MNRAS
519,
3154–3181
(2023)
https://doi.org/10.1093/mnras/stac3489
Advance
Access
publication
2022
December
2
FIRE-3:
updated
stellar
evolution
models,
yields,
and
microphysics
and
fitting
functions
for
applications
in
galaxy
simulations
Philip
F.
Hopkins
,
1
‹
Andrew
Wetzel
,
2
Coral
Wheeler,
3
Robyn
Sanderson
,
4
Michael
Y.
Grudi
́
c
,
5
†
Omid
Sameie
,
6
Michael
Boylan-Kolchin
,
6
Matthew
Orr
,
7
Xiangcheng
Ma
,
8
Claude-Andr
́
e
Faucher-Gigu
`
ere
,
9
Du
ˇ
san
Kere
ˇ
s,
10
Eliot
Quataert
,
11
Kung-Yi
Su,
7
Jorge
Moreno,
12
Robert
Feldmann
,
13
James
S.
Bullock
,
14
Sarah
R.
Loebman,
15
Daniel
Angl
́
es-Alc
́
azar,
7
,
16
Jonathan
Stern,
17
Lina
Necib,
18
Caleb
R.
Choban
10
and
Christopher
C.
Hayward
7
Affiliations
are
listed
at
the
end
of
the
paper
Accepted
2022
No
v
ember
24.
Received
2022
November
23;
in
original
form
2022
February
24
A
B
S
T
R
A
C
T
Increasingly,
uncertainties
in
predictions
from
galaxy
formation
simulations
(at
sub-Milky
Wa y
masses)
are
dominated
by
uncertainties
in
stellar
evolution
inputs.
In
this
paper,
we
present
the
full
set
of
updates
from
the
Feedback
In
Realistic
Environment
(FIRE)-2
version
of
the
FIRE
project
code,
to
the
next
version,
FIRE-3.
While
the
transition
from
FIRE-1
to
FIRE-2
focused
on
improving
numerical
methods,
here
we
update
the
stellar
evolution
tracks
used
to
determine
stellar
feedback
inputs,
e.g.
stellar
mass-loss
(O/B
and
AGB),
spectra
(luminosities
and
ionization
rates),
and
supernova
rates
(core-collapse
and
Ia),
as
well
as
detailed
mass-dependent
yields.
We
also
update
the
low-temperature
cooling
and
chemistry,
to
enable
impro
v
ed
accuracy
at
T
10
4
K
and
densities
n
1
cm
−
3
,
and
the
meta-galactic
ionizing
background.
All
of
these
synthesize
newer
empirical
constraints
on
these
quantities
and
updated
stellar
evolution
and
yield
models
from
a
number
of
groups,
addressing
different
aspects
of
stellar
evolution.
To
make
the
updated
models
as
accessible
as
possible,
we
provide
fitting
functions
for
all
of
the
rele
v
ant
updated
tracks,
yields,
etc,
in
a
form
specifically
designed
so
they
can
be
directly
‘plugged
in’
to
existing
galaxy
formation
simulations.
We
also
summarize
the
default
FIRE-3
implementations
of
‘optional’
physics,
including
spectrally
resolved
cosmic
rays
and
supermassive
black
hole
growth
and
feedback.
Key
words:
methods:
numerical
–stars:
formation
–ISM:
structure
– galaxies:
evolution
– galaxies:
formation.
1
INTRODUCTION
It
is
now
well
established
that
‘feedback’
from
stars
– e.g.
coupling
of
stellar
radiation,
outflows/mass-loss,
supernovae
(SNe),
etc.,
to
ambient
interstellar
medium
(ISM)
gas
–plays
an
essential
role
in
galaxy
formation.
In
the
absence
of
stellar
feedback,
most
of
the
gas
in
the
cosmic
web
would
rapidly
accrete
onto
galaxies,
cool
on
a
time-scale
short
compared
to
the
dynamical
time,
collapse,
and
fragment
and
turn
into
stars
or
brown
dwarfs
(Bournaud
et
al.
2010
;
Dobbs,
Burkert
&
Pringle
2011
;
Harper-Clark
&
Murray
2011
;
Hopkins,
Quataert
&
Murray
2011
;
Krumholz,
Klein
&
McKee
2011
;
Ta s ke r
2011
),
producing
galaxies
with
properties
grossly
discrepant
from
observations
(Katz,
Weinberg
&
Hernquist
1996
;
Somerville
&
Primack
1999
;
Cole
et
al.
2000
;
Springel
&
Hernquist
2003
;
Kere
ˇ
s
et
al.
2009
)
almost
independent
of
the
‘details’
of
star
formation
(White
&
Frenk
1991
;
Kere
ˇ
s
et
al.
2009
).
Meanwhile
observed
galaxies
are
seen
to
turn
their
gas
into
stars
at
a
rate
of
just
a
few
per
cent
per
dynamical
time
(Kennicutt
1998
),
with
molecular
clouds
disrupting
owing
to
feedback
after
just
a
few
per
cent
of
their
mass
becomes
stars
(Williams
&
McKee
1997
;
Evans
1999
;
E-mail:
phopkins@caltech.edu
†
NASA
Hubble
Fellow.
Evans
et
al.
2009
),
and
then
galaxies
appear
to
expel
a
large
fraction
of
their
mass
into
the
circumgalactic
medium
(CGM;
Aguirre
et
al.
2001
;
Pettini
et
al.
2003
;
Songaila
2005
;
Oppenheimer
&
Dav
́
e
2006
;
Martin
et
al.
2010
;
We r k
et
al.
2014
;
Tumlinson,
Peeples
&
We r k
2017
)
in
observed
galactic
winds
(Martin
1999
,
2006
;
Heckman
et
al.
2000
;
Sato
et
al.
2009
;
Chen
et
al.
2010
;
Steidel
et
al.
2010
;
Coil
et
al.
2011
;
Newman
et
al.
2012
)
In
the
past
decade,
there
has
been
remarkable
progress
capturing
these
feedback
processes
in
simulations
which
attempt
to
capture
the
multiphase
complexity
of
the
ISM
and
CGM
(Hopkins,
Quataert
&
Murray
2012
;
Kim
&
Ostriker
2017
;
Grudi
́
c
et
al.
2019b
;
Benincasa
et
al.
2020
;
Keating
et
al.
2020
).
These
simulations
have
begun
to
resolve
the
self-consistent
generation
of
galactic
outflows
and
fountains
alongside
accretion
on
to
galaxies
(Narayanan
et
al.
2006
;
Angl
́
es-Alc
́
azar
et
al.
2017b
;
Hayward
&
Hopkins
2017
;
Muratov
et
al.
2017
;
Hafen
et
al.
2019
,
2020
;
Ji
et
al.
2020
;
Hopkins
et
al.
2021a
)
and
the
turbulent
processes
within
the
ISM
(Hopkins
2013b
,
c
;
Guszejnov,
Hopkins
&
Ma
2017
;
Escala
et
al.
2018
;
Guszejnov,
Hopkins
&
Grudi
́
c
2018
;
Rennehan
et
al.
2019
;
Gurvich
et
al.
2020
).
Specifically,
there
have
been
major
advances
in
both
numerical
methods
and
understanding
the
key
physics
of
how
SNe
(Martizzi,
Faucher-Gigu
`
ere
&
Quataert
2015
;
Gentry
et
al.
2017
;
Rosdahl
et
al.
2017
;
Hopkins
et
al.
2018a
;
Smith,
Sijacki
&
Shen
2018
;
Kawakatu,
Wa d a
&
Ichikawa
2020
),
stellar
mass-loss
(Wiersma
et
al.
2009b
;
© 2022
The
Author(s)
Published
by
Oxford
University
Press
on
behalf
of
Royal
Astronomical
Society
Downloaded from https://academic.oup.com/mnras/article/519/2/3154/6865374 by California Institute of Technology user on 23 May 2023
FIRE-3
updates
3155
MNRAS
519,
3154–3181
(2023)
Conroy,
van
Dokkum
&
Kravtsov
2015
;
H
̈
ofner
&
Olofsson
2018
),
and
stellar
radiation
(Hopkins
et
al.
2011
;
Wise
et
al.
2012
;
Rosdahl
&
Teyssier
2015
;
Emerick,
Bryan
&
Mac
Low
2018
;
Kim,
Kim
&
Ostriker
2018
;
Hopkins
&
Grudi
́
c
2019
;
Hopkins
et
al.
2020a
)
couple
to
the
ISM.
One
such
example
is
the
Feedback
In
Realistic
Environ-
ments
(FIRE)
project
(Hopkins
et
al.
2014
),
1
which
represents
an
attempt
to
synthesize
all
of
the
major
known
stellar
feedback
channels
directly
from
stellar
evolution
models,
to
combine
them
with
the
known
ISM
thermochemical
cooling
processes
and
cosmological
initial
conditions
into
predictive
simulations
of
galaxy
formation.
Ho
we
ver,
most
of
the
galactic
models
abo
v
e,
including
FIRE,
utilize
stellar
evolution
‘inputs’
– e.g.
stellar
population-synthesis
models
that
predict
key
inputs
for
feedback
such
as
supernova
rates
and
mass-loss
rates
and
yields
stellar
population
spectra
–which
themselves
rely
on
stellar
evolution
libraries
whose
isochrones
and
mass-loss
assumptions
are,
in
turn,
calibrated
to
observations
that
are
often
several
decades
old.
Common
basic
assumptions
(that
all
massive
stars
are
single
and
non-rotating)
are
almost
certainly
incorrect,
and
some
key
empirical
ingredients
(e.g.
calibration
of
giant-star
mass-loss
rates)
have
been
revised
by
more
than
an
order
of
magnitude
in
the
past
decade.
Indeed,
stellar
astrophysics
has
seen
a
revolution
in
the
last
decade,
with
truly
transformati
ve,
qualitati
vely
new
data
(and
orders-of-magnitude
increase
in
data
volume)
coming
from
time-domain
surv
e
ys
such
as
Kepler
(Howell
et
al.
2014
;
Silva
Aguirre
et
al.
2015
),
astrometric
distances
from
Gaia
(Huber
et
al.
2017
;
Gaia
Collaboration
2019
),
massive
spectroscopic
surveys
such
as
SDSS
and
APOGEE
(Ahn
et
al.
2014
;
Majewski
et
al.
2017
;
Jofr
́
e,
Heiter
&
Soubiran
2019
),
alongside
an
enormous
number
of
time-domain
studies
focused
on
binaries
and
explosions
such
as
PTF/ZTF
and
ASAS-SN
(Jayasinghe
et
al.
2018
;
Bellm
et
al.
2019
),
and
now
accompanying
gra
vitational-wa
ve
constraints
from
LIGO.
These
radical
impro
v
ements
in
observations
have
been
accompanied
by
e
xplosiv
e
growth
in
theory
and
modelling,
utilizing
new
codes
and
techniques
and
data
to
fundamentally
revise
our
understanding
of
e.g.
basic
stellar
evolution
and
interior
dynamics,
rotation,
binarity,
mass-
loss,
and
the
pre-explosion
physics
that
is
crucial
for
nucleosynthetic
yields
(Paxton
et
al.
2015
;
Ness
et
al.
2016
;
Marigo
et
al.
2017
;
Silva
Aguirre
et
al.
2017
;
Aerts
2021
;
Aerts,
Mathis
&
Rogers
2019
)
Meanwhile
there
have
been
a
number
of
other
important
advances
for
galactic
‘inputs,’
e.g.
better
constraints
on
the
redshift
evolution
and
shape
of
the
meta-galactic
ultraviolet
background
(UVB;
e.g.
Khaire
et
al.
2019
;
Wo r s e c k
et
al.
2019
;
Gaikwad
et
al.
2021
)
and
timing
of
reionization
from
Cosmic
Microwave
Background
(CMB)
and
Gunn–Peterson
measurements
(e.g.
Hoag
et
al.
2019
;
Planck
Collaboration
2020a
,
and
references
therein),
and
an
explosion
of
data
on
cold
atomic
and
molecular
gas
from
facilities
such
as
ALMA
that
can
probe
the
detailed
thermo-chemical
state
of
ISM
metals
and
gas
at
densities
n
1
cm
−
3
and
T
10
4
K
(e.g.
Combes
2018
).
The
collection
of
these
advances
warrants
updating
the
basic
input
assumptions
of
our
previous
galaxy
formation
simulations.
In
this
paper,
we
therefore
synthesize
and
present
the
updated
set
of
stellar
evolution
libraries,
yields,
cooling
functions,
and
other
assumptions
that
underpin
the
FIRE
simulations,
constituting
the
‘FIRE-3’
version
of
the
FIRE
simulation
code.
In
Section
2
,
we
provide
a
brief
overview
of
previous
FIRE
versions
and
motiva-
tions
for
this
study,
and
describe
the
range
of
applicability
of
the
fitting
functions
provided
here.
Section
3
describes
the
updates
to
treatment
of
fluid
dynamics
(Section
3.1
),
the
UV
background
1
ht
tp://fire.nort
hwestern.edu
(Section
3.2
),
stellar
evolution
tracks
(Section
3.3
),
Solar
abun-
dances
(Section
3.4
),
cooling
physics
(Section
3.5
),
treatment
of
HII
regions
(Section
3.6
)
and
other
radiative
feedback
channels
(Section
3.7
),
SNe
(Section
3.8
),
stellar
mass-loss
(Section
3.9
),
star
formation
criteria
(Section
3.10
),
and
nucleosynthetic
yields
(Section
3.11
).
We
then
summarize
the
FIRE-3
implementations
of
‘optional’
physics
which
will
be
used
in
some
(but
not
all)
FIRE-3
runs,
specifically
explicit
evolution
of
cosmic
rays
(CRs;
Section
3.12
)
and
black
hole
accretion/feedback
(Section
3.13
).
In
Section
4
,
we
briefly
compare
the
effects
of
the
updates
to
the
default
FIRE-3
model
on
galaxy
formation
simulations,
and
conclude
in
Section
5
.
Some
additional
tests
are
presented
in
Appendix
A
,
and
additional
details
of
the
mechanical
feedback
implementation
are
in
Appendix
B
.
2
OVERVIEW
AND
B
ACKGR
OUND
2.1
FIRE-1
and
FIRE-2
The
first
version
of
the
FIRE
code
–FIRE-1
– attempted
to
synthesize
the
core
physics
of
stellar
feedback
from
SNe
(Ia
and
II),
stellar
mass-loss
(O/B
and
AGB
mass-loss),
and
radiation
(photoheating,
ionization,
and
pressure)
together
with
detailed
cooling
physics
from
∼
10
−
10
10
K,
into
fully
cosmological
simulations
of
galaxy
formation
(Hopkins
et
al.
2014
).
Subsequent
work
used
this
code
to
study
a
wide
variety
of
topics
ranging
from
detailed
properties
of
dwarf
galaxies
(Chan
et
al.
2015
;
O
̃
norbe
et
al.
2015
;
El-Badry
et
al.
2016
),
elemental
abundance
patterns
in
galaxies
(Ma
et
al.
2016
),
galactic
outflows
and
the
CGM
(Faucher-Giguere
et
al.
2015
;
Muratov
et
al.
2015
;
Va n
de
Voort
et
al.
2016
;
Hafen
et
al.
2017
),
the
origins
of
star
formation
scaling
relations
(Orr
et
al.
2017
;
Sparre
et
al.
2017
),
and
high-redshift
and
massive
galaxy
populations
(Feldmann
et
al.
2016
,
2017
;
Oklop
ˇ
ci
́
c
et
al.
2017
).
These
and
related
papers
which
developed
the
numerical
methods
(Hopkins
et
al.
2011
,
2012
)
represented
an
initial
attempt
to
directly
take
the
outputs
of
stellar
evolution
models
and
apply
them
in
galaxy-scale
simulations
to
model
salient
stellar
feedback
rates.
Ho
we
ver,
the
input
stellar
evolution
models
from
STARBURST99
(Leitherer
et
al.
1999
),
while
widely
used
at
the
time,
generally
relied
on
stellar
evolution
tracks
(e.g.
Bruzual
&
Charlot
2003
),
which
assumed
non-rotating,
non-binary
stellar
populations,
with
mass-
loss
rates
for
massive
stars
at
early
and
intermediate
ages
which
were
calibrated
to
older
observations
that
have
since
been
revised
significantly
(to
wards
slo
wer
mass-loss
rates;
see
e.g.
Smith
2014
).
Many
of
the
key
isochrone
inputs
and
outputs
were
calibrated
to
stellar
observations
from
before
1990
(e.g.
Weidemann
&
Koester
1983
;
Olnon
et
al.
1984
;
Knapp
&
Morris
1985
;
Weidemann
1987
;
Bloecker
1995
).
As
a
result,
even
some
of
the
first
FIRE-1
studies
noted
that
some
quantities:
e.g.
detailed
elemental
abundance
patterns
within
galaxies,
or
the
escape
fraction
of
ionization
photons,
were
quite
different
if
one
considered
more
‘state-of-the-art’
stellar
evolution
and/or
yield
models
instead
(see
Ma
et
al.
2015
,
2017
).
Moreo
v
er,
some
of
the
approximations
used
in
FIRE-1
(and
FIRE-2)
–for
example,
treating
core-collapse
SNe
(CCSNe)
yields
as
IMF-
averaged,
as
opposed
to
tracking
different
yields
from
different
stellar
mass
progenitors,
or
treating
low-temperature
molecular
gas
with
a
simple
subgrid
molecular
fraction
estimator
– significantly
limit
the
predictive
power
for
modelling
e.g.
different
detailed
internal
abundance
spreads,
or
cold
gas
observables
like
CO
(Bonaca
et
al.
2017
;
Escala
et
al.
2018
;
Keating
et
al.
2020
;
Muley
et
al.
2021
).
Downloaded from https://academic.oup.com/mnras/article/519/2/3154/6865374 by California Institute of Technology user on 23 May 2023
3156
P
.
F
.
Hopkins
et
al.
MNRAS
519,
3154–3181
(2023)
None
the
less,
the
next
version
of
FIRE,
FIRE-2
(Hopkins
et
al.
2018b
),
2
attempted
for
the
sake
of
consistency
to
keep
the
physical
inputs
(e.g.
stellar
evolution
assumptions,
feedback
rates,
etc.)
fixed
to
FIRE-1
values
as
much
as
possible,
while
updating
instead
the
numerical
methods.
This
represented
a
major
update
to
numerical
accuracy,
utilizing
a
new
hydrodynamic
(HD)
solver
with
a
flexible
arbitrary
Lagrangian–Eulerian
method
(Hopkins
2015
)
as
opposed
to
the
Pressure-SPH
method
used
for
FIRE-1
(Hopkins
2013a
),
enabling
the
accurate
numerical
addition
of
new
physics
such
as
magnetic
fields
(Hopkins
&
Raives
2016
),
anisotropic
diffusion
(Hopkins
2017b
),
and
CRs
(Chan
et
al.
2019
;
Su
et
al.
2020
;
Hopkins
et
al.
2020b
).
It
also
made
major
impro
v
ements
to
the
accurac
y
of
the
gravitational
force
integration,
and
numerical
treatment/coupling
of
both
mechanical
(SNe
and
mass-loss;
Hopkins
et
al.
2018a
)
and
radiative
(Hopkins
et
al.
2020a
)
feedback
‘injection’
which
improved
convergence.
Together
these
enabled
order-of-magnitude
higher-
resolution
simulations
reaching
∼
30
M
resolution
in
small
dwarfs
and
∼
3000
M
resolution
in
Local-Group
(MW
+
Andromeda)
haloes
(Wetzel
et
al.
2016
;
Garrison-Kimmel
et
al.
2019a
;
Wheeler
et
al.
2019
),
and
produced
a
large
number
of
detailed
results
studying
a
wide
variety
of
galaxy
properties,
including
behaviour
of
different
ISM
phases
(e.g.
El-Badry
et
al.
2018b
;
Moreno
et
al.
2019
,
2021
;
Benincasa
et
al.
2020
;
Gurvich
et
al.
2020
),
detailed
galactic
structure
and
scaling
relations
(e.g.
Orr
et
al.
2018
;
El-Badry
et
al.
2018a
;
Sanderson
et
al.
2020
;
Wellons
et
al.
2020
;
Bellardini
et
al.
2021
;
Yu
et
al.
2021
),
satellite
and
dark
matter
properties
(e.g.
Garrison-
Kimmel
et
al.
2019b
,
a
;
Necib
et
al.
2019
;
Samuel
et
al.
2020
;
Santiste
v
an
et
al.
2020
),
and
properties
of
the
CGM
(e.g.
Hafen
et
al.
2019
,
2020
;
Ji
et
al.
2020
;
Li
et
al.
2021
;
Stern
et
al.
2021
).
2.2
FIRE-3
In
this
new
version
of
FIRE,
FIRE-3,
our
goal
is
not
to
change
our
core
numerical
methods,
nor
to
change
the
fundamental
physics
being
simulated.
Instead,
it
is
to
update
the
‘known’
microphysics,
particularly
the
treatment
of
stellar
evolution,
yields,
ISM
cooling
and
chemistry,
to
more
accurate
and
complete
inputs
that
enable
more
detailed
observational
predictions.
We
stress
that
this
is
not
a
numerical
methods
paper.
All
of
the
rele
v
ant
numerical
methods
for
the
default
version
of
FIRE-3
are
described
and
e
xtensiv
ely
tested
in
the
lengthy
FIRE-2
numerical
methods
papers,
specifically
the
series
Hopkins
et
al.
(
2018b
,
2018a
,
2020a
),
but
also
including
the
updates
for
various
‘non-default’
physics
(e.g.
black
holes,
CRs,
etc.)
where
rele
v
ant
in
Angl
́
es-Alc
́
azar
et
al.
(
2017c
),
Ma
et
al.
(
2018
),
Su
et
al.
(
2018b
),
Garrison-Kimmel
et
al.
(
2019a
),
Chan
et
al.
(
2019
),
and
Hopkins
et
al.
(
2020b
).
We
do
make
some
minor
numerical
modifications
to
the
‘default’
treatment
in
FIRE-3
for
impro
v
ed
accurac
y
,
all
of
which
we
describe
below
,
but
even
these
are
all
modifications
proposed
and
tested
specifically
in
the
FIRE-2
numerical
methods
papers
abo
v
e.
We
also
stress
that
all
of
these
updates
are
driven
by
more
accurate
theoretical
and
empirical
inputs
to
the
‘microphysics,’
rather
than
any
‘desired’
result
on
galactic
scales.
As
with
previous
versions
of
FIRE,
we
implement
all
physics
here
in
the
GIZMO
code
(Hopkins
2015
).
2
The
FIRE-2
simulations
are
publicly
available
via
Wetzel
et
al.
(
2022
)
at
http://flathub.f
latironinstitute.org/f
ire
2.3
Utility
and
range
of
applicability
of
inputs
For
the
sake
of
making
the
updated
models
public
in
the
most
useful
form
for
other
galaxy
formation
simulations,
we
have
endeavored
to
reduce
as
much
as
possible
the
new
stellar
evolution
and
cooling
physics
to
simple,
easily
implemented
fitting
functions,
which
can
be
immediately
inserted
into
different
numerical
simulation
codes
and/or
semi-analytic
models
for
galaxy
and
star
formation.
The
types
of
models
for
which
these
are
applicable
are
those
with
resolution
broadly
in
the
range
∼
10
−
10
6
M
(mass)
or
∼
0
.
1
−
100
pc
(force/spatial),
which
attempt
to
explicitly
treat/resolve
some
of
the
multiphase
structure
of
the
ISM
and/or
CGM
(e.g.
the
existence
of
giant
molecular
cloud
complexes),
and
spatially
and
time-resolved
galactic
star
formation
and
SNe/stellar
feedback
events
(e.g.
the
time
between
individual
SNe
in
a
single-star
particle
is
gen-
erally
much
larger
than
its
numerical
time-step),
but
with
insufficient
resolution
to
actually
model/predict
individual
(proto)stellar
collapse
and
masses
and
accretion/evolution
tracks
(aka
forward-modelling
stellar
accretion
and
the
IMF).
For
significantly
higher-resolution
simulations,
different
algorithms
such
as
those
in
e.g.
STARFORGE
(Grudi
́
c
et
al.
2021
;
Guszejnov
et
al.
2021
)
are
required,
which
can
correctly
treat
every
star
as
an
evolving-mass
sink
particle
and
deal
with
accretion
and
feedback
from
individual
proto
and
main-sequence
stars
along
independent
stellar
evolution
tracks.
For
significantly
lower-resolution
simulations,
different
algorithms
like
those
in
MUFASA
or
SIMBA
(Dav
́
e,
Thompson
&
Hopkins
2016
;
Dav
́
e
et
al.
2017
,
2019
;
Thomas
et
al.
2019
),
which
treat
stellar
feedback
as
a
continuous,
collective
processes
integrating
implicitly
o
v
er
(rather
than
trying
to
directly
numerically
resolve)
different
star-forming
regions
and
ISM
phases,
are
more
appropriate.
3
UPDATES
FROM
FIRE-2
TO
FIRE-3
We
now
describe
all
updates
from
FIRE-2
to
FIRE-3.
‘Default’
FIRE-
2
should
be
understood
to
be
the
version
presented
and
studied
e
xtensiv
ely
in
Hopkins
et
al.
(
2018b
).
Any
details
that
we
do
not
explicitly
describe
as
modified
in
this
section
remain
identical
in
FIRE-3
and
FIRE-2.
3.1
Fluid
dynamics,
magnetic
fields,
conduction,
and
viscosity
3.1.1
Improved
face
error
detection
The
fluid
dynamics
solver
is
largely
unchanged,
using
the
same
meshless
finite-mass
method
as
FIRE-2.
The
only
change
is
a
slightly
impro
v
ed
treatment
of
special
cases,
discussed
in
Hopkins
(
2015
),
where
one
simultaneously
has
(1)
strong
fluxes,
(2)
elements
with
extremely
different
densities
(hence
kernel/cell
sizes)
interacting
directly,
and
(3)
a
pathological
spatial
configuration
of
neighbouring
elements
(e.g.
all
∼
32
nearest
cell
centres
nearly
aligned
in
a
plane).
In
this
case,
the
matrix
inversion
procedure
needed
to
determine
the
ef
fecti
ve
faces
for
HD
fluxes
becomes
ill-conditioned,
and
floating
point
errors
can
lead
to
artificially
large
(or
small)
fluxes.
We
have
impro
v
ed
the
procedure
from
Hopkins
(
2015
)
for
dealing
with
such
cases,
by
simultaneously
(1)
pre-conditioning
the
matrices
to
reduce
floating-point
errors,
(2)
adaptively
expanding
the
neighbour
search
to
ensure
dimensionless
condition
numbers
(as
defined
in
appendix
C
of
Hopkins
2015
)
10
3
,
and
(3)
limiting
the
ef
fecti
ve
face
area
to
the
maximum
possible
geometric
area
between
neighbour
cells.
This
newer
treatment
is
the
default
behaviour
in
the
public
GIZMO
code
Downloaded from https://academic.oup.com/mnras/article/519/2/3154/6865374 by California Institute of Technology user on 23 May 2023
FIRE-3
updates
3157
MNRAS
519,
3154–3181
(2023)
(see
Hopkins
2017a
),
3
but
was
not
implemented
in
FIRE-2
to
ensure
code-consistency.
But
in
any
case,
this
occurs
rarely
and
has
small
effects.
3.1.2
Additional
‘default’
physics
In
FIRE-3,
certain
physics
that
were
‘optional’
in
FIRE-2
are
now
‘default.’
This
includes
turbulent
transport
of
passive
scalars
(e.g.
metals),
following
Hopkins
et
al.
(
2018b
)
Appendix
F3
(see
Colbrook
et
al.
2017
;
Su
et
al.
2017
;
Escala
et
al.
2018
,
for
validation
tests),
and
kinetic
magnetohydrodynamics
(MHD),
i.e.
magnetic
fields
(see
Hopkins
&
Raives
2016
;
Hopkins
2016
,
for
the
GIZMO
MHD
methods)
with
fully
anisotropic
Spitzer–Braginskii
conduction
and
viscosity
as
in
Hopkins
(
2017b
;
with
the
coef
ficients
gi
ven
in
Hopkins
et
al.
2020b
,
scaling
appropriately
with
the
plasma
state,
accounting
for
saturation
and
limitation
by
plasma
instabilities;
Cowie
&
McKee
1977
;
Squire,
Schekochihin
&
Quataert
2017
;
Komarov
et
al.
2018
).
As
shown
in
Su
et
al.
(
2017
,
2018a
),
Hopkins
et
al.
(
2020b
),
Chan
et
al.
(
2019
),
and
Ji
et
al.
(
2020
),
while
important
for
predicting
certain
properties
and
regulating
e.g.
CR
transport,
these
physics
generally
have
small
effects
on
bulk
galaxy
properties.
FIRE-3
also
includes,
by
default,
the
abundance
‘tracers’
(following
an
additional
set
of
passive
scalars
to
model
trace
metal
species)
model
of
We t z e l
et
al.
(in
preparation).
3.2
UV
background
FIRE-2
adopted
the
meta-galactic
UVB
spectrum,
as
a
function
of
redshift,
from
Faucher-Gigu
`
ere
et
al.
(
2009
).
Since
this
time
observational
constraints
on
the
UVB
have
greatly
improved.
For
example,
the
tabulation
used
in
FIRE-2
produced
a
redshift
of
HI
reionization
z
∼
10,
consistent
with
WMAP-7
constraints
(Komatsu
et
al.
2011
)
but
too
high
compared
to
more
recent
Planck
data
and
other
measurements
that
imply
a
lower
reionization
mid-point
z
8
(e.g.
Planck
Collaboration
2020b
).
For
FIRE-3,
we
update
the
assumed
UVB
to
the
more
recent
Faucher-Gigu
`
ere
(
2020
)
model,
which
synthesizes
and
better
reproduces
a
number
of
different
empirical
constraints
including
updated
luminosity
functions,
stellar
spectra
including
binary
stars,
obscured
and
non-obscured
active
galactic
nucleus
(AGN;
following
Shen
et
al.
2020
),
intergalactic
HI
and
He
II
photoionization
rates
measured
at
z
∼
0
−
6,
and
the
local
X-ray
background.
As
demonstrated
in
O
̃
norbe,
Hennawi
&
Luki
́
c
(
2017
)
and
Puchwein
et
al.
(
2019
),
UVB
models
that
assume
the
intergalactic
medium
(IGM)
is
optically
thin
will
produce
earlier
reionization
than
intended
if
applied
without
correction
in
simula-
tions
like
ours
to
the
pre-reionization
Universe
(which
is
opaque
to
ionizing
photons
and
not
yet
in
ionization
equilibrium).
The
UVB
tabulation
we
use
for
FIRE-3
therefore
adopts
the
‘ef
fecti
ve’
photoionization
and
photoheating
rates
calibrated
to
match
correctly
the
Planck
2018
reionization
optical
depth
and
recent
constraints
from
quasar
absorption
spectra
on
HeII
reionization
(Khaire
2017
;
Wo r s e c k
et
al.
2019
).
These
ef
fecti
ve
rates
produce
more
accurate
redshifts
of
HI
and
HeII
reionization
in
the
simulations;
these
correspond
to
the
sharp
drops
in
the
H
I
and
He
II
photoionization
rates
at
z
>
7
and
z
>
3
for
the
FIRE-3
model
in
Fig.
1
.
4
3
The
public
version
of
GIZMO
is
available
at
ht
tp://www.tapir.calt
ech.edu/
∼
phopkins/Site/GIZMO.html
4
The
updated
UVB
model
is
available
in
the
TREECOOL
file
with
the
public
GIZMO
code
(
ht
tp://www.tapir.calt
ech.edu/
∼
phopkins/Site/GIZMO.html
)
or
alternatively
at
ht
tps://galaxies.northwest
ern.edu/
uvb-fg20/
Figure
1.
Comparison
of
the
meta-galactic
UV
background
(UVB)
pho-
toionization
rates
versus
redshift
z
for
H
I
(
H
I
;
thick
)
and
He
II
(
He
II
,
multiplied
by
100
for
ease
of
comparison;
thin
),
in
FIRE-2
(
blue
dotted
)
and
FIRE-3
(
black
solid
),
per
Section
3.2
.
Updated
observational
constraints
modestly
increase
the
photoionization
rates
at
intermediate
redshifts.
The
sharp
drops
in
the
H
I
and
He
II
ionization
rates
in
FIRE-3
at
z
>
7
and
z
>
3,
respectively,
come
from
more
accurately
accounting
for
the
large
optical
depth
of
the
Universe
to
these
photons
when
H
I
or
He
II
ionization
is
not
yet
complete
(see
Faucher-Gigu
`
ere
2020
).
The
effects
of
the
changed
background
intensity
and
shape
are
generally
minor
except
for
e.g.
detailed
CGM
or
Ly
α
forest
studies.
The
most
important
effect
of
the
updated
UVB
is
later
reionization,
which
can
significantly
influence
dwarf
galaxy
star
formation
histo-
ries
in
the
ultra-faint
regime
especially.
These
effects
(and
the
role
of
remaining
uncertainties
in
the
UVB)
will
be
studied
in
detail
in
future
work.
3.3
Stellar
evolution
tables
3.3.1
Updated
isochrones
and
stellar
models
Our
stellar
feedback
models
take
their
inputs
in
the
form
of
SNe
and
stellar
mass-loss
rates,
stellar
luminosities,
and
spectra,
directly
from
standard
stellar
evolution
models
as
a
function
of
stellar
population
age,
mass,
and
metallicity.
The
tabulations
adopted
in
FIRE-2
are
all
described
in
detail
in
the
appendices
of
Hopkins
et
al.
(
2018b
).
We
have
re-fit
all
of
the
salient
stellar
evolution
tables
for
impro
v
ed
(1)
physical
accurac
y
(using
more
recent
and
detailed
stellar
evolution
models),
(2)
numerical
accuracy
(using
more
accurate
fitting
functions),
and
(3)
consistency
(using
more
recent
models
that
make
simultaneous
predictions
for
more
diverse
quantities).
Wherever
possible,
we
use
the
results
from
the
2021
January
version
of
STARBURST99
(Leitherer
et
al.
2014
),
adopt-
ing
a
three-part
Kroupa
(
2001
)
IMF
(with
slopes
(0
.
3
,
1
.
3
,
2
.
3)
from
(0
.
01
−
0
.
08
,
0
.
08
−
0
.
5
,
0
.
5
−
100)
M
),
an
8
M
SNe
cut-
of
f
(120
M
BH
formation
cutof
f),
the
preferred
‘e
volution’
wind
model
for
O/B
and
‘empirical’
model
for
AGB
mass-loss,
using
the
updated
Gene
v
a
2013
rotating
stellar
model
isochrones
(which
are
designed
to
reproduce
many
of
the
effects
attributed
to
binarity
as
well,
in
massive
stellar
populations,
and
therefore
show
much
smaller
differences
compared
to
models
like
BPASS
from
Eldridge
et
al.
2017
,
as
compared
to
the
older
models),
sampled
as
densely
as
possible
at
all
metallicities
available.
We
have
also
compared
the
results
using
all
available
isochrone
sets
published
in
the
last
decade
in
either
STARBURST99
or
BPASS,
to
ensure
we
do
not
fit
any
spurious
features.
All
quantities
below
are
IMF-integrated,
with
M
∗
=
M
∗
(
t
)
the
star-particle
mass
at
time
t
(
t
its
age
at
a
given
Downloaded from https://academic.oup.com/mnras/article/519/2/3154/6865374 by California Institute of Technology user on 23 May 2023