CHALLENGES OF COSMOLOGICAL

FUZZY DARK MATTER S I M U L A T I O N S

Simon May
[email protected]

Max-Planck-Institut für Astrophysik

29th January 2020

Theoretical background

Simulations: Numerical solutions to the FDM equations

Fundamental equations
Simple examples: soliton solutions
Pseudo-spectral methods

Computational cost of FDM simulations

Initial conditions and parameters for cosmological simulations

Outlook

Simon May 1 / 29
What is “fuzzy dark matter”?

▶ F(C)DM, BECDM, ULDM, ELBDM, 𝜓DM, quantum-wave DM,

(ultra-light) axion(-like) DM (ULA, ALP)…
▶ New extremely light scalar particle (𝑚 ≈ 𝟣𝟢−𝟤𝟤 𝖾𝖵!)
▶ Non-thermal production mechanism (thus not ultra-hot)
▶ Aggregations of bosons can form a Bose–Einstein condensate
▶ Quantum effects counteract gravity at small scales (uncertainty
principle), erase structure
▶ Tiny mass
⇒ large de Broglie wavelength (𝜆 ∼ 1/𝑚)
⇒ macroscopic quantum effects on kpc scales

Simon May 2 / 29
Motivation for fuzzy dark matter
Particle physics perspective:
▶ Original concept – strong CP problem:
Why doesn’t QCD violate CP symmetry?
▶ Solved by Peccei–Quinn U(1) symmetry
and (pseudo-)scalar field (axion!)
Peccei and Quinn (1977)!
▶ Fuzzy dark matter is not the QCD axion, but axion-like particles
are a common feature of early-universe theories
Astrophysics perspective:
▶ Small-scale challenges (cusp–core, missing satellites, …)
→ Ultra-light scalars: WIMP alternative, could improve this
𝑚𝜒
wrong LSS CDM-like
10−23 eV 10−22 eV 10−21 eV
▶ No sign of (WIMP) CDM
Simon May 3 / 29
Motivation for fuzzy dark matter
Particle physics perspective:
▶ Original concept – strong CP problem:
Why doesn’t QCD violate CP symmetry?
▶ Solved by Peccei–Quinn U(1) symmetry
and (pseudo-)scalar field (axion!)
Peccei and Quinn (1977)!
▶ Fuzzy dark matter is not the QCD axion, but axion-like particles
are a common feature of early-universe theories
Astrophysics perspective:
▶ Small-scale challenges (cusp–core, missing satellites, …)
→ Ultra-light scalars: WIMP alternative, could improve this
𝑚𝜒
wrong LSS CDM-like
10−23 eV 10−22 eV 10−21 eV
▶ No sign of (WIMP) CDM
Simon May 3 / 29
Derivation of the fuzzy dark matter equations
▶ Start with simple scalar field action
1 4 √ 1 𝜇𝜈 1 𝑚2 𝑐2 2 𝜆
𝑆= 2
∫ d 𝑥 −𝑔( 𝑔 (𝜕𝜇 𝜙)(𝜕 𝜈 𝜙) − 2
𝜙 − 2 2 𝜙4 )
ℏ𝑐 2 2 ℏ ℏ 𝑐
Note: corresponds to superfluid dark matter without
self-interaction (𝜆 = 0 or 𝑇 → 0)
▶ QCD axion case: originates from periodic potential
𝑉 (𝜙) ∼ Λ4 (1 − cos(𝜙/𝑓𝑎 )) for 𝜙 ≪ 𝑓𝑎
▶ Rewrite
ℏ3 𝑐 1 ℏ3 𝑐
𝜙=√ Re(𝜓𝑒−𝑖𝑐 /ℏ 𝑚𝑡 ) = √
2 2 2
(𝜓𝑒−𝑖𝑐 /ℏ 𝑚𝑡 + 𝜓∗ 𝑒𝑖𝑐 /ℏ 𝑚𝑡 )
2𝑚 2 2𝑚
and take non-relativistic limit with perturbed FRW metric
d𝑠2 = (1 + 2Φ 2 2 2 2Φ
𝑐2 )𝑐 d𝑡 − 𝑎(𝑡) (1 − 𝑐2 )d𝑥 ⃗
2

▶ Result: Schrödinger equation

3 ℏ2 2
𝑖ℏ(𝜕𝑡 𝜓 + 𝐻𝜓) = − ∇ 𝜓 + 𝑚Φ𝜓
2 2𝑚
Simon May 4 / 29
Derivation of the fuzzy dark matter equations
▶ Start with simple scalar field action
1 4 √ 1 𝜇𝜈 1 𝑚2 𝑐2 2 𝜆 
𝑆= 2
∫ d 𝑥 −𝑔( 𝑔 (𝜕𝜇 𝜙)(𝜕 𝜈 𝜙) − 2
𝜙 − 22𝜙4 )
ℏ𝑐 2 2 ℏ ℏ 𝑐

Note: corresponds to superfluid dark matter without
self-interaction (𝜆 = 0 or 𝑇 → 0)
▶ QCD axion case: originates from periodic potential
𝑉 (𝜙) ∼ Λ4 (1 − cos(𝜙/𝑓𝑎 )) for 𝜙 ≪ 𝑓𝑎
▶ Rewrite
ℏ3 𝑐 1 ℏ3 𝑐
𝜙=√ Re(𝜓𝑒−𝑖𝑐 /ℏ 𝑚𝑡 ) = √
2 2 2
(𝜓𝑒−𝑖𝑐 /ℏ 𝑚𝑡 + 𝜓∗ 𝑒𝑖𝑐 /ℏ 𝑚𝑡 )
2𝑚 2 2𝑚
and take non-relativistic limit with perturbed FRW metric
d𝑠2 = (1 + 2Φ 2 2 2 2Φ
𝑐2 )𝑐 d𝑡 − 𝑎(𝑡) (1 − 𝑐2 )d𝑥 ⃗
2

▶ Result: Schrödinger equation

3 ℏ2 2
𝑖ℏ(𝜕𝑡 𝜓 + 𝐻𝜓) = − ∇ 𝜓 + 𝑚Φ𝜓
2 2𝑚
Simon May 4 / 29
Derivation of the fuzzy dark matter equations

3 ℏ2 2
𝑖ℏ(𝜕𝑡 𝜓 + 𝐻𝜓) = − ∇ 𝜓 + 𝑚Φ𝜓
2 2𝑚

▶ Mean field approximation: interpretation as the single

macroscopic wave function of Bose-Einstein condensate with
density 𝜌 = 𝑚|𝜓|2
▶ “FDM equations” are the nonlinear Schrödinger–Poisson
system of equations

ℏ2
𝑖ℏ𝜕𝑡 𝜓c = − ∇2 𝜓 + 𝑚Φ𝜓c
2𝑚𝑎2 c c
4𝜋𝐺
∇2c Φ = 𝑚(|𝜓c |2 − |𝜓c̄ |2 )
𝑎

▶ Only a single scale, determined by ℏ

𝑚 (→ wavelength)
Simon May 5 / 29
Approaches to fuzzy dark matter simulations
I. Schrödinger–Poisson equations
ℏ2 2
𝑖ℏ𝜕𝑡 𝜓 = − ∇ 𝜓 + 𝑚Φ𝜓
2𝑚
∇2 Φ = 4𝜋𝐺𝑚(|𝜓|2 − |𝜓|̄ 2 )

II. Madelung formulation (fluid dynamics representation)

𝜕𝑡 𝜌 + ∇ ⋅ 𝜌𝑣 ⃗ = 0 𝜌 𝑖𝛼
√ 𝜓=√ 𝑒
1 ℏ2 ∇2 𝜌 𝑚
𝜕𝑡 𝑣 ⃗ + (𝑣 ⃗ ⋅ ∇)𝑣 ⃗ = − ∇( − √ + Φ)
𝑚 2𝑚
⏟⏟⏟⏟⏟ 𝜌 𝜌 = 𝑚|𝜓|2
=𝑄 ℏ
2
𝑣 ⃗ = ∇𝛼
∇ Φ = 4𝜋𝐺(𝜌 − 𝜌)̄ 𝑚
▶ Phase is undefined for 𝜌 = 0
⇒ significant effects on overall evolution

Simon May 6 / 29
Approaches to fuzzy dark matter simulations
I. Schrödinger–Poisson equations
ℏ2 2
𝑖ℏ𝜕𝑡 𝜓 = − ∇ 𝜓 + 𝑚Φ𝜓
2𝑚
∇2 Φ = 4𝜋𝐺𝑚(|𝜓|2 − |𝜓|̄ 2 )

II. Madelung formulation (fluid dynamics representation)

𝜕𝑡 𝜌 + ∇ ⋅ 𝜌𝑣 ⃗ = 0 𝜌 𝑖𝛼
√ 𝜓=√ 𝑒
1 ℏ2 ∇2 𝜌 𝑚
𝜕𝑡 𝑣 ⃗ + (𝑣 ⃗ ⋅ ∇)𝑣 ⃗ = − ∇( − √ + Φ)
𝑚 2𝑚
⏟⏟⏟⏟⏟ 𝜌 𝜌 = 𝑚|𝜓|2
=𝑄 ℏ
2
𝑣 ⃗ = ∇𝛼
∇ Φ = 4𝜋𝐺(𝜌 − 𝜌)̄ 𝑚
▶ Phase is undefined for 𝜌 = 0 “quantum potential”
⇒ significant effects on overall evolution “quantum pressure”

Simon May 6 / 29
Schrödinger–Poisson vs. Madelung formulation

Li, Hui, and Bryan (2018)

Simon May 7 / 29
Schrödinger–Poisson vs. Madelung formulation

Li, Hui, and Bryan (2018)

Simon May 7 / 29
Schrödinger–Poisson vs. Madelung formulation
The choice

Madelung formulation:
▶ Simple to add to existing hydrodynamics codes
▶ Hydrodynamic interpretation
But:
▶ Validity unclear
▶ Quantum potential/pressure is computationally challenging
(third derivative of density!)
▶ Schrödinger–Poisson implementation actually not so difficult
(depending on the method)
→ stick with Schrödinger–Poisson wave function

Simon May 8 / 29
Soliton solutions to the FDM equations
▶ Spherical symmetry & time-independent density profile:

𝜓(𝑡, 𝑥)⃗ → 𝑒𝑖𝛽𝑡 𝑓(𝑟), Φ(𝑡, 𝑥)⃗ → 𝜑(𝑟) (𝜑(𝑟)

̃ = 𝜑(𝑟) + 𝛽)

▶ SP equations reduce to

1 1
0 = − 𝑓 ″ (𝑟) − 𝑓 ′ (𝑟) + 𝜑(𝑟)𝑓(𝑟)
̃
2 𝑟
2
0 = 𝜑̃ ″ (𝑟) + 𝜑̃ ′ (𝑟) − 4𝜋𝑓(𝑟)2
𝑟
All constants can be “absorbed”!
▶ Solve numerically using 𝑓(0) = 𝛼, 𝑓 ′ (0) = 𝜑̃ ′ (0) = 0,
𝑟→∞
𝜑(𝑟) −−−→ − 𝑟𝑐
√
▶ Given solution 𝑒𝑖𝛽𝑡 𝑓(𝑟), 𝑔(𝑟) = 𝑒𝑖𝛼𝛽𝑡 𝛼𝑓( 𝛼𝑟) is also a solution
√ ⃗ 1/2 𝑣2 𝑡+𝛿)
⃗ 𝑖(𝛼𝛽𝑡+𝑣⋅⃗ 𝑥−
𝜓(𝑡, 𝑥)⃗ = 𝛼𝑓( 𝛼|𝑥⃗ − 𝑣𝑡|)𝑒

Simon May 9 / 29
Properties of solitonic cores
√ ⃗ 1/2 𝑣2 𝑡+𝛿)
⃗ 𝑖(𝛼𝛽𝑡+𝑣⋅⃗ 𝑥−
𝜓(𝑡, 𝑥)⃗ = 𝛼𝑓( 𝛼|𝑥⃗ − 𝑣𝑡|)𝑒

▶ Scaling symmetry of SP equations:

(𝑡, 𝑥, Φ, 𝜓) → (𝜆−2 𝑡, 𝜆−1 𝑥, 𝜆2 Φ, 𝜆2 𝜓)
▶ Solitons only have a single parameter: 𝛼
Can be expressed through soliton mass or radius
▶ 𝑀c ∼ 𝑚12 𝑟 ⇒ more massive cores are smaller
c

▶ Profile approximated by analytical expression

2 −8
𝑟 𝑟→0
𝜌s (𝑟) ≈ 𝜌0 (𝑚, 𝑟c )(1 + 0.091( ) ) −−−→ const.
𝑟c 𝑟→∞
−−−→ 𝑟−16

▶ Virialized FDM halos form soliton(-like) cores

Simon May 10 / 29
Simple soliton simulations (→ PyUltraLight)
Soliton collision interference pattern

Phase difference = 0

Phase difference = 𝜋

Simon May 11 / 29
Using pseudo-spectral methods to simulate FDM (in AREPO)
ℏ2 2
𝑖ℏ𝜕𝑡 𝜓 = − ∇ 𝜓 + 𝑚Φ𝜓
▶ Symmetrized split-step Fourier 2𝑚
∇ Φ = 4𝜋𝐺𝑚(|𝜓|2 − |𝜓|̄ 2 )
2
method (“kick–drift–kick”)
▶ Small time step Δ𝑡:
𝑡+Δ𝑡 ℏ
∇2 + 𝑚 ′ ′
𝜓(𝑡 + Δ𝑡, 𝑥)⃗ = 𝒯𝑒−𝑖 ∫𝑡 (− 2𝑚 ℏ Φ(𝑡 ,𝑥))d𝑡
⃗
𝜓(𝑡, 𝑥)⃗
𝑖 Δ𝑡 ℏ 2 𝑚
⃗ 𝑚
2 ( 𝑚 ∇ − ℏ Φ(𝑡+Δ𝑡,𝑥)− ℏ Φ(𝑡,𝑥))
⃗
≈𝑒 𝜓(𝑡, 𝑥)⃗
−𝑖 𝑚 Δ𝑡
ℏ 2 Φ(𝑡+Δ𝑡,𝑥)
⃗ ℏ Δ𝑡
𝑖𝑚 2 ∇
2 −𝑖 𝑚 Δ𝑡
ℏ 2 Φ(𝑡,𝑥)
⃗
≈ 𝑒⏟⏟⏟⏟⏟⏟⏟ 𝑒⏟ ⏟⏟⏟⏟⏟
𝑒 𝜓(𝑡, 𝑥)⃗
“kick” “drift” “kick”
⇒ Algorithm:
𝑚 Δ𝑡
𝜓 ← 𝑒−𝑖 ℏ 2 Φ 𝜓 kick
ℏ Δ𝑡 2
−𝑖 𝑚 2 𝑘
𝜓 ← IFFT(𝑒 FFT(𝜓)) drift
1
Φ ← IFFT(− FFT(4𝜋𝐺𝑚(|𝜓|2 − |𝜓|̄ 2 ))) update potential
𝑘2
𝑚 Δ𝑡
𝜓 ← 𝑒−𝑖 ℏ 2 Φ 𝜓 kick

Simon May 12 / 29
Pseudo-spectral methods for FDM
Advantages
Algorithm:
𝑚 Δ𝑡
𝜓 ← 𝑒−𝑖 ℏ 2 Φ 𝜓 kick
ℏ Δ𝑡 2
−𝑖 𝑚 2 𝑘
𝜓 ← IFFT(𝑒 FFT(𝜓)) drift
1
Φ ← IFFT(− FFT(4𝜋𝐺𝑚(|𝜓|2 − |𝜓|̄ 2 ))) update potential
𝑘2
𝑚 Δ𝑡
𝜓 ← 𝑒−𝑖 ℏ 2 Φ 𝜓 kick

Choice of time step: Δ𝑡 ≤ min( 𝜋1 𝑚 2

ℏ 𝑎 Δ𝑥
2
, ℏ 1
𝑚 𝑎 |Φmax | )
▶ “Exact” solution
▶ Automatic conservation of mass
▶ Can adapt existing PM code
▶ Simple implementation
Simon May 13 / 29
Pseudo-spectral methods for FDM
Advantages
Algorithm:
𝑚 Δ𝑡
𝜓 ← 𝑒−𝑖 ℏ 2 Φ 𝜓 kick
ℏ Δ𝑡 2
−𝑖 𝑚 2 𝑘
𝜓 ← IFFT(𝑒 FFT(𝜓)) drift
1
Φ ← IFFT(− FFT(4𝜋𝐺𝑚(|𝜓|2 − |𝜓|̄ 2 ))) update potential
𝑘2
𝑚 Δ𝑡
𝜓 ← 𝑒−𝑖 ℏ 2 Φ 𝜓 kick

Choice of time step: Δ𝑡 ≤ min( 𝜋1 𝑚 2 2

ℏ 𝑎 𝚫𝒙 ,
ℏ 1
𝑚 𝑎 |Φmax | )
▶ “Exact” solution
▶ Automatic conservation of mass
▶ Can adapt existing PM code
▶ Simple implementation
Simon May 13 / 29
Why is it hard to simulate FDM?
Computational challenges

▶ Tiny mass ↔ macroscopic quantum

effects, de Broglie wavelength of
galactic scale
▶ Both large scales and de Broglie scale
must be resolved for correct evolution
(sub-kpc cores can form)
▶ Time step criterion: Δ𝑡 ∼ Δ𝑥2
(seems to be approach-independent)

▶ Tooling: Hydrodynamics codes are

designed for 𝑁-body simulations

Schive, Chiueh, and Broadhurst (2014)

Simon May 14 / 29
Plans for future numerical methods

▶ Disadvantage of pseudo-spectral method: uniform mesh, lacks

adaptivity
→ large high-resolution simulations are expensive/infeasible
▶ Plan for improved methods:
▶ Investigate hybrid methods
▶ Perhaps adapt AREPO’s moving mesh for FDM
→ need second derivative on irregular mesh with minimal noise
▶ Goal: full cosmological simulations with baryons
(hydrodynamics)

Simon May 15 / 29
Correspondence of CDM and FDM initial conditions
Constructing a wave function 𝜓 from a phase space distribution
function 𝑓:
⃗ 𝑖 /ℏ 𝑥⋅⃗ 𝑣+𝑅
⃗
𝑚
𝜓(𝑥)⃗ ∼ ∑ √𝑓(𝑥,⃗ 𝑣)𝑒 𝑣⃗

𝑣⃗

For “cold”/single-stream distribution function:

√ 𝑖𝛼
𝜓= 𝜌𝑒
ℏ
𝑣 ⃗ = ∇𝛼
𝑚
Grid discretization implies a maximum velocity which can be
represented
Mocz, Lancaster, et al. (2018), Mocz, Fialkov, et al. (2019)
ℏ 𝜋
𝑣<
𝑚 Δ𝑥

Simon May 16 / 29
CPU time vs. mass (𝐿 = 1 Mpc)

104

103
CPU time / h

102

101
1283
100 2563
[3.26e-20 x 1] × 1.2
10 1 5123
10243
10 24 10 23 10 22 10 21
m / eV
Simon May 17 / 29
CPU time vs. resolution (𝐿 = 1 Mpc)

104

103
CPU time / h

102

101

100 m = 1e-24
m = 1e-23
[8.84e-12 x5] × 1.2
10 1 m = 1e-22
m = 1e-21
27 28 29 210 211 212
PMGRID
Simon May 18 / 29
CPU time vs. resolution (𝑁 = 2563 )

102

101
CPU time / h

100

m = 1e-24
m = 1e-23
10 1
[14.8 x 2] × 1.2
m = 1e-22
m = 1e-21
2 × 100 3 × 100 4 × 100 6 × 100 101
L / Mpch 1

Simon May 19 / 29
Velocity criterion
ℏ 𝜋
𝑣<
𝑚 Δ𝑥
108
m = 1e-23
m = 2.5e-23
m = 5e-23
m = 1e-22
m = 2.5e-22
107 CPU h ≙
two full months
CPU time / h

107 of entire MPA

cluster (FREYA)!

106

100 101
L / Mpch 1
Simon May 20 / 29
Velocity criterion

107
CPU time at Lmax / h

m = 1e-23
m = 2.5e-23
m = 5e-23
m = 1e-22
m = 2.5e-22
106 1
10 6 × 100 4 × 100 3 × 100 2 × 100 100
zend
Simon May 21 / 29
Cosmological 1 Mpc boxes
𝑧 = 0, 𝑚 = 𝟤.𝟧 × 𝟣𝟢−𝟤𝟥 𝖾𝖵, 𝑣ΛCDM = 𝟫𝟩.𝟤 𝗄𝗆/𝗌

Simon May 22 / 29
Cosmological 1 Mpc boxes
𝑧 = 0, 𝑚 = 𝟤.𝟧 × 𝟣𝟢−𝟤𝟥 𝖾𝖵, 𝑣ΛCDM = 𝟫𝟩.𝟤 𝗄𝗆/𝗌

Simon May 22 / 29
Matter power spectra
1 Mpc boxes
FDM, 643
FDM, 963
FDM, 1283
10 1 FDM, 2563
FDM, 3683
FDM, 4003
FDM, 4323
FDM, 4963
10 2 FDM, 5123
CDM, 1283
P(k)

CDM, 2563

10 3

10 4
10 2 10 1 100
Kbin
Simon May 23 / 29
Matter power spectra
10 Mpc boxes 28803 grid already takes > 530 GB RAM!
103
linear theory
FDM, 10243
102 FDM, 20483
FDM, 25923
101 FDM, 28803
CDM, 2563
100 CDM, 5123
P(k) / h 3 Mpc3

10 1

10 2

10 3

10 4

10 5

100 101 102 103 104

→ difficult to achieve converged
resolution (≈ de Broglie wavelength)
k / h Mpc 1
Simon May 24 / 29
Halo finding: halo mass function
10 Mpc, z = 0.00
10 3 CDM 5123
FDM 28803
10 5
1

10 7
dn/dM / h 4 Mpc 3 M

10 9

10 11

10 13

10 15

106 107 108 109 1010 1011 1012 1013

M/h M
1

Simon May 25 / 29
Halo finding: halo mass function
z = 0.00

101
CDM
(dn/dM)FDM / (dn/dM)

100

106 107 108 109 1010 1011 1012 1013

M/h M
1

Simon May 26 / 29
First combined simulation of FDM with baryons & stars
Mocz, Fialkov, et al. (2019)

▶ Uniform pseudo-spectral method

▶ 𝐿 = 𝟣.𝟩ℎ−1 𝖬𝗉𝖼, 𝑧 = 127…5.5
▶ 10243 FDM cells ⇒ Δ𝑥 = 𝟣.𝟨𝟨ℎ−1 𝗄𝗉𝖼
▶ 5123 baryon cells
▶ Runtime: ≈ 𝟥 × 𝟣𝟢𝟨 CPU h, 𝟤𝟢× CDM run with
5123 + 5123 particles/cells
▶ Baryonic models still calibrated for CDM!

Simon May 27 / 29
Outlook
Conclusions from my own work so far:
▶ Cosmological FDM simulations seem to give sensible results, in
agreement with CDM for limiting cases
▶ Working with “big” data (simulations) requires time/effort
▶ Central problem of static pseudo-spectral method: limited
resolution/box size!
▶ Both memory and run-time cost

Still to do:
▶ Pushing the pseudo-spectral method to its limits with the
largest cosmological FDM simulations so far (current largest:
≈ 2.5 Mpc)
▶ Surpassing the limitations of uniform grids using hybrid or
adaptive schemes?
▶ Combination with baryons

Simon May 28 / 29
Short summary of the main problems for simulations

1. Time integration Δ𝑡 ∼ Δ𝑥2

2. Rapid oscillations even in low-density regions since velocity
corresponds to the gradient of the phase (→ velocity criterion)
3. Large dynamic range: “large”-scale structure simulations still
require resolving de Broglie wavelength
4. “New” field without decades of experience or refined
codes/methods as for CDM

Simon May 29 / 29
References

Peccei, R. D. and Helen R. Quinn (June 1977). “CP conservation in the presence of
pseudoparticles”. In: Phys. Rev. Lett. 38, pp. 1440–1443. DOI:
10.1103/PhysRevLett.38.1440.
Li, Xinyu, Lam Hui, and Greg L. Bryan (Oct. 2018). “Numerical and Perturbative
Computations of the Fuzzy Dark Matter Model”. In: arXiv: 1810.01915 [astro-ph.CO].
Schive, Hsi-Yu,Tzihong Chiueh, andTom Broadhurst (July 2014). “Cosmic structure as the
quantum interference of a coherent dark wave”. In: Nature Physics 10, pp. 496–499. DOI:
10.1038/nphys2996. arXiv: 1406.6586 [astro-ph.GA].
Mocz, Philip, Lachlan Lancaster, et al. (Apr. 2018). “Schrödinger-Poisson-Vlasov-Poisson
correspondence”. In: Phys. Rev. D 97.8, 083519, p. 083519. DOI:
10.1103/PhysRevD.97.083519. arXiv: 1801.03507 [astro-ph.CO].
Mocz, Philip, Anastasia Fialkov, et al. (Nov. 2019). “Galaxy Formation with BECDM – II.
Cosmic Filaments and First Galaxies”. In: arXiv e-prints, arXiv:1911.05746,
arXiv:1911.05746. arXiv: 1911.05746 [astro-ph.CO].
Zhang, Jiajun, Hantao Liu, and Ming-Chung Chu (Dec. 2018). “Cosmological Simulation for
Fuzzy Dark Matter Model”. In: Frontiers in Astronomy and Space Sciences 5, 48, p. 48. DOI:
10.3389/fspas.2018.00048. arXiv: 1809.09848 [astro-ph.CO].

Simon May 1/1