Single and merger soliton dynamics in scalar field dark matter

with and without self-interactions

Matthias Stallovits1∗ and Tanja Rindler-Daller1,2,3†
1
Institut für Astrophysik, Universitätssternwarte Wien,
Fakultät für Geowissenschaften, Geographie und Astronomie,
Universität Wien, Türkenschanzstr.17, A-1180 Vienna, Austria
2
Vienna International School of Earth and Space Sciences,
Universität Wien, Josef-Holaubek-Platz 2, A-1090 Vienna, Austria
3
Wolfgang Pauli Institut, Oskar-Morgenstern-Platz 1, A-1090 Vienna, Austria
(Dated: June 12, 2024)
arXiv:2406.07419v1 [astro-ph.CO] 11 Jun 2024

Scalar field dark matter (SFDM) made of bosons has become a popular alternative to the cold dark
matter (CDM) paradigm, especially for its potential to cure the so-called ”small-scale problems”
of CDM structure formation. Cosmological simulations have determined that SFDM halos exhibit
a core-envelope structure, but they are computationally expensive. Halo cores have been found
to be well approximated by ”solitons”, i.e. equilibrium structures of the underlying equations
of motion. The study of single soliton and multiple soliton merger dynamics constitutes a more
feasible approach to investigate in detail the genuine quantum dynamics of SFDM and its interplay
with self-gravity for a multitude of free boson parameters. In this paper, we present dedicated
simulations of single solitons and binary soliton mergers, for models without and with a 2-boson,
repulsive, weak to intermediate self-interaction (SI), as well as multiple soliton mergers without SI.
We adapt the open-source code PyUltraLight to simulate solitons with SI. We derive numerical
scaling relations between the central density and mass of solitons for several values of SI and find
deviations from the monotonic relations known from fuzzy dark matter (no SI), or the strongly
repulsive Thomas-Fermi regime. Solitons with SI exemplify larger cores and lower central densities,
compared to solitons without SI. Using our simulations, we extract numerical density profiles for
solitons and post-merger objects, and fit them to analytic functions of previous literature. We find
a mild preference for Gaussian cores for objects with SI, while the envelopes of post-mergers can
be fit to NFW profiles albeit with some caution as we discuss. Similar to previous work, we find
global, persistent oscillations for solitons as well as post-mergers, confirming that self-gravitating
SFDM has very long relaxation times, although objects with SI exhibit oscillations of comparatively
smaller amplitude.

I. INTRODUCTION typically much larger than the physics-based Jeans scale

(roughly one order of magnitude even smaller than the
The nature of dark matter (DM) remains one of Jeans scale is the minimum size of objects in virial or
the most pressing open problems in astronomy and hydrostatic equilibrium in a given DM model). This lim-
physics. Many particles beyond the established stan- itation is one reason why the interpretation of CDM sim-
dard model of particle physics have been conceived as ulations and their comparison with observations (before
possible DM candidates. The current cold dark matter and after the addition of baryons) remains a challenge.
(CDM) paradigm is based upon weakly-interacting mas- Discrepancies have been reported between CDM simula-
sive particles (WIMPs), which are (much) heavier than tions and observations of DM-dominated dwarf galaxies,
the proton, or upon the very light quantum chromody- which go under the header of ”small-scale problems” for
namics (QCD) axion with a mass of ∼ 10−6 −10−5 eV/c2 . their appearance at scales of a few hundred parsecs to a
In either case, the DM dynamics on galactic and cos- few kpcs, see e.g. [1, 2]. In the context of this paper, the
mological scales has been numerically modelled by per- ”cusp-core-problem” is of special interest to us: CDM
forming simulations of the N -body gravitational dynam- universally predicts cuspy DM centers, with the density
ics of minimum-size objects, which assemble structure going as ∼ r−1 for r → 0, within virialized objects from
from galaxies to the cosmic web hierarchically over cos- dwarf galaxies to galaxy clusters, whereas observations
mic time. While the minimum size of objects from Jeans seem to indicate in many cases (nearly-)constant central
instability within static or expanding backgrounds is de- cores. It remains unclear whether (the proper simula-
termined by the concrete DM particle physics model, in tion and physical feasibility of) baryon feedback at all
practice, the minimum size used in simulations is dictated scales is able to ”explain away” the empirically found
by limitations on resolution and computation times. As discrepancy. Given these issues and the hitherto non-
a result, that minimum size of objects in simulations is detections of above CDM particle candidates, the com-
munity has been considering alternative DM particles,
altogether. In order to explain the small-scale problems
∗ purely from the DM side, a larger Jeans or characteris-
[email protected]
† [email protected] tic scale, below which structure formation is suppressed,
 2

is required. One family of such models comprises scalar literature. On the other hand, in the FDM regime of
field dark matter (SFDM), which consists of (ultra-)light large de Broglie lengths, more simulations have been per-
bosons in a Bose-Einstein-condensed many-particle state. formed to date, although cosmological simulations suffer
This type of model has been considered since the ∼ 2000s the same issues of resolution and computational feasibil-
using many names: if the ultralight bosons have no self- ity. The work by [10] is a rare example, where FDM
interaction (SI), they have been known as fuzzy dark structure formation simulations with baryons down to a
matter (”FDM”) or (quantum-) wave DM, see [3–8], al- redshift of z ∼ 5 have been presented. The overwhelm-
though the term ”BECDM” has been also commonly ing number of studies have focused on FDM-only simula-
used, see e.g. [9, 10]. If an (attractive) axion-like SI tions, either in a cosmological setting, see e.g. [6, 7, 9, 20–
is included with a much smaller boson mass than the 22] and e.g. the review by [23], or within static back-
QCD axion, the model is also called ultralight axion DM grounds by merging single so-called ”solitons” to create
(”ULAs”), see e.g. [11]. If a strongly repulsive SI is con- larger halo-like structures, see e.g. [24] and [25] and more
sidered, instead, the model has been also called ”SFDM- references below. Solitons refer to minimum-energy ob-
TF” in [12, 13] or ”SIBEC-DM” in [14, 15], to highlight jects in (near-)equilibrium, which occur generically in
the ”Thomas-Fermi (TF) regime” or the BEC nature, SFDM and which will be also the focus of our work.
respectively, although other terms for this model have Given the fact that cosmological simulations of SFDM
been used by [16], [17] or [18]. For models without SI, (with or without baryons) are extremely computationally
the characteristic scale is given by the de Broglie length expensive, the approach of studying gravitational dynam-
of the bosons: if a galactic minimum scale of ∼ 1 kpc is ics at the level of single solitons remains an important
desired, the bosons of SFDM have to have a tiny mass of means of investigation. Many details in the unfolding
m ∼ 10−23 − 10−22 eV/c2 , i.e. SFDM is often associated dynamics of gravitationally bound SFDM structures are
with ultralight bosons. However, at the other extreme not yet understood, and the few existing sophisticated
of the TF regime, the characteristic scale depends upon simulations often lack the resolution to draw conclusions
the parameter combination g/m2 , where g denotes the SI with respect to the full spatial range of those objects, let
coupling strength to be introduced below. In this regime, alone to probe the SFDM parameter space in a thorough
a minimum scale of order 1 kpc is feasible for much higher way. Therefore, dedicated ”small-scale” simulations are
boson masses m ≫ 10−22 eV/c2 . Overall, the potential required to study various dynamical aspects in more de-
of SFDM to make a cure to the CDM ”small-scale prob- tail, or probe a larger range of parameter space. Indeed,
lems” has been a motivation to consider the astrophysical it is also important to understand the dynamics of indi-
implications of this family of models, using various theo- vidual solitons, given that simulations have found that
retical and computational methods. they serve as approximate models for cores or central
Generally, the equations of motion of SFDM are a cou- parts of evolving SFDM galactic halos. By the same to-
pled set of Klein-Gordon-Einstein equations which reduce ken, mergers of solitons have been studied to investigate
in the Newtonian regime to a nonlinear Schrödinger equa- how structure formation at small scales may proceed. In
tion (NLSE) or Gross-Pitaevskii (GP) equation, coupled general, merger simulations are computationally costly,
to the Poisson (P) equation; we will refer to this set of and even in static backgrounds issues and open ques-
differential equations as the GPP system. The Newto- tions remain, as we will discuss. Thus, the work in this
nian regime is appropriate for the study of single galactic paper is motivated by all these considerations and shares
SFDM halos (except in the vicinity of, say, central super- its scope with similar previous investigations in the lit-
massive black holes), which are of interest to us for our erature, where dedicated ”small-scale” simulations are
study here. used to help shed light on the dynamics of SFDM soli-
tons. We will present our simulations of single soliton
Although the characteristic scales in SFDM can be
dynamics, as well as binary and multiple soliton merger
much larger than in the CDM case, depending upon the dynamics in the forthcoming sections, and expand upon
SI regime and the choice of m, g or their combination
previous literature in several regards, notably the study
g/m2 , those scales are still small in the context of fully- of the density profiles of simulated solitons, and the in-
fledged cosmological simulations. Therefore, issues re-
clusion of models with weak repulsive SI. Yet, this paper
lated to resolution and computation time have not dis- shall be considered explorative in nature: we are not con-
appeared. In fact, in the TF regime of strongly repulsive
cerned with model fitting to observational galactic data
SI, the SI scale given by g/m2 is much larger than the de at this point. Instead, we hope to enhance our theoretical
Broglie scale. For computational efficiency, any dynam-
understanding of SFDM halos and halo cores to inform
ics at that latter scale needs to be coarse-grained, typi- for the future more sophisticated procedures of fitting
cally resulting in fluid approximations of the fundamen-
models to observational data.
tal equations of SFDM, or rather SFDM-TF, see [12, 13]
for the theoretical and numerical foundational work, and Finally, we think that our analysis here, as well as sim-
follow-up studies in [15, 19]. While this approach in the ilar ones in the literature, albeit entirely theoretical in
TF regime is appropriate to study halo formation and their approach, will also contribute towards a better un-
evolution at the level of gross features, the addition of derstanding of self-gravitating quantum systems, in gen-
baryons represents a difficulty not yet pursued in the eral. Laboratory experiments are currently attempting
 3

to investigate the connection between quantum mechan- the Gross-Pitaevskii-Poisson (GPP) system. The Gross-
ics and gravity, but are limited to the environment of Pitaevskii (GP) equation in (2) is essentially a ”mean-
the Earth gravitational field (either co-moving on ground field” equation for the field ψ (see e.g. [29]), while the
labs, or in free fall in space labs). On the other hand, the- Poisson equation in (3) has the usual form, determining
oretical investigations of the dynamics of self-gravitating the gravitational potential of the object self-consistently
solitons under various SI and gravity regimes enable to via its mass density as follows
probe physical regimes, which are unattainable in the lab.
The study of SFDM solitons has this further motivation, ∂ψ(~x, t) ~2
i~ =− ∆ψ(~x, t) +mΦ(~x, t)ψ(~x, t)
apart from the DM problem. ∂t 2m
This paper is organized as follows. In Section II, we +g|ψ(~x, t)|2 ψ(~x, t), (2)
collect all the fundamental equations and the numerical
routines, which are used to solve the equations of mo- ∆Φ(~x, t) = 4πGm|ψ(~x, t)|2 . (3)
tion of soliton dynamics. It also includes a subsection
on analytical density profiles for SFDM from the litera- Thus, ψ is supposed to represent the complex wave func-
ture, which we will use to fit our density data. Another tion of N dark matter particles, having the same mass
subsection is reserved for the code implementation. We m and being in the same state, which make up a gravi-
use and modify the open-source code PyUltraLight[26] tationally bound object of volume V , such that
by [27], in order to include models with SI as well as to Z
enable a fitting routine on the density data of our simu- |ψ(~x, t)|2 d3 ~x = N. (4)
lations. Section III presents our results of single-soliton V
dynamics, for models without and with SI, focusing on
Φ is the gravitational potential of the object, and G is the
long-time behavior, density profiles and the relationship
gravitational constant. The GP equation in (2) has this
between central density and mass of solitons. As of Sec-
form, upon using the standard first Born approximation
tion IV, we move to merger simulations, where we first
to describe an essentially contact-like s-wave 2-particle
present binary mergers in that section, again, without
SI, whose strength is characterized by the constant self-
and with SI included. We discuss various simulation set-
coupling strength g. It is related to the scattering length
tings and density profiles of post-merger objects. In Sec-
as via
tion V, we present results on multiple soliton mergers
without SI, where we discuss long-time behavior, the im- as
g = 4π~2 . (5)
pact of the initial configurations, and the issues that must m
be overcome, if SI is included. Finally, we summarize and The SI is of repulsive nature for g > 0, and attractive for
discuss our results, and put them into perspective with g < 0. Models without SI, for which g = 0, have been
previous literature in Section VI. A discussion of techni- the focus of the overwhelming part of the astrophysics
cal aspects concerning simulation parameters, issues and literature on SFDM. Our paper is a rare exception, in
configurations are deferred into two appendices. that we will simulate SFDM gravitational dynamics in
the wave approach, i.e. by directly solving GPP (2-3),
also for SFDM models with SI. We will focus on zero and
II. FUNDAMENTAL EQUATIONS AND
NUMERICAL ROUTINES
repulsive SI, because these models generically support
gravitationally stable objects.
A. Equations of motion
B. Rescalings and code units
In the nonrelativistic regime, the dynamics of galac-
tic halos or halo cores composed of a single species of Before the equations of motion are solved, it is recom-
ultralight bosonic SFDM is described by a single wave mended to write them in dimensionless units. We will
function, which obeys the coupled Schrödinger-Poisson use and modify for our purpose the open-source code
(SP) equations, if the bosons of mass m have no self- PyUltraLight by [27], hence we adopt its code units,
interaction (see e.g. the early account in [28] for a dif- as follows. To write the GPP system in dimensionless
ferent motivation). The wave function is a complex vari- form, three scales for length (6), time (7) and mass (8)
able ψ(~x, t) of space and time, whose modulus squared are needed. These originate from [7] and are commonly
describes the local probability (number) density of the used for SFDM (further examples are [27, 30, 31]). The
SFDM particles; upon multiplication by m, we can inter- code units are defined as
pret
 14 ! 21
2
8π~2 10−23 eV/c

ρ(~x, t) = m|ψ(~x, t)|2 , (1) L = ≈ 121 · kpc,
(6)
3m2 H02 Ωm0 m
as the corresponding local mass density of SFDM. If   12
a particle self-interaction (SI) is included, the coupled 8π
T = ≈ 75, 5 Gyr and (7)
differential equation system has been also known as 3H02 Ωm0
 4
− 41   23
and analytic models of the literature, which have been

1 8π ~
M= devised from various considerations. Importantly, it has
G 3H02 Ωm0 m
! 23 been established from theoretical and simulation work
2
10−23 eV/c that SFDM halos are characterized by a core-envelope
≈ 7 · 107 · M⊙ . (8) structure, see e.g. [6, 7, 9, 12, 32–37]. As such, SFDM
m
has been considered to resolve several small-scale prob-
Here, m is again the boson mass, G the gravitational lems of CDM, notably the ”cusp-core-problem”. It is the
constant, H0 the present-day Hubble constant, ~ the re- pressure contributions of SFDM, either in the form of the
duced Planck’s constant and Ωm0 the present-day matter quantum pressure (for models without SI), or in the form
density parameter. Constants and scaling parameters are of a (n = 1)-polytropic pressure (for models with strong,
shown in table I. In fact, the global mass and length scales repulsive SI) which avoid the central density cusps of
depend upon the particle mass, but since we will fix this CDM halos, and produce density cores, instead.
mass to m = 10−22 eV/c2 in our simulations, these quan- However, the outer halo parts (”envelope”) of simu-
tities are fixed, as well. We choose this particle mass in lated SFDM halos have been found, on average, to be
order to compare our results with previous ones in the close to CDM halos, see e.g. [10, 12, 13, 19]. This simi-
literature. larity in the dynamics of SFDM and CDM halo envelopes
Additionally to these re-scalings, we introduce a di- was convincingly explained by [12].
mensionless self-coupling strength Λ̂, which depends on Now, the profiles found in the halo central parts differ,
the dimensional self-coupling strength g, according to depending upon the SI regime of SFDM. As a result,
different analytical profiles have been considered. In the
c2 g regime of strongly repulsive SI, the TF regime of SFDM
Λ̂ = . (9)
4πG~2 (i.e. SFDM-TF or SIBEC-DM), the central density is
Here, c is the speed of light in vacuum. The derivation very close to a (n = 1)-polytrope. Since we will not
of this dimensionless self-coupling strength can be found consider the TF regime in this paper, we refer the reader
e.g. in [32], from which we also took the notation (ex- to the literature, e.g. [12, 17, 35].
cept for the 4π difference in the Poisson equation). The Our focus in this paper are models without SI (such
dimensionless GPP equations read thus as as FDM) and models with a weak SI. In these cases, the
de Broglie wavelength is the characteristic scale which
ˆ, t̂)
∂ˆψ̂(~x 1ˆ ˆ determines the size of galactic halo cores. In this regime,
i =− ∆ ˆ, t̂)ψ̂(~x
ψ̂(~x, t̂)+Φ̂(~x ˆ, t̂)+Λ̂|ψ̂(~x ˆ, t̂)
ˆ, t̂)|2 ψ̂(~x
ˆ
∂ t̂ 2 ”solitons” are the minimum energy solutions to the GPP
(10) equations, and they have been found to approximate the
cores of fully evolved ”core-envelope” halos (see cited
ˆ, t̂) = 4π|ψ̂(~x
ˆ Φ̂(~x
∆ ˆ, t̂)|2 . (11) literature above). These solitons or ”solitonic cores”
The hats refer to dimensionless quantities, but we will of SFDM halos can be modelled as coherent standing
suppress them in the forthcoming for brevity, except for wave packets, for which a Gaussian profile has been of-
Λ̂. ten adopted for simplicity, see e.g. [38, 39]. Similar to
other applications in quantum physics, the Gaussian pro-
file has been also used for models with weak SI, especially
C. Analytical density profiles for SFDM solitons as variational ansatz in energy calculations. Assuming
and halos spherical symmetry, the radial density profile of a soli-
tonic halo core as a Gaussian can thus be modelled as
A main result in our paper will be the comparison of ρ(r) = ae−r
2
/2σ2
, (12)
density profiles between our simulated SFDM solitons
with the central density a = ρ(0) and the width of the
Gaussian σ (we use the notation ”a” for the central den-
TABLE I. Constants and parameters. The global length and sity to avoid confusion with other models below).
mass scale in code units (marked with a star) were calculated The results of dedicated simulations of FDM (i.e.
for a boson mass of 10−22 eV/c2 . SFDM without SI) by [6] prompted them to devise an
Parameter Value Unit empirical fitting formula, which is able to model the cen-
~ 1.0545718 · 10−34 Js tral profiles of halos in this regime, as well. For simplicity,
H0 67.7 km/s we call it Schive profile; it is given by
Mpc
Ωm0 0.31 1
m3
δs
G 6.67 · 10−11 kg s2
ρS (r) =   2 8 , (13)
1 pc 3.0857 · 1016 m r
1 + rsol
M⊙ 1.989 · 1030 kg
1 length-code unit⋆ ∼ 38.26 kpc
1 mass-code unit⋆ ∼ 2213594 M⊙ where δs denotes the central density, and rsol describes
the width of the profile (the index ”s” stands for ”Schive”
 5

to distinguish it from other models). For a comparison which also has some additional features, notably the pos-
between Schive and Gaussian profiles, we refer to [40] sibility to fit density profiles to simulated data. We will
and [32]. provide our source code to the public in due course.
In this paper, we will use the Gaussian and Schive
profiles as fitting formulae for the density data of our
simulated solitons at rest, as well as for post-merger ob- 1. Dynamical evolution
jects. For single solitons, we have found that the two core
parameters are related roughly as The calculation of the dynamical evolution is described
in [27], which is also the basis of the modification by [31].
rsol ≈ 4σ. (14) In that paper, the SI-term is treated as being part of the
Post-merger objects result upon colliding and merging of potential term of the Schrödinger equation, ”transform-
single solitons, and the final core of a post-merger object ing” it basically into the GP equation. Since the SI can
is expected to retain a solitonic nature, a finding which be viewed as another potential term, it can be treated in
we also confirm, using our simulations. We will present the pseudo-spectral method like the gravitational poten-
our results, also with regard to the frequency of preferred tial. Indeed, we came up independently with this very
profiles in later sections. natural approach to include the SI. Therefore, the dis-
Our post-merger objects are found to develop a core- cretized evolution can be schematically written as
envelope structure, similar to results in previous litera- 
ih

ture. In order to fit the envelope of these objects, we use ψ(~x, t + h) = exp − Φ(~x, t + h)
2
a standard Navarro-Frenk-White (NFW) profile as an an- " #
alytic model. The NFW profile was first introduced in ihΛ̂ 2
· exp − |ψ(~x, t + h)|
the context of fitting CDM halo density data in [41] and 2
has the following form    
−1 ih 2 ih
δN F W ·F exp − k F exp − Φ(~x, t)
ρN F W (r) =  2 , (15) 2 2
r
1 + rrs
" #
rs ihΛ̂
· exp − |ψ(~x, t)|2 ψ(~x, t), (16)
2
where δN F W is the characteristic density and rs is the
scale radius. While the NFW profile is not adequate for  
SFDM halo central parts, it has been found to describe −1 1
Φ(~x, t + h) = F − 2 F 4π|ψ(~x, ti )|2 . (17)
the outer parts of such halos in an averaged sense (see k
cited references above). Therefore, in our investigation,
a combination of a core profile plus NFW will be used to Here is h the time step, ψ(~x, ti ) the field configuration
fit the density data of our simulated halo(s)/cores. at half the full time step ti , k the wave number of the
Fourier domain, F and F −1 denote the discrete Fourier
transform and its inverse. The order of the operations
D. Implementation goes from right to left. Thus, we start in position space,
where a half step with the potential operator is taken.
Then, we change to Fourier space, where a full step is
To solve the GPP system of equations (10-11), we use
taken. Afterwards, the inverse Fourier transform is used,
our modified version of the ultralight dark matter solver
to change back to position space, where the second half
PyUltraLight. This code uses a symmetric, pseudospec-
step of the potential operator is taken, completing a full
tral Fourier-split-step-algorithm, where a discretization
iteration. However, our code uses an additional simplifi-
of the equations is done by using a simulation box with a
cation, presented initially in [27], where the consecutive
N × N × N -grid. Here, N is the number of points along
half steps of two iterations are combined to a single full
an axis, which is determined by the resolution (not to
step. Therefore, only the very first and the very last op-
be confused with the above number of particles denoted
eration need an actual half step. Schematically this looks
N - we will not need it in the forthcoming). The equa-
like
tions are solved with periodic boundary conditions, i.e.    
if matter would exit the box at one side it would reenter ih ih
ψ(t + nh) = exp + Φ exp + Λ̂|ψ(t)|2
on the other side. The original PyUltraLight code was 2 2
developed to solve the Schrödinger-Poisson system, i.e. n  !
ultralight SFDM without SI. In order to study models
Y
2 ih 2
· exp[−ihΦ] exp[−ihΛ̂|ψ(t)| ] exp ∇
with (weak) SI, we modified the original code and created 2
our own version, which we call PyUltraLight_SI. Indeed, 
ih
 
ih

2
while our work was in progress, the authors of [31] came · exp − Φ exp − Λ̂|ψ(t)| ψ(t). (18)
2 2
up with their own modification of PyUltraLight to in-
clude SI, but their code is not public. Therefore, we pro- We stress that this is only a schematic representa-
ceeded to create our own modification PyUltraLight_SI, tion (there is no Fourier transformation in the equation
 6

above). Additionally, we want to point to the different which is loaded at the start of a simulation run. For
signs in the beginning and the end, which are only needed details and how the user-specified initial conditions are
for this representation. In the code itself, we need to add incorporated, the reader is referred to Section 3.2 of [27].
(not subtract like shown above) an additional half step, Additionally, a text file is produced, which stores infor-
due to the way the execution of the algorithm is per- mation about the calculation. While these are mostly in-
formed. cidental information, the information about Lambda hat
and beta are necessary for the main code PyUltra-
Light_SI, since they are needed for the evolution of the
2. Initial conditions initial profile. This process of loading the parameters is
done automatically, such that the user only has to specify
Similar to the original PyUltraLight code, simulation Λ̂ once. This is an improvement, since β depends on Λ̂
runs in PyUltraLight_SIrequire an initial soliton pro- and before our adjustment, these parameters would have
file as an input. In PyUltraLight_SI, an initial soliton to be specified in the main part of the code [43].
profile is loaded, which is stored in an external file. To- We also note that ”high” positive values of Λ̂ demand
gether with the initial conditions for the soliton - includ- a smaller maximum radius for the calculation of the ini-
ing its mass in CU - specified in the associated Jupyter tial soliton profile. The reason is that the wave function
Notebook, the initial field configuration is created which profile is shallower at the maximum radius, due to the
is then passed on to the solution algorithm. This ini- boundary conditions, and this effect is additionally en-
tial file is produced with an additional script called soli- hanced, the higher Λ̂. Without a proper adjustment of
ton solution.py. This script was part of the original this maximum radius, the solver can grind and fail to
PyUltraLight code (see [27] for details), and we adapted converge to a solution (see also section III B). Finally, we
it for our needs as follows. The initial soliton profile is cal- point out that, for all simulations discussed in this paper,
culated by imposing spherical symmetry onto the system we choose periodic boundary conditions and we disregard
of equations, and assuming that the gravitational poten- phase shifts in the wave functions of binary or multiple
tial and density are time-independent, i.e. by adopting solitons.
a stationarity ansatz. The resulting ordinary differential
equations can be solved with a 4th-order Runge-Kutta
solver and the shooting method. We adopt the same 3. Extraction and fitting of density profiles
ansatz, i.e. we assume
ψ(~x, t) → eiβt f (r), Φ(~x, t) → ϕ(r), (19) In order to analyze the density profiles of the simulated
single solitons, as well as multiple soliton post-merger
where r = |~x|, β is a constant and f (r) is the spatial objects, a new tool was created in the code, called den-
part of the wave function. Using the Laplace operator sity profile. This new additional feature of our PyUltra-
in spherical symmetry and denoting the derivative with Light_SIpackage uses the 3D simulation data to create
respect to r with a prime, the dimensionless GP equation radially averaged density profiles. It is incorporated into
of (10) transforms to the Jupyter notebook and can be controlled there [44]. It
1 1 is able to differentiate between single and multiple soliton
−βf (r) = − f ′′ (r) − f ′ (r) + f (r)ϕ(r) + Λ̂f 3 (r). (20) simulations.
2 r
For the specified state of the simulation, the 3D den-
Using (19) and defining ϕ̃(r) := ϕ(r)+β, the GP equation sity data from the main simulation is used to create the
can be simplified to profile. In order to do so, the code determines the center
of the soliton, or of the post-merger object, respectively,
2
f ′′ (r) = − f ′ (r) + 2f (r)ϕ̃(r) + 2Λ̂f 3 (r). (21) by searching for the grid point with the highest density.
r From this point, the minimum distance to the edge of
With respect to the Poisson equation, we have the box is determined. This is necessary, since the center
of a soliton or post-merger could be at any grid point,
2
ϕ̃′′ (r) = 4πf 2 (r) − ϕ̃′ (r). (22) in principle, which would cause problems in the averag-
r ing, when more data is available in a certain direction,
The boundary conditions are as follows. In order to if the center of the object were misplaced from the cen-
guarantee that the radial wave function is smooth, f ′ (r) ter of the box. By truncating the density data at the
as well as ϕ̃′ have to vanish at the origin, while the maximum distance that is possible in all directions, it
wave function and the gravitational potential go to 0 for can be assured, that there will be the same amount of
r → ∞. In practice, the gravitational potential is var- data in all directions. This maximum radial distance is
ied at the origin until a solution for f (r) is found, that then binned, depending on the resolution that was used.
approaches zero at some maximum radius. This creates spherical shells with different number of grid
This stationary GPP system of ordinary differential points. For each radial bin, the densities for all possible
equations is solved by soliton solution.py [42]. After ex- grid points in such a shell are averaged. As a result, a
ecution, the initial soliton profile is saved as numpy-file, discrete radially-averaged density profile is created.
 7

Our new package density profile also offers the possi- many cases, we found a posteriori the relation (14) but
bility of calculating the (cumulative) mass of the soliton in the fitting we use as a ”boundary” the model in [39],
or post-merger object from this profile, using two differ- where the spherical FDM halo core radius is defined by
ent routines: the trapezoid rule, or the integration with including 99 % of the mass of the Gaussian (called the
spherical shells. In most cases, the trapezoid rule proved ”Gaussian sphere”). Hence, we use
to be better, which is why we recommend to use it for
the mass determination. So, the routine can be used to size of soliton core ≈ 2.575829 σ, (23)
calculate that radius, which contains a certain percent of
the interior mass. This percentage can be specified by and pick that radial point that comes closest to this value
the user in the notebook. Especially the radius contain- as the connection point between core and outer envelope.
ing 99 % of the soliton mass, denoted R99 , is a useful To fit the data beyond the connection point, the NFW
indicator for the success of a simulation, because it can profile is used.
provide clues as to whether the box size has been chosen Our code part density profile discriminates automati-
too big or too small, for the problem at hand. cally between simulations which involve one or more soli-
However, the main point of density profile is the fit- tons. For a single soliton, the code will only use the core
ting of the discrete density profile [45] with analytic den- models to fit the radial density profile. For post-merger
sity models. We use the Python tool curve fit from objects, it will determine the connection point and cal-
scipy.optimize. As mentioned above, the radial density culate the core as well as the NFW fit.
profile may consist of two different sections: The core and
the outer envelope. For single solitons and for the core
part of a post-merger object, we allow both the Gaus- III. SINGLE SOLITON DYNAMICS
sian model (12) and the Schive model (13). Apart from
being convenient analytic models, the Gaussian profile One aspect of PyUltraLight is the study of the grav-
is motivated as a ”fiducial” wave packet solution to the itational dynamics of SFDM under various scenarios,
underlying differential equations, while the Schive model given the flexibility of the code environment. A gen-
was found by fitting density profiles to halos from cos- eral theme of interest concerns the long-time behavior
mological FDM simulations (see Section II C). Now, den- of single solitons at rest with zero velocity. These soli-
sity profile uses both to fit the density data by using a tons at rest serve as a test bed to study the fundamental
score system to decide which core model fits best. This nature and stability of ultralight FDM or SFDM objects
score system rates the profiles according to the following under self-gravity, without the complications due to grav-
three criteria: itational disturbances from other sources. In particular,
• smallest residual the impact of different SI (none, weak, strong; attrac-
tive vs. repulsive) can be studied in detail. Therefore,
• standard deviation of the parameter for the central one pillar of our paper is the dedicated study of certain
density aspects of single-soliton stability and dynamics, and the
comparison of our results with previous literature. Of
• standard deviation of the parameter which defines course, another reason is the easy environment provided
the width of the model by single solitons, in order to test our modified version
The fits are compared based on these criteria. That pro- PyUltraLight_SI, before using it for more sophisticated
file which ”wins” in two of these three criteria is then simulations. In order to present our results in a coher-
used and displayed. ent manner, we separate our discussion of simulations of
For a single soliton, it is appropriate to fit its density models without and with SI.
data using one of the above core profiles. However, for
mergers an outer envelope is expected, as a result of ac-
quiring a new equilibrium upon mass infall. The resulting A. Simulations without self-interaction
density data require more than a core profile. For lack of
a theoretically motivated global analytic model, we use These models have Λ̂ = 0, i.e. cases which tradition-
a core and an envelope profile and fit them together; for ally belong to the FDM category. We place solitons at
the envelope we use the NFW model (15). rest in the middle of the box and study their evolution
However, we first have to answer where the core ends for different soliton masses (while the boson mass stays
and where the envelope begins. We tested different ap- fixed). This way, we are able to investigate how low-
proaches, but in the end only one turned out to be useful. and high-mass solitons behave, and how simulation pa-
To find the connection point, we use the standard devia- rameters like resolution affect the results. During our
tion σ from the fit to the Gaussian (even if the Gaussian simulations, we encountered various problems which are
is not used as the core fit). In fact, our results suggest related to issues of numerical resolution and box size, be-
that the numerical differences between the Schive and cause these influence the overall spatial resolution of the
Gauss profile within the core region are actually small; simulation. We will mention some in passing, but defer
for a comparison between these profiles see also [32]. In details into Appendix A.
 8

1. General set-up get a good approximation for the central density as well
as for the soliton mass, given the competing constraints
During the first test runs with PyUltraLight, we en- from resolution and computation time.
countered already a problem which is well known in the
community, let us call it the ”high-mass problem”. In
FDM, the radius of (approximate) equilibrium objects 2. Oscillations at rest
[46], such as solitons, goes inversely proportional to their
mass, i.e. solitons get smaller with higher mass, leading We confirm previous results in the literature, concern-
to resolution problems for the central (= smallest) re- ing the oscillatory nature of solitons at rest (see e.g.
gions of solitons. Effectively, the central density regions [25, 47]), though we recognize it preferentially for higher
of solitons suffer a cutoff, resulting in lower values than soliton masses. We observe that solitons start oscillat-
expected. The only way to avoid this problem seems to ing, expanding and contracting over time. To demon-
require a suite of dedicated simulations to find the ”op- strate this phenomenon, we present a simulation of a
timum” parameters, in order to observe the competing soliton with 80 CU (∼ 8 · 107 M⊙ ), which ran for 0, 13 CU
requirements between resolution and computation time, (∼ 10 Gyr) in a 5 CU box with a resolution of 384. We
which change with changing soliton mass. Generally, we show a selection of density profiles at different snapshots
initialized a soliton with a given mass in the middle of in figure 1. We can see how the numerical density pro-
the box with zero initial velocities. These solitons at rest file changes over time, especially in the outer regions.
were then simulated for 0, 013 CU (∼ 1 Gyr). The expansion and contraction can be also noticed based
The spatial resolution of a simulation is based upon on the change of the fitting parameters of the Schive
the numerical resolution and the box size. As a com- model, though the changes are small. An initial expan-
promise, for a soliton of given mass, we increased the sion, which causes the central density to decrease and the
spatial resolution by increasing the numerical resolution width parameter of the Schive model rsol to increase, is
and decreasing the box size, until the relative change in followed by a stagnation phase during ∼ 4 − 6 Gyr. From
the central density was around 1 %. Since this proce- 8 Gyr onward, we see a contraction, causing an increase
dure would be too much effort for every single case, we in the central density and a decreasing rsol . At the end of
performed simulations for representative cases, which are the simulation, the soliton has become slightly broader,
shown in Table II. For other masses, we used the ”next but the central density is lower in comparison to the
best” parameters. initial snapshot. The oscillations seem to be caused by
While this way, the central densities are correctly rep- the interplay between quantum pressure and self-gravity
resented, the soliton masses, as calculated from their den- (first and second term on the right-hand side of (10), re-
sity profiles, deviate from the initial input values in the spectively), and the effect may be just more visible for
code. The culprit is the box size which needs to be large higher-mass solitons than for lower-mass solitons, as fol-
enough, given the periodic boundary conditions. As a lows. Solitons with higher mass have a higher density. As
result, the calculated soliton masses tend to be higher a result, the quantum pressure is higher as well (it can be
than the input numbers. However, the central density as shown that the equation of state for soliton models with-
a key quantity of interest to us is not affected by this off- out SI is well approximated by a (n = 2)-polytrope (see
set in input vs. calculated soliton mass. In Appendix A, [39, 48]), thus the pressure increases with increasing den-
we discuss a generic procedure, which can be followed to sity). While the quantum pressure leads to expansion,
self-gravity counteracts efficiently, the higher the mass,
causing a contraction. Hence, we can see how the self-
TABLE II. This table shows the lowest numerical resolution gravity of the soliton avoids its ”dispersion”, which would
(second column) and the biggest possible box length (third otherwise happen like for any other unconfied quantum
column) for the investigated soliton masses, to ensure that mechanical wave packet. This phenomenon of ”gravita-
we do not run into the ”high-mass problem”. We note that tionally induced inhibition of dispersion” has been also
these parameters do not reproduce exactly the soliton masses predicted for laboratory quantum systems, see [49].
as calculated from their density profile (see also main text). Let us also note that these oscillations due to com-
For soliton masses in between the listed ones, we always used peting dynamical terms as seen in figure 1 are different
the next best set-up, e.g. for 30 CU we used the parameters
in nature from the oscillations which have been reported
for 40 CU.
elsewhere, especially with regard to real scalar fields, or
soliton mass [CU] lowest resolution biggest box size [CU] real SFDM such as axion-like DM, see e.g. [50]. In these
20 256 8 models, solitons are often called ”oscillatons” for this rea-
40 320 8 son. Those oscillations refer to the inherent field oscilla-
60 384 7
tions about a mean, whose amplitude is higher, the lower
80 384 5
100 384 4 the particle mass, and they should not be confused with
200 384 2 the dynamical oscillations. Let us also note that in the
300 320 1 presence of SI or for complex fields, the inherent field
oscillations are much smaller in amplitude, compared to
 9

)=0 Gyr 107 )=2 Gyr

106
105
log of density ρ(r) [M⊙/kpc 3]

log of densit, ρ(r) [M⊙/kpc 3]

104
103

102 101

100 10−1

10−3
10−2

Densi)y profile 10−5 Densi)y profile

10−4
Sc ive: δs=6690413.244, rsol=4.85 Sc ive: δs=6676287.059, rsol=4.854
100 101 102 100 101 102
log of radius r [kpc] log of radius r [kpc]

107 )=4 Gyr 107 )=6 Gyr

105 105
log of densit, ρ(r) [M⊙/kpc 3]

log of densit, ρ(r) [M⊙/kpc 3]

103 103

101 101

10−1 10−1

10−3 10−3

10−5 Densi)y profile 10−5 Densi)y profile

Sc ive: δs=6663608.01, rsol=4.857 Sc ive: δs=6663056.326, rsol=4.857
100 101 102 100 101 102
log of radius r [kpc] log of radius r [kpc]

107 )=8 Gyr 107 )=10 Gyr

105 105
log of densit, ρ(r) [M⊙/kpc 3]

log of densit, ρ(r) [M⊙/kpc 3]

103 103

101 101

10−1 10−1

10−3 10−3

10−5 Densi)y profile 10−5 Densi)y profile

Sc ive: δs=6673513.121, rsol=4.855 Sc ive: δs=6683931.244, rsol=4.853
100 101 102 100 101 102
log of radius r [kpc] log of radius r [kpc]

FIG. 1. These plots show the density profiles for six selected snapshots (see legend) of a simulation of a 80 CU soliton (Λ̂ = 0),
which was initialized at rest. The box size was 5 CU and the resolution 384. The orange, solid curves are fits according to
equation (13), with units for δs in M⊙ /kpc3 and rsol in kpc.

the real-field case for the same particle mass. 3. Central-density-mass relation

An important key quantity with respect to the nature

of DM is the central density of self-gravitating DM struc-
tures. The run of the central density, as well as the pres-
 10

Masses[Msun] 1e8 is quoted in [39] (see also [51]),

log of central den ity [[M⊙/kpc 3] 1 2 3 4 5 6
3
109 Gm2

ρc = ρj M 4, (26)
~2
108
where ρj is a dimensionless constant depending on the
107 eigenstate and calculated numerically. Using the value
for the ground state from the numerical calculation of
106 [8], which is 0.0044, we get
ρc=0.16359⋅M 4
105 Central den ity Gm2
 
1
b ≈ 0.0044 ≈ 7.05094 · 10−27 3 , (27)
~ 2
M⊙ kpc3
50 100 150 200 250 300
Masses [CU] i.e. very close to (25). We can see from figure 2, and the
calculation of the fit parameter b, that our simulations
FIG. 2. The central-density-mass relation without SI (Λ̂ = 0) with PyUltraLight_SI yield the expected scaling rela-
from the simulation data (blue dots). The simulation pa- tion. Moreover, this result confirms that our modified
rameters can be found in Table II. The central densities are code reproduces the soliton behavior for Λ̂ = 0.
extracted from the density profiles at the respective final snap-
shots of the evolution at 0, 013 CU (∼ 1 Gyr). The solid, or-
ange curve is the fit, according to equation (24). B. Simulations with self-interaction

ence or absence of constant central density cores, affect In this subsection, we consider single-soliton simula-
the dynamics of baryonic tracers, and so the DM den- tions for Λ̂ 6= 0. Because the work by [31] came out while
sity is subject to indirect observational constraints, e.g. ours was under progress, we took the opportunity to com-
from galactic velocity data. Since solitons have been rec- pare simulations performed with our modified code, with
ognized as approximate models for the core regions of those of [31]. We reproduced their binary mergers, using
simulated galactic FDM/SFDM halos, we are interested the same simulation and SFDM parameters, for Λ̂ = 0,
in their central density as a function of model parame- Λ̂ = −0.02 and Λ̂ = 0.02; see their figures 6, 7 and 8, re-
ters. spectively, in [31]. The results were in very good mutual
We use the density data we acquired during the anal- agreement. We will present our results of binary merger
ysis of the high-mass problem and extracted the central simulations in section IV and multiple soliton mergers in
densities associated with the respective soliton mass, in section V, respectively.
order to quantify their scaling relation. This is useful, not
only to compare our results with the literature, but also
to predict, which central densities and thus which spatial 1. Range for Λ̂
resolution we need for certain soliton masses. It has been
known from analytic considerations, confirmed later by As a matter of fact, previous literature which solved
simulations, that the central density of solitons for Λ̂ = 0 the GPP equations using the wave approach (e.g. by
should scale with M 4 (see e.g. [12], [39] and references using pseudo-spectral methods), including [27] and [31],
therein). In figure 2, we plot the central densities over have mainly focused on models without SI, or on mod-
the soliton masses (as blue dots) for the parameters from els with weak SI. Motivated by the QCD or string ax-
Table II. ions, attractive SI has been generally considered more
We fit these data points with the theoretical expecta- frequently than repulsive SI. However, for strongly repul-
tion, sive SI, there is another well-known regime, the so-called
Thomas-Fermi (TF) regime, which has been explored as
ρc = bM 4 , (24) well, see Section I for references. As described there, in
that case, the stability of single ”TF solitons” against
with the central density ρc , the soliton mass M and our
gravitational collapse is being attained solely by the SI
fitting parameter b. The central density is in physical
pressure, while quantum pressure is very much subdom-
units, while the soliton mass is in mass code units, thus
inant and can be neglected. However, quantum pressure
the unit of b is M⊙ /(kpc3 CU4 ). In units of astronomical
does play a role in the envelopes of post-merger objects,
interest to us, our fit yields
such as fully-fledged galactic halos, also in the TF regime;
M⊙ 1 these models being often called SFDM-TF or SIBEC-DM
b ≈ 0.16359 ≈ 6.81341 · 10−27 3 . (25) in the literature. Since models with Λ̂ > 0 offer another
kpc3 CU4 M⊙ kpc3
pressure contribution to oppose gravity (in contrast to
To compare this result with the literature, we need the attractive SI), we focus in this paper on simulations with
exact definition of b, or rather the whole relation, which positive, repulsive SI. Furthermore, our initial goal has
 11

been to carry out simulations in the TF regime, in order 1.0

to compare the outcome with previous literature, which Wave f nction profile for Λ̂ = 5
usually adopts a fluid approach to study SFDM-TF halo 0.8 Wave f nction profile for Λ̂ = 10
formation and evolution, as e.g. in [12, 13, 15, 19].
Now, in order to be in the TF regime, we require 0.6
Λ̂ ≫ 17; see [32]. However, we encountered serious issues
with PyUltraLight_SIfor sufficiently high Λ̂, which pre- 0.4
vented us from studying the TF regime. Given our quest
of performing small, but dedicated simulations, we in- 0.2
stead report in this paper our results of models with ”in-
termediate” range in repulsive SI, having Λ̂ ≤ 10. For one
0.0
thing, this intermediate regime has not been explored in
0 1 2 3 4 5 6 7
previous literature. Furthermore, we know from [12] that radius [CU]
the severe resolution limitations in SFDM-TF, where the
de Broglie length is much smaller than the characteristic FIG. 3. A comparison between two wave function profiles
TF radius or length scale, lend themselves to the adop- with Λ̂ = 5 (blue curve), calculated with a maximum radius
tion of a fluid approach, rather than the wave equation of 9 CU, whereas Λ̂ = 10 (orange curve), calculated with a
approach which is used in PyUltraLight and PyUltra- maximum radius of 7.6 CU. For comparison’s sake, the blue
Light_SI. It remains to be seen whether the numerical curve is plotted to the maximum radius of the orange curve
algorithms, or the specifics of the codes may be extended (notice the linear axis scales).
such that the TF regime can be covered, but this task
turned out to be well beyond the scope of this paper.
The mentioned issues come at play already at the level 2. Low Λ̂
of the initial soliton profiles, as follows. We have de-
scribed above how the initial profile is determined, via As mentioned already, a repulsive SI leads to solitons
its boundary conditions and numerical method: in order with larger width and we expect them to spread more
to guarantee that the radial wave function is smooth, readily (or ”disperse more readily”) over time, given the
f ′ (r) as well as ϕ̃′ have to vanish at the origin, while the repulsive SI pressure. This should lead, in turn, to de-
wave function and the gravitational potential go to 0 for creased central densities, compared to solitons without
r → ∞. In practice, the gravitational potential is var- SI. Since higher Λ̂ leads to lower central density, the de-
ied at the origin until a solution for f (r) is found, that mands on the spatial resolution are less severe, and the
approaches zero at some maximum radius. For small Λ̂, requirements to avoid the ”high-mass problem” are not
no issue is encountered. However, for sufficiently high so strict. On the other hand, more readily ”dispersing”
Λ̂’s, and even for the intermediate cases considered here, solitons take up more space over time within the simu-
the wave function becomes significantly broader (i.e. the lation box. The latter is thus required to be chosen big
associated density profile flatter), because the repulsive enough to prevent the loss of scalar field mass beyond the
SI becomes stronger. Examples are presented in figure box boundaries. At the same time, the choice of the box
3, where two wave functions for different Λ̂ are shown. size must ensure that the calculated soliton mass is close
The wave function for Λ̂ = 10 has a larger width than to its input initial value (for resolution limits mentioned
that with Λ̂ = 5. The solver requires to find solutions earlier).
whose profiles go to zero at the maximum radius, but To illustrate how the repulsive SI influences the density
this process becomes increasingly difficult, the broader profiles, we set up a soliton at rest with a mass of 20 CU
the profiles get. As a result, the would-be solution oscil- for Λ̂ = 0.02. This value for the SI is inspired by the tests
lates at a certain radius, ultimately causing the solver to for the binary merger simulations made by [31], which we
grind. To overcome this problem, we decrease the max- reproduced. Furthermore, for this weak SI, we expected
imum radius for higher Λ̂’s (Λ̂ ≥ 6), such that the code the outcome to be comparable to the case with Λ̂ = 0
never reaches that radius, where the solution starts os- (see figure 1, but note the difference in soliton mass). For
cillating. We deem this approach legitimate, because in the simulation parameters, we choose 0.013 CU (∼ 1 Gyr)
the outer regions the profile goes quickly towards zero, and a resolution of 256. For the box size, we first chose
and hence the introduced numerical error is small. 2 CU to ensure that the soliton stays inside the box within
Because of this and other issues to be mentioned below, the evolution time. Then, we reduced the box size, such
we discuss in the forthcoming two subsets of range in Λ̂, that the calculated soliton mass resulted in 20 CU as it
and call them should be. This was accomplished, once 1, 03 CU was
chosen for the box size [52]. The final density profile can
be seen in figure 4. As expected, the density profile looks
• low Λ̂ in the range 0 ≤ Λ̂ ≤ 1
similar to the Λ̂ = 0 case, except for the lower central
density and greater width, exemplified by the larger core
• intermediate Λ̂ in the range 1 < Λ̂ ≤ 10. parameter of the Schive profile, and the fact that the
 12

105

R_99 104
log of density ρ(r) [M⊙/kpc 3]

log of density ρ(r) [M⊙/kpc 3]

104
R_99

103
103

102

Den)ity profile Densi)y profile

101 Schive: δs=24673.023, rsol=18.276 102 Gauß: a=3751.138, σ=10.853
10−1 100 101 10−1 100 101
log of radius r [kpc] log of radius r [kpc]

FIG. 4. The density profile of the final state (∼ 1 Gyr) for a FIG. 5. The density profile of the final state (∼ 1 Gyr) for
20 CU soliton which was initialized in a 1.03 CU box with a a 20 CU soliton which was initialized in a 1.38 CU box with
resolution of 256 and Λ̂ = 0.02. The black vertical represents a resolution of 256 and Λ̂ = 1. The black vertical represents
that radius, which contains 99 % of the mass of the soliton, that radius, which contains 99 % of the mass of the soliton,
while the orange, solid curve is the fit according to equation while the orange, solid curve is the fit according to equation
(13), with units for δs in M⊙ /kpc3 and rsol in kpc. (12), with units for a in M⊙ /kpc3 and σ in kpc. In comparison
to figure 4, the core is much more extended, leading to a lower
central density and a larger R99 .
soliton mass is different.
Now, we run the same soliton simulation, but we set
Λ̂ = 1. While the duration and the resolution stayed the range for Λ̂, because for the simulated models with Λ̂ > 1
same, the box size had to be increased to 1.38 CU due to we have no accurate mass determination. Yet, we pro-
the factor 50 stronger SI. The density profile can be seen ceed with the evolution of the solitons and their density
in figure 5. As expected from the trend seen in the wave profiles over time with various Λ̂’s. We simulated a 20 CU
functions, and the previous simulation, the profile has soliton at rest, but we initialized it with a duration of
become even more uniform with a lower central density. 0.013 CU (∼ 1 Gyr), a box size of 8 CU and a resolution
Also, the radius R99 , which includes 99 % of the mass of of 256. We increased the time resolution by increasing
the soliton, is larger. This trend will play a role for the the save number to 10, such that we get 10 snapshots
intermediate Λ̂-case. with a resolution of 0.1 Gyr; we show 3 snapshots of each
We conclude that a higher Λ̂ yields a more uniform den- simulation: the initial state at t = 0 Gyr, the middle
sity distribution of matter within the soliton, compared state at t = 0.5 Gyr and the final state at t = 1 Gyr. For
to Λ̂ = 0, as a higher repulsive SI pressure pushes against each snapshot, we look at the radial density profile as
the central pileup of matter due to gravity, resulting also well as the 2D density profiles in the x-y-plane. We ran
in a lower central density for the same soliton mass. For simulations for Λ̂ = 5 and Λ̂ = 10, respectively.
the 20 CU solitons plotted in figures 4 and 5, we recog- In figure 6, we show the radial soliton density profiles
nize a core parameter which is ≈ 2.4 times larger in the for Λ̂ = 5, while the associated 2D density profiles in the
case with Λ̂ = 1, compared to the case with Λ̂ = 0.02 x−y plane can be seen in figure 7. We recognize the same
(remember rsol ≈ 4σ). However, we stress that the size trend as seen in the previous runs with lower Λ̂: wider,
of the fitted core parameter depends not only on Λ̂, but more uniform profiles with lower central density result.
also on the soliton mass. The middle snapshot seems to indicate the formation of a
”ring”, or rather a spherical ”shell”, which merely exem-
plifies the redistribution of matter in the soliton. Upon
3. Intermediate Λ̂ looking at the SI term in the GP equation (10), we see
that it depends not only on the SI strength (given by Λ̂),
We have seen above that the profiles become more uni- but also on the density (given by |ψ̂(~x, t)|2 ). Thus, for a
form and favor larger fitted core parameters, as Λ̂ in- given Λ̂, the density affects the ensuing dynamics. Since
creases. Also, R99 gets larger. This trend exacerbates the SI pressure is more prevalent in denser regions, it
with higher Λ̂. Thus, when we want to determine an ac- pushes matter outwards and this ”expansion” is faster in
curate mass from the profile, in order to calculate R99 , the dense inner regions, compared to the outer regions.
we would need to reduce the box size and ”cut into” the Eventually, the inner ”shells” catch up with the outer
core-like, extended density profile, which affects the mass layers, forming the ”ring-like” structure in the 2D plot.
determination, in turn. In effect, too high or too low a Now, in figure 8, we show the radial soliton density pro-
calculated soliton mass results. files for Λ̂ = 10, while the associated 2D density profiles
This issue is one reason why we decided to split the in the x − y plane can be seen in figure 9. We stress that
 13

105
t=0 Gyr 4
t=0 Gyr
103
lo of densit− ρ(r) [M⊙/kpc 3]

x-distance [CU]
101

10−1
0
10−3

10−5 −2
Den)ity profile
10−7
Gauß: a=28471.069, σ=7.469
100 101 102 −4−4 −2 0 2 4
log of radius r [kpc]
y-distance [CU]
104 t=0.5 Gyr 4
t=0.5 Gyr
101
2
log of density ρ(r) [M⊙/kpc 3]

x-distance [CU]
10−2

10−5 0
10−8
−2
10 −11
Den)ity profile
Gauß: a=5432.723, σ=14.229
100 101 102 −4−4 −2 0 2 4
lo of radiu) r [kpc]
y-distance [CU]
4
103
t=1 Gyr t=1 Gyr
101
2
log of density ρ(r) [M⊙/kpc 3]

x-distance [CU]

10−1

10−3 0
10−5

10−7 −2
10−9 Den)ity profile
Schive: δ =1318.541,
−4−4
s rsol =71.106

100 101
lo of radiu) r [kpc]
102 −2 0 2 4
y-distance [CU]
FIG. 6. The evolution of the radial density profile of a 20 CU-
soliton, which was initialized in a 8 CU box with a resolution FIG. 7. The corresponding 2D density profiles to figure 6.
of 256 and Λ̂ = 5. The orange, solid curves are fits according Shown are contour lines of constant density.
to equations (12) and (13), with units for a and δs in M⊙ /kpc3
and σ and rsol in kpc, respectively.
 14

the maximum radius for the initial soliton profile was re-
t=0 Gyr
duced from 9 CU (the initial maximum radius for Λ̂ < 6) 104
to 7.6 CU (see the explanation to Fig.3). The overall evo-

log of dens ty ρ(r) [M⊙/kpc 3]

lution of the profiles is very similar to the ones for Λ̂ = 5. 102
Here, we encounter some numerical artifacts in the 2D
profiles which, however, have no consequence for the ra-
dial profiles. Their origin remains unclear, although they 100
arise generically, see e.g. figure 8 in [25], or figure 10 in
[53]. 10−2
We conclude that solitons exhibit a very uniform den-
sity profile for increasing values of Λ̂, which we can un- 10−4
derstand as the interplay of repulsive SI pressure with
self-gravity. Also, it appears that the outer parts of soli-
tons experience fewer oscillations, the higher Λ̂, maybe 10−6 Dens ty prof le
as a result of being ”washed out” due to the matter re-
Ga)ß: a=28933.203, σ=10.086
distribution. Yet, this statement also depends upon the 100 101 102
log of rad us r [kpc]
soliton mass. In general, oscillations in the tails seem to
be more prominent for higher soliton mass, independent
of the given value of Λ̂. t=0.5 Gyr
103

101
log of dens ty ρ(r) [M⊙/kpc 3]
4. Central-density-mass relation for Λ̂ > 0

In the previous subsection, we have seen that the cen- 10⊙1

tral density gets lower with increasing Λ̂, for fixed value
of the soliton mass. Now, we want to study how the 10⊙3
central density might change with varying soliton mass
at fixed Λ̂, in order to produce a similar plot as in fig- 10⊙5
ure 2. However, the issues described above prevented us
from studying this central-density-mass relation for in- 10⊙7
termediate values of Λ̂. Instead, we focus on two cases,
10⊙9 Dens ty prof le
Λ̂ = 0.02 and Λ̂ = 1. Given the lack of literature to guide Ga)ß: a=5188.075, σ=18.921
us, our expectations were as follows. For very small Λ̂,
we would expect no big qualitative change, compared to 100 101 102
log of rad )s r [kpc]
Λ̂ = 0. On the other hand, since our range of studied Λ̂’s
is well below the TF regime of strongly repulsive SI, we
would not expect to find a scaling similar to SFDM-TF, t=1 Gyr
either. In fact, in the latter case, analytic considera- 103
tions predict that the central density is proportional to
log of dens ty ρ(r) [M⊙/kpc 3]

the soliton mass, ρc ∝ M , see e.g. [12] and references 101

therein, i.e. a much milder growth with mass, compared
to equation (24)[54].
10⊙1
In our simulation runs, we set up solitons at rest with
different masses and use the simulation parameters ac-
cording to Table II, just as for the Λ̂ = 0 case. Yet, our 10⊙3
runs do not cover the same range of soliton masses. The
difficulties with respect to the broadening (”dispersion”) 10⊙5
of solitons with Λ̂ 6= 0 described above become exacer-
bated for higher-mass solitons. In many cases, solitons Dens ty prof le
became unstable, prohibiting the extraction of meaning- 10⊙7
Ga)ß: a=1297.063, σ=29.701
ful radially averaged density profiles from the simula-
tions. As a result, we present our findings for a different 100 101 102
log of rad )s r [kpc]
range in soliton mass.
Figure 10 shows the central-density-mass relation at FIG. 8. The evolution of the radial density profile of a 20 CU-
fixed Λ̂ = 0.02 for solitons with a mass of up to 150 CU. soliton, which was initialized in a 8 CU box with a resolution
The central densities were extracted from the density pro- of 256 and Λ̂ = 10. The orange, solid curves are fits according
files at the final snapshot (∼ 1 Gyr). We can recognize to equation (12), with units for a in M⊙ /kpc3 and σ in kpc.
that ρc increases monotonically with soliton mass with
 15

4 Masses[Msun] 1e8
t=0 Gyr 0.5 1.0 1.5 2.0 2.5 3.0
140000 Central densities with ̂⊙ = 0.02

Central density [[M⊙/kpc 3]

120000
2
x-distance [CU]

100000
80000
0 60000
40000
20000
−2 0
20 40 60 80 100 120 140
Masses [CU]
−4−4 −2 0 2 4 FIG. 10. The central-density-mass relation for Λ̂ = 0.02. The
y-distance [CU] central densities are extracted from the density profiles at the
4 respective final snapshots of the evolution (at ∼ 1 Gyr).
t=0.5 Gyr
2 a slope that is shallower than in the case without SI.
x-distance [CU]

However, at around 50 CU the central density peaks, be-

yond which ρc decreases with mass. In light of the results
of the previous subsection, we think we can understand
0 this behavior as follows. The broadening of the solitons is
caused by the SI pressure, or SI term in the GP equation
(10), which depends on Λ̂ and on the density (wave func-
tion squared). This term opposes the self-gravity in the
−2 equation, which attempts to contract the soliton. The
gravitational potential Φ(r) also depends on the density
via the Poisson equation (11). Thus, for fixed Λ̂ 6= 0,
−4−4 −2 0 2 4
the dynamical importance of each term (self-gravity vs
SI repulsion) is affected by the density of the soliton in
y-distance [CU] characteristic ways [55]. Our simulations suggest that
4 the initially monotonic increase of ρc with soliton mass
t=1 Gyr can be attributed to the subdominance of the SI term
for the chosen time frame[56]. Once the soliton mass is
chosen big enough, however, the SI term ”takes over” or
2 rather is ”fast enough”, such that we notice the broad-
x-distance [CU]

ening and uniformation of the soliton profile within the

given simulation time in form of a decrease in the cen-
tral density. This statement applies for the static and
0 dynamic case: an initialized soliton with high mass will
already have a lower central density, than a correspond-
ing low-mass soliton, at given Λ̂. During the course of its
”evolution” -i.e. sitting in the box at rest-, the tendency
−2 of soliton broadening will make its impact onto the final
snapshot of the simulation from which the central den-
sity is extracted. As a result, beyond the turnover to the
−4−4 −2 0 2 4
prevalence of the SI term, we get lower central densities
for bigger soliton masses.
y-distance [CU] In order to investigate this further, we also studied
the central-density-mass relation for a stronger SI with
FIG. 9. The corresponding 2D density profiles to figure 8. Λ̂ = 1. In this case, the maximum soliton mass with
Shown are contour lines of constant density. The symmetric useful density profiles was around 55 CU. Because of this
structure in the center of the bottom panel is a numerical limitation, we ran more simulations for solitons at lower
artifact. masses, < 20 CU, using a higher resolution. The other
 16

Masses[Msun] 1e8 ready at lower soliton masses.

0.2 0.4 0.6 0.8 1.0 1.2
Central densities with ̂⊙ = 1
Central density [[M⊙/kpc 3]

4000
IV. BINARY SOLITON MERGERS

3000
Apart from the single-soliton dynamics studied in the
previous section, one further focus of this paper is the
2000 merger dynamics of solitons. In this section, we dis-
cuss our binary merger results, while the next section will
1000 present mergers consisting of more than two solitons.
The most basic interaction between two solitons is a
0 direct collision. This kind of scenario is a main testbed
10 20 30 40 50 for analysing the behavior of SFDM, see e.g. [58]. Col-
Masses [CU] lisions have been also performed using PyUltraLight in
the original work by [27], as well as in [31]. In [31], such
FIG. 11. The central-density-mass relation for Λ̂ = 1. The collisions were simulated, for models with and without
central densities are extracted from the density profiles at the SI. However, their choice of parameters did not lead to
respective final snapshots of the evolution (at ∼ 1 Gyr). More successful mergers. In contrast, we are particularly in-
data points (i.e. simulation runs) at lower soliton mass are terested in creating mergers, which meant to run many
included, compared to figure 10. simulations in order to test setup and model parameters.
We investigated the unfolding merging process, and stud-
ied the density profiles of the post-merger objects at the
simulation parameters were the same as for the previous final snapshots of our simulations.
runs of the central-density-mass relation with Λ̂ = 0.02. For the mergers we consider two basic scenarios: i)
In figure 11, we plot the obtained central densities of the The solitons are supposed to collide head-on along two
final states (again at ∼ 1 Gyr) versus soliton mass. As in different lines, one is the x-axis and the other along the
main diagonal line in the x-y-plane. These scenarios are
the previous Λ̂ = 0.02 case, we get a similar behavior: for
basically the same, for we could show, that the results of
sufficiently low soliton mass, the central densities grow
mergers is the same when the whole initial configuration
with mass, peak at a maximum and then decrease with
is rotated about the z-axis. Therefore, it would be enough
increasing soliton mass. We added more simulations, i.e.
to only look at the x-axis-case. However, we decided to
data points, at the lower-mass side of the prospective
use the diagonal variant for a different reason. Using the
maximum, in order to improve the visibility of the trend
longer diagonal of our square box is useful to evade the
that we see.
ejection of SFDM mass from the main point of interaction
To conclude we find a very similar run in the central- (i.e. to prevent exiting and re-entering of mass in the
density-mass relation of solitons at low, repulsive SI: the box), given the requirements of box size on an accurate
central density grows with mass, reaches a maximum and mass determination.
then declines with increasing soliton mass. This behavior The simulation parameters were set to avoid running
is very different from the monotonic relationships known into the ”high-mass problem”. The simulation time was
in the FDM regime (no SI) and the TF regime (highly set to 0.18 CU (∼ 13.59 Gyr), i.e. roughly one Hubble
repulsive SI). In lack of a firm analytic prediction, we re- time. All solitons were placed 0.4 CU from the center of
frained from presenting a fit [57] to our simulation data. the box in the x-y-plane. The velocities were varied until
Yet, we think we can understand our findings as a result a merger happens and a final end product was visible
of the dynamical competition between self-gravity and within the simulation time. Since we are working within
SI pressure, which depends on the chosen values for Λ̂, the x-y-plane, the z-component of all quantities is zero.
soliton mass and simulation time. The importance of the The relative velocity between solitons is a key param-
SI term is further illustrated by the fact that the central eter, which determines whether a merger is successful, or
densities of solitons of a given mass decrease markedly, for not. As in previous literature, our choice made in this
increasing Λ̂. For example, a soliton of 50 CU ∼ 108 M⊙ paper may appear somewhat artificial, in that we make
has a central density of ρc ∼ 5 × 105 M⊙ kpc−3 if Λ̂ = 0 sure that mergers happen. The chosen values also reflect
(figure 2), while ρc ∼ 105 M⊙ kpc−3 for Λ̂ = 0.02 (figure computational demands, in order to guarantee mergers
10) and ρc ∼ 500 M⊙ kpc−3 for Λ̂ = 1 (figure 11), respec- within a Hubble time, given our simulation setup. Cos-
tively. Of course, details depend on the fitting profiles mological simulations of FDM, using the wave approach,
which have been used to extract the values for ρc . By as well as using fluid variants for SFDM-TF, have estab-
comparing figure 10 and figure 11, we also recognize that lished that mergers do happen and that structure for-
the soliton mass, at which ρc peaks, is shifted toward mation above the characteristic Jeans scales proceeds in
lower values if Λ̂ is increased, in accordance with an ever similar fashions than in CDM, see e.g. [9, 15, 19]. Nev-
more important SI term that counteracts self-gravity al- ertheless, the many details of the merger dynamics in
 17

TABLE III. The simulation parameters for the equal-mass TABLE IV. The soliton parameters for the equal-mass merger
merger without self-interaction, which was initialized, such without self-interaction, which was initialized such that the
that the two solitons collide head-on along the x-axis. These two solitons collide head-on along the x-axis, each with mass
simulation parameters are also used for the case, where the of 20 CU ∼ 44.5·106 M⊙ . The initial separation of the solitons
two solitons collide along the main diagonal of the x-y-plane. is ∼ 30.7 kpc and the relative velocity is ∼ 5 km/s.
The box size is ∼ 100 kpc on a side, the simulation time is
∼ 13.59 Gyr, which results -together with the saving number soliton nr. mass x y vx vy
that gives the number of snapshots- in a temporal resolution 1 20 0.4 0 −5 0
of 1.359 Gyr. 2 20 −0.4 0 5 0

Parameter Value Unit

box size 2.65 CU
resolution 256 /
simulation time 0.18 CU
105 R_99

log of density ρ(r) [M⊙/kpc 3]

saving number 10 /
104

103
different SI regimes are far from being fully understood
and are subject of ongoing research. Importantly, the low
102
and intermediate SI regime has not been simulated pre-
viously. Therefore, we investigate mergers in the frame- D nsit/ )rofil
work of dedicated simulations here to probe fundamental 101 Schiv : δs=75075.453, rsol=13.723
NFW: δNFW=6925.266, rs=9.583
characteristics, deferring more realistic scenarios to fu-
ture work. In the following sections, we will discuss the 100 101
log of radius r [kpc]
different configurations and our results.
FIG. 12. The density profile of the final snapshot (∼
13.59 Gyr) of a merger of two 20 CU-solitons which collide
A. Mergers without self-interaction
head-on along the x-axis: Λ̂ = 0 and the remaining param-
eters can be found in tables III and IV. The black vertical
First, we consider mergers without self-interaction marks that radius, which contains 99 % of the mass of the
(Λ̂ = 0). We ran a lot of test simulations to deter- post-merger object. The orange, solid curve is the fit accord-
mine those relative velocities which work with our re- ing to equ.(13), with units for δs in M⊙ /kpc3 and rsol in kpc,
quirements. Anticipating our later results, we note that while the green, solid curve is the fit according to equ.(15),
higher velocities are possible, compared to cases with re- with units for δNF W in M⊙ /kpc3 and rs in kpc.
pulsive SI, without running the risk of not receiving a
merger within a Hubble time. We found that a rela-
tive velocity of 5 km/s works well, which are 5 CU for ton parameters in table V. The density profile of the post-
the absolute value for the initial velocity for each soliton. merger object is shown in figure 13. We mentioned al-
We discuss two kinds of simulations: mergers with equal ready that the density profile should in principle look the
mass ratio and mergers with a mass ratio of 2 : 1. same as in the x-axis head-on collision, due to rotational
symmetry. However, the fit parameters are different, and
also the density tail of the data look a little different.
1. Mergers with equal mass ratio The reason are the rounding errors which are introduced
when the irrational initial parameters for position and
Here the solitons have the same mass of 20 CU. The velocity are truncated. Nevertheless, the density profiles
simulation parameters and the soliton parameters for look very much the same, with a dominant Schive-fit core,
the x-axis variant can be found in tables III and IV. surrounded by a poorly developed envelope.
The resulting density profile of the post-merger object
is shown in figure 12. We can see that the core is well
described with the Schive model, while the outer parts 2. Mergers with 2:1-mass ratio
are not well described with a NFW profile. This comes
with no surprise, because post-merger objects made of a Here, one soliton has twice the mass of the other soli-
small number of solitons typically end up with density ton. Since the total mass is now bigger, this would enable
profiles which look very similar to single solitons, namely us to use higher velocities. But to compare the different
core-dominated. The NFW-profile can fit fully-developed merger simulations with each other, we decided to leave
post-merger envelopes, which arise for multiple (N > 2) the relative velocity at 5 km/s. We start again with the
soliton mergers, as we will see in the next section. x-axis head-on collision. The simulation and soliton pa-
Now, we look at the diagonal variant of the merger, rameters for this case can be found in tables VI and VII,
with the simulation parameters in table III and the soli- respectively. We note that we used a bigger box, in or-
 18

TABLE V. The soliton parameters for the equal-mass merger TABLE VI. The simulation parameters for the merger with-
without self-interaction, which was initialized such that the out self-interaction, where the solitons had a mass ratio of 2:1.
two solitons collide head-on along the main diagonal of the They were initialized such that the two solitons collide head-
x-y-plane, each with mass of 20 CU ∼ 44.5 · 106 M⊙ . The on along the x-axis. The box size is ∼ 101.5 kpc on a side, the
initial separation of the solitons is ∼ 30.5 kpc and the relative simulation time is ∼ 13.59 Gyr, which results together with
velocity is ∼ 5 km/s. Note that due to the separation in the the saving number (gives the number of snapshots) in a tem-
x- and y-component the values look different to the x-axis poral resolution of 1.359 Gyr.
head-on collision, but are the same if we look at the absolute
value (and ignore rounding errors). Parameter Value Unit
box size 2.67 CU
soliton nr. mass x y vx vy resolution 256 /
1 20 0.28284 0.28284 −3.53553 −3.53553 simulation time 0.18 CU
2 20 −0.28284 −0.28284 3.53553 3.53553 saving number 10 /

TABLE VII. The soliton parameters for the merger without

105 R_99 self-interaction, where the solitons had a mass ratio of 2:1.
log of density ρ(r) [M⊙/kpc 3]

They were initialized such that the two solitons collide head-
104
on along the x-axis, with mass of 20 CU ∼ 44.5 · 106 M⊙ and
40 CU ∼ 89 · 106 M⊙ , respectively, an initial separation of
∼ 30.7 kpc and a relative velocity of ∼ 5 km/s.
103
soliton nr. mass x y vx vy
102
1 40 0.4 0 −5 0
2 20 −0.4 0 5 0
D nsit/ )rofil
101 Schiv : δs=73581.861, rsol=13.801
NFW: δNFW=7000.324, rs=9.594
100 101 resolution for this merger may not be good enough to
log of radius r [kpc]
resolve the central regions of the final object. We were
FIG. 13. The density profile of the final snapshot (∼
prevented from increasing the spatial resolution, since
13.59 Gyr) of a merger of two 20 CU-solitons which collide the box size was set such that the calculated mass agreed
head-on along the main diagonal of the x-y-plane. Λ̂ = 0 and with the initial mass. The highest resolution we tested
the remaining parameters can be found in tables III and V. was 512, but much higher resolution would be required to
The black vertical marks that radius, which contains 99 % of resolve the central densities. Given the purpose for which
the mass of the post-merger object. The orange, solid curve PyUltraLight - and our code modification of it-, was de-
is the fit according to equ.(13), with units for δs in M⊙ /kpc3 signed in the first place (to investigate SFDM dynamics
and rsol in kpc, while the green, solid curve is the fit accord- on laptop workstations), we did not pursue further HPC
ing to equ.(15), with units for δNF W in M⊙ /kpc3 and rs in resources for this project. For the dedicated simulations
kpc. here, we can say that the limited spatial resolution in
the unequal-mass mergers (at least with the masses we
used) will tend to produce smaller values for the central
der not to miss any SFDM mass that might otherwise densities than expected. Still, the analysis of the density
leave and reenter the box. The density profile of the profiles is not much affected by this limitation. Finally,
post-merger object for this merger variant is shown in we recognize that the envelope does not show as many
figure 14. At first glance, the profile looks very similar global oscillations, as in the equal-mass merger cases. We
to equal-mass mergers, yet there are some notable dif- think that the now higher total mass helps the merger to
ferences. First, the connection point between core and get more rapidly to a (near-)equilibrium state.
envelope, that was determined by the fitting routine, lies Next, we turn our attention to the diagonal collision.
closer toward the origin. This makes sense, since a soli- Here we used the simulation and soliton parameters from
ton with a higher mass is smaller in size (for Λ̂ = 0). As a tables VIII and IX, respectively. In this case, we were
result, the core, which still dominates the profile, is also able to reduce the box size in comparison to the x-axis
smaller. Furthermore, the central density is now much collision, since we had more room in the diagonal di-
higher, since more mass is able to accumulate in the cen- rection (as intended). The final density profile of the
ter. The smaller core ”leaves more room” for the envelope post-merger object can be seen in figure 15. There are
which is now bigger, compared to the previous case, but no big differences, compared to the x-axis case. Again,
still not developed well enough to make the NFW model rounding errors are to be blamed that the fit parameters
a good fit to the data. However, also the core is not fit as and the tail of the data points differ. However, there is a
well as in the equal-mass merger case. One explanation slightly better spatial resolution since the box is smaller,
could be again the ”high-mass problem”, i.e. the spatial which explains why the central density is bigger for the
 19

TABLE IX. The soliton parameters for the merger without

106 R_99 self-interaction, where the solitons had a mass ratio of 2:1.
log of density ρ(r) [M⊙/kpc 3]

They were initialized such that the two solitons collide head-
105
on along the main diagonal of the x-y-plane, with mass of
20 CU ∼ 44.5 · 106 M⊙ and 40 CU ∼ 89 · 106 M⊙ , respectively,
104 an initial separation of ∼ 30.7 kpc and a relative velocity of
∼ 5 km/s. Note that due to the separation in the x- and
103 y-component the values look different to the x-axis collision,
but are the same if we look at the absolute value (and ignore
102 rounding errors).
Density )rofile
Gauß: a=639999.588, σ=2.092
101 NFW: δNFW=58840.926, rs=5.707
soliton nr. mass x y vx vy
1 40 0.28284 0.28284 −3.53553 −3.53553
100 101
log of radius r [kpc] 2 20 −0.28284 −0.28284 3.53553 3.53553

FIG. 14. The density profile of the final snapshot (∼

13.59 Gyr) of a merger with mass ratio of 2:1, where the soli-
106 R_99
tons collided head-on along the x-axis. Λ̂ = 0 and the re-

log of density ρ(r) [M⊙/kpc 3]

maining parameters can be found in tables VI and VII. The
105
black vertical marks that radius, which contains 99 % of the
mass of the post-merger object. The orange, solid curve is
the fit according to equ.(12), with units for a in M⊙ /kpc3 104
and σ in kpc, while the green, solid curve is the fit according
to equ.(15), with units for δNF W in M⊙ /kpc3 and rs in kpc. 103

102
Density )rofile
TABLE VIII. The simulation parameters for the merger with- Gauß: a=656211.539, σ=2.068
101 NFW: δNFW=70811.917, rs=5.49
out self-interaction, where the solitons had a mass ratio of 2:1.
They were initialized such that the two solitons collide head- 100 101
log of radius r [kpc]
on along the main diagonal of the x-y-plane. The box size is
∼ 97 kpc on a side, the simulation time is ∼ 13.59 Gyr, which
results together with the saving number (gives the number of FIG. 15. The density profile of the final snapshot (∼
snapshots) in a temporal resolution of 1.359 Gyr. 13.59 Gyr) of the merger with mass ratio of 2:1, where the
solitons collided head-on along the main diagonal of the x-
Parameter Value Unit y-plane. Λ̂ = 0 and the remaining parameters can be found
box size 2.56 CU in tables VIII and IX. The black vertical marks that radius,
resolution 256 / which contains 99 % of the mass of the post-merger object.
simulation time 0.18 CU The orange, solid curve is the fit according to equ.(12), with
saving number 10 / units for a in M⊙ /kpc3 and σ in kpc, while the green, solid
curve is the fit according to equ.(15), with units for δNF W in
M⊙ /kpc3 and rs in kpc.
diagonal merger variant. Still, the limitations mentioned
above prevail, leading to lower accuracy in the central
pressure that is always at play. Therefore, successful bi-
region. We expect the ”real” central density to be higher
nary soliton mergers for Λ̂ = 0 may fail to merge, once
than what can be seen in figure 15.
Λ̂ 6= 0. This problem gets worse for multiple (N > 2)
mergers (see next section).
For Λ̂ = 1, we failed to produce any mergers within a
B. Mergers with self-interaction Hubble time. Thus, we focus on Λ̂ = 0.02. Moreover, for
equal-mass mergers we also failed to get mergers within
Our main interest has been the impact of SI onto soli- a Hubble time, even for small relative velocity: if the
ton and merger dynamics. However, our tests have shown velocity is small, the solitons will not have enough time
that simulation runs of mergers with SI are troubled with to come close enough to interact, or the merger is de-
competing requirements, in terms of physics and simula- layed due to the repulsive SI which pushes the solitons
tion parameters. Therefore, we present a good case of apart again, after their first encounter (if one happens),
a binary merger with SI in this subsection, but consider preventing a successful merger, after all. We found that
the subject far from being settled. As a general rule, we mergers can happen within 2 − 3 Hubble times, at best.
observe that the addition of repulsive SI makes success- Inspecting the corresponding density profiles at earlier
ful mergers much more difficult, for the same dynamical snapshots reveals that the ”binary-soliton-objects” are
reasons as discussed in the previous section: a repulsive far from any (near-)equilibrium, let alone to show the
SI works against self-gravity, in addition to the quantum characteristics of a final merger object. In principle, there
 20

TABLE X. The simulation parameters for the merger with TABLE XI. The soliton parameters for the merger with Λ̂ =
Λ̂ = 0.02, where the solitons had a mass ratio of 2:1. They 0.02, where the solitons had a mass ratio of 2:1. They were
were initialized such that the two solitons collide head-on initialized such that the two solitons collide head-on along the
along the main diagonal of the x-y-plane. The box size is main diagonal of the x-y-plane, with mass of 20 CU ∼ 44.5 ·
∼ 53 kpc on a side, the simulation time is ∼ 13.59 Gyr, which 106 M⊙ and 40 CU ∼ 89·106 M⊙ . The initial separation of the
results together with the saving number (gives the number of solitons is ∼ 30.7 kpc and the relative velocity is ∼ 0.25 km/s.
snapshots) in a temporal resolution of 1.359 Gyr.
soliton nr. mass x y vx vy
Parameter Value Unit 1 40 0.28284 0.28284 −0.17678 −0.17678
box size 1.4 CU 2 20 −0.28284 −0.28284 0.17678 0.17678
resolution 320 /
simulation time 0.18 CU
saving number 10 /
106
R_99

log of densi−y ρ(r) [M⊙/kpc 3]

105
are three knobs to turn to create successful mergers: i)
increase the simulation time well beyond a Hubble time; 104
ii) choose a tiny value for Λ̂ 6= 0; iii) reduce the ini-
tial separation, which enables a smaller relative velocity, 103
in turn. Option (ii) is unattractive, given our research
scope. Option (iii) is plagued by the issue mentioned in
Section III A 1 and A, and would demand a very large 102 Densit/ pr)file
Gauß: a=232242.554, σ=2.922
suite of dedicated, tailored simulations, to find the best NFW: δNFW=18157.366, rs=8.39
set of parameters. This approach is also questionable 101
10−1 100 101
from a physics’ perspective, for the best set of simula- log of radius r [kp ]
tion parameters are unlikely to be realized in nature. In
addition, option (iii) bears the danger that solitons start FIG. 16. The density profile of the final snapshot (∼
overlapping too early in the process [59], circumventing 13.59 Gyr) of the merger with mass ratio of 2:1, where the
a real merger simulation in the first place. This leaves solitons collided head-on along the main diagonal of the x-y-
option (i), a possibility which is often pursued in simula- plane. Λ̂ = 0.02 and the remaining parameters can be found
tions. However, we keep our demand to seek for mergers in tables X and XI. The black vertical marks that radius,
within a Hubble time for two reasons: first, we want to which contains 99 % of the mass of the post-merger object.
The orange, solid curve is the fit according to equ.(12), with
compare ”apples with apples” by choosing the same fi-
units for a in M⊙ /kpc3 and σ in kpc, while the green, solid
nal snapshots in our simulations. Second, we deem it curve is the fit according to equ.(15), with units for δNF W in
worthwhile to succumb SFDM to the Gretchenfrage: if M⊙ /kpc3 and rs in kpc.
mergers (eventually structure assembly) with certain soli-
ton masses at given Λ̂ 6= 0 fail to happen within a Hub-
ble time, we might as well conclude that SFDM models, dominant core and small envelope. In fact, this over-
which predict such soliton mass spectra, are disfavored all feature is more pronounced, now that repulsive SI
as reasonable DM candidates. has been added, for SI tends to produce bigger cores.
In fact, this leaves the choice of soliton mass to come Unsurprisingly, the NFW model provides a poor fit to
to the rescue. A higher total mass helps to accelerate the numerical data in the small envelope. Now, interest-
the merging process and keeps the repulsive SI at bay. ingly, a wider core usefully favors the Schive model as
However, the issue with the initial separation does not core fit, since this model does not fall off as rapidly as
go away with a higher mass, since the repulsive SI works the Gauss model. Nevertheless, in this specific case, the
in favor of exacerbating the overlap problem by expand- Gauss model has been preferred by the fitting routine,
ing the solitons, compared to their brethrens without providing a perfect fit to the core. We should mention,
SI. Thus, we focus on unequal-mass mergers for their though, that in this simulation we increased the reso-
higher total mass and consider diagonal collisions. We lution, given the computational resources that go into
succeeded in producing mergers within a Hubble time merger simulations, in order to provide satisfactory re-
and for a box size appropriate for a good (calculated) sults. The value was chosen according to test simula-
mass resolution, if we choose a low enough relative veloc- tions, where we used lower resolutions (at smaller box
ity. The simulation and the soliton parameters are listed size) to see what central densities we should expect. The
in tables X and XI, respectively. The resulting density increased spatial resolution would still not be enough to
profile of the post-merger object of this successful simu- avoid the high-mass problem; for that we would have to
lation can be seen in figure 16. increase the resolution above 384. However, since this
Similarly to the previous cases of Λ̂ = 0, the profile problem is common for the unequal-mass mergers we did
of the post-binary-merger object has a very extended, not redo the simulation, since, again, the trend of the
 21

profile would barely change.

TABLE XII. The soliton parameters for the (N = 5)-merger
While a higher total mass helps to produce mergers
for Λ̂ = 0, where the solitons were initialized such that the
(as demonstrated here), we still think there is reason to
five solitons collide in the center of the box, starting in pen-
be cautious: higher soliton masses lead to higher cen- tagon/star shape, each with mass of 20 CU ∼ 44.5 · 106 M⊙ ,
tral densities, which results in a higher impact of the an initial distance between soliton and center of the box of
repulsive SI over self-gravity. That means, SI may yet ∼ 23 kpc, and a velocity of ∼ 2.5 km/s. Notice that due to
prevent a merger from happening, or at least decelerate the separation in the x- and y-component, the values look dif-
it, such that one Hubble time is not enough to produce a ferent but are the same if we look at the absolute value.
(near-)equilibrium final merger state, even for high soli- soliton nr. mass x y vx vy
ton masses. Therefore, we expect that mergers with re- 1 20 0 0.6 0 −5
pulsive SI are only possible within a certain mass range of 2 20 0.57063 −0.18541 −4.75525 −1.54508
solitons, as already discussed earlier. How these conclu- 3 20 0.35267 −0.48541 −2.93893 4.04508
sions fit with the results of some cosmological simulations 4 20 −0.35267 −0.48541 2.93893 4.04508
of FDM (e.g. by [9]) as well as SFDM-TF (e.g. by [15]), 5 20 −0.57063 0.18541 4.75525 −1.54508
which both have produced successful mergers, albeit at
a much higher mass scale, remains to be seen. Certainly,
more future work on SFDM structure formation in dif- sist during many Hubble times, a characteristic feature
ferent SI regimes will be required. for any SI regime, though more pronounced for repulsive
SI. So, the question arises how long it actually takes for
the merger to settle down to a final (near-) equilibrium
V. MULTIPLE SOLITON MERGERS
state.
As an example, we simulated a 5-soliton merger for a
In this section, we consider finally multiple mergers, ”short” (roughly one Hubble time) and a long simulation
where more than two solitons are involved. However, time and compared the results. To be specific, we used
the difficulties mentioned previously for binary mergers a 5-soliton merger which is oriented in the x-y-plane in
with repulsive SI are exacerbated for multiple mergers. a pentagon (or star formation as we will call it later),
Therefore, we present here our results for Λ̂ = 0, and where each soliton had the same mass of 20 CU and the
defer the analysis of repulsive SI to future work. The same distance to the center of 0.6 CU. For the velocity,
main advantage we have is that we can draw on experi- we ensured that -in absolute value- all solitons move with
ence gained from many simulations that we performed, a velocity of 5 CU (∼ 2.5 km/s), inspired by the choice
and the possibility to compare our results to previous lit- of [24] (though their simulation times span hundreds of
erature concerning similar studies. Our search using a Hubble times, while our ”long-time” simulation spans
variety of parameters helped to find good initial condi- roughly ten Hubble times, see below). A full overview
tions for all mergers with more than two solitons. We of the parameters of the solitons can be found in table
highlight here some of our results. XII. Since we are in the x-y-plane, all z-components are
zero.
For the ”short-time” simulation we used a box size
A. Long-term behavior of 4 CU (∼ 150 kpc) and a resolution of 384, which is
enough to avoid the high-mass problem, although it is
We already addressed the generic problem in SFDM not tailored to yield an accurate total mass resolution,
to produce successful mergers within a Hubble time, since this would have caused a very long calculation time
given the highly oscillatory nature of the system which and the mass accuracy is not so important for the trend
makes relaxation harder to attain, compared to standard of the density profile, as we discussed above [60]. The
DM models. After all, identifying a meaningful final duration of the ”short-term” simulation was set to the
merger end state requires some (near-) equilibrium to usual 0.18 CU (∼ 13.59 Gyr), i.e. roughly a Hubble time.
be achieved. Indeed, in the course of performing many The resulting density profile of the post-merger object at
test runs for the merger parameters, we found that, even the final snapshot can be seen in figure 17.
with good enough parameters where the merging pro- We clearly recognize oscillations in the tail of the pro-
cess was completed after half of the simulation time, the file. Apart from that, we also see that this merger de-
post-merger object at the final snapshot still oscillates veloped a more pronounced envelope, due to the higher
significantly. This oscillation is potent enough that it af- number of solitons (N = 5, compared to N = 2 before).
fects the density profiles such to yield changes over time As a result, and in accordance with expectations, the
(the central density can change, accompanied by persis- NFW model fits better now, in terms of the slope over
tent oscillations in the other parts). In this sense, it most of the envelope.
is not quite correct to speak of a final or end state of Now, for the long-term simulation we increase the du-
the post-merger object - it rather signifies the state at ration to 2 CU (∼ 150 Gyr), i.e. roughly ten times longer
the final chosen snapshot. Depending on the soliton and than a Hubble time. Owing to the long simulation time,
merger parameters, this non-relaxed final state may per- we needed to reduce the spatial resolution for computa-
 22

106
106
105
log o density ρ(r) [M⊙/kpc 3]

log of density ρ(r) [M⊙/kpc 3]

105
104
104
103
103
102
102
101
Density p)o ile 101 Den)ity profile
100
Schive: δs=589540.969, rsol=9.189 Schive: δs=1009948.83, rsol=8.303
NFW: δNFW=41704.566, rs=7.505 100 NFW: δNFW=1079.312, rs=17.983
10−1
100 101 102 100 101 102
log o radius r [kpc] log of radius r [kpc]

FIG. 17. The density profile of the final snapshot (∼ FIG. 18. The density profile of the final snapshot (∼ 150 Gyr)
13.59 Gyr) of the ”short-term” 5-soliton merger without SI, of the long-term 5-soliton merger without SI, where the soli-
where the solitons collided in the center of the box, initially tons collided in the center of the box, initially configured in a
configured in a pentagon/star shape in the x-y-plane. The pentagon/star shape in the x-y-plane. The soliton parameters
soliton parameters for this particular density profile can be are the same as in figure 17 and can be found in table XII.
found in table XII. The box size was 4 CU and the resolution The box size was 8 CU and the resolution 256. The orange,
384. The orange, solid curve is the fit according to equ.(13), solid curve is the fit according to equ.(13), with units for δs in
with units for δs in M⊙ /kpc3 and rsol in kpc, while the green, M⊙ /kpc3 and rsol in kpc, while the green, solid curve is the
solid curve is the fit according to equ.(15), with units for fit according to equ.(15), with units for δNF W in M⊙ /kpc3
δNF W in M⊙ /kpc3 and rs in kpc. and rs in kpc. For this fit, the connection point in the fitting
routine was moved manually three data points towards the
right side.

tional feasibility, so we used a box size of 8 CU and a

resolution of 256. These latter two parameters have been
commonly used by the simulations of [27], but these au-
thors were not concerned with soliton mergers. In doing
so, we encounter the regime of the high-mass problem, pens to be in the central parts. This issue can be also
yielding a reduced central density, but since we are inter- seen in the middle part of the profiles: in the short-term
ested in the outer envelope/tail, this is no major restric- simulation we have a less-pronounced transition between
tion. The other parameters stayed the same as in the central and outer parts, while in the long-term simula-
previous run. The resulting profile of the post-merger tion we have a more clear-cut distinction between core
object for the long-term simulation is shown in figure and envelope parts of the merger object.
18. To fit this profile, we manually moved the connec-
tion point three data points towards the right side, such Finally, with respect to the envelope, we recognize that
that the NFW model fits the data better. Of course, in both runs the NFW model fits the outer data well, at
this comes with the price of loosing accuracy in the core least the slope is well represented, as a result of the multi-
model fit. The reason for moving the connection point ple merger. Nevertheless, the oscillations in the envelope
is that we can see a clear distinction between core and have not ceased after 150 Gyr, instead they are persistent
envelope, while the automatic search for the connection in terms of amplitude. Thus, we could show that not only
point failed in recognizing this feature in this multiple the core region but also the envelope undergoes some sig-
merger case. (Remember that the determination crite- nificant changes, well after one Hubble time. This char-
rion is tailored for single cores.) acteristic feature exemplified for FDM (i.e. SFDM with-
Moreover, we recognize the following. Although we out SI) is well in accordance with the findings of [47],
expected a lower central density, due to the loss of spa- confirming that SFDM - in this case FDM as a special
tial resolution, the central density is about 1.75 times case- has a very long relaxation time. It remains to be
higher than in the short-term simulation. The explana- seen whether this feature presents a potential problem for
tion is the non-relaxed, oscillating merger: the merging SFDM [61]. In terms of the fitting results, we also stress
object keeps oscillating well after one Hubble time, i.e. that a direct comparison of the resulting fit parameters
the density profile also keeps undergoing some changes. for the Schive and the NFW model between figures 17
Thanks to the oscillations, the object expands and con- and 18 seems not very useful, given the complications of
tracts, changing the central density in turn. It just so the non-relaxed state of the post-merger object (apart
happens that figure 17 displays a snapshot, where more from the very different snapshot in time between the fig-
mass happened to be in the middle part of the merger, ures). On the contrary, the difference in the fit results is
while figure 18 shows a snapshot where more mass hap- yet another signature of the non-relaxed state.
 23

B. Impact of the initial configuration onto the old and new. Old simulations are those where the high-
density profile mass problem had not yet been identified. Therefore, the
central densities of those simulations are lower than ex-
We have seen above that mergers keep oscillating well pected. For those simulations, a box size of 8 CU and 256
after the initial collision and mass assembly, with signif- for the resolution was used. New simulations took care
icant effects onto the density profile. Now, we might ask of the high-mass problem, but do not necessarily have an
how the oscillations themselves depend upon the initial accurate mass resolution. As already argued, this restric-
soliton configuration prior to their collision. In this sub- tion does not influence the central density much, which
section, we want to take a closer look on how the initial is why we did not pursue fine-tuned simulation runs. As
configuration, which determines the form of the collision for the simulation parameters, we use the ones from table
depending on number and position of the solitons, might II for the total mass.
impact the density profiles in the final state.
In order to do so, we focus our investigation on the cen-
tral densities of the profiles and analyse them, according
to their initial configuration. In our merger simulations, In figure 19, we collect the central densities from all
we only used scenarios where the solitons start in a given our multiple soliton merger simulations and plot them
initial position and then move toward the center of the over the respective configuration. The configurations are
box. In order for the merger to proceed evenly, all soli- sorted by total mass and, if equal, by number of soli-
tons have the same distance to the center, the same mass tons. In addition, an attempt was made to place the old
and move with a velocity of 5 CU (∼ 2.5 km/s). Again, and new simulations quite close to each other in order to
these parameters were chosen to guarantee that a merger illustrate the rise in density as a result of the better spa-
happens within a Hubble time, i.e. the simulation time tial resolution of the latter (Appendix A provides some
was also the same with 0.18 CU (∼ 13.59 Gyr). To keep more discussion, concerning the configurations.). We rec-
track of the configurations for each profile, we introduced ognize two main trends: i) Mergers with fewer solitons
a convention that looks like this: tend to produce a higher central density, until it is out-
weighed by the mass increase (inside the same volume)
A(B), Note. (28) due to higher number of solitons; see e.g. the 2−, 3− and
4−soliton mergers. Although the 3-soliton merger has a
Here is higher total mass, the central densities are lower than
for the 2−soliton mergers. This effect can be also seen
• A the number of merging solitons. if we compare the 2−soliton merger and the 4−soliton
merger with a total mass of 80 CU. This explains why we
• B denotes the mass of the solitons. If unspecified, usually get a drop in central density, whenever another
it can be easily inferred from the total mass and soliton is added.[62] ii) Head-on collisions tend to pro-
the number of solitons, since all solitons have the duce lower central densities, though there are some ex-
same mass. ceptions. They result in mergers with more pronounced
oscillations, and therefore an arbitrary choice of snapshot
• Note: an abbreviation to indicate the configura-
has a low probability to show a maximum central density.
tion, where
Higher central densities occur more frequently when the
– K denotes a frontal head-on collision along the merger is ”soft”, since then the oscillations are weaker,
given axis (e.g. Kx stands for a head-on col- leading also to a ”faster” approach to (near-)equilibrium.
lision along the x-axis). With a head-on colli-
sion we mean that two solitons approach each
other on an axis by moving directly towards
To showcase some results in more detail, we present
each other.
simulations of higher accuracy for dedicated cases below.
– x, y, and z: the three axes of the Cartesian Although the high-mass problem can not be avoided,
coordinate system in which the solitons are since the central densities are too high, we were able to
initialized. find box sizes such that the mass accuracy is satisfac-
– a hyphen: a separation between axes (e.g. x-y tory. We ran two different merger scenarios (with and
means, one soliton is initialized on the x-axis without head-on collision) for the N = 3 and N = 6 soli-
and one on the y-axis). ton merger. These cases illustrate the findings mentioned
above, in terms of the trend in the central density. To in-
We also use words to describe the configuration (e.g. crease the ”smoothness” of the final merger product, we
star shape or triangle). According to this convention chose individual soliton masses of 40 CU, which results
a ”4, Kx-y-z” stands for 4 solitons (with a given to- in a sufficiently high mass for the merger to settle earlier
tal mass), where two collided head-on along the x-axis, into a (near-)equilibrium, meaning that the envelope is
while the other two were initialized on the y and z axis, not so chaotic and produces not as many oscillations as
respectively. In our convention, we also use the term for smaller-mass solitons.
 24

107
log of central density [[M⊙/kpc 3]

106

105

6 (40), Rando (old)

2, Kx (new, 40)
2, )-3D (old, 40)
)
2, K) (ne(, 60)
3, Ky-) (old, 60)
3, )-y-z (old, 60)
3, Triangle (old, )

2, K) (ne(, 800))
3, )-y-z (ne(, 60)
3, Triangle (ne(, 60)

4, K)-Ky (ne(, 8 )
4, K)-Kz (old, 80)
4, K)-Kz (ne(, 8 )
4, Ky-)-z (old, 800)
4, Ky-)-z (ne(, 8 )
4, Square (old, 80)

5 (40), Rando (old)

4, Square (ne(, 0)
5, Star (old, 1000)
5, Star (ne(, 100)
)
6 (20), K)-Ky-Kz 100)
6 (40), K)-Ky-Kz (old)
(old)
2, )-3D (ne(, 40

3, K)-y (ne(, 60

4, K)-Ky (old, 80
0
6

5, Rando (old,
Configuration
FIG. 19. The central density from the final snapshots (∼ 13.59 Gyr) plotted versus the used geometric initial configuration.
The note indicates, according to the convention in (28), if the simulation data stem from old or new simulations and which
total mass was used. In Appendix B, we add more discussion.

1. In detail: The (N = 3)-case soliton parameters can be found in tables XV and XVI.
The resulting density profile of the post-merger object
For this case, we choose the Kx-y and the triangle sce- for this scenario is shown in figure 21.
nario. First, we discuss the Kx-y-scenario. Here we use In this case, the profile is still dominated by the core.
the simulation parameters from table XIII to get an ac- However, the oscillations in the envelope are much less
curate mass. The soliton parameters for this scenario are pronounced than in the previous case. This shows that
listed in table XIV, while the resulting density profile of the triangle-shaped merger is softer in comparison to the
the post-merger object can be found in figure 20. As ex- Kx-y merger. Also, the slope of the NFW model fits
pected, the profile is still mostly dominated by the core. better to the outer data. Additionally, we see that the
An envelope can be recognized, but the NFW model fits central density is lower than in the Kx-y-case. This sug-
the numerical data poorly. gests that we got a lucky snapshot in figure 20, where
Now, we draw our attention to the other (N = 3)- the central density happens to be higher. On the other
scenario, which is the triangle shape. The simulation and hand, in figure 21 we see a state, where the merger looks
 25

TABLE XIII. The simulation parameters for the (N = 3) Kx- TABLE XVI. The soliton parameters for the (N = 3) merger,
y merger. The box size is ∼ 115 kpc on a side, the simulation where the solitons are positioned in a triangle shape with
time is ∼ 13.59 Gyr, which results together with the saving individual mass of 40 CU ∼ 89 · 106 M⊙ , an initial distance of
number (gives the number of snapshots) in a temporal reso- solitons to the center of ∼ 15 kpc, and the absolute value of
lution of 1.359 Gyr. the velocity for each soliton is ∼ 2.5 km/s.
Parameter Value Unit soliton nr. mass x y z vx vy vz
box size 3 CU 1 40 0.4 0 0 −5 0 0
resolution 256 / 2 40 −0.27772 −0.27772 0 3.4715 3.4715 0
simulation time 0.18 CU 3 40 −0.27772 0.27772 0 3.4715 3.4715 0
saving number 10 /

107

R_99
TABLE XIV. The soliton parameters for the (N = 3) Kx-y 106

log of density ρ(r) [M⊙/kpc 3]

merger with individual soliton mass of 40 CU ∼ 89·106 M⊙ , an
initial distance of solitons to the center of ∼ 15 kpc, and the 105
absolute value of the velocity for each soliton is ∼ 2.5 km/s.
104
soliton nr. mass x y z vx vy vz
1 40 0.4 0 0 −5 0 0 103
2 40 0 0.4 0 0 −5 0
3 40 −0.4 0 0 5 0 0 102 D nsity p)ofil
Gauß: a=1493201.257, σ=1.941
101 NFW: δNFW=129602.124, rs=5.292
107 100 101
log of radius r [kpc]
R_99
106
log of density ρ(r) [M⊙/kpc 3]

FIG. 21. The density profile of the final snapshot (∼

105
13.59 Gyr) of the (N = 3) merger, where the solitons are
positioned in a triangle shape with parameters in tables XV
104
and XVI. The black vertical marks that radius which con-
tains 99 % of the mass of the post-merger object. The orange,
103
solid curve is the fit according to equ.(12), with units for a in
102
M⊙ /kpc3 and σ in kpc, while the green, solid curve is the fit
Density )rofile according to equ.(15), with units for δNF W in M⊙ /kpc3 and
Gauß: a=1550479.269, σ=1.903 rs in kpc.
101 NFW: δNFW=170453.782, rs=4.968
100 101
log of radius r [kpc]
close to equilibrium, exemplified by its smooth profile.
FIG. 20. The density profile of the final snapshot (∼
13.59 Gyr) of the (N = 3) Kx-y merger with simulation pa-
rameters in tables XIII and XIV. The black vertical marks 2. In detail: The (N = 6)-case
that radius which contains 99 % of the mass of the post-
merger object. The orange, solid curve is the fit according
to equ.(12), with units for a in M⊙ /kpc3 and σ in kpc, while Here we choose two scenarios, a pronounced head-on
the green, solid curve is the fit according to equ.(15), with collision and its opposite. For the first we use the Kx-Ky-
units for δNF W in M⊙ /kpc3 and rs in kpc. Kz-scenario, using the simulation parameters from table
XVII and the soliton parameters from table XVIII. With
these parameters, we end up with the density profile of
TABLE XV. The simulation parameters for the (N = 3) the post-merger object shown in figure 22.
merger, where the solitons are positioned in a triangle shape. This profile exemplifies what we were expecting: due
The box size is ∼ 119 kpc on a side, the simulation time is to the higher number of merging solitons we get a higher
∼ 13.59 Gyr, which results together with the saving number central density and a bigger envelope, compared to the
(gives the number of snapshots) in a temporal resolution of (N = 3)-cases. The envelope shows a lot of oscillations,
1.359 Gyr. thanks to the many violent head-on collisions, while its
Parameter Value Unit slope matches the slope of the NFW model.
box size 3.1 CU For the other (N = 6)-case, we choose a ”random con-
resolution 256 / figuration”, by which we mean that we picked a value for
simulation time 0.18 CU one positional coordinate from a random generator and
saving number 10 / adjusted the other parameters, such that they meet our
requirement of same distance and velocity of 5 CU. The
 26

TABLE XVII. The simulation parameters for the (N = 6) TABLE XIX. The simulation parameters for the (N = 6)
Kx-Ky-Kz merger. The box size is ∼ 142 kpc on a side, the merger, where the solitons are positioned ”randomly” in the
simulation time is ∼ 13.59 Gyr, resulting together with the simulation box. The box size is ∼ 145.5 kpc on a side, the
saving number (gives the number of snapshots) in a temporal simulation time is ∼ 13.59 Gyr, resulting together with the
resolution of 1.359 Gyr. saving number (gives the number of snapshots) in a temporal
resolution of 1.359 Gyr.
Parameter Value Unit
box size 3.72 CU Parameter Value Unit
resolution 256 / box size 3.8 CU
simulation time 0.18 CU resolution 256 /
saving number 10 / simulation time 0.18 CU
saving number 10 /

TABLE XVIII. The soliton parameters for the (N = 6) Kx- 108

Ky-Kz merger with individual mass of 40 CU ∼ 89·106 M⊙ , an
107 R_99
initial distance of solitons to the center of ∼ 23 kpc, and the

log of density ρ(r) [M⊙/kpc 3]

absolute value of the velocity for each soliton is ∼ 2.5 km/s.
106
soliton nr. mass x y z vx vy vz
105
1 40 0.6 0 0 −5 0 0
2 40 0 0.6 0 0 −5 0 104
3 40 0 0 0.6 0 0 −5
4 40 −0.6 0 0 5 0 0 103
5 40 0 −0.6 0 0 5 0
6 40 0 0 −0.6 0 0 5 102 D nsit/ )rofil
Schiv : δs=11574406.17, rsol=4.123
101 NFW: δNFW=1766486.545, rs=2.966
100 101
log of radius r [kpc]
simulation parameters for this scenario can be found in
table XIX, while the soliton parameters are listed in ta-
FIG. 23. The density profile of the final snapshot (∼
ble XX. The resulting density profile of the post-merger 13.59 Gyr) of the (N = 6) merger, where the solitons are po-
object is shown in figure 23. sitioned ”randomly” in the simulation box with parameters
As expected, the profile has a higher central density in tables XIX and XX. The black vertical marks that radius
than the Kx-Ky-Kz case. Also, we see a more pronounced which contains 99 % of the mass of the post-merger object.
core-envelope structure. The envelope itself shows less The orange, solid curve is the fit according to equ.(13), with
oscillatory behavior than in the previous case, and there- units for δs in M⊙ /kpc3 and rsol in kpc, while the green, solid
curve is the fit according to equ.(15), with units for δNF W in
M⊙ /kpc3 and rs in kpc.

107
R_99
fore the NFW model fits the data better.
log of density ρ(r) [M⊙/kpc 3]

106
In summary, the specific cases demonstrate once more
105 what we have seen in the trend plotted in figure 19. Head-
on collisions delay the merging process and its equilibra-
104
tion, exemplified by non-relaxed density profiles. Statis-
103 tically, we will end up with a lower central density in the
final snapshot. This was shown here explicitly, since we
102 Density )rofile even used a higher soliton mass in order to dampen the
Gauß: a=4940870.803, σ=1.578 oscillations. Thus, we conclude that a soft merging pro-
101 NFW: δNFW=213353.66, rs=4.308
cess helps to accomplish a successful merger, depending
100 101
log of radius r [kpc] on the relative velocity and the initial geometric configu-
ration; a result which is in good accordance with expec-
FIG. 22. The density profile of the final snapshot (∼ tation and with ”classic mergers”.
13.59 Gyr) of the (N = 6) Kx-Ky-Kz merger with parame- Also, we could show that the (N = 6)-case produces
ters in tables XVII and XVIII. The black vertical marks that a higher central density, even with the head-on-collision
radius which contains 99 % of the mass of the post-merger ob- scenario. This matches our prediction based upon the
ject. The orange, solid curve is the fit according to equ.(12), trend of figure 19. If we had used a smaller soliton num-
with units for a in M⊙ /kpc3 and σ in kpc, while the green, ber, we would have probably (depending on if we get a
solid curve is the fit according to equ.(15), with units for lucky or unlucky snapshot) seen a smaller central density,
δNF W in M⊙ /kpc3 and rs in kpc. especially if the scenario includes head-on collisions.
 27

TABLE XX. The soliton parameters for the (N = 6) merger, where the solitons are positioned ”randomly” in the simulation
box with individual mass of 40 CU ∼ 89 · 106 M⊙ , an initial distance of solitons to the center of ∼ 23 kpc, and the absolute
value of the velocity for each soliton is ∼ 2.5 km/s.
soliton nr. mass x y z vx vy vz
1 40 0.3 0.1 −0.51 −2.5 −0.83333 4.25
2 40 0.5 −0.3 0.14142 −4.16666 2.5 −1.1785
3 40 −0.2 0.26458 −0.5 1.66666 −2.20483 4.16666
4 40 0 0.6 0 0 −5 0
5 40 0.3 0.4 0.33166 −2.5 −3.33333 −2.76238
6 40 −0.1 −0.4 0.43589 0.83333 3.33333 −3.63242

VI. SUMMARY AND DISCUSSION merging solitons for models without SI (typically FDM),
and for models with a weak-to-intermediate, repulsive
SI included. For this purpose, we used and modified
Scalar field dark matter (”SFDM”) has become a popu-
the open-source code PyUltraLight, calling our modified
lar candidate for the cosmological DM, with its potential
variant PyUltraLight_SIwhich we aim to make public.
to cure several ”small-scale” structure formation prob-
The code solves the Gross-Pitaevskii-Poisson equations of
lems in the standard CDM paradigm. SFDM is com-
motion, using a symmetric, pseudospectral Fourier-split-
prised of (ultra-)light spin-zero bosons in a Bose-Einstein
step algorithm. We used periodic boundary conditions
condensate, without or with a self-interaction (SI) in-
for all our simulations.
cluded. The overwhelming astrophysics literature con-
siders models of ultralight bosons without SI, i.e. ”fuzzy The user can specify the value of a non-dimensionalized
dark matter” (FDM), or the opposite regime of a strongly SI coupling strength, according to the nonlinear SI term
repulsive SI, the so-called Thomas-Fermi (”TF”) regime, in equ.(10). Furthermore, a prime focus of ours is the
or SFDM-TF. Whereas FDM has been modelled, us- study of density profiles of single solitons, as well as of
ing the nonlinear Schrödinger or Gross-Pitaevskii wave post-merger objects made of multiple solitons, because
equation, along with the Poisson equation, SFDM-TF as the central cores of SFDM halos, they exhibit differ-
has been often modelled, using fluid approximations to ences to CDM predictions, and are eventually subject
the fundamental wave equation of motion. As such, the to observational constraints. Therefore, we included an-
numerical algorithms, strategies and computational de- other code feature to the original PyUltraLight soft-
mands vary, but either regime poses serious challenges ware, which allows to extract radially-averaged density
to realistic cosmological simulations. Therefore, previ- profiles from the numerical density data, in order to use
ous literature has often focused on more simplistic sce- them as input to a fitting routine. The latter is able
narios, where single or multiple (near-)equilibrium ob- to fit analytic (closed-form) density profiles, known from
jects in SFDM have been investigated. These so-called previous literature. In this paper, we used a Schive or
”solitons”, which are also the focus of this paper, exhibit Gauss profile, equ.(13) or equ.(12), respectively, to fit
the basic features of SFDM gravitational dynamics, and the profiles of single solitons and the cores of post-merger
have been found to approximate the central parts (the objects, while a NFW profile equ.(15) is used for the en-
”cores”) of larger, galactic-scale halos in actual cosmo- velope of post-merger objects. Surely, users may change
logical simulations, i.e. ”solitonic cores” in FDM, see or extend this choice in the future. Applying our fitting
[6, 7, 9, 10], or ”polytropic/TF cores” in SFDM-TF, see routines to our simulation results, we find that the Schive
[12, 13, 15, 19]. Given these findings and the fact, that profile is generally favored over the Gaussian for single
the details of soliton dynamics and SFDM structure for- solitons and the cores of post-merger objects, if Λ̂ = 0.
mation are far from being fully understood, the need for In this regard, we confirm the early work by [6], who in-
dedicated simulations of solitons and their merger dy- troduced what we call ”Schive profile” as a good fitting
namics remains. Moreover, realistic cosmological simula- model in the first place. However, when it comes to the
tions and their high-performance computing requests ex- central densities, there is not much difference between the
clude the possibility to probe the boson parameter space, Gaussian and Schive profile. Also, we find that the size
in terms of boson mass or SI coupling strength, in any ef- of cores does generally not only depend upon the value
ficient nor exhaustive way. On the other hand, ”smaller- of Λ̂, but also upon soliton mass. Overall, for Λ̂ 6= 0, we
scale” soliton simulations are computationally cheaper, find a mild preference for the Gaussian core profile.
especially for static backgrounds, and thus allow an eas- Another feature of interest to us in this paper was the
ier sampling of SFDM parameter spaces. In fact, the derivation of the scaling relation between central density
various SI regimes are still in parts poorly explored, if ρc and mass M of single solitons. Since solitons serve
at all, and low and intermediate SI strengths have been as approximate halo cores, this relationship is of impor-
rarely considered. tance when it comes to the comparisons to galactic data
In this paper, we study the dynamics of single and or CDM predictions. For Λ̂ = 0, our single soliton simu-
 28

lations confirmed the well-known relationship, ρc ∝ M 4 . periodic boundary conditions is not optimum for merger
Among other tests, this agreement also established that simulations. While it is true that many previous works
our code modifications to incorporate finite SI worked have made use of an absorbing sponge as boundary con-
correctly, upon setting Λ̂ = 0 as in the original code. dition in merger simulations, we stress that we have ob-
Performing our single soliton simulations, we were also served much care in our simulations to pick appropriate
able to derive numerical ρc − M relations for weak SI, parameters of, say, box size, in order to avoid the loss
Λ̂ = 0.02 and Λ̂ = 1, respectively. In contrast to models and re-entry of any SFDM mass in the box upon soli-
without SI (”FDM”, Λ̂ = 0), or models in the strongly ton collisions and merging. Our simulations had never
repulsive TF regime (”SFDM-TF”, Λ̂ ≫ 17), there is no more than 6 solitons involved, but if the latter exceed
analytic relationship available to use as a fitting function 10, an absorbing sponge is probably recommended. In
to our numerical results, as far as we know. In lack of fact, the recent work by [66] has considered this ques-
an analytic derivation, we refrained from presenting fits, tion in detail for Λ̂ = 0, and as far as density profiles
although it appears that a low-order polynomial multi- are concerned, the results are in agreement. However,
plied by a Gaussian can fit the data. This means that more merger simulations for Λ̂ 6= 0 are desirable in the
the initial increase of ρc with increasing M turns over af- future. As discussed in our paper, we experienced severe
ter reaching a peak at some soliton mass -which depends issues which prevented us from creating good mergers for
upon Λ̂- into the opposite behavior, where ρc decreases Λ̂ ≥ 1. While periodic boundary conditions and limited
with increasing M , in sharp contrast to the monotonic computational times are probably part of the problem,
scaling relations known for FDM or SFDM-TF solitons. we also suspect that other numerical algorithms might
We present arguments to explain this phenomenon, on be required to tackle the intermediate SI regime in an
the basis of the mutual competition between quantum efficient way.
and SI pressure vs self-gravity. However, more work will With respect to other recent work which is related to
be required and, ideally, analytic ρc −M relationships can ours, we mention some of them as follows. The authors
be derived in the future, for the weak and intermediate of [63] have also simulated soliton mergers, in order to
SI regimes. derive density profiles of post-merger objects, for Λ̂ = 0,
Furthermore, we confirmed in our work what has been and to study the implications for core-halo mass rela-
seen in previous literature, namely that a NFW fit to tions. They used an absorbing sponge, given that their
the outer envelope only has a chance to fair well, if that single solitons (before merging) are at least a factor of
envelope is well developed, as a result of the merger of five more massive than ours, and given their higher num-
many solitons. Binary soliton mergers have poorly de- bers of solitons (up to 14). As a result, they can use
veloped envelopes and exhibit poor fits. However, even smaller simulation boxes than we do, which translates
for our multiple soliton merger simulations (up to 6 soli- into much smaller radial ranges in their plotted density
tons), the outcome of a NFW fit as being good or bad profiles. In order to fit their density data, they use exclu-
is somewhat erratic, depending on whether the chosen sively a Schive profile for the core, along with a relation
snapshot is still undergoing equilibration, as they mostly to determine the size parameter of the core. Similarly to
do. Of course, our numbers of solitons is still small, com- us, at some outer radius a NFW profiles takes over.
pared to the soliton merger simulations by e.g., [9, 24, 63]. While our paper was in the process of being finished,
Yet, we think caution should be observed for several rea- we noticed the very recent work by [67], which is one
sons. First, in non-expanding backgrounds (such as used of the rare cases, where SFDM with a finite SI is con-
here and in most other soliton merger simulations), we sidered. Virialized single solitons are a focus of their
actually expect outer halo density profiles to fall off as work by investigating the impact of (mostly repulsive)
∝ r−12/7 , regardless of the DM model, see [12] and ref- SI onto soliton shape and oscillations. Since the authors
erences therein, rather than the NFW profile. Second, are interested in deriving analytic and scaling relations
the fact that cosmological CDM-only simulations predict in order to pinpoint the transitions between weak and
universal NFW profiles at all mass scales, does not im- strong SI, they use Gaussians, which take into account
ply that NFW should fit well for post-merger objects of the SI parameter, for this purpose (see also [37]), al-
arbitrary mass in SFDM. In fact, if NFW (or another though Schive profiles (no SI) and (modified) (n = 1)-
CDM-inspired outer model profile) does not fit well, it polytropes (strongly repulsive TF regime) are also con-
will speak against using it for this regime of SFDM struc- sidered. As discussed in previous literature (e.g. in [68]),
ture formation, and may rather be a sign that the in- the authors of [67] also find degeneracies between bo-
herent quantum and SI dynamics dominate over genuine son mass and SI coupling strength, such that constraints
gravitational dynamics in such post-merger objects. By (on mass only) from observations should be reconsid-
extension, this same caution should be observed when ered, for SFDM models with SI. Interestingly, they also
actual galactic velocity data of dwarf galaxies is fitted find similar results from their scaling analysis than we
to closed-form, ”ironclad” core-envelope density profiles do from our dynamical simulations of dedicated single
(see e.g., [40, 64, 65]). In addition to these considera- soliton cases: For increasing, repulsive SI strength, the
tions, there is the issue of boundary conditions and box solitons expand, i.e. the core radius grows, while their
sizes, and some readers may object that our choice of central density decreases, compared to solitons without
 29

SI. In this weak and intermediate SI regime, the Gaus- Nevertheless, a ”gross similarity” to CDM in the en-
sian core fares generally better than both the Schive or velopes of fully-fledged FDM or SFDM-TF galactic ha-
the TF profile, which is in accordance with our results. los has been established in several previous works, and
A more in-depth theoretical comparison and analysis of explained in [12] as a result of the dynamical similar-
their results with respect to ours is worthwhile, which we ity (or ”correspondence”) between large-scale, coarse-
defer to future work. grained quantum dynamics and large-scale velocity dis-
We might also speculate whether the ”core expansion” persion in collisionless CDM. In fact, even toy models
of single solitons with finite, repulsive SI (compared to respect this correspondence at the level of the pressure
Λ̂ = 0), which we can understand well as a result of terms, as shown in [12], strengthening the coarse-graining
the more prominent SI pressure, lies already at the heart argument. Still, the impact of the gravity term onto this
of the phenomenon of post-evolved, core-plus-envelope correspondence seems to matter, and we might say that
expansion seen in the 3D halo simulations of [19]. After our results here suggest that this dynamical similarity
all, the effect is more prevalent, the higher the repulsive fades below a certain mass scale of objects in gravita-
SI. In the TF regime considered in [19], primordial halo tional (near-)equilibrium. In this light, we might ques-
cores of radius ∼ 0.1 kpc were found to expand to 2 tion whether it is useful and justified to enforce a NFW
kpc within dwarf-galactic SFDM-TF halos of 109 M⊙ by tail upon SFDM profiles, in attempts to construct SFDM
redshift zero. While we should not confuse an in-built, wave functions as eigenstates with this property, as e.g.
”static” result with an unfolding, dynamical process, it pursued by [70], although they find deviations from NFW
remains to be seen how genuine SFDM physics blends in the outer parts of their massive model halos, upon
in detail with gravitational dynamics upon more future evolving their initial states with a Schwarzschild method.
work. Another approach to study equilibrium objects in SFDM
The authors of [69] recently considered the regime of is due to [71], in which the cores are ”fixed” as soli-
”TF solitons” within halos made of SFDM-TF. Their re- tonic solutions, but the envelopes or tails are dynam-
sults may be compared to the appropriate papers study- ically evolved upon the choice of outer boundary con-
ing the TF regime ([13, 19]), rather than our weakly re- ditions (possibly subject to different physical considera-
pulsive cases. In our context, we merely notice that [69] tions), and constrained by the total mass. In our paper,
find similar phenomena with respect to a smooth central the radially averaged density data are the result of pure
profile, with density oscillations in the envelopes, whose dynamics, without any in-built assumptions on how the
amplitude diminish, the larger the mass. Also, long-time profiles (or profile parts) should look like in the begin-
fluctuations persist, such that the system never really re- ning or the end. The choice of analytic fitting model is
laxes. The authors find that solitons form earlier, if the subject to scrutiny, of course, as elaborated above.
simulation boxes are bigger, which suggests that there is In any case, more simulations of solitons and halos
more mass available to help the formation process. We in weak and intermediate SI regimes, Λ̂ 6= 0, will be
also found higher masses in bigger boxes, compared to required in the future to make a more in-depth and fair
the input values, and this effect may be related (for one comparison to the results in this paper.
thing, periodic boundary conditions were also used in Before we close, let us mention a different perspec-
[69]). tive, altogether, as follows. The effect of gravity onto
Finally, we highlight again that we also see global, per- the dispersion of Gaussian wave packets has been also a
sistent oscillations in our post-merger objects, which have field of study in the laboratory, using atomic or molec-
been seen in early and recent multiple soliton merger ular systems, in quests to understand the interplay be-
simulations, see e.g.[25, 47], as well as in cosmological tween quantum mechanics and gravity. Of course, the
FDM simulations, such as [9], along with the papers analogy to SFDM solitons is limited by the fact that
cited above. This indicates a very long relaxation time the self-gravity of laboratory systems is highly negligible,
in SFDM mergers - at least for models without SI (previ- compared to the ambient gravitational field of the Earth.
ous and this work) and for weak repulsive SI (this work), Still, theoretical investigations have been pursued, as e.g.
questioning the usefulness of CDM-inspired profiles. In by [49], where the Schrödinger-Newton equation is used
general, we find more oscillations in the envelopes at to calculate numerically the onset of gravitationally in-
higher soliton mass, but a non-vanishing Λ̂ leads to some duced inhibitions of the dispersion of wave packets and
”wash-out” of these oscillations. In models with finite its mass dependence. For sufficiently high mass - in
SI, the SI pressure is more potent in the central parts of their case close to 1010 atomic units, the wave packet
(near-)equilibrium objects. However, during the process will not only resist dispersing, but actually starts con-
of equilibration our simulations suggest that the SI pres- tracting. On the other hand, smaller-mass wave pack-
sure does affect the outskirts as well, by smoothing over ets will expand i.e. disperse, unhampered by self-gravity
local wave disturbances and interferences, caused by the of the constituent particles (of, say, complex molecules).
quantum pressure. Thus, we notice already at the level of We mention this particular example of paper, because it
few-soliton mergers what has been observed in the more demonstrates theoretically for lab systems of a complete
sophisticated simulations of strongly repulsive SFDM-TF different physical scale what we see in our SFDM soli-
models by [13, 15, 19]. ton simulations, in line with our physical understanding:
 30

more massive solitons will benefit from their self-gravity be placed sufficiently far apart. Also, the velocities of
to form stable, non-dispersing equilibrium objects (either the solitons have to be set sufficiently low. But this com-
remain stable if single objects, or become stable upon bination is a problem. The solitons now have a greater
mergers), even if SI is added, as we showed. Of course, distance to travel while being slower and the merging
SFDM solitons of mass which surpass a certain maxi- process may not complete within a Hubble time. On the
mum mass can undergo gravitational collapse (we have other hand, a higher velocity implies a higher kinetic en-
not studied this regime in our paper). On the other hand, ergy of the solitons upon collision, which may also delay
SFDM solitons of small mass with comparatively little or inhibit a merger. Likewise, a smaller separation of
self-gravity are prone to dispersion upon gravitational solitons causes a (possibly significant) overlap, prevent-
interaction with their more massive brethren, similar to ing the unfolding of a ”classic merger” to begin with. All
what happens to small-mass subhalos in bigger hosts in these difficulties prevented us from simulating successful
standard CDM scenarios. mergers for all solitons with mass less than 5 CU, so we
focused on higher-mass solitons.
We presented a good ”work-around” solution for the
ACKNOWLEDGMENTS high-mass problem already in section III A 1. However,
while work on this project was well advanced, we found
We want to thank Shaun Hotchkiss and Emily Kendall a ”nicer” solution to determine at least a lower limit for
for their helpful comments regarding the main section the spatial resolution for each soliton mass, which does
of PyUltraLight. A special thanks goes out to Emily not depend on the numerical resolution or box size and
Kendall, which supported us in the development of den- still solves the high-mass problem by making use of the
sity profile. The authors acknowledge the support by the grid point length, i.e. the length between two grid points,
Austrian Science Fund FWF through an Elise Richter fel- along with the known central density for a 20 CU soliton.
lowship to T.R-D., project no. V 656-N28. T.R-D. fur- We use a 20 CU soliton as a benchmark, because this
thermore acknowledges the support by the FWF through mass regime is not heavily affected by the issues at high
the single-investigator grant (FWF-Einzelprojekt) no. or low mass (explained above). The grid point length can
P36331-N, and the support by the Wolfgang Pauli In- be calculated by dividing the box length on a side by the
stitute in hosting this grant. resolution. For example, let us assume we have a soliton
with an arbitrary mass, that should have a central density
of around 54 · 103 M⊙ /kpc3 . For our 20 CU soliton we
Appendix A: Details on spatial resolution and mass know we have a central density of 27 · 103 M⊙ /kpc3 , if we
accuracy, exemplified for Λ̂ = 0 use 8 CU for the box size, 256 for the numerical resolution
and Λ̂ = 0. These parameters are good enough, such that
In section III A 1, we mentioned the issues from the we do not cut off the central region significantly, hence
competing requests of resolution and computation time. avoid the high-mass problem. The grid point length in
There are basically two main problems for all simula- this case is 8 divided by 256 or 1/32 CU. This would be
tions: i) the high-mass problem and ii) inaccurate soli- the minimal spatial resolution we need. Anything above
ton masses. For mergers there is an additional problem, this would work, too, but is not necessary (such as not
namely the low-mass problem. to increase the computation time). To ensure now that
To reiterate, the high-mass and low-mass problem are our soliton with arbitrary mass does not suffer from the
related to the specific mass-size relationship of (single) high-mass problem, we need a spatial resolution that is
solitons: for Λ̂ = 0, the soliton radius (properly de- at least twice the spatial resolution of the 20 CU soliton.
fined, e.g. R99 ) goes inversely proportional to the soli- This can be achieved by halving the grid point length
ton mass. Thus, high-mass solitons are comparatively of the 20 CU soliton, yielding 1/64 CU. The fraction is
smaller in size than low-mass solitons. This gives us an very useful, since it tells us, which relation the box size
upper limit for soliton masses, since higher-mass solitons in CU has to the resolution. Thus, in this case we could
require a higher resolution, such that the central regions simulate our soliton with arbitrary mass with a box size
can be resolved, avoiding a ”cut-off” of the central den- of 1 CU and a resolution of 64. Options which require
sity. The increase in resolution will ultimately lead to a bigger box only have to hold the relation of box size
a significantly higher, and at some point, unrealizable, to resolution equal 1 to 64. This way, we can get an
computation time. In our case, the highest soliton mass estimate of the minimal spatial resolution for the single-
we simulated was 300 CU. soliton case.
One the other hand, the broader low-mass solitons are This is a useful approach, especially when test simu-
a problem for merger simulations, and for further rea- lations are performed, in order to inform in advance the
sons. Not only do mergers take significantly longer if the target simulations of higher accuracy or realism. The
individual soliton masses are low, but one also runs in a only disadvantage is the prior approximate knowledge of
vicious circle of competing demands, as follows. In or- the central density of our soliton. In fact, we can use
der to ensure that the bigger, low-mass solitons do not the central-density-mass relationship to infer the density.
overlap at the beginning of the simulation, they should And by extension our approach would also work for cases
 31

with finite SI, Λ̂ 6= 0, if analytic relationships will once light of the requests from the high-mass problem, mass
be derived. However, the case of mergers is more com- accuracy and computation time. Still, we caution the
plicated, although this approach gives some indication, reader that some cases defy a good solution for all issues;
if the spatial resolution is high enough to be not affected the ideal parameters may simply be not achievable, if
by the high-mass problem. Indeed, this is why we knew too high a spatial resolution (like in the unequal merger
in the cases of unequal mergers in sections IV A 2 and cases), or too big a box size (violent merger of solitons
IV B that the spatial resolution is too low, impacting the with repulsive SI) is required, for a given limited compu-
accuracy of the central densities that we got. tation time.
The approach of using grid point length has been also
a good starting point for getting accurate masses. To re-
iterate, the problem of ”inaccurate soliton masses” refers Appendix B: Configuration tables
to the fact that the mass calculated from spherically-
averaged density profiles, using e.g. the cutoff radius R99 , In section V B, figure 19, we used some abbreviations
of post-evolved solitons can deviate significantly from the for configurations which were not explained. For illus-
initial input soliton mass which is set in the input file tration, we will provide the tables for the initial soliton
soliton solution.py. This deviation is quite independent parameters for these cases in this appendix. The mass
of the numerical integration procedure to calculate the of the individual solitons can be calculated using their
cumulative mass up to R99 . In general, the calculated number and the total mass of the merger given in figure
masses turned out to be higher than the input value, al- 19.
though the opposite also happens, especially for Λ̂ 6= 0, The first scenario that was not discussed is the x-3D-
although mass does not leave the simulation box. case. This configuration describes a 2−soliton merger,
One reason for the inaccurate masses from the cumu- where one soliton moves along the x-axis, while the other
lative mass integration is the artificial radial cutoff (e.g. heads from diagonally above. For both solitons the initial
for R99 ), given that solitons for Λ̂ = 0 have no compact distance to the center of the box, where they collide,
support and the wavefunction needs to be truncated. In was 0.35 CU. The soliton parameters for this scenario are
the code, we have to set or rather change the predefined shown in table XXI.
”maximum radius”, which determines the ”edge” of the Another configuration for which we performed simula-
wavefunction profile. If the box size now happens to be tion runs is the square scenario. As the name suggests,
chosen too large, the mass gets overestimated. Thus, in the solitons are arranged in a square shape in the x-y-
general we had to reduce the box size to cut off the den- plane with a distance to the center of the box of 0.4 CU.
sity profile, such that the mass matches the total input The according soliton parameters can be found in table
mass of the soliton. In cases where the mass had been XXII.
underestimated, the box size was simply too small. In Finally, we mention the ”random merger” with 5 soli-
order to fix this issue, it is required to find the correct tons, which is similar to the random 6−soliton merger.
box size for each simulation case at hand. We tried many For illustration we include the soliton parameters in table
other different routes, e.g., an increase in the numerical XXIII. The initial distance to the center for all solitons
resolution can help for certain soliton masses. However, was 0.6 CU.
more simulations will be required in the future to pin- All of these configurations exemplify the same behavior
point general trends. Nevertheless, along with the use than the cases which we discussed in section V B, as fol-
of the grid point length, we were able to find ”ideal” pa- lows. The configurations here have a higher central den-
rameters in a decent amount of time, for cases of interest. sity than their respective counterparts with one or two
For each scenario, it is generally advisable to run some head-on collisions, except for the 2−soliton merger, which
test simulations with low (numerical) resolution, to get can be explained (again) by having picked a ”lucky”
an idea what the central density and the mass looks like. snapshot, where the oscillation yields a maximum in the
center. Also, the profiles look much smoother than those
Now, to summarize our procedure: For the central den-
sities which we got, we used the grid point length method of the head-on collisions, with fewer and lower-amplitude
oscillations in their envelopes. As elaborated in the main
from above to see which combinations of numerical res-
olutions and box sizes are feasible. Next, we reduced text, mergers without head-on collisions develop in a
more gentle manner. Additionally, mergers involving a
the box size to obtain a calculated soliton mass near the
total input mass (in fact, a little above it, such that higher number of solitons develop fewer oscillations, as
well, since the higher total mass helps the merger to calm
the increase in resolution gets us to the correct mass)
of the solitons and afterwards increased the numerical down more rapidly. Furthermore, such post-merger ob-
jects also have a more pronounced envelope, compared to
resolution.[72] To spare computation time, we chose the
minimum required numerical resolution. In most cases, those involving fewer solitons. Judging from our results,
it appears that mergers involving at least four solitons
a reduction in the box size alone got us so much over-
mark that transition, but this statement also depends
all spatial resolution, that we could refrain from using a
upon soliton mass and relative velocity.
numerical resolution higher than 256. Indeed, this reso-
lution has been sufficient for most of our simulations, in
 32

TABLE XXI. The initial soliton parameters for a (N = 2) soliton merger, where one soliton falls diagonally onto another,
which moves along the x-axis. For both solitons, the distance to the center is ∼ 13.5 kpc and the absolute value for the velocity
is ∼ 2.5 km/s. The abbreviation for this scenario is ”x-3D”.
soliton nr. x y z vx vy vz
1 0.20207 0.20207 0.20207 −2.88675 −2.88675 −2.88675
2 −0.35 0 0 5 0 0

TABLE XXII. The initial soliton parameters for a (N = 4) soliton merger, where the solitons are positioned in a two-dimensional
square shape. The distance to the center for all solitons is ∼ 15 kpc and all of them move with a velocity of ∼ 2.5 km/s to the
center. The abbreviation for this scenario is ”Square”.
soliton nr. x y z vx vy vz
1 0.27772 0.27772 0 −3.4715 −3.4715 0
2 0.27772 −0.27772 0 −3.4715 3.4715 0
3 −0.27772 −0.27772 0 3.4715 3.4715 0
4 −0.27772 0.27772 0 3.4715 −3.4715 0

TABLE XXIII. The soliton parameters for a (N = 5) soliton merger, where the solitons are positioned ”randomly” in the
simulation box. The initial distance to the center for all solitons is ∼ 23 kpc and the absolute value of the velocity for each
soliton is ∼ 2.5 km/s. The abbreviation for this scenario is ”Random”. Notice that this is the same abbreviation as for the
6−soliton merger, since the first 5 solitons have the same initial parameters.
soliton nr. x y z vx vy vz
1 0.3 0.1 −0.51 −2.5 −0.83333 4.25
2 0.5 −0.3 0.14142 −4.16666 2.5 −1.1785
3 −0.2 0.26458 −0.5 1.66666 −2.20483 4.16666
4 0 0.6 0 0 −5 0
5 0.3 0.4 0.33166 −2.5 −3.33333 −2.76238

[1] J. S. Bullock and M. Boylan-Kolchin, Small- [9] P. Mocz, M. Vogelsberger, V. H. Robles, J. Zavala,
Scale Challenges to the ΛCDM Paradigm, M. Boylan-Kolchin, A. Fialkov, and L. Hern-
Annual Review of Astronomy and Astrophysics 55, 343 (2017). quist, Galaxy formation with BECDM - I.
[2] A. Del Popolo and M. Le Delliou, Small Scale Prob- Turbulence and relaxation of idealized haloes,
lems of the ΛCDM Model: A Short Review, Galaxies Monthly Notices of the Royal Astronomical Society 471, 4559–4570
5, 10.3390/galaxies5010017 (2017). arXiv:1705.05845 [astro-ph.CO].
[3] S.-J. Sin, Late-time phase transition and the galactic halo [10] P. Mocz, A. Fialkov, M. Vogelsberger, F. Becerra,
as a Bose liquid, Phys. Rev. D 50, 3650 (1994). X. Shen, V. H. Robles, M. A. Amin, J. Zavala,
[4] W. Hu, R. Barkana, and A. Gruzinov, Fuzzy Cold Dark M. Boylan-Kolchin, S. Bose, F. Marinacci, P.-H. Chava-
Matter: The Wave Properties of Ultralight Particles, nis, L. Lancaster, and L. Hernquist, Galaxy formation
Phys. Rev. Lett. 85, 1158–1161 (2000). with BECDM – II. Cosmic filaments and first galaxies,
[5] T. Matos, F. S. Guzmán, and L. A. Ureña- Monthly Notices of the Royal Astronomical Society 494, 2027 (2020)
López, Scalar field as dark matter in the universe, https://academic.oup.com/mnras/article-pdf/494/2/2027/33051104/
Classical and Quantum Gravity 17, 1707–1712 (2000). [11] R. Hlozek, D. Grin, D. J. E. Marsh, and P. G. Ferreira, A
[6] H.-Y. Schive, T. Chiueh, and T. Broadhurst, Cos- search for ultralight axions using precision cosmological
mic structure as the quantum interference of a co- data, Phys. Rev. D 91, 103512 (2015).
herent dark wave, Nature Physics 10, 496–499 (2014a), [12] T. Dawoodbhoy, P. R. Shapiro, and T. Rindler-
arXiv:1406.6586 [astro-ph.GA]. Daller, Core-envelope haloes in scalar field dark
[7] H.-Y. Schive, M.-H. Liao, T.-P. Woo, S.-K. Wong, matter with repulsive self-interaction: fluid
T. Chiueh, T. Broadhurst, and W.-Y. P. Hwang, dynamics beyond the de Broglie wavelength,
Understanding the Core-Halo Relation of Quan- Monthly Notices of the Royal Astronomical Society 506, 2418–2444
tum Wave Dark Matter from 3D Simulations, arXiv:2104.07043 [astro-ph.CO].
Phys. Rev. Lett. 113, 261302 (2014b). [13] P. R. Shapiro, T. Dawoodbhoy, and T. Rindler-
[8] L. Hui, J. P. Ostriker, S. Tremaine, and E. Wit- Daller, Cosmological structure formation in
ten, Ultralight scalars as cosmological dark matter, scalar field dark matter with repulsive self-
Phys. Rev. D 95, 043541 (2017). interaction: the incredible shrinking Jeans mass,
 33

Monthly Notices of the Royal Astronomical Society 509, 145–173tion (2022),

in scalar field dark matter models and its conse-
arXiv:2106.13244 [astro-ph.CO]. quences for the formation of supermassive black holes,
[14] S. Hartman, H. Winther, and D. Mota, Con- Phys. Rev. D 103, 063012 (2021).
straints on self-interacting Bose-Einstein conden- [33] F. S. Guzmán and L. A. Ureña López, New-
sate dark matter using large-scale observables, tonian collapse of scalar field dark matter,
Journal of Cosmology and Astroparticle Physics 2022 (02), 005. Phys. Rev. D 68, 024023 (2003).
[15] S. T. H. Hartman, H. A. Winther, and D. F. [34] F. S. Guzmán and L. A. Ureña-López, Evolution of the
Mota, Cosmological simulations of self-interacting Bose- Schrödinger-Newton system for a self-gravitating scalar
Einstein condensate dark matter, aap 666, A95 (2022), field, Physical Review D 69, 10.1103/physrevd.69.124033
arXiv:2203.03946 [astro-ph.CO]. (2004).
[16] P. J. E. Peebles, Fluid Dark Matter, [35] T. Rindler-Daller and P. R. Shapiro, An-
The Astrophysical Journal 534, L127 (2000). gular momentum and vortex formation in
[17] J. Goodman, Repulsive dark matter, Bose–Einstein-condensed cold dark matter haloes,
New Astronomy 5, 103 (2000). Monthly Notices of the Royal Astronomical Society 422, 135–161 (20
[18] C. G. Böhmer and T. Harko, Can dark https://academic.oup.com/mnras/article-pdf/422/1/135/18591284/m
matter be a Bose–Einstein condensate?, [36] T. Rindler-Daller and P. R. Shapiro, Complex
Journal of Cosmology and Astroparticle Physics 2007 (06), 025. Scalar Field Dark Matter on Galactic Scales,
[19] H. Foidl, T. Rindler-Daller, and W. W. Zeilinger, Halo Modern Physics Letters A 29, 1430002 (2014),
formation and evolution in scalar field dark matter and arXiv:1312.1734 [astro-ph.CO].
cold dark matter: New insights from the fluid approach, [37] P.-H. Chavanis, Predictive model of BEC dark matter
Phys. Rev. D 108, 043012 (2023). halos with a solitonic core and an isothermal atmosphere,
[20] M. Nori and M. Baldi, AX-GADGET: a Phys. Rev. D 100, 083022 (2019).
new code for cosmological simulations of [38] P.-H. Chavanis, Collapse of a self-gravitating
Fuzzy Dark Matter and Axion models, Bose-Einstein condensate with attractive self-
Monthly Notices of the Royal Astronomical Society 478, 3935 (2018), interaction, Phys. Rev. D 94, 083007 (2016),
https://academic.oup.com/mnras/article-pdf/478/3/3935/25079820/sty1224.pdf.
arXiv:1604.05904 [astro-ph.CO].
[21] S. May and V. Springel, Structure formation in [39] S. O. Schobesberger, T. Rindler-Daller, and P. R.
large-volume cosmological simulations of fuzzy Shapiro, Angular momentum and the absence of
dark matter: impact of the non-linear dynamics, vortices in the cores of fuzzy dark matter haloes,
Monthly Notices of the Royal Astronomical Society 506, 2603 (2021), Monthly Notices of the Royal Astronomical Society 505, 802–829 (20
https://academic.oup.com/mnras/article-pdf/506/2/2603/39207201/stab1764.pdf.
https://academic.oup.com/mnras/article-pdf/505/1/802/38378689/st
[22] B. Schwabe, M. Gosenca, C. Behrens, J. C. Niemeyer, [40] D. J. E. Marsh and A.-R. Pop, Axion dark
and R. Easther, Simulating mixed fuzzy and cold dark matter, solitons and the cusp–core problem,
matter, Phys. Rev. D 102, 083518 (2020). Monthly Notices of the Royal Astronomical Society 451, 2479 (2015)
[23] J. Zhang, H. Liu, and M.-C. Chu, Cosmological Simu- https://academic.oup.com/mnras/article-pdf/451/3/2479/4001975/st
lation for Fuzzy Dark Matter Model, Frontiers in As- [41] J. F. Navarro, C. S. Frenk, and S. D. M. White,
tronomy and Space Sciences 5, 10.3389/fspas.2018.00048 A Universal Density Profile from Hierarchical Clus-
(2019). tering, The Astropyhsical Journal 490, 493–508 (1997),
[24] B. Schwabe, J. C. Niemeyer, and J. F. Engels, Simula- arXiv:astro-ph/9611107 [astro-ph].
tions of solitonic core mergers in ultralight axion dark [42] Notice that in the code soliton solution.py the variables
matter cosmologies, Phys. Rev. D 94, 043513 (2016). have different names than here in the paper.
[25] L. Hui, A. Joyce, M. J. Landry, and X. Li, [43] The dependence of β on Λ̂ is indirect: inclusion of SI
Vortices and waves in light dark matter, changes the overall potential term and hence affects β,
Journal of Cosmology and Astroparticle Physics 2021 (01), 011. see equation 3.12 in [27]. In practice, the effect on β is
[26] The code is available at small: we checked to see that it changes little between
https://github.com/auckland-cosmo/PyUltraLight. Λ̂ = 0 and Λ̂ 6= 0, for the choice of Λ̂ in this paper.
[27] F. Edwards, E. Kendall, S. Hotchkiss, and [44] This is different to the boson mass, which has to be spec-
R. Easther, PyUltraLight: a pseudo-spectral ified in the script density profile.
solver for ultralight dark matter dynamics, [45] The discrete density profile is in the end only the raw
Journal of Cosmology and Astroparticle Physics 2018 (10), 027. density data for the fit, which is why will use both terms.
[28] L. M. Widrow and N. Kaiser, Using the Schroedinger [46] Such inverse mass-size relationships are common in com-
Equation to Simulate Collisionless Matter, pact stellar objects as the end stages of stellar evolution.
The Astrophysical Journal Letters 416, L71 (1993). [47] J. Veltmaat, J. C. Niemeyer, and B. Schwabe,
[29] L. Pitaevskii and S. Stringari, Bose-Einstein Conden- Formation and structure of ultralight bosonic
sation (Clarendon Press, Oxford, 2003) ISBN: 0-19- dark matter halos, Phys. Rev. D 98, 043509 (2018),
850719-4. arXiv:1804.09647 [astro-ph.CO].
[30] A. Paredes and H. Michinel, Interference [48] P.-H. Chavanis, Derivation of the core mass-
of dark matter solitons and galactic offsets, halo mass relation of fermionic and bosonic
Physics of the Dark Universe 12, 50–55 (2016). dark matter halos from an effective thermody-
[31] N. Glennon and C. Prescod-Weinstein, Modifying namical model, Phys. Rev. D 100, 123506 (2019),
PyUltraLight to model scalar dark matter with self- arXiv:1905.08137 [astro-ph.CO].
interactions, Phys. Rev. D 104, 083532 (2021). [49] D. Giulini and A. Großardt, Gravitation-
[32] L. E. Padilla, T. Rindler-Daller, P. R. Shapiro, ally induced inhibitions of dispersion ac-
T. Matos, and J. A. Vázquez, Core-halo mass rela- cording to the Schrödinger–Newton equation,
 34

Classical and Quantum Gravity 28, 195026 (2011). sidered mass scale, e.g. galaxy clusters experience per-
[50] D. Cyncynates and T. Giurgica-Tiron, Structure of the manent infall, yet there are some calm clusters like the
oscillon: The dynamics of attractive self-interaction, Coma cluster whose observation may be in conflict with
Phys. Rev. D 103, 116011 (2021). a DM model having high relaxation times.
[51] M. Membrado, A. F. Pacheco, and J. Sañudo, [62] Of course, this has also something to do with the cho-
Hartree solutions for the self-Yukawian boson sphere, sen box size and resolution, and the fact that a poor
Phys. Rev. A 39, 4207 (1989). spatial resolution decreases the central density. This be-
[52] Actually, a more detailed trial-and-error process enables comes apparent when comparing the central densities of
an even more accurate mass adjustment, but since the new and old simulations (old simulations have a poorer
change in the density profile would be minuscule, we did spatial resolution).
not pursue such fine-tuning. [63] J. L. Zagorac, E. Kendall, N. Padmanabhan,
[53] A. Navarrete, A. Paredes, J. R. Salgueiro, and and R. Easther, Soliton formation and the core-
H. Michinel, Spatial solitons in thermo-optical media halo mass relation: An eigenstate perspective,
from the nonlinear Schrödinger-Poisson equation and Phys. Rev. D 107, 083513 (2023).
dark-matter analogs, Phys. Rev. A 95, 013844 (2017). [64] E. Calabrese and D. N. Spergel, Ultra-light
[54] In fact, this milder growth of the central density with soli- dark matter in ultra-faint dwarf galaxies,
ton (resp. halo core) mass may help SFDM-TF to better Monthly Notices of the Royal Astronomical Society 460, 4397 (2016)
fit with galactic velocity data, than the FDM regime with https://academic.oup.com/mnras/article-pdf/460/4/4397/8117919/st
Λ̂ = 0, see [12] for a comparison. [65] T. Bernal, L. M. Fernández-Hernández, T. Matos,
[55] In the TF regime, closed-form expressions for ρ(r) and and M. A. Rodrı́guez-Meza, Rotation curves of high-
Φ(r), and their limits for r → 0, can be derived which resolution LSB and SPARC galaxies with fuzzy and
make the dependence more explicit. multistate (ultralight boson) scalar field dark matter,
[56] With that we mean, that the density is too low for the Monthly Notices of the Royal Astronomical Society 475, 1447 (2017)
SI term to be able to counter the increase in density for https://academic.oup.com/mnras/article-pdf/475/2/1447/23602590/
higher masses in the given time span of the simulation. [66] I. Álvarez-Rios, F. S. Guzmán, and P. R. Shapiro, Effect
If we were to choose a later time to extract the central of boundary conditions on structure formation in fuzzy
density, the turnover point would happen at even lower dark matter, Phys. Rev. D 107, 123524 (2023).
soliton masses. [67] M. Indjin, I.-K. Liu, N. P. Proukakis, and G. Rigopoulos,
[57] It appears that a Gaussian multiplied by a low-order Virialized profiles and oscillations of self-interacting fuzzy
polynomial might fit both numerical data sets, but more dark matter solitons, Phys. Rev. D 109, 103518 (2024).
analytic investigations and simulations would be required [68] T. Rindler-Daller, On particle scattering in Gross-
to investigate this further. Surely, an analytical relation- Pitaevskii theory and implications for dark matter halos,
ship would also help to predict the soliton mass at which Frontiers in Astronomy and Space Sciences 10, 1121920 (2023),
the central density becomes a maximum, and its depen- arXiv:2212.05812 [astro-ph.GA].
dence on Λ̂. [69] R. Galazo Garcı́a, P. Brax, and P. Valageas, Soli-
[58] J. A. González and F. S. Guzmán, Interference pattern tons and halos for self-interacting scalar dark matter,
in the collision of structures in the Bose-Einstein con- Phys. Rev. D 109, 043516 (2024).
densate dark matter model: Comparison with fluids, [70] T. D. Yavetz, X. Li, and L. Hui, Construction of wave
Phys. Rev. D 83, 103513 (2011). dark matter halos: Numerical algorithm and analytical
[59] In fact, the quantum wave nature of SFDM solitons constraints, Phys. Rev. D 105, 023512 (2022).
changes the way how they interact (interfere!) and how [71] I. Álvarez-Rios and F. S. Guzmán, Spherical solutions of
the assembly of structure proceeds in detail. In other the Schrödinger-Poisson system with core-tail structure,
words, the long-range nature of gravity, which is always Phys. Rev. D 108, 063519 (2023).
at play in cosmology, is compounded with the long-range [72] We chose a tolerance of ±0.5 CU for the calculated mass
nature of quantum interference effects in SFDM models. to be allowed to deviate from the input mass.
[60] Since the mass is not accurate, we also do not include [73] D. J. E. Marsh and P. G. Ferreira, Ultralight scalar
the R99 boundary in the plot. fields and the growth of structure in the Universe,
[61] While galactic or galaxy cluster halos are never exactly in Phys. Rev. D 82, 103528 (2010).
equilibrium, due to infall of smaller-mass objects, the re- [74] S.-C. Lin, H.-Y. Schive, S.-K. Wong, and T. Chiueh, Self-
quirement of establishing some kind of (near-)equilibrium consistent construction of virialized wave dark matter ha-
seems to be a necessity for successful DM models. Of los, Phys. Rev. D 97, 103523 (2018).
course, the degree of equilibrium depends on the con-