Are these quasi-normal modes?

Esses são modos quase-normais?

Gustavo Godinho 1, ∗
and Willians Barreto 1, 2, †
1
Centro de Ciências Naturais e Humanas, Universidade Federal do ABC,
Av. dos Estados 5001, 09210-580 Santo André, São Paulo, Brazil
2
Centro de Física Fundamental, Universidad de Los Andes, Mérida 5101, Venezuela

Abstract: We discuss how to extract numerically the expected lowest quasi-normal mode
(QNM) associated with the pressure anisotropy for a Bjorken flux evolution to equilibrium.
This QNM was easily decoded subtracting the hydrodynamical attractors and compared
with other authors calculations. After evolutions with transients close to the expected lowest
QNM the system goes to a tail (pure imaginary frequency) for late times. We analyze the
relevance of Navier-Stokes, second order and Borel attractors at each stage of the evolution,
which begins far-from-equilibrium and ends close to equilibrium.
Keywords: Numerical Holography; Numerical Relativity; Computational Physics.

Resumo: Discutimos como extrair numericamente o modo quase-normal (MQN) esperado

mais baixo associado à pressão anisotrópica para um fluxo de Bjorken evoluindo para o
equilíbrio. Este MQN foi facilmente decodificado subtraindo os atratores hidrodinâmicos e
comparado com os calculados por outros autores. Após evoluções com transientes próximos
ao MQN mais baixo esperado, o sistema vai para uma cauda (frequência imaginária pura)
para tempos tardios. Analisamos a relevância dos atratores de Navier-Stokes, de segunda
ordem e de Borel em cada estágio da evolução, que começa longe do equilíbrio e termina
perto do equilíbrio.
Palavras-chave: Holografia Numérica; Relatividade Numérica; Física Computacional.

How does hydrodynamics emerge from the non-

equilibrium regime? The current view is that hy-
drodynamics is a universal attractor [3], with dissi-
I. INTRODUCTION pative contributions even when local gradients are
large [4]. In the simulation of a QGP the hydrody-
The main purpose of this work is to show namics can be modeled using kinetic theory [5], [6],
how we can extract extra physical information [7] and holographic duality [8], [9], [10], [11]. Ki-
from an available and yet squeezed data. What netic descriptions show purely exponential decay of
a naked eye does not see, and from a naïve (fresh) nonhydrodynamic modes [12], while in holography
point of view. In current times a huge quantity these modes also feature an oscillatory behavior
of certified data is in the cloud which can be an- [4], [13], [14].
alyzed using Artificial Intelligence (AI) task-force From a practical point of view, our in-
algorithms, for example. That is not the case here, put data is the output of a holographic numeri-
but we will keep it in mind for future work based on cal model studied recently in [15], [16], [17], for
the our results. Instead we will implement a very the simplest model. That is the case of a Su-
simple idea to go deep into the data and learn. persymmetric Yang-Mills (SYM) plasma undergo-
The context is the quark-gluon plasma ing a Bjorken flow [18]. Our motivation was a
(QGP) formed in ultrarelativistic heavy-ion colli- recent work on homogeneous isotropization [19],
sions [1], to study out of equilibrium properties [2]. with non-hydrodynamical homogeneous QNMs1

1 QNMs are the response of black holes to perturbations in

∗ [email protected] different channels; are damped oscillations characterized
† [email protected] by complex eigenfrequencies.
Revista Brasileira de Física, Vol. 4, Nº 3, e202411040302
www.revistabrasileiradefisica.com | DOI: https://doi.org/10.5281/zenodo.14198452
Licença Creative Commons CC BY-ND 4.0 2
√
once they trespass some threshold to thermalize, (τ, x, y, ξ), where τ = − is the proper time
t2 z2
correlated with a stairway to equilibrium entropy. and ξ = 12 ln{(z + t)/(z − t)} = arctanh(z/t) is the
In a general initial setting out of equilibrium, the spacetime rapidity. In these coordinates all the
Bjorken flux clearly first goes to hydrodynamiza- hydrodynamic fields are exclusive functions of the
tion and then to thermalization, with a transient proper time3 τ . In terms of the Milne coordinates
violation of the energy conditions in some cases. the metric of the 4D Minkowski spacetime where
The oscillatory behavior in the pressure anisotropy the fluid is defined reads,
is observed particularly strong when the energy
conditions are transiently violated to finally go ds2(4D) = −dτ 2 + dx2 + dy 2 + τ 2 dξ 2 . (1)
to an apparent exponential decay. Are those os-
cillatory features related with QNMs? Why the This setting is a first approximation to the expand-
oscillations decay to a tail for all evolved initial ing quark-gluon plasma (QGP) formed in high-
data? Why plateaus in entropy are formed for the energy heavy-ion collisions near mid-rapidity, i.e.
Bjorken flow? Why the stairway to equilibrium en- close to the collision axis (the transverse expan-
tropy is not observed for the Bjorken flow? Here sion to the collision axis is completely neglected).
we try to give some answers or at least report cred- This basic model can be extended to consider an
ible evidence, analyzing results produced with the anisotropic (viscous) holographic fluid [21] (see
available numerical code of a known precision. also [15], [16] and references therein).
We organize this work as follows. For the
sake of completeness we present in section II the B. Correspondence
most salient features of the studied model (includ-
ing initial conditions), the holographic description
of the Bjorken flux, a general notion of hydrody- The holographic Bjorken flow of a relativis-
namic attractors, as for a non expert reader but tic and strongly coupled quantum fluid can be im-
interested in the field. In section III we describe plemented by considering that the 4D flat space-
the numerical code developed as the robot used time (1), where the fluid lives, is (up to a global
in this work. We present our numerical results in conformal factor) the boundary of a 5D curved
section IV to finally discuss them and conclude in spacetime asymptotically anti-de Sitter, the bulk.
section V. The simplest holographic model is the conformal
Here we use plus metric signature and nat- and strongly coupled SYM plasma. In some limit
ural units ℏ = c = kB = 1. the gauge/gravity duality states that the SYM in
4D Minkowski spacetime is dual to classical (Ein-
II. HOLOGRAPHIC BJORKEN FLOW stein) gravity in 5D.
The Ansatz for the 5D bulk metric can be
written using infalling Eddington-Finkelstein (EF)
A. Model coordinates as follows [26], [27]
h
A foundational model to study the matter ds2 = 2dτ [dr − A(τ, r)dτ ] + Σ(τ, r)2 e−2B(τ,r) dξ 2
created in relativistic nucleus–nucleus collisions is i
the Bjorken flow [18]. In this model the relativis- + eB(τ,r) (dx2 + dy 2 ) , (2)
tic fluid possesses the following features (symme-
tries). Taking the collision axis to be the longi- where r is the radial holographic direction, τ is the
tudinal z-axis, it is assumed boost invariance in EF time which reduces to the proper time in (1) at
that direction where the fluid is inhomogeneous the boundary (r → ∞), x and y are the coordinates
and rapidly expanding at the speed of light. In in the plane transverse to the beamline and ξ is the
the transverse xy plane the nuclei is assumed to rapidity in the longitudinal direction, as described
be homogeneous2 and of infinite extent, remov- in section II A. In such a way
ing thus all the dependence on the coordinates
x and y [20]. Bjorken symmetry is more eas- lim ds2 = ds2(4D), (3)
r→∞
ily handled by changing from Cartesian coordi-
nates (t, x, y, z) to the so-called Milne coordinates

3 The fluid expanding longitudinally in the static

2 Transverse distances are considered much smaller than Minkowski spacetime is equivalent to at rest fluid in the
the nuclear radii. longitudinally expanding (flat) spacetime.
Revista Brasileira de Física, Vol. 4, Nº 3, e202411040302
www.revistabrasileiradefisica.com | DOI: https://doi.org/10.5281/zenodo.14198452
Licença Creative Commons CC BY-ND 4.0 3
leads to the following boundary conditions obtained the following formulas for observables:

lim A → 12 r2 , (4a) ϵ̂(τ ) = −3a2 (τ ), (6a)

r→∞
3
lim B → − 32 ln τ, (4b) p̂T τ ) = −3a2 (τ ) − τ ȧ2 (τ ), (6b)
r→∞ 2
lim Σ → rτ 1/3 . (4c) p̂L (τ ) = 3a2 (τ ) + 3τ ȧ2 (τ ), (6c)
r→∞

In the EF coordinates infalling radial null

geodesics satisfy τ = constant, while outgoing ra- where ϵ̂(τ ), pT (τ ), pL (τ ) are, respectively, the
dial null geodesics satisfy dr = Adτ . By foliating (normalized) energy density, the transverse pres-
the bulk spacetime in slices of constant τ (null hy- sure, and the longitudinal pressure of the SYM
persurfaces) one implements a time evolution of plasma. Note that an overdot represents a time
the system according to the so-called character- derivative with respect to time. Once determined
istic formulation of general relativity (for a de- the time evolution of a2 (τ ) UV coefficient we have
tailed review including numerical issues see [27]). the dynamical evolution of the physical observables
For the original characteristic formulation involv- at the boundary4 .
ing asymptotically flat spacetimes see [28], [29]
(and for a practical review see [30]). C. Attractors

The line element (2) still has a residual dif-

A Bjorken flow is an ideal playground to test
feomorphism invariance under radial shifts, that
the current notion of hydrodynamics: attractors.
is, r 7→ r + λ(τ ), with λ(τ ) an arbitrary function
They drive in some stage the behavior for observ-
of time. In order to integrate in the radial direc-
ables such as the pressure anisotropy, which is dic-
tion the Einstein equations, one must consider the
tated by the decay of the nonhydrodynamic QNM.
entire portion of the bulk geometry causally con-
As proposed in [3] the Borel resummation of the di-
nected to the boundary. Using the residual diffeo-
vergent asymptotic gradient series defines a hydro-
morphism invariance we can deal with black holes,
dynamic atrractor, to which far-from-equilibrium
requiring that the radial position of the apparent
solutions would coalesce before converging to the
horizon remains fixed for all the time slices.
corresponding limits associated with finite order
Einstein’s equations are obtained –straight truncations of the hydrodynamic gradient expan-
forward– from (2) and solved numerically consider- sion, such as Navier-Stokes theory or the second
ing the boundary conditions (4) and for arbitrary (or higher) order hydrodynamic [16].
initial conditions. This was routinely done in char- For the pressure anisotropy of the SYM
acteristic numerical relativity [30]. plasma undergoing a Bjorken flow, the correspond-
ing analytical hydrodynamic expressions for the
What it is most interesting to consider is the Navier-Stokes (NS) regime, the second-order gra-
ultraviolet (UV) near boundary expansions of the dient expansion, and the Borel resummation of the
metric coefficients (reminiscent of the Bondi for- divergent gradient expansion are given by, respec-
malism to deal with gravitational radiation [28]), tively [21], [22], [23], [24]
and their relation to the holographically normal-
ized one-point Green’s function of the energy mo-
 
∆p̂ 2
mentum tensor of the boundary SYM gauge the- = , (7a)
ϵ̂ NS 3πωΛ
ory. From this procedure we extract the physi-  
cal observables of the strongly coupled fluid under ∆p̂ 2 2(1 − ln(2))
= + 2 , (7b)
consideration. Thus, the minimalist metric fields ϵ̂ 2nd order 3πωΛ 9π 2 ωΛ
reads:
 
∆p̂ −276 + 2530ωΛ
ϵ̂ Borel
=
3(120 − 570ωΛ + 3975ωΛ 2 ) . (7c)
a2 (τ )
A(τ, r) = 21 [r + λ(τ )]2 + λ̇(τ ) + + O(r−3 ),
r2
(5a) with ∆p̂ ≡ p̂T − p̂L defined as the pressure
B(τ, r) = − 32 −1
ln τ + O(r ), (5b)
1 + 3τ λ(τ )
Σ(τ, r) = τ 1/3 r + + O(r−1 ), (5c) 4 In some way the holographic correspondence can be con-
3τ 2/3
sidered an algorithm, with the universe as a classical com-
puter doing a simulation of a dual quantum computer; the
where λ(τ ) and a2 (τ ) have to be specified at τ = played role of a black hole is fundamental for thermaliza-
τ0 . After the holographic correspondence task, are tion and for the holographic realization itself.
Revista Brasileira de Física, Vol. 4, Nº 3, e202411040302
www.revistabrasileiradefisica.com | DOI: https://doi.org/10.5281/zenodo.14198452
Licença Creative Commons CC BY-ND 4.0 4
anisotropy and ωΛ the dimensionless time measure, ➔ Post-processing: Calculate the
scaled by the energy parameter Λ (as fixed in [25]) energy parameter Λ;
which in turn depends on each initial condition.
These expressions are interpreted as hydrodynamic ✓ Input: Read the output and
attractors [16], adapted to our holographic model. adjust observables and time;
✓ Output: Write the scaled
D. Initial conditions data;
✓ Output: Generate and save
In order to prepare a code to extract the ob- plots;
servables we require define the compactified holo- ➔ Create labeled directories with data
graphic coordinate and plots for analysis and post-
post-processing;
1
u≡ , (8) ➔ Erase temporary data.
r
in terms of which are defined the subtracted5 field
variables:

up Xs (τ, u) ≡ X(τ, u) − XU V (τ, u), (9) Also an extra automate tool was developed
to analyze and visualize the whole set of results,
where X denotes any of the metric functions, p is presented in the next section. The features of the
an integer, and XU V is some truncation of X. At auxiliary script written in Python are:
this point, and without loss of generality, we set
λ(τ ) = 0 for any time and for this work purposes.
Thus we use here the set of initial conditions (IC) • Input: Read two tables of temporal win-
for Bs and a2 as reported in [15] by Table I (also dows for all ICs;
used in [16] and [31])
• Task One:

III. AUTOMATE CODES ➔ For each IC:

✓ Prepare data: first subtraction
The numerical code used in [15], [16] was and fit;
automatized for this work. The original code ✓ Calculate the putative QNM
implemented the pseudo-spectral method to imaginary frequency;
solve the characteristic hypersurface equations, ✓ Output: Append each imagi-
and the Adams-Bathforth method to evolve the nary frequency;
initial/boundary conditions. The features of the
robot written in Python/Fortran are: ➔ Generate and save a table.
• Task two:

• Input: Read a set of parameters for any ➔ For each IC:

initial condition IC; ✓ Prepare data: second subtrac-
tion and interpolate values;
• Input: Read a set of ICs;
✓ Output: Write a vector;
• Wrapper: ✓ Calculate the putative QNM
For each IC (serial or parallel): complex frequency;
➔ Driver: Integrate the Einstein ✓ Output: Append each com-
field equations up to some time; plex frequency;
✓ Output: Write data at each ➔ Generate and save a target plot in
time slice; the complex plane for the bracketed
ICs within some tolerance.

The first and second subtractions are explained in

5 This is a common practice in characteristic numerical rel- the next section.
ativity [30].
Revista Brasileira de Física, Vol. 4, Nº 3, e202411040302
www.revistabrasileiradefisica.com | DOI: https://doi.org/10.5281/zenodo.14198452
Licença Creative Commons CC BY-ND 4.0 5
10.0 5
7.5
0
5.0

NS)
2.5 5

p/ ]
p/

0.0

[
10

ln( p/
2.5
5.0 15
7.5
0.0 0.5 1.0 1.5 2.0 2.5 3.0
20
0.0 0.5 1.0 1.5 2.0 2.5 3.0
FIG. 1. Pressure anisotropy evolution for 25 IC as used
in [15], [16] and [31]. FIG. 3. Subtraction of the NS attractor to the pressure
anisotropy ∆p̂/ϵ̂ − [∆p̂/ϵ̂]N S (shifted c). All the curves
displayed fit with ln(a) − bωΛ for late times.
2
0
Model NS 2nd Order Borel
2
1 0.786(4) 1.73(2) 2.07(2)
4 2 0.786(4) 1.73(3) 2.17(1)
ln( p/ )

6 3 0.786(4) 1.73(2) 1.97(2)

8 4 0.786(4) 1.73(2) 1.98(2)
10 5 0.786(3) 1.73(2) 1.83(2)
12 6 0.786(3) 1.73(2) 1.84(2)
14 7 0.786(4) 1.73(2) 1.96(2)
0.0 0.5 1.0 1.5 2.0 2.5 3.0 8 0.786(4) 1.73(2) 2.03(2)
9 0.787(5) 1.73(3) 1.91(2)
FIG. 2. Same as Fig. 1 in log scale. 10 0.786(4) 1.73(2) 2.07(2)
11 0.786(4) 1.73(3) 2.15(2)
12 0.786(4) 1.73(2) 2.11(2)
IV. NUMERICAL RESULTS 13 0.786(4) 1.73(2) 1.97(2)
14 0.786(4) 1.73(2) 1.99(2)
In the following sequence of results we begin
15 0.786(3) 1.73(2) 1.80(2)
by analyzing the pressure anisotropy as reported in
[15], [16] and [31] (see figure 1), for the set of 25 16 0.786(3) 1.73(2) 1.71(2)
IC, now for late times. Figure 2 displays the same 17 0.786(3) 1.73(2) 1.86(2)
results in log scale. Clearly for late times all IC 18 0.786(4) 1.73(2) 1.88(2)
decay apparently to the same tail. We subtract 19 0.786(3) 1.73(2) 1.78(2)
the attractors (7), one at a time, to determine the 20 0.786(3) 1.73(2) 1.75(2)
law of the decay rate for late times (this simple
21 0.786(3) 1.73(2) 1.65(2)
operation is justified in section V). Doing the sub-
traction, the best trial fit function for the three 22 0.786(3) 1.73(2) 1.59(2)
attractors and for all IC with ωΛ ∈ [2.0, 3.5], was 23 0.786(4) 1.73(2) 1.96(2)
(ae−bωΛ + c), not (ae−bωΛ ). All IC have the same 24 0.786(4) 1.73(2) 1.83(2)
decay (b) for the subtracted NS and second order 25 0.786(3) 1.73(2) 1.78(2)
attractors, as shown in Table I, within the indi-
cated numerical error. Figures 3 and 4 display the TABLE I. Decay parameter (b) fitted for each subtrac-
subtraction for NS and Borel. These results are tion.
expected, discussed and explained in the next sec-
tion. This is the first subtraction as commented in
section III. the revelation of transient oscillations with period
What turns out to be more interesting is and decay (numerically close), for a significantly
Revista Brasileira de Física, Vol. 4, Nº 3, e202411040302
www.revistabrasileiradefisica.com | DOI: https://doi.org/10.5281/zenodo.14198452
Licença Creative Commons CC BY-ND 4.0 6
5

0
Borel)

5
p/ ]
[

10
ln( p/

15

20
0.0 0.5 1.0 1.5 2.0 2.5 3.0

FIG. 4. Subtraction of the Borel attractor to the pres-

FIG. 6. Complex frequencies for the Borel subtraction,
sure anisotropy ∆p̂/ϵ̂ − [∆p̂/ϵ̂]Borel (shifted c). All the
and for a number of IC computed with Harminv. In
curves displayed fit with ln(a) − bωΛ for late times.
this target plot the blue center choice is for IC#20, and
the gold circle is for the adjusted frequency for IC#23.
Other (red) IC are: 3-6, 12, 14-25 (≈ 70% of the IC
10 set).
log( p/ [ p/ ] NS)
5 log( p/ [ p/ ] 2nd Order)
log( p/ [ p/ ] Borel)
Attractor)

0 lected data we use Harminv [32], a well known and

robust software to extract QNMs. This in turn re-
5
]

quires the interpolation for equal time-step in ωΛ ,

p/

10 for which we choose the standard Spline method.

[

The results are displayed in figure 6. Thus, we have

ln( p/

15 collected numerical evidence of a collective behav-

ior for the studied set of IC. But, are these really
20
QNMs? If so, are they related to other theoreti-
25 cally expected QNM? In the following section we
0.0 0.5 1.0 1.5 2.0 discuss our results.
V. DISCUSSION
FIG. 5. Subtraction of attractors to the pressure
anisotropy ∆p̂/ϵ̂ − [∆p̂/ϵ̂]Attractor for IC#23: NS (vio- It was a matter of logical (and simple) oper-
let), Second order (salmon), Borel (yellow). They are ation subtracts an attractor to capture the hidden
compared with a modeled (guessed) QNM with com- decaying oscillations, like the QNM. In this sense
plex frequency 14.45 − i12.75, shifted to fit visually we subtracted hydrodynamical modes, homogeniz-
(green dashed curve); the top line (red dashed line) is ing the Bjorken flux. Doing that, late time tail
just a reference for the decay rate. The dashed curve and transient QNM emerge and characterized nu-
in blue is the pressure anisotropy without subtraction.
merically. Subtracting an equilibrium function, an
attractor, it is reasonable to think that the best
one for late time is the NS attractor. This leads
number of IC. That behavior is more persistent in us to the approximate value for the decay b ≈ 0.79
time for the Borel attractor subtraction (see fig- for all IC. This can be interpreted as an imaginary
ure 4). To show that clearly we isolate the IC#23, frequency in the context of QNM analysis. Then it
for instance. Figure 5 displays the subtracted pres- was manifest, before the transition to a tail, tran-
sure anisotropy for this IC, compared with the non- sient QNM signals with complex frequencies. Vi-
subtracted pressure anisotropy and with a crudely sually, the decay rate and the period appeared to
guessed QNM. In consequence we define a second be similar for a good number of evolved IC, at least
task, that is, the systematic extraction of complex in some window (see Figs. 3 and 4). To visualize a
frequencies in specific windows of time for each IC representative evolution we select the IC#23 evo-
and for the most persistent signal, that is, for the lution. It was clear enough that the subtraction
Borel subtraction. This is the second subtraction of the Borel attractor revealed a more persistent
as commented in section III. To analyze the se- QNM that for other attractors (see Fig. 5). As
Revista Brasileira de Física, Vol. 4, Nº 3, e202411040302
www.revistabrasileiradefisica.com | DOI: https://doi.org/10.5281/zenodo.14198452
Licença Creative Commons CC BY-ND 4.0 7
displayed in Fig. 6, they are not exactly the low- (NS’s regime). If there exits a second order regime
est QNM but they are significantly close to. it is intermediary.
Our results motivated us to establish con- It seems that the function of the attrac-
nections with the results of other authors (includ- tor is to drive the system to the threshold of hy-
ing own work). Without surprise, it is good to drodynamization from the initial state far-from-
extract numerical evidence of the expected lowest equilibrium. The system chase the QNM, particu-
QNM for the Bjorken flux, from a rapid calculation larly when the energy conditions are being (or close
and a relatively low precision code. In [33], Yanik to be) violated, looking for the exit and evolve to
and Peschanski obtain the QNM for a scalar per- equilibrium. This could explain the formation of
turbation of a background 5-d planar black hole plateaus for entropy far-from-equilibrium. In the
geometry, which is ωstatic /πT = 3.1194 − 2.74667i, way to equilibrium the QNM do not persist for
calculated previously by Starinets [34] as a fun- late times, where a pure decay dominates, with-
damental QNM near extremal black branes dual out a stairway formation. It is interesting to ob-
to strongly coupled N = 4 SYM plasma at fi- serve that the plateaus for entropy are not in phase
nite temperature. Using the dual gravity descrip- with the QNM, anticipating the violation of energy
tion in [24], the authors calculate numerically the conditions as observed in previous works [15], [16],
form of the stress tensor for a boost-invariant flow [19].
in a hydrodynamic expansion up to terms with QNMs and tails are signatures of gravita-
240 derivatives. They identify the leading sin- tional radiation from binary black holes (see [36]
gularity in the Borel transform of the hydrody- for a recent work, for example), in the astrophys-
namic energy density with the lowest nonhydrody- ical context. Chesler and Yaffe [37], in the con-
namic excitation corresponding to a ‘nonhydrody- text of horizon formation and far-from-equilibrium
namic’ quasinormal mode on the gravity side. Ex- isotropization in a SYM plasma, pointed that non-
plicit gravity calculation for the lowest mode yield monotonicity is unsurprising because the late time
ωqnm = 3.1195 − 2.7467i, which agrees with the response is dominated by the lowest quasi-normal
frequency of the lowest nonhydrodynamic scalar (±9.8 − 8.7i)T [34]. They also point out, in the
quasinormal mode as calculated in [33]. Then they same context, the presence of infalling gravita-
reproduce numerically that mode from the large tional radiation. Thus, the tails for late time in this
order behavior of the hydrodynamic series6 , ob- work is not a surprise, and neither a (transient)
taining ωBorel = 3.1193 − 2.7471i. In [19] for ho- lowest QNM. The computational procedure, as im-
mogeneous isotropization we obtain for the SYM plemented in this work, can be extended to the
model the lowest QNM for the quintuplet channel holographic model 1RCBH undergoing a Bjorken
ωHI /T ≈ 9.8 − 8.6i (in agree with [35]), which is flux [17]. Although for this last model the Borel
equal to (except by a scale factor of T and work- attractor is not known, as far as we know, the sub-
ing precision) the other authors QNM complex traction of the NS attractor catchs a significant
frequency, ≈ π(3.12 − 2.75) ≈ 9.8 − 8.6i. Now, part of the QNM before the final relaxation (de-
if we consider the ratio of the Borel frequency cay).
−ℜ(ωBorel )/ℑ(ωBorel ) ≈ 1.1345 and compare it Finally, we think that this work was a nice
with the same ratio for our centered complex fre- example of deep human learning, and could be in-
quency (for IC#20), calculated with Harminv teresting for new practitioners in the field.
(see Fig, 6) in post-post-processing, we get ≈
1.1369, which is close enough to extract a scale ACKNOWLEDGMENTS
factor for each evolved IC. For our centered com-
plex frequency in Fig, 6 (IC#20) the scale factor GG and WB thanks FAPESP, Scientific Ini-
is ≈ 0.219. Thus, 0.219 × (14.231 − 12.518i) ≈ tiation Program, under grant 2023/07953-5. WB
3.12 − 2.74i, in agree with the expected values, thanks FAPESP, Research Projects Program, un-
within numerical errors. der grant 2022/02503-9 and acknowledge the finan-
Shedding non-equilibrium (Borel’s regime), cial support by National Council for Scientific and
the system is driven to equilibrium, characterized Technological Development (CNPq) under grant
by transient QNM. For the most of IC is revealed number 407162/2023-2. Also we thank to Nairy
that the corresponding QNM is close to the ex- Villarreal for comments about the original version
pected lowest QNM. Then, any of the considered of this work. The authors thank to the Central de
IC evolves to the stable (pure imaginary) QNM Computação Multiusuário (CCM) at UFABC, for
support using the clusters Titânio and Carbono.

6 The working precision in [24] is 10−100 .

Revista Brasileira de Física, Vol. 4, Nº 3, e202411040302
www.revistabrasileiradefisica.com | DOI: https://doi.org/10.5281/zenodo.14198452
Licença Creative Commons CC BY-ND 4.0 8

[1] U. Heinz and R. Snellings, Ann. Rev. Nucl. Part. 126009 (2024); arXiv: 2402.04529.
Sci. 63, 123 (2013); arXiv:1301.2826. [20] P. Romatschke, U. Romatschke, Relativistic fluid
[2] J. Berges, M. P. Heller, A. Mazeliauskas, and R. dynamics in and out of equilibrium (Cambridge
Venugopalan, Rev. Mod. Phys. 93, 35003 (2021); University Press, 2019).
arXiv:2005.12299. [21] M. Spalinski, Phys. Lett. B 776, 468 (2018);
[3] M. P. Heller and M. Spalinski, Phys. Rev. Lett. arXiv:1708.01921.
115, 072501 (2015); arXiv:1503.07514. [22] P. Romatschke, Phys. Rev. Lett. 120, 012301
[4] W. Florkowski, M. P. Heller, and M. Spalin- (2018); arXiv:1704.08699.
ski, Rept. Prog. Phys. 81, 046001 (2018); [23] R. Baier, P. Romatschke, D. T. Son, A. O.
arXiv:1707.02282. Starinets, M. A. Stephanov, JHEP 04, 100 (2008);
[5] A. Kurkela, E. Lu, Phys. Rev. Lett. 113, 182301 arXiv:0712.2451.
(2014); arXiv:1405.6318. [24] M. Heller, R. Janik, P. Witaszczyk, Phys. Rev.
[6] A. Kurkela, A. Mazeliauskas, J. Paquet, S. Lett. 110, 211602 (2013); arXiv:1302.0697.
Schlichting, D. Teaney, Phys. Rev. Lett. 122, [25] R. Critelli, R. Rougemont, J. Noronha, Phys. Rev.
122302 (2019); arXiv:1805.01604. D 99, 066004 (2019); arXiv:1805.00882.
[7] D. Almaalol, A. Kurkela, M. Strickland, Phys. [26] P. Chesler, L. Yaffe, Phys. Rev. D 82, 026006
Rev. Lett. 125, 122302 (2020); arXiv:2004.05195. (2010); arXiv:0906.4426.
[8] J. Maldacena, Adv. Theor. Math. Phys. 2, 231 [27] P. Chesler, L. Yaffe, JHEP 07, 086;
(1998); arXiv:hep-th/9711200. arXiv:1309.1439.
[9] S. Gubser, I. Klebanov, A. Polyakov, Phys. Lett. [28] H. Bondi, M. van der Burg, A. Metzner, Proc.
B 428, 105 (1998); arXiv:hep-th/9802109. Roy. Soc. Lond. A 269, 21 (1962).
[10] E. Witten, Adv. Theor. Math. Phys. 2, 253 [29] R. Sachs, Proc. Roy. Soc. Lond. A 270, 103
(1998); arXiv:hep-th/9802150. (1962).
[11] E. Witten, Adv. Theor. Math. Phys. 2, 505 [30] J. Winicour, Living Reviews in Relativity 15, 2
(1998); arXiv:hep-th/9803131. (2012).
[12] G. Denicol, J. Noronha, H. Niemi, D. Rischke, [31] C. Cartwright, M. Kaminski, M. Knipfer Phys.
Phys. Rev. D 83, 074019 (2011); arXiv:1102.4780. Rev. D 107, 106016 (2023); arXiv: 2207.02875.
[13] P. Kovtun, A. Starinets, Phys. Rev. D 72, 086009 [32] S. Johnson, Harminv: a program to solve the har-
(2005); arXiv:hep-th/0506184. monic inversion problem via the filter diagonaliza-
[14] M. Heller, R. Janik, M. Spalinski, P. tion method (fdm), v1.4.2.
Witaszczyk, Phys. Rev. Lett. 113, 261601 [33] R. Janik, R. Peschanski, Phys. Rev. D 74, 046007
(2014); arXiv:1409.5087. (2006); arXiv:hep-th/0606149.
[15] R. Rougemont, J. Noronha, W. Barreto, G. Deni- [34] A. Starinets, Phys. Rev. D 66, 124013 (2002);
col, T. Dore, Phys. Rev. D 104, 126012 (2021); arXiv:hep-th/0207133.
arXiv:2105.02378. [35] R. Critelli, R. Rougemont, J. Noronha, JHEP 12,
[16] R. Rougemont, W. Barreto, J. Noronha, Phys. 029 (2017); arXiv:1709.03131.
Rev. D 105, 046009 (2022); arXiv:2111.08532. [36] T. Islam, G. Faggioli, G. Khanna, S. Field, M.
[17] R. Rougemont, W. Barreto, Phys. Rev. D 106, van de Meent, A. Buonanno, Phenomenology and
126023 (2022); arXiv: 2207.02411. origin of late-time tails in eccentric binary black
[18] J. Bjorken, Phys. Rev. D 27, 140 (1983). hole mergers, (2024); arXiv:2407.04682.
[19] R. Rougemont, W. Barreto, Phys. Rev. D 109, [37] P. Chesler, L. Yaffe, Phys. Rev. Lett. 102, 211601
(2009); arXiv: 0812.2053.