A concise introduction to trapped surface

formation in general relativity

Donato Bini, Istituto M. Picone, CNR, Rome, Italy
Giampiero Esposito, INFN Sezione di Napoli, Naples, Italy

Abstract
Trapped surface formation in general relativity can be studied through a coupled set of
nonlinear equations, where various terms can be neglected, as was proved by a rigorous math-
ematical analysis of Christodoulou. This paper is devoted to a pedagogical synthesis of the
mathematical formalism employed in this analysis, i.e. the optical structure of general rela-
tivity investigated by Christodoulou, Klainerman and other authors after them.
arXiv:1705.01706v2 [gr-qc] 10 May 2017

1 Introduction
In the sixties, when the global techniques of differential topology were applied to investigate causal
structure and singularities in gravitational collapse and cosmology, the concept of trapped surface
was elaborated for the first time. In Ref. [1], a trapped surface is defined to be a compact
spacelike two-surface with the property that the null geodesics which meet orthogonally, lo-
cally converge in future directions. More precisely, this means what follows: we consider in a
4-dimensional spacetime (V, g) the compact spacelike 2-surface (i.e. a compact Riemannian 2-
manifold) which is a submanifold of a 3-dimensional spacelike submanifold M . Let n and be the
future unit normal to M and , respectively, with tangent to M . One can then define two null
vectors n which contribute to form the spacetime tensor
1
h = g + n n = g + (+ + ). (1.1)
2
These tensors are then used to define the null mean curvatures (in our paper we always use explicit
summations to achieve a clear distinction between 2-, 3- and 4-dimensional concepts)
4
X 4
X
h
= . (1.2)
,=1 =1

The surface is said to be a trapped surface in the spacetime (V, g) if both + and are negative
on , i.e., if the null geodesic congruences both converge [2, 3, 4].
Penrose asked himself the question whether it is reasonable to expect trapped surfaces to
develop at all in our actual universe, and he found that there can be no reason of principle against
a trapped surface developing. However, the proof of theorems on the formation of trapped surfaces
for solutions of vacuum Einstein equations was a much harder task, and was accomplished only
in 2009 by Christodoulou [5], with an appropriate choice of initial conditions, called the short-
pulse method. Later on, a simpler proof of trapped-surface formation was obtained in Ref. [6]
by enlarging the admissible set of initial conditions and relaxing the corresponding propagation
estimates.
The basic tool in such investigations is the geometry of double null foliations, which leads
eventually to the so-called optical structure of vacuum Einstein equations. Although the physics-
oriented community is by now familiar with Newman-Penrose formalism and solutions of the eikonal
equation in curved spacetime, the techniques used in Refs. [5, 6] are not yet widespread, and hence
we find it appropriate to summarize here some key concepts and results in such References. We
therefore consider a region D = D(u , v ) of a vacuum spacetime (M, g) spanned by a double null
foliation generated by the optical functions (u, v) that solve the eikonal equation
4
X
(grad, grad) = (g 1 ) ( )( ) = 0, = u, v (1.3)
,=1

1
and are increasing towards the future, so that u takes values in the closed interval [0, u ] and v
takes values in the closed interval [0, v ]. Let Hu be the outgoing null hypersurfaces generated
by the level surfaces of u, and let Hv be the incoming null hypersurfaces generated by the level
surfaces of v. The two-dimensional surfaces obtained by intersection of Hu and Hv are

Suv Hu Hv , (1.4)

and we denote by Hu (v1 , v2 ) the portion of Hu defined by v [v1 , v2 ], while Hv (u1 , u2 ) is the
portion of Hv defined by u [u1 , u2 ]. The two optical functions u and v make it possible to define
the vector fields with components given by
4
X
L 2 (g 1 ) u, (1.5)
=1

4
X
M 2 (g 1 ) v. (1.6)
=1

These are future-directed null geodesic vector fields, in that

L L = 0, M M = 0, (1.7)

the integral curves of L being the generators of each Hu , and the integral curves of M being the
generators of each Hv . Indeed, one has in arbitrary local coordinates [5]
4
X 4
X 4
X
g L L = 2 L u = 2 L u
,=1 =1 =1
4 4
!
X X
1 1
= 4 (g ) ( u) u = 2 (g ) ( u)( u) = 0, (1.8)
,=1 ,=1

where we have exploited the commutation of covariant derivatives of functions when torsion van-
ishes, the condition g = 0, the Leibniz rule to express
4
X h i
(g 1 ) ( u)( u) + ( u)( u) = (gradu, gradu) (1.9)
,=1

and the eikonal equation (1.1) for = u. The same holds with u and L replaced by v and M ,
respectively.
Once the geodesic vector fields L and M are at our disposal, we can define
1 2 1 1
= P4 . (1.10)
2 g(L, M )
=1 g L M

For small values of u and v , the spacetime slab D(u , v ) is completely determined by data along
the null characteristic hypersurfaces H0 and H e 0 corresponding to v = 0 and u = 0, respectively.
We assume that H0 can be extended [6] to negative values of v, and that the spacetime (M, g) is
Minkowskian for v < 0 and for all u 0. The double null foliation can be chosen in such a way
that the function defined in (1.8) obeys the condition

(0, v) = 1 v [0, v ], (1.11)

and one defines the pair (e3 , e4 ) of null vector fields such that

e3 M, e4 L, (1.12)

2
 g(e3 , e4 ) = 2 g(L, M ) = 2. (1.13)
Given a two-surface S(u, v) and an arbitrary frame (e1 , e2 ) tangent to it, one can define the Ricci
coefficients  
()()() = g e() , De() e() , , , = 1, 2, 3, 4. (1.14)

With the understanding that lower-case indices a, b take only the values 1, 2, and that for covariant
derivatives with a subscript D3 De3 , D4 De4 , the Ricci coefficients are completely determined
by the components
ab g(Da e4 , eb ),
eab g(Da e3 , eb ), (1.15)
1 1
a g(D3 ea , e4 ), ea g(D4 ea , e3 ), (1.16)
2 2
1 1
g(D4 e3 , e4 ), e g(D3 e4 , e3 ), (1.17)
4 4
1
a g(Da e4 , e3 ). (1.18)
2
For example, one has
4
X
ab = g (L ); (ea ) (eb ) . (1.19)
,,=1

In order to display the null structure equations, we have to consider also the trace-free parts of
eab , here denoted by bab (e4 ) and bab (e3 ), where we exploit the possibility to define, for
ab and
any vector field X,
ab (X) g(Da X, eb ), (1.20)
so that
ab = ab (e4 ),
eab = ab (e3 ). (1.21)
Moreover, we need , the induced covariant derivative operator on the surface S(u, v), the projec-
tion 3 (respectively 4 ) to S(u, v) of the covariant derivative D3 with respect to the vector field
e3 (respectively D4 with respect to e4 ), and the null curvature components

ab R(ea , e4 , eb , e4 ),
eab R(ea , e3 , eb , e3 ), (1.22)
1 1
a R(ea , e4 , e3 , e4 ), ea R(ea , e3 , e3 , e4 ), (1.23)
2 2
1 1
R(Le4 , e3 , e4 , e3 ), R(e4 , e3 , e4 , e3 ), (1.24)
4 4
where R is the Hodge dual of R. One then arrives at the null structure equations [6]
2
X
4 ab = ac cb 2ab ab , (1.25)
c=1

2
X
3
eab = ecb 2e
eac
eab
eab , (1.26)
c=1
2
X
4 a = ac (c ec ) a , (1.27)
c=1
2
X
3 ea = eac (c ec ) + ea ,
(1.28)
c=1

3 1 1 1
4 + (a ea )( a ea ) (a ea )( a + ea ) (a + ea )( a + ea ) + ,
e = 2e (1.29)
4 4 8 2

3
 3 1 1 1
+ (a ea )( a ea ) + (a ea )( a + ea ) (a + ea )( a + ea ) + ,
3 = 2e (1.30)
4 4 8 2
supplemented by the constraint equations
 
b 1 1 b 1
div(e4 ) = tr ( e) (e4 ) tr , (1.31)
2 2 2
 
b 1 1 b 1 e
div(e3 ) = tr e + ( e) (e3 ) tre + , (1.32)
2 2 2
curl = curle b 3 ) (e
= + (e b 4 ), (1.33)
1b b 1
K = + (e 4 )(e3 ) (tr )(tr
e), (1.34)
2 4
where K is the Gauss curvature of the 2-surface S. From these equations one gets in particular a
pair of equations which play a key role in the formation of trapped surfaces, i.e. [6]
2
X
1
4 tr + (tr )2 = bab bab 2tr , (1.35)
2
a,b=1

1 b 4 ) 1 (tr )(e
3 b + (tr
e)b =
b + 2e (e b 3 ) +
b F, (1.36)
2 2
where the tensor products in Eq. (1.34) have components obtainable from the general formula for
P2 P2
pairs of 1-forms C = a=1 Ca dxa and E = b=1 Eb dxb on the 2-surface S [5]
2
X
b ab = Ca Eb + Cb Ea gab
(C E) Cf E f . (1.37)
f =1

Note that the above expression can be written also as a symmetric-tracefree part (modulo a factor
of 2)
b = 2[C E]STF .
C E (1.38)

2 Approximate form of nonlinear equation

Following Ref. [6], it is instructive to outline an approximate treatment of the nonlinear equa-
tion responsible for trapped-surface formation. For this purpose, we assume that spacetime is
Minkowskian for v < 0 and all non-negative values of u. The values of v are restricted to the
closed interval [0, ], where is positive and small. The radius of the 2-surface S = S(u, v) is
denoted by r = r(u, v), i.e. |S(u, v)| = 4r2 , and r(0, 0) r0 . Further assumptions are as follows.
1
(i) For small values of , u and v approach their flat-space values u 2 (t r + r0 ) and v
1 dr
2 (t + r r0 ), while 1 and du 1.

is close to r2 , corresponding to the imbedding in flat space.

(ii) The value of tre

(iii) The right-hand side F of Eq. (1.36) can be neglected in a first approximation, as well as
2tr on the right-hand side of Eq. (1.35).
In light of these assumptions, Eq. (1.35) reduces to
d b 2,
tr || (2.1)
dv
which, by integration, yields
Z v Z v
b 2 (u, v)dv = 2 b 2 (u, v)dv.
tr(u, v) tr(u, 0) || || (2.2)
0 r(u, 0) 0

4
 b finding
Now we can multiply the exact form of Eq. (1.36) by ,

d b2 b 2 = F,
b
|| + (tr
e)|| (2.3)
du
while, by application of the Leibniz rule, adding and subtracting terms that make it possible to
exploit the assumptions (i) and (ii), we find
 
d 2 b2 d b2 dr b 2 b 2 tre 2 dr b
(r || ) = r2 || + 2r || = r2 || + + r2 F
du du du r du
    
b 2 tre 2 2 dr b F,
= r2 || + + 1+ + r2 F (2.4)
r r du

which yields, upon integration,

Z u
2 b 2 (u, v) = r2 (0, v)||
r || b 2 (0, v) + F (u , v)du . (2.5)
0
Ru
By virtue of the assumptions (i) and (ii) and of Eq. (2.4), the integral 0 F (u , v)du is negligible
in the slab D(u, ), and hence one obtains the approximate relation

b 2 (u, v) r2 (0, v)||

r2 || b 2 (0, v). (2.6)

As a next step, one freely prescribes the trace-free part of the extrinsic curvature along the initial
hypersurface [6], so that
b v) = b0 (v)
(0, (2.7)
for some traceless 2-tensor b0 . Hence Eq. (2.6) becomes

b 2 (u, v) r2 (0, v) b 2
|| |0 | (v). (2.8)
r2 (u, v)

Furthermore, since |v| and r(u, v) = r0 + v u, Eq. (2.8) reduces to

b 2 (u, v) r02
|| |b0 |2 (v). (2.9)
(r0 u)2

This formula can be now inserted into the right-hand side of Eq. (2.2), and leads to
Z v
2 r02
tr(u, v) |b0 |2 (v )dv + error term. (2.10)
(r0 u) (r0 u)2 0

It is now clear that the trace of the extrinsic curvature is never positive provided that
Z
2(r0 u)
< |b0 |2 (v )dv . (2.11)
r02 0

On the other hand, from Eq. (1.35) the condition for the initial hypersurface not to contain trapped
surfaces is Z
2
|b0 |2 (v )dv < . (2.12)
0 r0

The joint effect of majorizations (2.11) and (2.12) is that formation of trapped surfaces is expected
provided that the condition
Z
2(r0 u) 2
< |b0 |2 (v )dv < (2.13)
r02 0 r0

5
is fulfilled. Such a condition requires an upper bound of the form [6]
1
|b0 | . (2.14)

In order to control the error term F in (2.4), we need for some positive c [6]

2 dr  1   1 
tr
e+ = O( c ), + 1 = O( c ), = O 2 + c , = O( 1+c ), = O 2 + c . (2.15)
r du
Many optical structure equations (see Sect. 3) have curvature components as sources, and hence
one has to derive bounds not just for all Ricci coefficients , , , e and e, but also for all null
e,
curvature components , , , , e and .e In Ref. [5] Christodoulou obtained such estimates by
making the short-pulse ansatz for the initial data. This means that initial data are taken to be
trivial and that b0 satisfies, relative to coordinates v and transported coordinates along H0
d
(transport being taken with respect to dv ), the condition

1
b0 (v, ) = f0 ( 1 v, ), (2.16)

where f0 denotes a fixed traceless, symmetric S-tangent 2-tensor along H0 .

3 The optical structure equations

The aim of this section is to summarize the conceptual and technical framework leading to the 16
optical structure equations of vacuum Einstein equations, since their knowledge is not widespread,
and the notation used in the literature is sometimes a bit cumbersome, so that its potentialities
are hidden rather than being fully appreciated. P
We consider a spacetime manifold (M, g) with boundary, where the metric g = 4,=1 g dx
dx is taken to be a smooth solution of the vacuum Einstein equations
1
R g R = 0 = R = 0.
2
The past boundary of M is the future null geodesic cone Cp of a point p, and the initial data are
assigned on Cp . The future-directed null geodesics issuing from p are the generators of Cp [7],
while a timelike geodesic from p with tangent vector T at p is denoted by p . With the notation
of the Introduction, let us define the vector fields Z e4 and W e3 . If is a 1-form on Mp
such that
2
X 2
X X2 X2
m n m
(Z) = h m dx | Z i= m Z = (W ) = n W n = 0, (3.1)
m=1 n=1
xn m=1 n=1

we then say that is a S 1-form, which is therefore the specification of a 1-form intrinsic to Suv
for each (u, v). The Lie derivative of with respect to Z can be restricted to the tangent space
T Suv , and such a restriction is here denoted by LZ |T Suv . This is a S 1-form as well as . Related
geometrical concepts are as follows.

(i) A S vector field is a vector field X defined on Mp such that, at each point x Mp , X is tangential
to the surface Suv through x. This is therefore a vector field intrinsic to Suv for each (u, v).
M M
(ii) A type Tsq S tensor field is a type Tsq tensor field defined on p such that, at each x p
and each X1 , ..., Xs Tx M , one has

(X1 , ..., Xs ) q Tx Suv ,

6
and (X1 , ..., Xs ) = 0 if one of X1 , ..., Xs is either Z or W . One therefore deals with a type Tsq
tensor field intrinsic to Suv for each (u, v).
The work in Ref. [5] proves that, for any given S vector field Y , the Lie derivatives

LZ Y = [Z, Y ] and LW Y = [W, Y ]

are also S vector fields. One can therefore define the restricted Lie derivatives

LZ Y |T Suv LZ Y, LW Y |T Suv LW Y. (3.2)

As a next step, for a S tensor field of type Tsq , the Lie derivative LZ |T Suv is defined by considering
on each Hu extended to the tangent space T Hu according to the condition that it vanishes if
one of the entries is Z, and setting the Lie derivative of with respect to Z, when restricted to
the tangent space T Suv , equal to the restriction to such a tangent space of the usual Lie derivative
with respect to Z of this extension. In analogous fashion, the restriction to the tangent space T Suv
of the Lie derivative of with respect to W is defined by considering on each Hv extended to
the tangent space T Hv in such a way that it vanishes if one of the entries is W , and setting

LW |T Suv = restriction to T Suv of the usual Lie derivative with respect to W of this extension.

This method yields Lie derivatives which are, themselves, S vector fields of type Tsq . We write
hereafter
e
LZ |T Suv D, LW |T Suv D. (3.3)
In particular, if is a 0-form, i.e. a function f , one has
2
X X2
f e f
Df = Zf = Za a
, Df = W f = Wb b. (3.4)
a=1
x x
b=1

For any function f defined on Mp , we denote by duv f the S 1-form obtained by restriction to each
surface Suv of the differential df , i.e.

X2
f
duv f df |Suv = a
dxa . (3.5)
a=1
x

e derivatives, i.e. [5]

This operation commutes with the D and D

e uv f = duv Df.
Dduv f = duv Df, Dd e

b e4 = L, L
Set now (cf. Sect. 1) L b e3 = M . The tangent hyperplane Tp Hu to a given
bp ,
null hypersurface Hu at a point p Hu is given by all vectors X at p which are orthogonal to L
i.e. n o
Tp Hu X Tp M : g(X, L bp) = 0 , (3.6)

while the tangent hyperplane Tp Hv is given by

n o
b )=0 .
Tp Hv X Tp M : g(X, L (3.7)
p

Since Hu and Hv are null hypersurfaces in spacetime, their induced metrics are degenerate, while
the induced metric h on each surface Suv is Riemannian (i.e. positive-definite), and is a symmetric
2-covariant tensor field
X2
h= hab dxa dxb , hab = h(ab) .
a,b=1

7
 b p and a vector
Any vector X Tp Hu can be uniquely decomposed into a vector collinear to L
tangent to the surface Suv , i.e. (a being a real number)

b p + P X, P X Tp Suv .
X Tp Hu = X = aL (3.8)

If X and Y are any two vectors tangent to Hu at p, one has a simple relation between spacetime
metric g and induced metric h, i.e.

g(X, Y ) = h(P X, P Y ). (3.9)

Similarly, one has

b + X, X Tp Suv ,
X Tp Hv = X = aL (3.10)
p

and
g(X, Y ) = h(X, Y ) (3.11)
for any pair of vectors X and Y tangent to Hv at p.
The second fundamental form of a null hypersurface Hu is a bilinear form

u : T p Hu T p Hu R

defined by (see components in (1.15) and (1.19))

b Y ).
u (X, Y ) g(X L, (3.12)

It can be shown to be symmetric, because [5]

b [X, Y ]) = 0,
u (X, Y ) u (Y, X) = g(L, (3.13)

where X and Y are extended to vector fields along Hu which are tangential to Hu . It should be
b is tangential to Hu . One has
stressed that is intrinsic to Hu , because the vector field L

u (X, Y ) = u (P X, P Y ), (3.14)

and hence u is a symmetric 2-covariant S tensor field. Similarly, for the null hypersurfaces Hv
one defines
v : T p H v T p Hv R
such that
v (X, Y ) b Y ) = v (Y, X) = v (X, Y ).
e(X, Y ) g(X L, (3.15)

If is any 2-covariant S tensor field, we denote by the S tensor field of type (1, 1) (i.e. once
covariant and once contravariant) such that

h( X, Y ) = (X, Y ) X, Y Tp Suv . (3.16)

If e1 , e2 is an arbitrary basis for the tangent space Tp Suv , one has (unlike Ref. [5], we do not
use block capital letters for tensor components here, so as to avoid confusion with two-component
spinors for which A, B are a standard notation for unprimed spinor indices [1])
2
X 2
X
ea = ( )ba eb , ( )ba = ac (h1 )cb , a, b = 1, 2, (3.17)
b=1 c=1

which verify indeed the explicit form of Eq. (3.16), i.e.

2
X 2
X
hab ( )ac X c Y b = ab X a Y b . (3.18)
a,b,c=1 a,b=1

8
 If and T are symmetric 2-covariant S tensor fields, their product T is defined by

( T )(X, Y ) h( X, T Y ), X, Y Tp Suv . (3.19)

In arbitrary local coordinates for Suv , this formula reads as

2
X 2
X
( T )ab = hrs ( )ra (T )sb = (h1 )rs ar Tbs . (3.20)
r,s=1 r,s=1

The optical structure equations will involve (see below) the rescaled tensor fields (see (1.15)
and (1.19)-(1.21))
ab 1 ab ,
eab 1
eab , (3.21)
and a hypersurface version of divergence, curl and covariant derivative. More precisely, one defines
the covariant derivative intrinsic to Suv , for any pair X, Y of S vector fields, with the help of a
projection operator to the surfaces Suv , as given by [5]
1 1
(uv )X Y = X Y, V = V + g(V, e3 )e4 + g(V, e4 )e3 Tq Suv , (3.22)
2 2
 
for all V Tq Mp . Furthermore, the intrinsic divergence is defined by the formula

2
X
(uv div)a (uv )c ca , (3.23)
c=1

and one denotes by uv the area 2-form of Suv , with components

(uv )ab = (uv ) (ea , eb ), a, b = 1, 2. (3.24)

The latter concept is used to define the intrinsic curl of a S 1-form according to
2
X
1
uv curl (uv )cd (h1 )ac (h1 )bd (uv a b uv b a ) . (3.25)
2
a,b,c,d=1

Out of the area 2-form of Suv one can also build the twice sharp , defined as
2
X


ab
uv (uv )cd (h1 )ac (h1 )bd , (3.26)
c,d=1

and hence the wedge product of symmetric 2-covariant S tensor fields and T , i.e. [5]
2
X

ab
T uv (h1 )cd ac Tbd . (3.27)
a,b,c,d=1

Last, one considers

2
X

(, ) (h1 )ab a b , (3.28)
a,b=1
p
|| (, ). (3.29)
Since we have defined all concepts that are needed, we can now write down from Ref. [5], but
with our notation, the 16 equations expressing the optical structure of vacuum Einstein equations.
They read as follows:
Dh = 2, Dh e = 2e , (3.30)

9
 D = 2 , (3.31)
e
D e = 2
e e e, (3.32)


D = e , (3.33)
 
e =
De e + e , (3.34)
h i
= 2 2(, e) ||2 ,
De (3.35)
h i
e = 2 2(, e) ||2 ,
D (3.36)
1 1
K + (tr)(tr
e) (,
e) = , (3.37)
2 2

uv div uv d(tr ) + (tr ) = 1 , (3.38)


uv dive e ) + e e (tre
uv d(tr )e e
= 1 , (3.39)
1
uv curle
= e , (3.40)
2
uv curl = uv curl = uv curle
, (3.41)
h 1 i
) = 2 uv e
D(e e + 2e
+ uv e e + ( e+ e ) + h , (3.42)
2
h 1 i
e
D() e + 2 + (
= 2 uv + uv e + ) + h , (3.43)
2
 
e =
D e + e + 2duv

e, (3.44)
 
= e + 2duv .
De (3.45)

The first of Eqs. (3.30), and Eqs. (3.31), (3.33), (3.35), are propagation equations along the
generators of each Hu ; the second of Eqs. (3.30), and Eqs. (3.32), (3.34), (3.36), are propagation
equations along the generators of each Hv . Moreover, Eq. (3.37) is the Gauss equation of the
embedding of the surfaces Suv , with Gauss curvature K, in the spacetime manifold (M, g), while
Eqs. (3.38) and (3.39) are the Codazzi equations of such an embedding.

4 Open problems
As far as we can see, at least two outstanding problems deserve careful consideration:

(i) Despite the elegant proof obtained in Refs. [3] and [4] that the non-linear equations leading
to trapped surface formation can be studied by discarding some terms, it would be interesting to
solve them numerically without discarding any term, no matter how small it can be.

(ii) It would be interesting to apply optical-structure methods to extended theories of classical

gravity, and possibly to the quantum theory of black holes.

Acknowledgments
G. E. is grateful to the Dipartimento di Fisica Ettore Pancini of Federico II University for
hospitality and support.

10
References
[1] R. Penrose, An analysis of the structure of spacetime, Adams Prize Essay, Cambridge Uni-
versity (1965); Structure of space-time, in Battelle Rencontres, 1967 Lectures in Mathematics
and Physics, edited by C.M. DeWitt and J.A. Wheeler, pp. 121-235, Benjamin, New York,
(1968).
[2] Y. Choquet-Bruhat, General Relativity and the Einstein Equations, (Oxford University Press,
Oxford, 2009).
[3] J. M. M. Senovilla, Trapped surfaces, in Black Holes, New Horizons (World Scientific, Singa-
pore, 2013).
[4] I. Bengtsson, Some examples of trapped surfaces, arXiv:1112.5318 [gr-qc], in Black Holes, New
Horizons (World Scientific, Singapore, 2013).
[5] D. Christodoulou, The Formation of Black Holes in General Relativity (European Mathemat-
ical Society, Zurich, 2009).
[6] S. Klainerman and I. Rodnianski, On the formation of trapped surfaces, Acta Math. 208, 211
(2012).
[7] Y. Four`es-Bruhat, Existence theorem for certain systems of nonlinear partial differential equa-
tions, Acta Math. 88, 141 (1952); reprinted in Max Planck Institute for History of Science,
Preprint Series n. 480 (2016).

11