A continuous Riemann-Hilbert problem for colliding

plane gravitational waves

arXiv:1701.01296v1 [gr-qc] 5 Jan 2017

Stefan Palenta and Reinhard Meinel

Theoretisch-Physikalisches Institut, Friedrich-Schiller-Universitat Jena,
Max-Wien-Platz 1, 07743 Jena, Germany
E-mail: spalenta@gmail.com, meinel@tpi.uni-jena.de
Abstract. We present the foundations of a new solution technique for the
characteristic initial value problem (IVP) of colliding plane gravitational waves. It
has extensive similarities to the approach of Alekseev and Griffiths in 2001, but we use
an inverse scattering method with a Riemann-Hilbert problem (RHP), which allows for
a transformation to a continuous RHP with a solution given in terms of non-singular
integral equations. Ambiguities in this procedure lead to the construction of a family
of spacetimes containing the solution to the IVP. Therefore the described technique
may also serve as an interesting solution generating method.

Keywords: gravitational waves, characteristic initial value problem, inverse scattering

method, Riemann-Hilbert problem
1. Introduction
Recent observations confirm the existence of gravitational waves (GW) emitted in
strongly gravitating binaries, where the nonlinearity of the Einstein equations plays an
important role. The observational data are convincingly reproducible with numerical
models, however the performance of analytic descriptions of the strong gravity regime
is limited.
On the other hand, the model of colliding plane GW represents the simplest method
to study nonlinear wave interactions. Many features of nonliearity as well as conceptual
issues like focussing properties and arising singularities have been discussed on the
basis of colliding plane waves. For their interaction region many solution generating
techniques and exact solutions have been described (cf. the overview of Griffiths [1]
or [2]) where the shape of the incoming waves has to be derived afterwards. Hauser
and Ernst [3, 4, 5, 6, 7] pioneered the search for a method to address the characteristic
initial value problem and proved the existence and uniqueness of its solution. Alekseev
and Griffiths [8, 9] described a more practical procedure for colliding gravitational and
electromagnetic waves leading to singular integral equations. Our treatment of the
characteristic initial value problem features many similarities to this approach, but
allows for additional transformations to non-singular integral equations. It is supposed

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

to be better suited for approximations using spectral methods, however it is still too
early to clearly compare the performance of the two approaches related to this goal.
Finally we aim for a systematic study of interaction properties of colliding plane GW
depending on the initial data.
In this paper we consider purely gravitational plane waves with distinct wavefronts
and arbitrary polarisation colliding in a Minkowski background. The planes of symmetry
are spanned by two commuting spacelike Killing vectors and this symmetry reduces
the Einstein equations to the hyperbolic Ernst equation. We make use of the strong
formal analogy to axially symmetric and stationary spacetimes (with one timelike and
one spacelike Killing vector) governed by the elliptic Ernst equation by formulating a
linear problem (LP) in the Neugebauer form, cf. [10]. Its solution can be represented
by the solution of a Riemann-Hilbert problem (RHP) whose jump matrix is defined
by the characteristic initial values. This procedure is known as the inverse scattering
method [10]. Due to the inevitable non-analytic behaviour of the initial data on the
wavefronts, the jump matrix is discontinuous. Adapting a general method of Vekua
[11], we implement a transformation to a continuous RHP (cRHP) which can be solved
using non-singular integral equations.
Because of the non-analytic behaviour of the initial data the RHP does not uniquely
define the solution to the LP and we need to impose regularity conditions of the LP
coefficient matrices. Remaining degrees of freedom may lead to families of solutions,
therefore our procedure could also serve as solution generating technique.
2. The characteristic initial value problem for colliding plane waves
2.1. Ernst equation
We write the metric in the Rosen-form with the parametrisation as given in [1] :
2eU
|dx + iEdy|2 .
(1)

E+E
It contains the two real functions M (u, v) and U (u, v) and the complex Ernst potential
E(u, v) only depending on the two lightlike coordinates u and v. The spacelike
coordinates x and y parametrise the planes spanned by the Killing vectors. The vacuum
Einstein equations reduce to the essential relations
ds2 = 2eM dudv

Uuv = 0,

(E + E)(2E
uv Uu Ev Uv Eu ) = 4Eu Ev ,

(2)

(2Uuu Uu2 + 2Mu Uu ) = 4Eu Eu ,

E)
2 (2Uvv Uv2 + 2Mv Uv ) = 4Ev Ev ,
E)

(4)

(E +
(E +

(3)
(5)

where by coordinate indices u and v as well as f and g below we denote partial

derivatives. Equation (2) has the general solution
eU = f (u) + g(v)

(6)

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

containing two arbitrary functions f (u) and g(v). In accordance to Griffiths [1] we
choose
f=

1
2

for u 0,

g=

1
2

for v 0,

f 0 (0) = 0 = g 0 (0),

(7)

whereby (4) and (5) show that f and g are monotonically decreasing for u, v 0. Using
them as coordinates, (3) becomes the hyperbolic Ernst equation


Ef + Eg
Re(E) 2Ef g +
= 4Ef Eg .
(8)
f +g
Having determined E and U we can afterwards obtain the function M by integration
of the field equations (4) and (5). Together with E also the function E 0 = aE + ib
(a, b R) is a solution to the Ernst equation (8). We fix this freedom by demanding as
connection to the Minkowski background the normalisation
E( 21 , 21 ) = 1.

(9)

2.2. Spacetime regions

As illustrated in figure 1, it is appropriate to divide a colliding wave spacetime into four
regions [12] with the following coordinate dependencies of the metric functions:
I
II
III
IV

:
:
:
:

u < 0, v < 0 :
u 0, v < 0 :
u < 0, v 0 :
u 0, v 0 :

E = 1,
E = E(u, 0),
E = E(0, v),
E(u, v),

M = 0,
M = MII (u),
M = MIII (v),
M (u, v),

eU
eU
eU
eU

= 1,
= 21 + f (u),
= 21 + g(v),
= f (u) + g(v).

(10)

Figure 1. Identification of the 4 spacetime regions of colliding GW after Griffiths [1]

The physical interpretation is that on a Minkowski background (I) two plane waves
propagate undisturbed in opposite direction (II and III, there is always a frame where
the collision happens head on) until their collision and nonlinear interaction(IV).
Using the functions f and g as coordinates in the interaction region IV, the initial
value problem of colliding plane GW corresponds to finding a solution E to the Ernst
equation (8) with given initial values E(f, g = 21 ) and E(f = 12 , g) respecting the

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

normalisation (9). Due to the fact that these boundaries are the characteristic curves
of the Ernst equation, it is sufficient to provide initial values of E without additionally
giving its derivatives.
As indicated by the f + g in the denominator within (8) (and discussed in detail
in [1]), the colliding wave spacetime features a generic scalar curvature singularity at
f + g = 0. This can be understood by the mutual focussing properties of waves in GR.
For a large variety of exceptional cases this singularity is replaced by a Killing-Cauchy
horizon, but this horizon is conjectured to be unstable [13]. For linear polarised waves
this instability has been rigorously proven [14]. The regions II and III are confined
by coordinate degeneracies on lightlike hypersurfaces. They can be identified with the
points f = g = 21 and f = g = 12 and inherit their singular character. Nevertheless,
for the vacuum case considered here they are no scalar curvature singularities on their
own and so the term fold singularity has been established. As a consequence of this
singularity structure, the colliding plane wave spacetime is not globally hyperbolic [1].
2.3. Colliding wave conditions
Using the junction conditions of OBrien and Synge [15] for lightlike boundaries (shown
to be appropriate by Robson [16]) the metric has to meet the following demands:
E, M C 0 ,

U C 1.

(11)

This allows us to perform C 1 -transformations u u0 (u) and v v 0 (v) to arrange

f=

1
2

un1 (u),

g=

1
2

v n1 (v).

(12)

Alternatively, we may use such C 1 -transformations to achieve MII (u) = 0 and MIII (v) =
0. Then f (u) and g(v) are determined by the field equations (4), (5) and the junction
conditions (7) with the Minkowski background. The exponents n1/2 describe also in
this case the first order behaviour of f (u) and g(v) because they cannot be changed by
C 1 -transformations. The field equations (4), (5) impose the restriction
n1/2 2.

(13)

where n1/2 = 2 implies an impulsive wavefront. Furthermore, together with the

continuity of M , they lead to the so-called colliding wave conditions for E first
formulated by Hauser and Ernst [3], which within the normalisation (9) have the form




lim1 1 ( 12 g)Eg Eg = 2k2 , (14)
lim1 1 ( 21 f )Ef Ef = 2k1 ,
(f,g)( 2 , 2 )

(f,g)( 2 , 2 )

with
k1/2 = 1

1
n1/2

1
k1/2 < 1.
2

(15)

In the context of the characteristic IVP the colliding wave conditions are a matter of
choosing suitable initial values for E featuring divergent derivatives at (f = 21 , g = 12 ).

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

3. Inverse scattering method for parallel polarisation

3.1. Scheme of inverse scattering

Ernst
equation

integrability
condition of

boundaries
Initial values

linear
problem

conditions for the

solution of the LP

(i)

guessing (difficult)

solution of
Ernst eq.

(iv)

solution is also
solution of

solution
fitting to
initial values
of LP

RiemannHilbert
problem

transformation

contour
matrix jump

(iia)

("scattering data")

singular integral equation

(bad convergence)

(iii)

regularity
conditions

solution
of RHP

(iib)

continous
RHP
regular
integral
equation
(good
convergence)

backtransformation
solution
and linear
of
cRHP
combination

Figure 2. Scheme of the inverse scattering method with additional transformation to

a continuous Riemann-Hilbert problem

In the course of the inverse scattering method, the nonlinear Ernst equation is
expressed as the integrability condition of a system of linear partial differential equations,
the so-called linear problem (LP). This system is in turn solved by constructing
an appropriate Riemann-Hilbert problem (RHP) which has the same solution. In
correspondence with the boundaries of the IVP, the contour of the RHP has to be chosen
as two specific parts of the real axis in the complex plane of a spectral parameter. The
RHP relates inner and outer limit of its solution at the contour by a purely multiplicative
matrix-valued jump. It is possible to connect the ends of the contour arbitrarily and
equate the jump matrix to 1 on the added part in order to regard the RHP contour as
closed. Regrettably, the singularity in the divergence of the Ernst potential demanded
by the colliding wave equations (14) leads to a jump matrix which doesnt tend to 1 at
the ends of the initial contour. Therefore the RHP is discontinuous and its solutions
feature singularities at the ends of the initial contour. In fact, there are even two different
appearances of these singularities, so that at each of the 2 initial contours an ambiguity
in the RHP solution arises.
In this article we present a transformation to a continuous Riemann-Hilbert problem
(cRHP) which clarifies these ambiguities and which is supposed to have better numerical
properties than the (discontinuous) RHP.
For given initial data E(f, g = 21 ) and E(f = 21 , g) the following 4 steps indicated
in figure 2 have to be carried out:
(i) Translating the initial data into the jump matrix by solving a system of ordinary
differential equations (ODE). For special cases an analytical treatment is possible.

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

(ii) Solving the Riemann-Hilbert problem by expansion of an additive jump function

in Chebyshev polynomials
(a) via singular integral equations with bad numerical properties
(b) via transformation to the cRHP and its non-singular integral equations with
better numerical properties
(iii) Evaluating regularity conditions, which assure that the RHP solution fulfills the
LP. These are purely algebraic equations to determine the linear combination
coefficients of the RHPs 4 basic solutions.
(iv) Fixing the remaining degrees of freedom to adapt the solution to its initial data
The whole process of the inverse scattering method shall be illustrated by examining
the case of parallel polarised gravitational waves. The contour of the parallel polarised
case will be directly transferred to the RHP for general polarised GW.
3.2. Linear problem for parallel polarisation
Within the Newman-Penrose formalism the singular waves in the spacetime regions II
and III are described by the Weyl tensor components 0 (v) and 4 (u) respectively.
Parallel polarised colliding GW are characterised by a constant difference between the
phases 0 (v) and 4 (u) of these components,
0 (v) 4 (u).

(16)

Since the phases depend on different variables, (16) implies 0 = const and 4 = const.
Therefore parallel polarised GW are lacking a whole functional degree of freedom
compared to the general case, which makes them a very special setup.
The general solution for parallel polarisation has been derived by Hauser and Ernst
[3] in terms of generalized Abel transformations. It has been reformulated in order to
obtain an initial point for the generalisation to arbitrary polarisation [4]. We will follow
the same line here to establish our methods through the parallel case.
The LP for parallel polarisation is to find the function LPpp (f, g; ) satisfying
s
LPpp
kg
LPpp
=
(1
+
)A
f
=
,
(17)
1
LPpp ,
LPpp
k+f
g
= (1 + )B
where A and B are real functions of f and g whereas k is an independent spectral
parameter, which enters the equations through the spectral parameter depending on
f , g and k. The partial derivatives f and g are taken with constant k rather than
constant . The solution pp can be thought of either a function on the complex -plane
or as a function of k defined on a two-sheeted Riemann surface with branch cut along
the segment [f, g], a twofold covering of the complex k-plane. We call the sheet with
1 for k the upper one and the sheet with 1 for k the lower one,
cf. figure 3.

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

The integrability conditions of the LP (17) assure the existence of a potential pp

fulfilling
fpp = A,

(18)

gpp

(19)

= B,
fpp + gpp
2fppg +
= 0.
f +g

(20)

The linear equation (20) is indeed the Euler-Poisson-Darboux equation, which one can
derive from the Ernst equation for real E by setting
pp = 21 ln E.

(21)

The LP solution LPpp is only defined up to multiplication with a function of k.

We now fix this freedom by demanding the normalisation


k.
(22)
LPpp 12 , 21 = 1
Here and in the rest of the paper we use the following convention for an arbitrary
function F depending on f , g and : Where F is displayed with 2 arguments, these
should be understood as the values of f and g, but where F is displayed with a single
argument, this should be taken as the value of . Evaluating the LP (17) at = 1
and = 1 equation (22) easily leads to
LPpp (1) = 1,
LPpp (1) = e

(23)

2 pp ( 21 , 12 )

2 pp

(24)

The Ernst potential with the normalisation (9) is therefore given by

E = e2

pp

= LPpp (1).

(25)

3.3. Riemann-Hilbert problem for parallel polarisation

The RHP connected to (17) is to find a function pp (f, g; ) which is analytic everywhere
in the complex Riemann k-surface except on the contour k , where it has a jump
described by the equation
pp
pp
+ = (k) .

(26)

pp
Herein pp
+ is the inner (left to the contour) and the outer (right to the contour)
limit of pp . In addition, we fix the freedom of multiplying pp with a function of f
and g by demanding the normalisation
pp

0 (1) = 1

f, g.

(27)

The derivatives of pp with respect to the coordinates can be written as

pp
pp
pp
f = (f )g,=const + f ,

with the partial derivatives

f =
(2 1),
2(f + g)

g =

pp
pp
pp
g = (g )f,=const + g

1
(2 1),
2(f + g)

(28)

(29)

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

which are singular at and 0 respectively. Because the multiplicative jump is a

function only depending on the spectral parameter k, we can deduce by calculating
pp 1
pp 1
exhibit no jump on the
derivatives of (26) that the terms pp
and pp
g ( )
f ( )
contour k . Within the treatment of the cRHP we even show that there exists a solution
pp 1
pp 1
holomorphic in C \{} and pp
holomorphic in
of the RHP with pp
g ( )
f ( )
C \{0}. A power series expansion leads to
pp 1
= c(1) (f, g) + c(0) (f, g),
pp
f ( )

pp 1
pp
= d(1) (f, g)1 + d(0) (f, g) (30)
g ( )

and with the normalisation (27) we get the LP (17).

(1)
We choose the contour k in the k-surface to consist of a first part k directed
from k = 12 in the upper sheet through the branch point k = f to k = 21 in the
lower sheet and a second part directed from k = 12 in the lower sheet through the branch
point k = g to k = 21 in the upper sheet.

(1)

(2)

Figure 3. The 2 parts k and k of the contour k in the upper (left) and lower
(right) sheet of the two-sheeted Riemann k-surface. At the branch cut [f, g] bright
area is connected to bright area and dark area to dark area

By setting k = 21 we get the contour end points in the -plane:

s
s
1
1
+
g
g
2
1 = 1
,
2 = 12
.
f
+f
2
2

(31)

They also lie on the real axis and satisfy 1 > 1 > 2 > 0. The contour in the -plane
(1)
(2)
is divided into (1) corresponding to k and (2) corresponding to k . The first part
(1) is directed from 1 through = to 1 and the second part (2) is directed from
2 through = 0 to 2 .
The contour vanishes for f = 12 = g, which leads with (27) to a reproduction of the
pp 1
LP normalisation pp ( 21 , 12 ) = 1. Therefore normalised RHP solutions with pp
f ( )
pp 1
holomorphic in C \{} and pp
holomorphic in C \{0} are also solutions LPpp
g ( )
of the normalised LP.
In the plane parallel case we can rewrite (26) as the additive jump equation
pp
pp
ln pp
+ ln = ln =: i .

(32)

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

-f

Figure 4. The 2 parts (1) and (2) of the contour k in the -plane.

The solution can be given in terms of a Cauchy integral in the -plane as



1
1
1
pp
ln =
0
pp (k 0 )d0 ,
0
2 + 1

(33)

where the second term under the integral assures the normalisation (27). Evaluating
(33) at = 1 leads with (21) and (25) to
f
1
0
0
2
1
pp
1
pp
(k 0 )dk 0
1 (k )dk
pp
p
p 2
=

,
(34)
2 21
(k 0 g)(k 0 + f ) 2 g
(k 0 g)(k 0 + f )
1
pp
where pp
1/2 = |(k) . For f = 2 the second line of the linear problem (17) reads
1/2





ln pp 12 , g g = (1 + 1 )gpp 21 , g .
(35)

With (22) the integration from g 0 = 1/2 to g 0 = g gives

s
1/2
k + 21 pp 0
ln pp ( 21 , g) = pp ( 12 , g)
0 dg .
k g0 g
g

(36)

(2)

From this the additive jump on k can be shown to be

1
q
2
pp ( 1 , g 0 )
pp
pp
pp
0 g0 2
1
2 = i(ln + ln ) = 2 2 +
dg 0
g

(37)

(1)

and analogous the additive jump on k is

pp
1

i(ln pp
+

ln pp
)

q
1
= 2 2

fpp0 (f 0 , 12 )
df
.
+ f0

1
2

(38)

Note that these jump functions both are defined on the interval [ 12 , 12 ], but for given
(1)
(2)
pp
f and g only the values of pp
1 on k and the values of 2 on k are used in (34).
Obviously real initial values fpp (f, 21 ), gpp ( 12 , g) lead to real pp
1/2 and via (34) to a real
pp
solution . The combination of (37), (38) and (34) constitutes the general solution of
the IVP of parallel polarised colliding plane waves.
In [3] this solution is derived with a generalised Abel transformation. Furthermore,
a RHP similar to (26) is presented, where the spectral parameter lies in a simple complex
plane. Its solution H is related to pp by
ln pp (k) = (k + f )H (2k) + pp

(39)

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

10

and it uses the jump functions

pp
1 (/2)
,
g3 () =
2 1

pp
2 (/2)
g2 () =
.
2 1+

(40)

.
4. Inverse scattering method for arbitrary polarisation
4.1. Linear problem for arbitrary polarisation
The LP for colliding plane waves of arbitrary polarisation is to find the matrix
LP (f, g; ) satisfying
!
!
LP

A A
B
1 B
LP
f = U
U=
, V =
.
(41)
LP ,
A A
1 B
B
LP
g = V
Herein A and B are complex functions of f and g and the spectral parameter is defined
as in (17). In addition, we fix the freedom of right-multiplying a matrix function of k
to the solution, LP LP C(k), by demanding the normalisation
!
1
1
LP ( 21 , 12 ) =
k.
(42)
1 1
The integrability conditions of the LP (41) assure the existence of a potential E
fulfilling the equations
Eg
Ef
= A,
=B
(43)

E+E
E + E
and the Ernst equation (8).
To motivate the design of the RHP, we derive some properties of the LP. We proceed
analogous to [10]. Using the Pauli matrices
!
!
0 1
1 0
1 =
, 3 =
(44)
1 0
0 1
we can state the following Relations between the matrices U () and V () and their
respectively:
values at and
W () = 3 W ()3 ,

= 1 W ()1
()
W

with W {U, V }.

(45)

Due to these relations, from a given vector v() solving the LP we can derive the new
Therefore we can construct a matrix solution of the LP,
solutions 3 v() and 1 v().
!
LP
LP

(
)

(
)
3 v() + 3 1 v())

LP = v() + 1 v(),
=
,
(46)
LP () LP ()
depending only on a single scalar function which we will call the scalar solution of the
LP. The representation (46) is consistent with the normalisation (42) providing that
LP ( 12 , 21 ) = 1 k,

(47)

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

11

and so we will assume the matrix solution of the LP and the RHP later on to have
this structure (46), which we will abbreviate by saying LP is in normal form with the
scalar function LP . We denote the negative determinant of a matrix in normal form
with LP as
DLP := det LP .

(48)

The (2, 1)-elements of the LP equations yield for = 1 with (43):

Eg
Ef
(LP (1) + LP (1)),
LP
(LP (1) + LP (1)).
(49)
LP
g (1) =
f (1) =

E+E
E + E
Integration leads to LP (1) = aE + ib, a, b R. Considering (47) the Ernst potential
with the normalisation (9) is given by
E := LP (1).

(50)

Evaluation of the LP at = 1 in the same way leads to LP (1) = 1.

With the identity
(ln det M )x = Tr(Mx M 1 ),

(51)

holding for an arbitrary square matrix M , and the normalisations (9), (47) of the Ernst
potential and LP we can derive from the LP the relation
LP () + LP ()
LP () = E + E
DLP = LP ()
f, g, k.
(52)
In particular it states that the determinant of the LP solution is a function depending
only on the coordinates f and g.
4.2. Riemann-Hilbert problem for arbitrary polarisation
The RHP for arbitrary polarisation is to find the matrix (f, g; ) analytic in C \ and
satisfying on the jump equation
+ = J(k).

(53)

The jump matrix J(k) has the form

J(k) =

(k) (k)
(k)
(k)

!
with R,
+ 2 = 1,

(54)

exhibiting only one complex degree of freedom . It is sufficient to consider the same
jump matrix in both sheets of the k-surface and so we set J() = J(). Fixing the
freedom of left-multiplying an arbitrary matrix M (f, g) we demand the normalisation
!

1 E
(1) =
.
(55)
1 E
As the investigation of the cRHP will show, there exists a solution of the RHP
in normal form (46) with a scalar function and with f 1 holomorphic in C \{}
as well as g 1 holomorphic in C \{0}. Similar to the parallel polarised case a power
series expansion leads to
f ()1 = M(1) (f, g) + M(0) (f, g),

g 1 = N(1) (f, g)1 + N(0) (f, g).

(56)

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

12

The representation (46) is consistent with the normalisation (55) under the
conditions
(1) = E,

(57)

(1) = 1 f, g.

(58)

Evaluating (56) at = 1 and = 1 using (58) leads exactly to the LP (41).

For f = 21 = g the contour vanishes and so the solution is independent of k.
Considering (58) we find the normalisation (42) of the LP reproduced. In consequence,
normalised RHP solutions with f 1 holomorphic in C \{} and g 1 holomorphic
in C \{0} are also solutions LP of the normalised LP. Equation (58) can be regarded
as the single normalisation condition for the RHP solution in normal form with ,
whereas (57) defines the associated solution of the Ernst equation with the required
normalisation (9).
For = 0 the RHP (53) reduces to the case of parallel polarisation with pp :=
i ln , pp := and pp := 21 ln E.
By the proper choice of the scalar jump function we still have to assure that the
Ernst potential matches the initial values at f = 12 and g = 12 .
4.3. Calculation of the jump matrix from initial data

At first we want to note that reading off the values of the scalar solution at or
means in our setup changing the site of the contour:





[()]+/ = /+ (),
()
=

(
),
(
)
= /+ (),
(59)
/+
+/
+/
on the real axis R With (52) we can convert the jump equation (53) to
+ () + + () ()

+ () () ()

()
+ ()
=
,

=
+ () + ()
()
+ () + ()
() .
+ ()
+ ()
Using 1/2 := | (k) and 1/2 := |
It is now convenient to introduce = (
).
1/2

(60)
(k)

1/2

the

evaluation of (60) at = and = 0 yields

2+ ()+ ()
,
|+ ()|2 + |+ ()|2
2+ (0)+ (0)
,
2 =
|+ (0)|2 + |+ (0)|2

1 =

|+ ()|2 |+ ()|2
,
|+ ()|2 + |+ ()|2
|+ (0)|2 |+ (0)|2
2 =
.
|+ (0)|2 + |+ (0)|2

1 =

(61)
(62)

For a given k [ 21 , 21 ] we will now calculate + and + at 0 and by integration

of the LP in the (f, g)-plane. The starting point of the integration is ( 12 , 21 ), where the
normalisation (47) defines = 1 = . We achieve = at (k, 12 ) and = 0 at
( 21 , k). Choosing the integration path along g = 12 and f = 12 respectively now leads to
2 tremendous simplifications. First of all, the LP can be reduced to a single ODE in
both cases, and secondly only values on the boundaries of the IVP are used. Therefore
1/2 and 1/2 can be calculated from the initial values alone.

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

13

For f < k < g the value of is purely imaginary, and its imaginary part is
positive on the inner side of . Hence the ODEs read
s
!
!
!
1
f

k
1
+
A i A
+
f
2
> 0, g = ;
(63)
=
,

:=
i
=
f
+
i A A
+
f +k
2
f
s
!
!
!
1
e

+k
1
+
B
i B
+
e
> 0, f = , (64)
=
, := i/ = 2
e
+
i B
B
+
gk
2
g

where (63) is integrated from f = 21 to f = k and (64) is integrated from g =

g = k respectively. This corresponds to step (i) of the scheme in figure 2.

1
2

to

4.4. The boundary values of the jump matrix

We define the boundary coefficients Ab , Bb C as well as their amplitudes and phases
1/2 , A , B R by
q

q

iA
iB
1
1
Ab := 1 e :=
lim1 1
f A , Bb := 2 e :=
lim1 1
gB . (65)
2
2
(f,g)( 2 , 2 )

(f,g)( 2 , 2 )

Considering (43) we can thus state the colliding wave conditions (14) as
r
r
k1
k2
,
2 =
.
1 =
2
2
From the domain (15) of k1/2 we get
1
2

1/2 < 2 2 .

(66)

(67)

In order to calculate the boundary values of the jump matrix, J(1 ) and J(2 ),
we examine (63) for k = 21 + . Substituting f = 21 , the ODE system in leading
order in reads
!
!
!
f

1
1
1
+
A
i
A

b
+
b
= 2
, f = , g = ,
(68)
f
i Ab
Ab
+
+
2
2

which has to be integrated from = 0 to = . For 0 < <   1 we have f  1 and

can reduce (68) in leading order to
1
1
+ = i[( )] 2 Ab + , + = i[( )] 2 Ab + .
(69)
p
Substituting s = 2 arcsin( /) we obtain with + = 1 = + at s = 0 the solution
!
!
b|
cos(|Ab |s) i|A
sin(|A
|s)
+
b
Ab
=
.
(70)
iAb
cos(|Ab |s) |A
sin(|A
|s)
+
b
|
b

From the value at s = corresponding to = we get in the trivial limit  0 the

boundary values of the jump matrix elements at 1 :
1b := 1 (k = 21 ) = cos(21 ) i cos(A ) sin(21 ),

(71)

1b := 1 (k = 21 ) = sin(A ) sin(21 ).

(72)

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

14

In the same way we can derive

2b := 2 (k = 21 ) = cos(22 ) + i cos(B ) sin(22 ),

(73)

2b := 2 (k = 21 ) = sin(B ) sin(22 ).

(74)

With relation (67) the domain of <(1b ) and <(2b ) is

1 <(1/2b ) < cos( 2) < 0.

(75)

In particular 1/2b = 1, which would be necessary for a continuous connection to

the jump matrix 1 on a continued contour, is not consistent with the colliding wave
equations. The equality in (75) is reached for impulsive waves:
1/2b = 1

1/2 =

1
2

1
2

k1/2 =

n1/2 = 2.

(76)

4.5. Integral equations for the RHP

From the jump equation (26) and the representation (46) we can derive the scalar jump
equation
+ = + + ().

(77)

Using the additive jump function (k 0 , f, g) we can express as the Cauchy integral


1
1
1
() = 1 +

(k 0 , f, g)d0 .
(78)
2i 0 0 + 1
ffl
With the Cauchy principal value the inner and outer limit of an integral

1
(0 )d0
I() =
(79)
2i C 0
can be represented as
I+ () =

1
2i

(0 )d0 1
+ (),
0
2

I () =
0

1
2i

(0 )d0 1
().
0
2

Insertion into (77) yields with 1/2 := |(1/2) and F (, ) := ( )1 (0 + 1)1 for
(1) the integral equation

1
1
1
0
0
0
1+
F (, )1 ( )d + 1 () +
F (, 0 )2 (0 )d0
2i (1)
2
2i (2)



1
1
1
0
0
0
0
0
0
= 1+
F (, )1 ( )d 1 () +
F (, )2 ( )d
(80)
2i (1)
2
2i (2)



1
1
1
0
0
0
0
0
0
+ 1 +
F (, )1 ( )d + 1 () +
F (, )2 ( )d
2i (1)
2
2i (2)
and a similar relation for (2) . These integral equations can be evaluated by an
expansion in Chebyshev polynomials, which corresponds to step (iia) in figure 2. But
due to the discontinuities of the jump matrix J, the scalar solution has divergences at
the contour endpoints which recur also in the additive jump functions 1/2 . Therefore
such an expansion has a bad convergence and the transformation to a continuous RHP
becomes contemplable.

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

15

5. Transformation to a continuous Riemann-Hilbert problem

5.1. Concept of transformation
The transformation to a continuous Riemann-Hilbert problem is inspired by a recipe
described in [11], where a jump matrix discontinuity is removed through multiplication
with an appropriate branch cut perpendicular to the contour. In our RHP, we can
simultaneously remove the two discontinuities at the endpoints of a partial contour
1/2 1
1/2
featuring a branch cut along (1/2) . They
and L1/2
(1/2) using the functions L1/2
contain the fractions
+ 2
1 +
, L2 :=
.
(81)
L1 :=
1
2

1/2
1/2
We use L1/2
and L1/2
as well as their inverses as functions only in the -sheet with
real value at = 1, where we regard them to have a jump on the contour (1/2) . The
inner and outer limits at the contour are

(L11 )+ = ei1 |L11 |,

(L11 ) = ei1 |L11 |,

(1) ,

(82)

(L22 )+ = ei2 |L22 |,

(L22 ) = ei2 |L22 |,

(2) ,

(83)

1/2 1
L1/2
.

1/2
L1/2
()

L1/21/2 .

and analogous for

This implies
=
We restrict ourselves for
the majority of this paper to non-impulsive waves by demanding
1
2

< 1/2 < 2 2

(84)

1/2
1/2
become the
and L1/2
and hence excluding the case 1 = 21 2 = 21 , where L1/2
inverse of each other.
As in [11] we demand the jump matrix J of the initial RHP to be Lipschitz
continuous at the endpoints of . Thus we can state for later reference

lim ( 1/2 )x (J J(1/2 )) = 0 for |x| < 1.

1/2

(85)

5.2. The extended Riemann-Hilbert problem

We introduce the extended RHP (eRHP)
+ = g

(86)

with the slightly modified jump matrix

!

+
g=
, , R

(87)

defined on the whole real -axis, denoted by < , and featuring the properties
det g =
+ 2 2 = 1,

() = , () = , () = .

(88)

For 6= 0 the eRHP jump matrix g is neither unitary nor symmetric in any more.
The eRHP jump induced by the RHP jump matrix J is denoted as
(
J
,
gJ :=
(89)
1
else.

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

16

In the following subsections we will describe transformations like g g 0 by stating the

new jump functions 0 , 0 and 0 as functions of the old ones. The relations (88) remain
valid in all cases.
Note that already the special case |(2) = const, |< \(2) = 1, = 0 = features
two independent scalar solutions = L22 and = L22 1 . This ambiguity is connected
with the discontinuities of the jump function at 2 . Within the transformation
to the cRHP, at each partial contour there arise two different ways of removing the
discontinuities, and so this ambiguity can be explained.
For clarity of notation, we treat the eRHP without normalisation. We will derive a
scalar RHP solution and calculate E = (1)/(1) with the right normalisation (9).
5.3. Rotation transformation
We define a rotation transformation, which converts (86) to 0+ = 0 g 0 by
!
cos

i
sin

0 = R, g 0 = R1 gR ,
R =
.
i sin cos

(90)

The scalar solution and the Ernst potential transform as

0 = cos i sin (),

E 0 = cos E i sin .

(91)

Note that if was normalised and we normalise 0 according to (55) by

1
1
,
(92)
00 = T 0 , T = diag{
},
cos + i sin E cos i sin E
we get
cos E i sin
.
(93)
E 00 =
cos i sin E
This is the corresponding Ernst potential for a metric of the form (1) after a clockwise
rotation of the x-y-plane by an angle ,
!
!
!
00
x
cos sin
x
=
.
(94)
00
y
sin cos
y
Secondly, (93) is exactly the result of an Ehlers transformation E 00 = E/(1 i tan E)
with subsequent normalisation in virtue of (9).
The jump functions transform under (90) as
=0 = cos(2)= + sin(2),
0 = sin(2)= + cos(2),

<0 = <0 ,
0 = .

(95)

With (72) and (74) the boundary values of the RHP jump functions transform to
0
1b
= cos(21 ) i cos(A + 2) sin(21 ),

0
1b
= + sin(A + 2) sin(21 ),

(96)

0
2b
= cos(22 ) + i cos(B + 2) sin(22 ),

0
2b
= sin(B + 2) sin(22 ).

(97)

Thus the clockwise coordinate rotation in the x-y-plane by an angle corresponds to

a counterclockwise rotation of Ab and Bb in the complex plane by an angle 2. If

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

17

A B = n, n R, then the RHP jump matrix can be diagonalized at all 4 contour

endpoints simultaneously, which leads to tremendous simplifications in the transition to
a continuous RHP. We will call this case initially parallel polarised GW.
By convention we choose for the diagonalisation of J1/2b the rotation matrices R1
and R2 with
1 := ( A )/2,

2 := ( B )/2.

With these transformations we can achieve

!
2i1
e
0
g 0 |(1 +0) =
or g 0 |(2 0) =
0
e2i1

(98)
e2i2
0
2i2
0
e

!
.

(99)

From now on we use our freedom of a rotation in the x-y-plane to choose coordinates so
that the jump matrix is initially diagonal at 2 , i.e. gJ |(2 0) = diag(e2i2 , e2i2 ).
5.4. Singularity transformation
+ =
g by
We define a singularity transformation, which converts (86) to
I
= S1/2

I
I
g = (S1/2
)1 gS1/2+
.

(100)

Herein I is an index which takes the values e and o designating the two possibilities
of using either an even and an odd singularity transformation matrix,
!
!
11/2
11/2
L
0
0
L1/2
1/2
e
.
(101)
S1/2
or So1/2 :=
:=
1/2
1/2 1
0
L1/2
0
L1/2
Evaluation of the inner and outer limits similar to (83) leads to
i
1h
xI1/2
xI1/2
xI1/2
xI1/2
) ,
|L1/2 |
) + (|L1/2 |
+ |L1/2 |
(|L1/2 |
=
2
h
i
1
xI1/2
xI1/2
xI1/2
xI1/2

=
(|L1/2 |
|L1/2 |
) + (|L1/2 |
+ |L1/2 |
) ,
2
(
e2i1/2
(1/2) ,

=
(- associated with index 1)

else

(102)
(103)
(104)

with xe1/2 = 21/2 2 and xo1/2 = 21/2 1.

Considering (85) due to |x+ | < 1 we get by applying the singularity transformation
to a RHP jump matrix diagonalised at 2 :
I 1
I
gJI := (S2
) gJ S2+
,

gJI (2 ) = 1.

(105)

The jump matrix gJI is continuous at 2 , but not necessarily Lipschitz continuous,
whereas at 1 the jump matrix is still Lipschitz continuous. However, gJI is no longer
unitary, so another type of transformation is necessary to restore the unitarity of the
jump matrix at 1 so that it can be diagonalised by a rotation transformation and
made continuous by a singularity transformation.

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

18

5.5. Unitarisation transformation

+ =
g by
We define a unitarisation transformation, which converts (86) to
!
I I
w

0
= U I , g = (U I )1 (1 )gU I , U I :=

+
0
(wI I )1

(106)

where the constituents of the unitarisation matrix U I are defined as

I

Iu
i arg I+/ (1 )
u
I
I

I (),
,

:=
=

w
:=
e
. (107)
I :=
+
+/
I
I
+ u
+
u
The constant phase factor wI compensates the phase of I in 1 . The jump functions
are mapped to


= 12 (|I |2 + |I |2 ) + (|I |2 |I |2 ) ,
I 2 2i arg I+
 I2

)e

=
(w
.
(108)
+
1
I
2
I
2
I
2
= 2 (| | | | ) + (| | + | | ) ,
If we choose Iu so that |I (1 )|2 = |L2 (1 )|x+ , the unitarisation transformation
applied after removing the discontinuities at 2 yields
gJI := (UI )1 gJI U+I ,

gJI |(1 0) = 1,

gJI |(1 +0) = J(1 ).

(109)

Hence the unitarisation transformation reproduces the initial settings at 1 with gJI
still Lipschitz continuous at these points. Furthermore gJI 6= 1 almost everywhere on <
is no longer expressible in a closed form.
and so the matrix solution
5.6. The full transformation formula
We can now diagonalise the jump matrix gJI at 1 by the rotation transformation
I

gJ0 := R1
gJI R1
1

(110)

and remove the discontinuities there by a singularity transformation analogous to the

procedure at 2 . In summary, we get the cRHP
IK IK
IK
+ = gc ,

gcIK C 0

(111)

from the eRHP (86) by applying the transformation

IK := S2K U K R1 S1I ,

(112)

K 1
K
I
I 1 1
) gJ S2+
U+K R1 S1+
.
gcIK := (S1
) R1 (UK )1 (S2

(113)

The jump matrix gcIK of the cRHP depends in contrast to gJ on the coordinates f and
g. This is an interesting similarity to the treatment of Alekseev and Griffiths [9], where
the non-analytic behaviour of the solution at the wavefronts is handled by dynamical
monodromy data and generalised integral evolution equations.

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

19

5.7. The degree of the solution row vectors

We fix a point p on the imaginary axis of the -plane (one may think of p = i) and
define a p -regular function as a function which is only allowed to have poles or zeros
in p . Furthermore, we define the degree of a p -regular function f () as

f () has pole of order n in p ,

n
degree of f () := 0
0 6= f (p ) 6= ,

n
f () has zero of order n in p .
The same definition applies to matrices, keeping in mind that a matrix has a pole where
one element has pole and zero where all elements have a zero.
According to [11] a two-dimensional cRHP has a fundamental matrix = (1 ; 2 )
characterized by the p -regular and linearly independent solution row vectors 1 and
2 having minimal degree 1 and 2 , respectively. From the fundamental matrix all
solution vectors can be constructed as linear combinations. It is shown in [11] to have
the following two properties:
det 6= 0 6= p ;
0 < ( p )

1 +2

(114)

det < .

(115)

Thus we can conclude that det is p -regular with degree D = 1 + 2 . Furthermore,

due to det gcIK = 1 the determinant of the cRHP solution has no jump on < . In
consequence D = 0 = 1 + 2 .
5.8. The fundamental matrix of the cRHP
From (87) with (88) we can derive on < the identities
1 = 1 gcIK ()1 gcIK ,

1 = 3 gcIK 3 gcIK .

(116)

3 fulfil the
IK ()
Inserting (111) for gcIK () and gcIK we see that IK ()1 and
same jump equation as IK . This statement holds already for row vectors. Therefore
from an arbitrary p -regular solution vector 1 of the cRHP with minimal degree 1 we
can construct another p -regular solution vector with degree 1 represented by a single
scalar function IK
11 ,


p +
IK
1 IK
IK
1 IK
.
(117)
IK
Lp :=
1 = 11 , Lp 11 () := 1 + Lp 1 ()1 ,
p
Within the summation, no new zeros can arise because 1 has already minimal degree.


1 IK
From defining the matrix IK := IK
1 ; Lp 1 ()3 and calculating


2
IK 2
det IK (0) = |IK
|
+
|
|
6= 0,
(118)
11i
11a

IK
we see that IK
and Lp 1
1
1 ()3 are both linearly independent p -regular solution
we can write the
IK ()
vectors with degree 1 . Thus 1 = 2 = 0 and with IK :=
11

fundamental matrix of the cRHP in normal form with the scalar function IK :
!

IK ()
IK ()
IK
=
.
IK
IK ()

(119)

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

20

The row vectors constituting the fundamental matrix feature neither zeros nor poles.
IK IK
is equivalent to the scalar jump equation
Due to (88) the jump equation IK
+ = gc
IK
IK
IK
+ = + ( )+ ().

(120)

5.9. The normal form solution of the RHP

We now gradually revert the transformation (112) and ensure in each step the normal
form of the matrix solution. In the even case the first partial inverse transformation
0 eK := eK (S1e )1 yields directly a matrix in normal form with
0

eK

= L11 1 eK .

(121)

In the odd case, oK (S1o )1 is not in normal form, but we can obtain a solution 0 oK
0 oK 0 o
to the jump equation 0 oK
+ = gJ in normal form with the scalar function
0

oK

= L11 1 (1 + L1 )oK

(122)

via the linear combination 0 oK := oK (S1o )1 + 3 oK ()(S1o )1 ()1 . Note that if

we had defined the odd transformation by the alternative matrix
!
1
L
0
o
1
S1o :=
= L1
(123)
1 S1
0
L11 1
instead of S1o , we would end up with the same normal form solution after inverse
transformation.
1 yields directly a matrix
IK = 0 IK R
The second partial inverse transformation
IK in normal form with

IK
IK
IK = cos 1 0 + i sin 1 0 ().

(124)

IK =
IK (U K )1 yields a matrix
IK in normal
Likewise the inverse transformation
form with
IK = (wK K )1 IK .

(125)

At last, after inverse transformation with (S2K )1 we obtain, analogous to (121) and
(122), a solution IK to the initial RHP in normal form with scalar function
Ie = L22 1 Ie ,

Io = L22 1 (1 + L2 )Io .

(126)

In summary, via the cRHP we obtain the 4 matrix solutions ee , oe , eo and oo in

normal form with the scalar functions given in terms of the cRHP solutions IK as


1 ee
ee () = L22 1 (we e )1 cos 1 L11 1 ee + i sin 1 L1
() ,
1


oe
1
oe () = L22 1 (we e )1 cos 1 L11 1 (1 + L1 )oe + i sin 1 L1
(1 + L1
1
1 ) () ,


1 eo
eo () = L22 1 (1 + L2 )(wo o )1 cos 1 L11 1 eo + i sin 1 L1
() ,
(127)
1
oo () = L22 1 (1 + L2 )(wo o )1


oo
1
(1 + L1
cos 1 L11 1 (1 + L1 )oo + i sin 1 L1
1
1 ) () .
The solution of the four scalar jump equations (120) and the construction of these RHP
solutions is subsumed in step (iia) in the solution scheme figure 2.

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

21

6. Regularity conditions for solutions of the linear problem

6.1. Construction of the solution to the linear problem
Out of the RHP solutions (127) we construct the linear combination
= oo + diag(
p, p)eo + diag(
q , q)oe + diag(
r, r)ee ,
oo

eo

oe

ee

= + p + q + r ,

(128)
(129)

and the LP matrices

U := f 1 ,

V := g 1 .

(130)

The -coefficients p, q and r are functions of the coordinates f and g. By arranging these
coefficients to make U holomorphic in C \{} and g 1 holomorphic in C \{0}, we
obtain after normalisation a solution LP of the normalised LP.
From the property (52) of the LP follows as a necessary condition, that the negative
determinant D of must be independent of . Because of det gJ = 1 and the absence
of poles in IK , this is the case if D has no poles in 1/2 . Since D () = D it is
sufficient to examine the points 1/2 .
6.2. Regularity condition for the determinant det at 2
In order to derive a first condition for the -coefficients from the -independence of D ,
we collect the constituents of regular in 2 as


2e = (we e )1 {cos 1 L11 1 (1 + L1 )oe + q 1 ree


oe
1
ee
1
+i sin 1 L1
(1 + L1
(131)
1
1 ) () + q r () },
2o = (wo o )1 {cos 1 L11 1 [(1 + L1 )oo + peo ]


oo
eo
1
+i sin 1 L1
(1 + L1
1
1 ) () + p () }.

(132)

The corresponding normal form matrices e2 and o2 are solutions of I2+ = I2 gJI .
Because of gJI (2 ) = 1 these scalar solutions have no jump at 2 :
I
I
2+
(2 ) = 2
(2 ).

(133)

The scalar solution of the initial RHP can now be expressed as

() = L22 1 [(1 + L2 )2o + q2e ] .

(134)

The negative determinant D is regular (i.e. non-singular) in 2 if and only if the

prefactor of L2 in D vanishes at 2 . Since (133) this is equivalent to
(2 +
2 )|2o (2 )|2 = 0,

2 := (2o (2 ))1 [2o (2 ) + q2e (2 )] .

(135)

6.3. Regularity condition for the determinant det at 1

We introduce for the prefactors of the scalar solutions IK the notation
H e := L22 1 (we e )1 ,

H o := L22 1 (1 + L2 )(wo o )1 .

(136)

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

22

Due to the definition of wI and I we have at 1 :

H e (1 ) = 1 = H e (1 )
,
1/2
1/2
H o (1 ) = L12 + L12 = H o (1 )

L12 := L1 (2 ) = L2 (1 ) =

1 + 2
.
1 2

(137)

In particular H e and H o have no jump at 1 . We define the following constituents of

regular in 1 :
e
1p
:= H o eo + pr H e ee ,
o
1p
:= H o oo + qH e oe ,

e
:= H o eo () + pr H e ee (),
1m
o
1m
:= H o oo () + qH e oe ().

(138)

Because of gcIK = 1 the scalar cRHP solutions IK have no jump in 1 and with (137)
we obtain
I
I
I
I
(1 ).
1m+
(1 ) = 1m
(1 ) = 1p+
(1 ) = 1p

(139)

I
the scalar RHP solution can be expressed as
Using 1p/m
o
o
1
() = cos L11 1 (1 + L1 )1p
+ i sin L1
(1 + L1
1
1 )1m


e
e
1
+p cos L11 1 1p
+ i sin L1
1m
.
1

(140)

Considering (139) the vanishing of the prefactor of L1 in D at the point 1 can be

shown to be equivalent to
 o

o
o
e
(1 +
1 )|1p
(1 )|2 = 0, 1 := (1p
(1 ))1 1p
(1 ) + p1p
(1 ) .
(141)
Unlike the situation at 2 , during the calculation of D out of (140), in principle terms
21 1
1
proportional to L2
and L1221 could occur. However, these terms with non1 , L1
integer exponent are associated with branch cuts, which has to lie on < since IK and
the discussed transformations were continuous everywhere else in the -plane. But since
det gJ = 1 the negative determinant D has no jump on < and such a branch cut is
excluded. In summary, (135) and (141) are the conditions on the -coefficients p, q and
r assuring that D doesnt depend on .
6.4. Construction of the LP matrices U and V
From the normal form of we derive
!

(
f ()
(

f ()()
+ f ()()
f ()
)
)

D U =
+ f ()(

f ()() f ()() f ()(

)
)

(142)

and an analogous expression for V .

Because gJ is only a function of k, the LP matrices U and V have no jump on
the contour. Taking also into account the absence of poles in IK ,
Taking into account the absence of poles in IK , the LP matrices U and V can
only have poles in 1/2 as well as = and = 0 respectively. Because gJ is only a
function of k, the LP matrices have no jump on < , and so they become holomorphic in
C \{} and C \{0} respectively, if we can arrange the -coefficients so that poles at
1/2 are prevented. Due to the symmetries of (142) it is again sufficient to investigate

only the points 1/2 . Thanks to U11
= U22
() and U12
= U21
() we only have to

consider U22 and U21 , the same applies to V .

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

23

We note that the exponents of L1/2 and hence the divergent behaviour at 1/2 is
preserved under the derivatives
L2f = f+g2 L2 ,
L1f = f+g1 L1 ,
L1g = 1 (f1 +g) L1 , L2g = 2 (f1 +g) L2 .

(143)

6.5. Regularity condition for the LP matrix U at 2

The vanishing of the prefactor of L2 in D U21

can be shown to be equivalent to


2

(144)
2f + (22 1) 2 2 (2o (2 ))2 = 0.
f +g
For 2o (2 ) 6= 0 this leads together with the analogous calculation for D V21 to

22
2
,
(ln 2 )g = (1 22 ) 2 .
(145)
f +g
f +g
The restriction to = 2 is enforced by setting k = 12 so that (145) can be read as
differential equations for all f and g. The system is integrable and has the solution
 1
22 1
( 2 + f )( 12 g)
2 = iC2
, C2 C
(146)
f +g
where (135) yields even C2 R. An exceptional solution to (144) is given by 2o (2 ) = 0
for all f and g.

The vanishing of the prefactor of L2 in D U22

at 2 can be shown to be equivalent
to

 o

2f (2 )
22
22
2f + (22 1)
2 + (2 +
2)
2
|2o (2 )|2 = 0.
(147)
f +g
2o (2 )
f +g
This equation is automatically fulfilled if (135) and (144) hold.
(ln 2 )f = (1 22 )

6.6. Regularity condition for the LP matrix U at 1

The evaluation of the LP matrix elements at 1 is similar, but we have to use additionally
(139) and its derivatives after identifying the L1 -prefactors and setting k = 21 . We get
the analogous regularity conditions


21
o
1f + (21 1)
1 (1p
(1 ))2 = 0,
(148)
f +g

 o

1pf (1 )
21
21
o
1f + (21 1)
1 + (1 +
1)
1
|1p
(1 )|2 = 0.
(149)
o
f +g
1p
(1 )
f +g
Because the jump matrix gJ depends only on k, the LP matrices U and V have no
jump across < and hence singularities associated with branch cuts are excluded as
discussed in section 6.3. Again, (149) is automatically fulfilled if (141) and (148) hold.
The combined regularity conditions at 2 , (135), (144) and (147), have the trivial
solution 2o (2 ) = 0 and similar the combined regularity conditions at 1 , (141), (148)
o
and (149), have the trivial solution 1p
(1 ) = 0. The non-trivial solutions are
 1

 1
21 1
22 1
( 2 + f )( 12 g)
( 2 f )( 21 + g)
2 = iC2
,
1 = iC1
,
C1 , C2 R. (150)
f +g
f +g

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

24

In each case 2 purely algebraic equations result for the 3 -coefficients p, q and r, which
have to be solved in step (iii) of the solution scheme in figure 2. Therefore a function of
f and g may be left free to choose in the LP solution LP .
6.7. Colliding wave conditions revisited
In a last step (iv) of the solution scheme, within the generated family of solutions to the
LP the Ernst potential matching the initial data has to be identified. For this solution
the colliding wave conditions (14) are already fulfilled because of the appropriate choice
of the initial data. Beyond this proper IVP solution, the the family of solutions with
the same jump matrix that results from the LP is also interesting. The fraction of the
induced family of colliding wave spacetimes which obeys the colliding wave conditions
generalises the proper IVP solution.
A full treatment of the resulting colliding wave conditions is beyond the scope of
this article, but we will investigate in some more detail the trivial solution 2o (2 ) = 0
of the regularity conditions at 2 . For f = 21 the first partial contour (1) vanishes and
setting 2o (2 ) = 0 the scalar LP-solution is due to (134) of the form
2 1 LP
LP
0 ,
2 = L2

C < : |LP
0 | < C .

(151)

For g = 21 ,   1 the contour in the k-surface contracts to the twofold covering of

[ 12 , 12 ]. There the difference ||J(k) J(k = 12 )|| is bounded due to the Lipschitz
continuity of J(k). In the limit g 21 the deviation ||J(k) J(k = 21 )|| of the
jump matrix from its boundary value J(k = 21 ) goes to zero, hence we conjecture LP
T
for
constant
jump
matrix
to have the same colliding wave limit as the solution LP
Tc
J(k) J(k = 21 ). Calculating E = (1)/(1) we get
r
1
1
lim1 [
gEg ( , g)] = 2(2 1).
(152)
2
2
g 2
This is not matching the initial values where we had defined 2 := |Bb | and this is
not even compatible with the allowed domain (15) of a colliding wave limit since we
started with (84). Therefore the trivial solution 2o (2 ) = 0 and in the same way
o
1p
(1 ) = 0 (where the corresponding scalar solutions LP
has to be examined in
1
coordinates diagonalising J(1 )) do not lead to correct colliding wave spacetimes and
should be excluded.
7. Conclusions
With the inverse scattering method and the subsequent transformation to a continuous
RHP we were able to construct a solution to the characteristic initial value problem of
colliding plane waves.
For a given set of initial values the crucial problem consists in the solution to
the integral equation belonging to (120), whereas the derivation of the jump matrix
from the initial data via the ODE (61)-(64) is possible with high numerical accuracy,

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

25

if not analytical. Although the jump matrix can be only approximated numerically
for general initial data, the transformation to the cRHP only depending on J(1/2 )
is given analytically by (112) and (113). The regularity conditions (150) adapting the
RHP solution to the LP are algebraic and finally left over degrees of freedom have to
be fixed by comparison with the initial data.
In special cases where a fully analytic treatment is possible, the fourfold ambiguity
contained in the solution to the discontinuous RHP and the possible remnant functional
degree of freedom in the LP solution should lead to the construction of families of
exact solutions. In this sense the described procedure serves as a solution generating
technique which may generalise existing colliding wave solutions and lead to insights
into the structure of colliding plane waves. A more rigorous treatment of the colliding
wave conditions for the family of spacetimes induced by the LP solutions is subject of
ongoing investigations.
For an impulsive wave the boundary value of the corresponding RHP jump matrix
takes the value J(i ) 1 (i = 1 for impulsive wave in region II, i = 2 for
impulsive wave in region III) which is invariant under rotation transformation and
unitarisation transformation. Hence these types of transformation can be used to set
the derivatives of the jump functions and to zero at i instead of their values.
After appropriate preparation the discontinuities in the eRHP can be removed by the
alternative singularity transformations Sie and (Sie )1 instead of Sie and Sio . The inverse
transformation leads directly (i.e. without linear combinations) to the construction of
four RHP solutions out of the sRHP solutions. Since i = 1 i = 12 the derivation of
the regularity conditions has to be recapitulated carefully for a spacetime with at least
one impulsive wave, but we expect a simplification in the end. Massive simplifications
in the described solution procedure occur also for initially parallel polarised waves.
Acknowledgments
This research was supported by the Konrad-Adenauer-Stiftung and the Deutsche
Forschungsgemeinschaft (DFG) through the Graduiertenkolleg 1532 Quantum and
Gravitational Fields.
References
[1] Griffiths J B 1991 Colliding Plane Waves in General Relativity (Oxford: Oxford Univ. Press)
[2] Griffiths J B and Podolsk
y 2009 Exact Spacetimes in Einsteins General Relativity (Cambridge:
Cambridge Univ. Press)
[3] Hauser I and Ernst F J 1989 J. Math. Phys. 30 872-87
[4] Hauser I and Ernst F J 1989 J. Math. Phys. 30 2322-36
[5] Hauser I and Ernst F J 1990 J. Math. Phys. 31 871-81
[6] Hauser I and Ernst F J 1991 J. Math. Phys. 32 198-209
[7] Hauser I and Ernst F J 2001 Gen. Relativ. Gravit. 33 195-293
[8] Alekseev G A and Griffiths J B 2001 Phys. Rev. Lett. 87 221101
[9] Alekseev G A and Griffiths J B 2004 Class. Quantum Grav. 21 5623-54

A continuous Riemann-Hilbert problem for colliding plane gravitational waves

[10]
[11]
[12]
[13]
[14]
[15]
[16]

Neugebauer G and Meinel R 2003 J. Math. Phys. 44 3407-29

Vekua 1967 Systems of Singular Integral Equations (Groningen: P. Noordhoff)
Khan K A and Penrose R 1971 Nature 229 185-6
Yurtsever U 1989 Phys. Rev. D 40 329-59
Griffiths J B 2005 Gen. Relativ. Gravit. 37 111928
OBrien S and Synge J L 1952. Commun. Dublin Inst. Adv. Stud. A, 9
Robson E H 1973 Ann. Inst. Henri Poincare 18 77-88

26