Proceedings of Institute of Mathematics of NAS of Ukraine 2004, Vol.

50, Part 1, 131–141

Applications of Symmetry to General Relativity

B. Kent HARRISON

Physics and Astronomy Department, Brigham Young University, Provo, Utah 84602, USA
E-mail: [email protected], [email protected]

The equations of general relativity are highly nonlinear partial differential equations and
require special techniques to solve exactly. Symmetry considerations have sometimes helped
to find solutions. Three examples will be considered here. The first one considers a scaling
symmetry in the equations for perfect fluid with a spherical symmetry, which is of historical
importance in understanding neutron stars and black holes. The second involves symme-
tries in the Ernst equation for stationary axially symmetric fields, which are related to the
methods worked out in the 1970s for finding solutions to that equation. The third example
is an ongoing investigation into critical gravitational collapse, a topic of current interest,
using a symmetry of the equations and other analytic techniques.

1 Introduction
The Einstein equations of general relativity are, in general, second order highly nonlinear partial
differential equations. It has been necessary to employ many different techniques in order to
obtain exact solutions of these equations, and many are now known. I will discuss here a few
examples of the use of symmetry in such investigations. First I will present some cases of
historical interest, and then I will treat, more extensively, a current ongoing research problem.

2 Spherically symmetric configurations

My first example has to do with spherically symmetric gravitational fields for a perfect fluid.
The first solution of the vacuum Einstein’s equations to be found was the Schwarzschild solution,

ds2 = −(1 − 2m/r)dt2 + (1 − 2m/r)−1 dr2 + r2 dθ2 + r2 sin2 θdφ2

which is the spherically symmetric solution for a point mass m and which later proved the
foundation for the discovery of black holes. In this equation G = c = 1. For a perfect fluid, the
equations for pressure P , density ρ, and mass m out to radius r take the Oppenheimer–Volkoff
form [1],


dP/dr = −(P + ρ) m + 4πr3 P /r(r − 2m), (1)
2
dm/dr = 4πr ρ (2)

in which there is assumed an equation of state P = P (ρ). For high densities we expect the
equation of state to take the form P = kρ, where k is a constant, which takes the value 1/3
for a Fermi gas of neutrons, protons, and electrons. In this case, one can see that equations (1)
and (2) admit a scaling symmetry. If we write a = m/r and b = 4πr2 ρ/3m and take k = 1/3,
these equations become [2]

rda/dr = a(3b − 1),

rdb/dr = b[3 − 3b − 4a(3 + b)/(1 − 2a)].
132 B.K. Harrison

Division of the second of these by the first now gives a first order equation for b(a). If x =
4πρ0 r2 /3, where ρ0 is the density at r = 0, then the boundary condition near r = 0 (x = 0) is
a = x + · · · and b = 1 − 8x/5 + · · · = 1 − 8a/5 + · · · . Phase plane analysis of this equation shows
that there is a stable focus and saddle points. It is clear that 2a < 1 always, ensuring that there
is not a black hole event horizon outside the star. Numerical solution (done here with MAPLE)
gives Fig. 1, where the focus is obvious.

0.9

0.8

0.7

b
0.6

0.5

0.4

0.3

0.02 0.04 0.06 0.08 0.1 0.12 0.14

a

Figure 1. b vs. a for large density.

Density decreases as one moves out through the star and one must use a different equation
of state, which means that the a–b trajectory must leave the focus and move down close to
the a-axis, ending at the point where the pressure equals zero at the outer edge of the star.
At that point one can determine the radius and mass of the star. But the oscillation at the
focus is reflected in the final values. The total mass of the star is an oscillatory function of
the central density and there is a maximum mass in the neighborhood of 1.4 solar masses, the
Chandrasekhar limit. Thus for masses greater than that there will be collapse. There are two
families of stable equilibria, white dwarfs and neutron stars, and this points the way toward the
existence of collapsed configurations or black holes [2].

3 Axially symmetric configurations

For my second example, I consider stationary (rotating) axially symmetric fields, which include
the Schwarzschild solution and the Kerr solution for a spinning point mass. The standard
approach uses the metric


ds2 = −f (dt + ωdφ)2 + ρ2 f −1 dφ2 + hf −1 dρ2 + dz 2 ,

where f , ω, and h are functions of ρ and z. We can define a new, conjugate variable ψ, often
called the twist potential, by (subscripts denote derivatives):

ωρ = ρf −2 φz , ωz = −ρf −2 ψρ .

The field equations now are

f (fρρ + fρ /ρ + fzz ) = fρ2 + fz2 − ψρ2 − ψz2 ,

f (ψρρ + ψρ /ρ + ψzz ) = 2fρ ψρ + 2fz ψz (3)

or, where E = f + iψ (the Ernst potential),

(E)∇2 E = (∇E)2 ,
Applications of Symmetry to General Relativity 133

the Ernst equation. One can easily calculate the symmetries of equations (3). (Reference [3]
has a similar calculation for the Einstein–Maxwell equations.) There are five symmetries: two
on the independent variables alone,

∂z , ρ∂ρ + z∂z (4)

and three on the dependent variables alone,


∂ψ , f ∂f + ψ∂ψ , f ψ∂f + (1/2) ψ 2 − f 2 ∂ψ . (5)

The symmetries (4), in particular the second (scale) symmetry, can be used to look for special
solutions of the Ernst equation (which probably have little physical significance.) The symme-
tries (5), combined and exponentiated, yield the expression

E
= (aE + ib)/(icE + d)

the Ehlers transformation [4, 5], where a, b, c, and d are real constants. If E is a solution of the
Ernst equation, then E
is also. Certain combinations of these variables and their derivatives are
potentials for the Ernst equation. W. Kinnersley and his colleagues, through extensive research
in the 1970s, pursued relations involving these potentials and higher symmetries of the Ernst
equation in the jet space, eventually leading to a method of generating new solutions of the
Ernst equation [6]. Alternate methods include the use of Bäcklund transformations, found by
the present author and by G. Neugebauer [7, 8], and several other techniques found in the late
1970s.

4 Critical gravitational collapse

The third, and main, example is that of critical gravitational collapse. This work is still un-
finished, but several interesting results have been obtained. About ten years ago M. Choptuik
published a paper considering a massless scalar field in spherically symmetric gravity [9]. The
aim was to investigate possible collapse to a black hole. He assumed a metric as follows:


ds2 = −α2 (r, t)dt2 + a2 (r, t)dr2 + r2 dθ2 + sin2 θdφ2 ,
t
with a time variable suitable near the origin T0 = 0 α(0, t)dt. If φ(r, t) is the scalar field and
where

Φ = φr , Π = aφt /α,

the Einstein equations are


Φt = (αΠ/a)r , Πt = r−2 r2 αΦ/a r ,


αr /α − ar /a + 1 − a2 /r = 0, (6)



ar /a + a2 − 1 /(2r) − 2πr Φ2 + Π2 = 0,
at /α = 4πrΠΦ,

where subscripts denote derivatives. Choptuik does not use the last equation, since it is consis-
tent with the others in the sense that calculation of the cross derivative of a from the last two
equations, with use of the other equations, gives the same value. By analogy to the Schwarzschild
case, he takes a → ∞ to be the criterion for a black hole.
134 B.K. Harrison

4.1 Results from numerical integration

Choptuik integrated the equations numerically, scaling the integration steps carefully, and found
several interesting features. (1) For some values of a parameter p in the initial conditions for the
scalar field black holes form, but for other values they do not. There is a critical parameter, p∗ ,
for which black holes first form. (2) The precisely critical behavior, in which p = p∗ , is universal,
√ √
holding for all types of initial data. He uses variables X = (2π)rΦ/a, Y = (2π)rΠ/a,
ρ = ln(kr), τ = ln[k(T0∗ − T0 )], and finds the existence of the universal critical solution X ∗ , Y ∗ ,
which occurs in the neighborhood of a = ∞ and r = 0. (3) There is a scaling behavior. If Z =
either X or Y , then this is Z ∗ (ρ − ∆, τ − ∆) = Z ∗ (ρ, τ ), or Z ∗ (ρ − τ, τ − ∆) = Z ∗ (ρ − τ, τ ). Thus
there is a periodicity in τ with period ∆, where ∆ is a universal constant, approximately equal
to 3.43, independent of initial conditions. This indicates that the critical solution is somehow
dependent only on the original equations, in which there is no natural length or time parameter;
∆ thus should naturally come out of the equations. Furthermore, the mass M of the black hole
is given by a power law M = c|p − p∗ |γ , where γ is another universal constant, approximately
equal to 0.37, similar to critical phenomena in thermodynamics.

4.2 Symmetry treatment of equations

The scaling behavior suggests that an analytical approach, using scaled variables, might be
useful. Choptuik already made the observation that the original equations (6) are invariant
under the scaling r → kr, t → kt. We first define
√ √ √
A = (2π)Φ, B = (2π)Π, C = α/a, σ = (2π)φ, U = a2 .
Then the basic equations become
A = σr , B = σt /C,
At = (BC)r , Bt = (AC)r + 2AC/r, Cr /C = (U − 1)/r, (7)

2 2

Ur /U + (U − 1)/r = 2r A + B , Ut /U = 4rABC.
We calculate the symmetries of equations (7) and find that the following are invariant variables:
ξ = rh(t), U , F = rA, G = rB, and P = Ch2 /h
, where h(t) is an arbitrary function of t. In
2
terms of Choptuik’s variables,


we 2have F = aX and G = aY . We note that α , which is −g 00 , is
(aC) , which has a factor h /h . So if we define a new time coordinate t0 with dt0 = −h
dt/h2 ,
2 2

we see that h = 1/t0 . We could make this definition of h, but it is more convenient to define it
in terms of Choptuik’s function T0 (t) as h−1 (t) = T0∗ − T0 (t) , where the star indicates critical
value. (This then gives the convenient boundary condition aP = 1 at r, or ξ = 0.) We now
express the equations in terms of the variables ξ = rh(t) and v = t. (We introduce a new name
for t to avoid complications with partial derivative notation.) For any function H, we have
Hr = hHξ and Ht = Hv + ξ(h
/h)Hξ . We also find it useful to define P = −ξQ, z = ln ξ,
τ = const − ln h. We now note that τ is the same as Choptuik’s τ and z = ρ − τ , so we can look
for the periodicity with period 3.43. In terms of derivatives with respect to z and τ (which do
not appear explicitly), the equations take the following nice form
Fτ = (F + QG)z , Gτ = (G + QF )z + 2QF, Qz = Q(U − 2),
2 2
Uz /U = 2F + 2G + 1 − U, Uτ /U = 4QF G + 2F 2 + 2G2 + 1 − U. (8)

4.3 Approaches to solution

We now comment on several avenues of approach. (1) If we assume no τ dependence, then the
last equation of (8) gives
U = 1 + 2F 2 + 2G2 + 4QF G,
Applications of Symmetry to General Relativity 135

and the first equation gives

F = k − QG,

where k is a constant. The fourth equation is satisfied automatically. We get

Qz = Q(U − 2)

and

−1
Gz = Q 1 − Q2 [U (2QG − k) − 2QG]

so we can write an equation for G(Q):

−1
dG/dQ = 1 − Q2 (U − 2)−1 [U (2QG − k) − 2QG], (9)

where U is given above in terms of G and Q. Equation (9) for G(Q) can be put in various forms.
However, I will not pursue those here, since we really want to search for the periodicity in τ .
In our search for periodic behavior in τ , we note that it can be exhibited either in linear
equations resembling simple harmonic oscillator equations – which would have variable ampli-
tude – or in nonlinear cases with a limit cycle. The nonlinear character of the equations suggests
that the latter is the case; indeed, C. Gundlach in a review paper [10], sketches qualitative limit
cycles – which, however, may actually change to nonperiodic behavior.
(2) In the Schwarzschild solution, α2 = 1/a2 = 1 − 2m/r, for constant geometrized mass m.
That suggests that near that solution, we might suspect that α ∝ 1/a, or that C ∝ α/a ∝
1/a2 = 1/U . For U large, then, C (or P or Q) should be small. So let us consider an expansion
of equations (8) in powers of Q. We first define V = U Q, which satisfies


Vz = V 2F 2 + 2G2 − 1 ,

with

Qz = V − 2Q.

We ignore the Uτ equation. Expand G, F , V :

G = a + bQ + · · · , F = c + dQ + · · · , V = f + gQ + · · · ,

where a, b, . . . , g are functions of τ . We get, from the expansion,

ċ = f (a + d), d˙ = (a + d)(g − 2), ȧ = f (b + c),

2 2
ḃ = (b + c)(g − 2) + 2c, g = 2a + 2c − 1,

where the dot represents differentiation with respect to τ . There is another equation for g which
we ignore. We will also assume f to be a constant. We solve the first and third equations for d
and b and substitute them and g into the second and fourth equations. We get


ä = ȧ 2a2 + 2c2 − 3 + f (ċ + 2c),


c̈ = ċ 2a2 + 2c2 − 3 + f ȧ.

We integrate these with MAPLE. For f = 20, we get oscillatory behavior with a period of
about 2.5, but not a perfect limit cycle in a and c (Fig. 2). For f = 200, we are closer (Fig. 3).
For f = 225, we get what appears to be a perfect limit cycle, with period about 1.8 (Figs. 4
and 5).
136 B.K. Harrison

10
4

c
c
5
2

–15 –10 –5 0 5 10 15

0 a
–4 –2 2 4
a –5

–2

–10

Figure 2. c vs. a for f = 20. Figure 3. c vs. a for f = 200.

10 10

c c

5 5

–15 –10 –5 0 5 10 15 –12 –10 –8 –6 –4 –2 0

a t

–5 –5

–10 –10

Figure 4. c vs. a for f = 225. Figure 5. c vs. τ (shown as t) for f = 225.

There are several problems associated with this calculation. We ignored the Uτ equation.
A closely related matter is that we ignored any τ dependence in Q. We also ignored the second
equation for g, which appears to be inconsistent with other equations, and we assumed f to be
constant. There is no particular obvious reason why the value of 225 for f should be special,
and the period obtained is not 3.4 but is roughly half that. Nevertheless, the periodic behavior
of a and c strongly suggests that we may be on the right track.
(3) In order to clear up some of the problems of the last treatment, we take our functions F ,
etc., to be functions of Q(z, τ ) and v (= τ ). Then for any function H, we have Hz = HQ Qz =
HQ Q(U − 2) and Hτ = HQ Qτ + Hv . If we put

β = 2F 2 + 2G2 + 1 − U,

then from the U equations we can find

UQ = βU/[Q(U − 2)]

and

Qτ = Q(U − 2)(4QF G + β − Uv /U )/β.

We substitute for Qτ in the two equations for F and G and get rather messy equations involving
Q, F , G, and the derivatives of F and G with respect to v and Q. We expand F and G in
powers of Q, with coefficients as functions of v. We write

U = a/Q + b + · · · , F = d + eQ + · · · , G = h + mQ + · · · .
Applications of Symmetry to General Relativity 137

We find, where the dot indicates d/dv = d/dτ ,

b = 2d2 + 2h2 − 1, d˙ = ha, ḣ = da,

and equations for e and m,

aė + eȧ = ha(b − 2) + 2ma2 ,

aṁ + mȧ = dab + 2ea2 .

Thus d, for example, obeys

˙ − a2 d = 0,
d¨ − ȧd/a

indicating exponential instead of oscillatory behavior. Inspection of the equations indicates that
the change results precisely from the inclusion of Qτ .
(4) We now note that it is more natural to consider an expansion in 1/U than in Q, since U
is the variable considered to be large near a black hole. Furthermore, it is very convenient to do
so since we have expressions for both Uz and Uτ (8). For any H(z, τ ), we convert it to H(U, v),
where v = τ . This gives


Hz = HU Uz = HU U 2F 2 + 2G2 + 1 − U

and


Hτ = Hv + HU Uτ = Hv + HU U 4QF G + 2F 2 + 2G2 + 1 − U .

The equation for Qz now becomes


QU U 2F 2 + 2G2 + 1 − U = Q(U − 2).

This expression can be used to replace QU in the F and G equations when it occurs. Those
equations now become


Fv + 4QU F GFU = QU 2F 2 + 2G2 + 1 − U GU + QG(U − 2),


Gv + 4QU F GGτ = QU 2F 2 + 2G2 + 1 − U FU + QF U.

It will be noted that Q appears as a multiplier in all terms except the v derivatives, and U
appears in most terms. We define R = QU and rewrite the equations as



RU /R = 2F 2 + 2G2 − 1 /[U 2F 2 + 2G2 + 1 − U ],


Fv /R = 2F 2 + 2G2 + 1 − U GU − 4F GFU + G(1 − 2/U ),


Gv /R = 2F 2 + 2G2 + 1 − U FU − 4F GGU + F.

We see that the integration of the first equation gives an additive function ln b(v) in ln R, or
a multiplicative function b(v) in R. This can
 be removed in the second and third equations simply
by redefining the time variable as w = b(v)dv. This removes the ambiguity in the v (= τ )
dependence of Q (= R/U ). Inspection now shows that it is advantageous to define x = F + G
and y = F − G. The equations are now simplified to



RU /R = x2 + y 2 − 1 /[U x2 + y 2 + 1 − U ],
xv /R = 2y 2 xU + (1 − U ) (xU − x/U ) + y/U, (10)
yv /R = −2x2 yU − (1 − U ) (yU − y/U ) − x/U.
138 B.K. Harrison

These are still exact. If they could be solved, then x, y, and R could be substituted into the Uz
equation to find the z dependence of U , then the τ (= v) dependence could be found from
the Uτ equation. The obvious approach now is to expand in powers of 1/U . Unfortunately, this
does not give oscillatory behavior for x and y, but exponential behavior instead, just as in the
expansion for small Q. Variations of approach do not solve the problem. It may be that we
have an essential singularity, or to say it another way, that we should assume x and y involve,
say, exp(−U ). It is not clear how to investigate this possibility.
We try to solve the x and y equations (10) by a brute force approach. We assume

x = φ(w)S(U ), y = ψ(w)Y (U ),

substitute into (10), and assume that φ and ψ satisfy equations that are purely functions of w,
as follows (prime indicates d/dw):

φ
= aψ 2 φ + bφ + cψ,
ψ
= dφ2 ψ + f ψ + gφ,

where a, b, c, d, f , g are constants. We substitute these equations into (10) and equate coeffi-
cients, thus obtaining six equations for the two functions S and Y . If we assume x2 + y 2 large,
we get R = U , and then we find some consistency in these six equations, provided that Y = cS,
gc = −1, a = −dc2 , and f = −b. If we solve for d, f , and g and substitute in the second
equation, we find

ψ
= −(a/c2 )φ2 ψ − bψ − φ/c.

√ √
If we let φ = c/ aφ0 and ψ = 1/ aψ0 , we see that we may scale a and c away. So just put
a = c = 1, and we get

φ
= ψ 2 φ + bφ + ψ,
ψ
= −φ2 ψ − bψ − φ. (11)

We get inconsistent equations for S(U ). One of these is

SU /S = (1 − U + b)/[U (1 − U )],

yielding

S = kU [U/(U − 1)]b

for some constant k. The other is

2SSU = 1/U,

yielding S 2 = ln U + const. We ignore the second equation and note that for large U , the first
gives approximately S = kU .
We explore the φ, ψ equations (11). In the phase plane of φ and ψ the origin is a critical
point, a center. We can actually solve (11) in the phase plane exactly, finding

φ2 + ψ 2 + 2b ln(1 + φψ) = λ2 ,

where λ is a constant. If b = 0 this is exactly a circle. Integration of the time equations

(Figs. 6, 7) shows that we get oscillations in w and a limit cycle, which is nearly circular for
small b. λ measures the amplitude. For small b and λ = 1 (fixed by the initial conditions),
the period is about 7.2. Calculation from the exact solution indicates that the period should
Applications of Symmetry to General Relativity 139

1 1

x 0.5 y 0.5

–10 –5 5 10 –1 –0.5 0.5 1

w x

–0.5 –0.5

–1 –1

Figure 6. φ (shown as x) vs. t for b = 0.1. Figure 7. ψ (shown as y) vs. φ (shown as x) for
b = 0.1.

be about 2π. The cubic terms in (10) appear to be proportional to both U and λ, suggesting
that λ scales with 1/U . These oscillations are for constant U . We really want constant z, the
original scaled variable. Attempts to include the time dependence of U produce the exponential
behavior seen before.
Another approach proceeds as follows. Let x and y in equation (10) be functions of w only
and write, temporarily, α = x2 + y 2 . Treat α as a constant and integrate the R equation. We
get, exactly,

R = (1 − (α + 1)/U )β ,

where

β = (1 − α)/(1 + α).

Expansion for large U gives approximately

R = 1 + (α − 1)/U.

Put 1/U = b, which is now to be treated as a parameter, and put this into the x and y equations
in (10) along with the expression for α, keep terms of order b, and get:

 
xw = x + b x2 + y 2 − 2 + y ,

 
yw = −y − b x2 + y 2 − 2 + x .

These equations have interesting behavior. If x2 + y 2 is small, the equations may be treated
as linear. We find then that the origin in (x, y) phase space is a center if 1 > b > 1/3. In
that region, if we start with small x and y they remain small and we get linear equations,
with period depending on b, and arbitrary amplitude. For b > 1, x and y grow and we get
a nonlinear oscillation. For large enough b, the period is approximately proportional to 15/b
and the maximum amplitude of x and y is unity (Figs. 8, 9).
Instead of using the approximate value of R, we can use the exact value (being careful to
take the absolute value of the quantity in parentheses before raising to the power. Then

xw = R[x + b(y − x)],

yw = −R[y + b(x − y)].

Integration of these equations gives some very nice limit cycles, but the period varies with the
choice of b and of the initial conditions (Figs. 10–12).
140 B.K. Harrison

0.015
0.8

0.01 0.6
x x
0.4
0.005
0.2

–20 –10 10 20 –20 –10 0 10 20

w w
–0.2
–0.005
–0.4

–0.01 –0.6

–0.8
–0.015

Figure 8. x vs. w, b = 0.8, approximate R. Figure 9. x vs. w, b = 5, approximate R.

0.1 0.1

x y
0.05 0.05

–4 –2 0 2 4 –0.1 –0.05 0 0.05 0.1

w x

–0.05 –0.05

–0.1 –0.1

Figure 10. x vs. w, b = 2, y0 = 0.1, exact R. Figure 11. y vs. x, b = 2, y0 = 0.1, exact R.

x
0.5

–4 –2 2 4
w

–0.5

–1

Figure 12. x vs. w, b = 2, y0 = 1, exact R.

4.4 Summary of results

In summary, we note several features in this treatment. First, the original equations have
been expressed with τ and the scaled variable z as independent variables. Second, in the τ -
independent case, the equations may be reduced to a single first order equation. The significance
of this case is still unclear. Third, changing the independent variables to w (= τ ) and U separates
the equations into three equations for R, x, and y and two for the derivatives of U – a convenient
form. Fourth, oscillatory behavior is suggested in four different ways. The amplitude in these
may scale as 1/U . Each way has unacceptable assumptions. Trying to remove the assumptions
produces exponential behavior. It may be that, as indicated by Gundlach [10] and as mentioned
Applications of Symmetry to General Relativity 141

earlier, there is an epoch of oscillation followed by an epoch of final nonoscillatory behavior,

which might be exponential. One would think, however, that there still should be a way of
finding the period ∆ = 3.43 analytically. This has not yet been done successfully.

Acknowledgements
The author expresses appreciation to Eric W. Hirschmann of the Brigham Young University
Department of Physics and Astronomy, who brought this problem to his attention and with
whom a number of useful discussions have been held.

[1] Oppenheimer J.R. and Volkoff G., On massive neutron cores, Phys. Rev., 1939, V.55, 374–381.
[2] Harrison B.K., Asymptotic behavior of cold superdense stars, Phys. Rev. B, 1965, V.137, 1644–1652.
[3] Harrison B.K. and Estabrook F.B., Geometric approach to invariance groups and solution of partial differ-
ential systems, J. Math. Phys., 1971, V.12, 653–666.
[4] Harrison B.K., New solutions of the Einstein–Maxwell equations from old, J. Math. Phys., 1968, V.9, 1744–
1752.
[5] Ehlers J., Les theories relativistes de la gravitation, Paris, CNRS, 1959.
[6] Hoenselaers C., Kinnersley W. and Xanthopoulos B., Symmetries of the stationary Einstein–Maxwell equa-
tions. VI. Transformations which generate asymptotically flat spacetimes with arbitrary multipole moments,
J. Math. Phys., 1979, V.20, 2530–2536.
[7] Harrison B.K., Bäcklund transformation for the Ernst equation of general relativity, Phys. Rev. Lett., 1978,
V.41, 1197–2000.
[8] Neugebauer G., Bäcklund transformations of axially symmetric gravitational fields, J. Phys. A: Math. Gen.,
1969, V.12, L67–L70.
[9] Choptuik M.W., Universality and scaling in gravitational collapse of a massless scalar field, Phys. Rev. Lett.,
1993, V.70, 9–12.
[10] Gundlach C., Critical phenomena in gravitational collapse, gr-qc/0210101.