement. The FOR- flexibility inherent to an analytic method could allow
TRAN implementation of our code is found to be faster for more sophisticated calculations in the future, which
 Analytic Kerr Photon Orbits 9

Constants of the motion are required and may be speci-

fied either as the impact parameters at infinity (α, β), as
is most convenient in ray tracing applications, or as the
dimensionless angular momentum, l, and Carter’s con-
stant, q 2 . When any other information is not provided,
the program assumes geodesics which trace out the entire
domain of u, from the starting point and back or until
the event horizon is reached.
The program calls the main subroutine, geokerr,
which calls geomu to fill in missing inputs and calcu-
late µf . Alternatively, geokerr can solve for uf using
geor. Subsequently, geophitime calculates the φ and t
integrals using the Carlson routines. The program loops
over constants of the motion for a chosen initial polar an-
gle and black hole spin. Results are written to standard
output by default, and should be redirected from the ter-
minal to a text file in most cases. It is also possible to
Fig. 1.— Change in time vs. radial coordinate in the input the name of the desired output file. The subroutine
Schwarzschild metric for geodesics near the circular photon orbit geokerr encapsulates most of the code functionality, and
(dashed line), as described in Section 6. can fairly easily be adapted to another front end other
than the program used here. See the README file ac-
companying the code or online for more detail.
As an example use of the code, consider tracing rays
over a rectangular grid in −4 ≤ α ≤ 8, −6 ≤ β ≤ 6 for
a near extreme black hole, a = .998. The observer is at
infinity in the equatorial plane (µ0 = 0), and 20 rays will
be traced over each dimension. The input file for this
situation can be found online.1
Output is arranged as follows. The constants of the
motion are listed for each geodesic in the top line, fol-
lowed by columns giving uf , µf , ∆t, ∆φ, λ. The format
used for output can be changed with a tiny modification
to the source code. See the README file for more de-
tails. Plotting the affine parameter evaluated at either
the event horizon, or once the geodesic returns to its ini-
tial radius, as a function of impact parameters for this
data with 160, 000 geodesics produces Fig. 2 as explained
below.
Fig. 2.— Image of a near extreme (a = .998) Kerr black hole For less standard batch runs, it may be necessary to
viewed from the equatorial plane. Images intensities are taken to generate the input file from a simple program. Consider
be the affine parameter evaluated upon termination at the black a set of geodesics in the Schwarzschild metric (a = 0)
hole or after returning to the starting radius. Intensities are scaled
linearly from the minimum value outside the shadow to the maxi- to study the unstable circular photon orbits. The given
mum. parameters are chosen to be u0 = 1/30, uf = u+ √ = .5,
µ0 = .9, β = 0 and an array of values for α near 27.
wouldn’t be possible with a numerical code. The main The piece of code to write an appropriate input file
disadvantage of using an analytic code is that the affine is available online.2 Plotting the change in time as a
parameter cannot be used as an independent variable, function of final radial coordinate produces Fig. 1.
which may be desirable for adaptive integration tech-
niques in radiative transfer applications, for example. 7. APPLICATIONS/VALIDATION
6. IMPLEMENTATION We next describe a couple of relatively simple appli-
cations of the code to ray tracing problems as further
This section provides an overview of the various rou- validation and as examples of its utility. The first is the
tines used by the code described above, and examples of simplest illustration of the black hole shadow, which tests
their use. The README file online covers everything in the determination of the roots of U (u) and qualitatively
this section in greater detail. The FORTRAN 77 source parts of the time integral. Next are examples from the
file geokerr.f contains the main program as well as the standard model of thin disk accretion. The disk image
key routines, geokerr, geomu, geor and geophitime, and simple spectrum from line emission test the routine
and supporting functions. Inputs are given through com- that solves for uf . The projection of a uniform grid at
mand line prompt or a text file. Inputs from previ- infinity onto the equatorial plane of the black hole also
ous command line runs may be saved for future use. tests the calculation of φ, and hot spot emission pro-
These inputs include constants of motion for the desired vides a time-dependent test. Finally, spectra and images
geodesics, initial and final u and initial µ, the number
of turning points in u (ignored if the constants do not 1 http://www.astro.washington.edu/agol/geokerr/exfiles/abgrid.in
admit physical turning points), and the sign of u̇ and µ̇. 2 http://www.astro.washington.edu/agol/geokerr/exfiles/inputex.f
10 Dexter and Agol

Fig. 4.— Image of an optically thick standard relativistic accre-

tion disk around a near extremal black hole (a=.998). The disk
has outer radius rout = 18, and the observer’s inclination is 85◦ .

Fig. 3.— Projection of a uniform Cartesian grid in the image

plane to the equatorial plane of the black hole for µ0 = 1 (top)
and µ0 = .5 (bottom). Black hole spin is a = 0 (left) and a = .95
(right), and the area inside the horizon is removed from each image.
Compare to Fig. 2 of Schnittman & Bertschinger (2004b).

of synchrotron radiation from spherical accretion quan-

titatively test our radiative transfer routines.
Ray tracing utilizes the simple relationship between
points on an observer’s instrument and the constants
of motion of null geodesics. Consider the photographic
plate at infinity as a function of the impact parameters
α, β, perpendicular and parallel to the black hole spin
axis respectively. Images can be created by tracing rays
backwards from points on the plate to the black hole. Fig. 5.— Normalized spectra of line emission from a thin ac-
cretion disk at an inclination of 30◦ for various black hole spins.
The parameters α, β are easily expressed in terms of q 2 , l The emissivity is taken to be proportional to u2f between the
using (Cunningham & Bardeen 1973) marginally stable orbit and Rout = 15. Compare to Schnittman &
Bertschinger (2004b) Fig. 3.
l = −α(1 − µ20 )1/2 (67) rial plane of the black hole (e.g., Page & Thorne 1974,
2
q =β + 2
µ20 (α2 2
− ã ), (68) Shakura & Sunyaev 1973).
so that each point on the observer’s photographic plate 7.2.1. Grid Projection
corresponds to a unique geodesic. The first check of the code for this case is in visualiz-
ing the projection of a uniform grid at infinity onto the
7.1. Image in Affine Parameter
equatorial plane of the black hole. This is done by solving
As a first application of ray tracing, we can deter- for the final radius, uf , and azimuth where the geodesic
mine the appearance of the simplest possible black hole intersects µf = 0. Then, the new grid points are calcu-
shadow. The image “intensities” are taken to be the lated using pseudo-Cartesian coordinates (Schnittman &
affine parameter evaluated at the termination of the Bertschinger 2004b):
geodesic–either when it terminates at the black hole or
reaches a turning point and re-emerges to the starting p p
radius. Affine parameter is a good proxy for the emis- x= r2 + a2 cos φ, y = r2 + a2 sin φ. (69)
sion in this case, since it is related to the proper length The result of this projection for two different initial
along a geodesic, which would be the observed inten- observer inclinations and black hole spins is shown in Fig.
sity for constant emissivity and neglecting absorption. 3, and agrees with Fig. 2 of Schnittman & Bertschinger
The dimensionless affine parameter, λ′ , is given by (49). (2004b). The gravitational lensing effect can be seen in
The equatorial plane result for a Kerr black hole with the pictures with µ0 = .5 as the bunching of grid points
a = .998, to be compared to Bardeen (1973) Fig. 6, is behind the black hole, while frame dragging is evident in
shown in Fig. 2. The image shown here is 400 × 400. those with a = .95
7.2. Thin Disk Accretion 7.2.2. Thermal Disk Images
The next set of applications imagine the emitting As a next step, we can use the standard thin disk re-
source as an infinitesimally thin disk in the equato- sults for the radial temperature profile (e.g., Krolik 1998)
 Analytic Kerr Photon Orbits 11

to produce images of the disk at various inclinations as-

suming it is optically thick everywhere, so that the inten-
sity is that of a blackbody. Finding the radii of emission
from a grid in impact parameters and calculating the in-
tensity at each of these points produces an image of the
disk as seen by a distant observer. The result for an in-
clination of 85◦ and black hole spin a = .998 is shown in
Fig. 4. The image shows the effects of relativistic beam-
ing of the emission from gas moving towards the observer
versus the redshift of that moving away, as well as the
bending of the light from gas behind the black hole.

7.2.3. Line Emission

Next, following Schnittman & Bertschinger (2004b)
and Bromley et al. (1997) we consider monochromatic
emission from the disk, and give it an inner (outer) ra-
dius, Rin = Rms (Rout = 15), where Rms is the loca-
tion of the marginally stable circular orbit (e.g., Page &
Thorne 1974). The emissivity is weighted by u2f , physi-
cally motivated by the fact that we expect the tempera-
ture of gas in the disk to increase with decreasing radius.
The observed intensity is computed by exploiting the in-
variance of Iν /ν 3 (Misner et al. 1973),
Fig. 6.— Spectrogram of a circular hot spot of radius Rspot = .5
Iν0 = g 3 Iν (70) at the marginally stable orbit of a Schwarzschild black hole. The
observer is inclined at θ0 = 60◦ . Compare to Fig. 4 of Schnittman
where g ≡ ν0 /ν is the redshift, and ν0 (ν) is the ob- & Bertschinger (2004b).
served (emitted) frequency. To see the effect of black
hole spin on the emission in this case, we calculate Iν0 as
a function of g for several values of a by calculating the
intensity of rays at a location with redshift in a certain
range of g, and integrating them over the photographic
plate. The result is plotted in Fig. 5, and is in excel-
lent agreement with Fig. 3 of Schnittman & Bertschinger
(2004b). At higher black hole spin, the marginally stable
orbit is much closer to the black hole where the redshift
is much stronger, leading to a higher relative magnitude
and broadening of the low frequency peak (“red wing”).

7.2.4. Rotating Hot Spot

Finally, to test the time-dependence of the code, con-
sider a circular hot spot of finite radius Rspot = .5 orbit-
ing in the equatorial plane of a Schwarzschild black hole
at its marginally stable radius (Rms = 6). The emis-
sivity of the spot is taken to be Gaussian in the locally
flat space near the hot spot (for details, see Schnittman Fig. 7.— Light curves of the hot spot described in Fig. 6 for var-
2006), ious inclination angles. Intensities are normalized individually to
the integrated intensity over each orbit and scaled to the maximum
intensity from all inclinations. Compare to Fig. 6 of Schnittman
|x − xspot (t)|2
 
j(x) ∝ exp − , (71) & Bertschinger (2004b).
2
2Rspot
where j is the monochromatic emissivity. For some ob- shows the light curves of the hotspot for several inclina-
server coordinate time, t, the time delay and azimuthal tion angles. As the observer approaches edge-on viewing,
position from the observer to points on the disk are used the light curve becomes sharply peaked by a combination
to determine where on the photographic plate the sepa- of the Doppler beaming of the spot as it moves toward
ration between geodesic and hotspot are less than 4Rspot . the observer and the large gravitational lensing of the
For these points, the Gaussian emissivity and observed spot as it goes behind the black hole. The plot here is
frequency (redshift) are tabulated. Repeating this proce- in excellent agreement with Schnittman & Bertschinger
dure over a period of the motion gives a time-dependent (2004b).
spectrum, which is shown in Fig. 6 for an observer incli-
nation of 60◦ (µ0 = .5). This figure is in good agreement 7.3. Radiative Transfer
with Fig. 4 of Schnittman & Bertschinger (2004b). In more realistic astrophysical applications, the source
Integrating over frequency (redshift), or equivalently is not a delta function at a given inclination, and the
over the impact parameters, gives the light curve. Fig. 7 intensity along a ray can be written more generally as
12 Dexter and Agol

Fig. 8.— The spectrum of synchrotron radiation from optically Fig. 9.— Spectrum of synchrotron radiation from spherical accre-
thin spherical accretion onto a stellar mass black hole. The solid tion onto a stellar mass black hole. The solid line is the ray tracing
line is the ray tracing result, and the plotted points are the analytic result including absorption. The spectrum is heavily attenuated at
results. The two curves agree to within 5% at low frequencies, ν0 . 1011 Hz, and in this region follows the optically thick approx-
where the radiation originates at larger radii and the bending of imation of thermal emission from the τ = 1 surface. The spectrum
light should be unimportant. agrees well with the emission only model for ν0 & 1012 Hz

Z  ν 3
0
Iν0 = dIν . (72)
ray ν
If absorption can be neglected, dIν = jν dl where
dl = −pα uα dλ is the proper length differential measured
along the ray, pα is the photon four-momentum, uα is the
four-velocity of the emitting particle and λ is an affine
parameter.
The observed intensity is then
Z λ
Iν0 = jν g 2 dλ (73)
λ0

where jν is the emission coefficient in the rest frame of

the gas, and λ is now the dimensionless affine parameter
used above. It is calculated from (49) and used as the
independent variable along the ray. When absorption is
included, the solution to the radiative transfer equation
between affine parameters λ0 and λ reads (Fuerst & Wu
2004)
Z λ
′
3 −τν (λ0 )
Iν0 (λ) = g Iν (λ0 )e + e−(τν (λ )−τν (λ0 )) g 2 jν dλ′ ,
λ0
R (74)
where τν ≡ αν dl is the optical depth. Throughout this
paper we neglect scattering contributions to the emission
and absorption coefficients.
7.4. Synchrotron Radiation from Spherical Accretion
The code described above in conjunction with a routine
to perform radiative transfer along rays is now applied to
the particularly simple case of a stellar mass black hole Fig. 10.— Image of a spherically accreting Schwarzschild black
at rest with respect to the interstellar medium with a hole at ν = 1012 Hz as a contour plot and a 1-d profile.
temperature at infinity of 104 K and a density at infinity
of 1cm−3 . Ionized hydrogen accretes onto the black hole,
and the magnetic field threading the gas effectively cre-
ates collisions, so that the accreting gas can be considered B2 GM ρ
a perfect fluid. In the model, magnetic turbulence es- = , (75)
tablishes an equipartition of magnetic and gravitational 8π r
energy (Zeldovich & Novikov 1971). Then and cgs units are most convenient in the analytic calcu-
 Analytic Kerr Photon Orbits 13

lation. We assume an adiabatic equation of state with a

piecewise adiabatic index (Shapiro & Teukolsky 1983),  x  4.0505
!
M 0.40 0.5316 1/3
I ≃ 1/6 1+ 1/4
+ 1/2
exp(−1.8899xM ).
5 3 mp sin θ xM xM xM
γ= , T ≤1
3 2 me (83)
13 3 mp Note that this function is denoted I ′ (x) by Mahadevan
= , T > 1, (76) et al. (1996), and has a maximum error of ≈ 2.7%. The
9 2 me
spectrum is calculated by integrating (77) numerically.
where mp , me are the proton and electron mass and T To compare with these results, the ray tracing code
is the temperature in units of proton rest energy. Then is used to create an image of the synchrotron radiation
the fluid equations are non-linear, and can be solved nu- from the infalling gas in the same way as done previously
merically (Michel 1972) to find the temperature and fluid with affine parameter. To create an image, one specifies
velocity as functions of coordinate radius. a grid of points in α, β and calculates q 2 and l. This fully
The dominant form of radiation produced is syn- specifies the geodesic, and we can calculate the spacetime
chrotron radiation from the inner part of the accreting coordinates at which it intersects the accreting gas. The
sphere, where the electrons are ultrarelativistic (Shapiro intensity along each geodesic represents a point in the
1973b). In this case, the emissivity can be well approxi- image, which is why it is so important to be able to
mated analytically. Shapiro (1973a) performed the rela- calculate geodesics rapidly.
tivistic radiative transfer by approximating the photons The redshift is calculated using Viergutz (1993). Here,
as traveling on null geodesics in Minkowski spacetime, the flow is spherically symmetric and,
and calculating gravitational redshifts as well as the pho-
ton Doppler shifts along these paths.
 √ −1
g = γe−η [1 − eµ1 +η v r ρ−2 rsgn R] , (84)
Shapiro’s formula for the radiated spectrum is
Z r∗
with
Lν0 = 8π 2 dr r2 ×
2m
e2η ≡ ∆ρ2 Σ−1 , e2µ1 = ρ2 ∆−1 , γ = (1 − v r 2 )−1/2 ,
cos Θc 2
1−v
Z
(85)
d(cos Θ′ ) jν (77)
−1 (1 − v cos Θ′ )2 and v r is the radial component of the four-velocity. Writ-
p ten in terms of u in the Schwarzschild metric, this sim-
(1 − v 2 )(1 − 2m/r) plifies to
ν0 = ν ,
1 − v cos Θ′ √
where v(r) is the proper velocity seen by a stationary 1 − 2u
g= √ . (86)
observer and γ[1 + su (−1)Nu v u U ]
" #1/2 When the u-component of the four-velocity, v u , vanishes,
2 
this reproduces the standard gravitational redshift (Har-
 
27 2m 2m
| cos Θc | = −1 +1 (78) tle 2003).
4 r r
We first ignore absorption and compare radiated spec-
is the critical angle at which the light is recaptured by tra with the analytic calculation. The result is in Fig. 8.
the black hole. Shapiro (1973b) points out that the synchrotron radia-
The synchrotron emissivity for thermal, ultrarelativis- tion is dominated by a thin spherical shell of gas with
tic electrons averaged over polarization and solid angle ν ≃ νc . Then the first part of the spectrum, where
assuming isotropic emission in the rest frame is given by Lν0 ∼ ν 1/3 , originates from the outer part of the sphere.
(Pacholczyk 1970), The bending of light should be negligible in that region
and the ray tracing should agree with the analytic result,
2  which it does to within ≃ 5%. At higher frequencies, the
ne2 me c2

xM  radiation is originating in the innermost radii, and the
jν (T ) = ν √ I , (79)
2 3c kT sin θ bending of light becomes significant. The difference is
1 1 ∞ x ≃ 15% at high frequencies.
Z Z
I(x) ≡ dΩ dz z 2 exp(−z)F , (80) Next, absorption is included. Fig. 9 compares the spec-
4π x 0 z2
tra calculated with and without absorption. The radia-
with xM = ννc , tion is heavily attenuated at frequencies . 1011 Hz. At
  2 these frequencies, the luminosity is dominated by the in-
3eB kT nermost optically thin radius, which we take to be the
νc = , (81)
4πme c me c2 radius where τ = 1. Blackbody emission at the tem-
perature of gas at this radius, converted to a luminosity
and where
by integrating over impact parameter, is labeled ‘Ther-
Z ∞ mal’ in Fig. 9 and is a decent approximation to the full
F (x) ≡ x K5/3 (y)dy (82) spectrum when the fluid is optically thick.
x From ν0 ≃ 108 Hz to ν0 ≃ 1010 Hz, the gas is op-
is the synchrotron function. Mahadevan et al. (1996) tically thick everywhere. Then only thermal emission
have approximated I(x) above analytically by matching from the outermost radius is seen, and the spectrum
the asymptotic forms for large and small x. They find follows a Rayleigh-Jeans curve with Lν0 ∼ ν02 . From
14 Dexter and Agol

ν0 ≃ 1010 Hz to ν0 ≃ 1012 , the innermost optically thin transfer parts are about equally expensive in creating
radius is changing, and the luminosity begins to turn Figs. 10 and geodesic speed is relatively unimportant in
over. Starting at ν0 ≃ 1012 Hz, the gas is optically thin to the calculations leading to Figs. 5-9. In the latter cases,
the synchrotron radiation, and the spectrum reduces to this is because the same geodesics can be re-used at many
that of emission only (labeled ‘Emission’ in Fig. 9). This time steps, frequencies or both.
result agrees reasonably well with the assertion made The trend from these toy problems is that the sim-
by Shapiro (1973b) that absorption is negligible when pler cases benefit most from rapid geodesic calculation.
ν & 1011 Hz. However, there is reason to expect that for more realistic
Also of interest is the black hole shadow produced calculations rapid geodesic construction will again be im-
by various accretion models (Falcke et al. 2000). Fig. portant. Most accretion flows transition from optically
10 shows the shadow of the spherically accreting thin to thick. To accurately compute radiative transfer
Schwarzschild black hole as a 2-d contour plot and a 1-d from such flows, it is often necessary to take small steps
profile. The shadow is produced at α2 + β 2 = 27, and in the vicinity where the optical depth is about unity.
is caused by the difference in proper length of geodesics This requires calculating extra geodesic trajectories in
which intersect the horizon and return to infinity, as well this region. Since the region where the optical depth
as the blueshift of radiation from infalling gas behind the changes rapidly depends on frequency, and for a time-
black hole relative to the red shift of that nearest the ob- dependent accretion model on time as well, new geodesics
server. The asymmetry in Fig. 2 is not seen here due to must be computed at each time step and frequency, and
the spherical symmetry of the Schwarzschild metric. hence they cannot be re-used as is the case for the time-
independent, optically thin models considered in almost
8. FUTURE WORK all our examples.
The code presented here is the first to calculate all co- The precision of our code is also extremely high over
ordinates of Kerr null geodesics semi-analytically. This a broad range of geodesic parameters. This is currently
work’s natural extension is to timelike geodesics, which less important in radiative transfer applications where
involves many more cases, but only straightforward gen- the dynamical models are uncertain, but it is important
eralizations of the formulas given here (RB94, Appendix in caustic calculations such as those in RB94 and Bozza
A). The main challenge is that for bound orbits it is dif- (2008). Finally, our code can compute arbitrary sections
ficult to specify the number of radial turning points in of geodesics in any direction. This flexibility allows extra
advance. Ideally the affine parameter could be used as points to be calculated in regions where the optical depth
an independent variable to indicate how far along the is changing rapidly or to check convergence on the fly. It
geodesic to trace. However, it is a function of u and µ may also be useful in a future method for computing
which cannot be inverted. Compton scattering, in which rays are traced outwards
from each point on the geodesic to calculate the scattered
8.1. Advantages of Analyticity intensity into that point.
Unlike previous analytic work, our code makes no as-
The main advantages of using a semi-analytic code sumption of time-independence or axisymmetry in the
such as that presented here for tracing geodesics are accretion flow and is therefore well suited to the geome-
speed, accuracy and flexibility. The speed increase from tries used in 3D GRMHD simulations. Computationally
our code depends greatly on the application considered. expensive observables such as polarization and variability
For ray tracing applications, a lower bound is a factor of will be much more tractable given the speed and flexibil-
5 in the case where all coordinates are being calculated, ity of this code.
and the entire ray is being traced. The maximum speed
increase is probably a factor between 100-500 in the case
where the code is solving for geodesic coordinates at a J.D. thanks Jeremy Schnittman for help with the
specific point. geodesic hot spot model, and the referee, Avery Brod-
The importance of speed in tracing geodesics depends erick, whose comments improved this paper and led
on the computational expense of their construction rela- to substantial rewriting of the code. This work was
tive to that of the rest of the desired calculation. For the partially supported by NASA grants 05-ATP05-96 and
simple radiative transfer applications considered here, NNX08AX59H, and a graduate fellowship from the Kavli
time spent computing geodesics dominates in creating Institute of Theoretical Physics at the University of Cal-
Figs. 2,3,4. The construction of geodesics and radiative ifornia, Santa Barbara under NSF grant PHY05-51164.
REFERENCES
Abramowitz, M., & Stegun, I. A. 1965, Handbook of mathematical Bromley, B. C., Melia, F., & Liu, S. 2001, ApJ, 555, L83
functions with formulas, graphs, and mathematical tables (Dover Byrd, P. F., & Friedman, M. D. 1971, Handbook of elliptic integrals
Books on Advanced Mathematics, New York: Dover) for engineers and scientists (Second edition, Springer, New York)
Agol, E. 1997, PhD thesis, University of California, Santa Barbara Cadez, A., Fanton, C., & Calvani, M. 1998, New Astronomy, 3, 647
Bardeen, J. M. 1973, in Black holes (Les astres occlus), 215–239 Carlson, B. C. 1988, Mathematics of Computation, 51, 267
Beckwith, K., & Done, C. 2005, MNRAS, 359, 1217 —. 1989, Math. Comp., 53, 327
Beloborodov, A. M. 2002, ApJ, 566, L85 —. 1991, Math. Comp., 56, 267
Bozza, V. 2002, Phys. Rev. D, 66, 103001 —. 1992, Mathematics of Computation, 59, 165
—. 2008, ArXiv e-prints, astro-ph/0806.4102 —. 2005, J. Comput. Appl. Math., 174, 355
Braje, T. M., Romani, R. W., & Rauch, K. P. 2000, ApJ, 531, 447 Carter, B. 1968, Physical Review, 174, 1559
Broderick, A., & Blandford, R. 2003, MNRAS, 342, 1280 Chandrasekhar, S. 1983, The mathematical theory of black holes
—. 2004, MNRAS, 349, 994 (Oxford/New York, Clarendon Press/Oxford University Press)
Broderick, A. E. 2006, MNRAS, 366, L10 Cunningham, C. T. 1975, ApJ, 202, 788
Broderick, A. E., & Loeb, A. 2006, ApJ, 636, L109 Cunningham, J. M., & Bardeen, C. T. 1973, ApJ, 183, 237
Bromley, B. C., Chen, K., & Miller, W. A. 1997, ApJ, 475, 57
 Analytic Kerr Photon Orbits 15

Erdélyi, A., Magnus, W., Oberhettinger, F., & Tricomi, F., eds. Press, W. H., Teukolsky, S. A., Vetterling, W. T., & Flannery,
1981, Higher Transcendental Functions, Vol. II (Malabar, U.S.A.: B. P. 1992, Numerical recipes in FORTRAN. The art of scientific
Robert E. Krieger Publishing Company) computing (Cambridge: University Press, —c1992, 2nd ed.)
Falcke, H., Melia, F., & Agol, E. 2000, ApJ, 528, L13 Rauch, K. P., & Blandford, R. D. 1994, ApJ, 421, 46
Fanton, C., Calvani, M., de Felice, F., & Cadez, A. 1997, PASJ, Reid, M. J., Broderick, A. E., Loeb, A., Honma, M., & Brunthaler,
49, 159 A. 2008, ArXiv e-prints, astro-ph/0801.4505
Font, J. A., Ibáñez, J. M., & Papadopoulos, P. 1999, MNRAS, 305, Schnittman, J. D. 2006, ArXiv Astrophysics e-prints, astro-
920 ph/0601406
Fuerst, S. V., & Wu, K. 2004, A&A, 424, 733 Schnittman, J. D., & Bertschinger, E. 2004a, ApJ, 606, 1098
Hartle, J. B. 2003, Gravity : an introduction to Einstein’s general —. 2004b, ApJ, 606, 1098
relativity (San Francisco, CA, USA: Addison Wesley) Schnittman, J. D., Krolik, J. H., & Hawley, J. F. 2006, ApJ, 651,
Jaroszynski, M., & Kurpiewski, A. 1997, A&A, 326, 419 1031
Krolik, J. H. 1998, Active Galactic Nuclei: From the Central Shakura, N. I., & Sunyaev, R. A. 1973, in IAU Symposium, Vol. 55,
Black Hole to the Galactic Environment (Princeton: Princeton X- and Gamma-Ray Astronomy, ed. H. Bradt & R. Giacconi,
University Press) 155–+
Li, L.-X., Zimmerman, E. R., Narayan, R., & McClintock, J. E. Shapiro, S. L. 1973a, ApJ, 180, 531
2005, ApJS, 157, 335 —. 1973b, ApJ, 185, 69
Luminet, J.-P. 1979, A&A, 75, 228 Shapiro, S. L., & Teukolsky, S. A. 1983, Black holes, white dwarfs,
Mahadevan, R., Narayan, R., & Yi, I. 1996, ApJ, 465, 327 and neutron stars: The physics of compact objects (New York,
Michel, F. C. 1972, Ap&SS, 15, 153 Wiley-Interscience, 663 p.)
Misner, C. W., Thorne, K. S., & Wheeler, J. A. 1973, Gravitation Viergutz, S. U. 1993, A&A, 272, 355
(San Francisco: W.H. Freeman and Co.) Wu, S.-M., & Wang, T.-G. 2007, MNRAS, 378, 841
Noble, S. C., Leung, P. K., Gammie, C. F., & Book, L. G. 2007, Zeldovich, Y. B., & Novikov, I. D. 1971, Relativistic astrophysics.
Class. and Quant. Gravity, 24, 259 Vol.1: Stars and relativity (Chicago: University of Chicago Press)
Pacholczyk, A. G. 1970, Radio astrophysics. Nonthermal processes
in galactic and extragalactic sources (San Francisco: Freeman)
Page, D. N., & Thorne, K. S. 1974, ApJ, 191, 499